Current cereal crop yield in smallholder farmers’ field is below 1 t/ha in many parts of Sub-Saharan Africa (SSA) largely due to the degraded soil quality and lack of adequate management practices and investment. Worse, with very limited resources to cope with weather variability, such as this year’s devastating drought in the Horn of Africa region, smallholder farmers in Sub-Saharan Africa (SSA) face not only the low levels of yield but also highly unreliable crop production from year to year. In this modeling exercise we attempted to quantify how much area is under such unreliable condition and what are their potential, in terms of the average yield as well as its reliability, under the production intensification – once inorganic fertilizer and hybrid varieties become available to those areas.

Data and Method

Simulation was run for the current maize growing areas in SSA countries at 5 arc-minute spatial resolution, based on Spatial Production Allocation Model (SPAM; see http://mapspam.info for more detail) for 50-year period (1951-2000). HarvestChoice’s spatial datasets of synthesized long-term daily weather (SLATE; see previous post for more detail) and the gridded soil distribution database (see previous post for more detail) were used as the model input. The extent of maize growing areas was fixed over the simulated time period. To assess the yield variability under different crop management intensification assumptions, two input systems were defined as follows:

• Low-input: Farmers rely on the land’s inherent and no fertilizer or manure is applied; soil fertility degrades over time. OPV variety is used.
• High-input: Farmers use inorganic fertilizer at 40 kg[N]/ha rate and hybrid variety.

As we’re interested in the yield variability caused by rainfall variability, we only focused on the rainfed systems; supplementary irrigation was not simulated in this study. For each grid cell 50-year sequential simulation of maize cultivation was run for each soil. In post-processing of the result, area share of each soil was used to compute the area-weighted average of simulated yield for each grid cell per year.

We defined the “reliability” based on the probability of achieving a predefined threshold level of yield for at least three out of five years (i.e., > 60% of years). A range of the threshold yield level was set from 1, 2, 3, …, 7 t/ha to construct the probability function across the region. In addition, we set the target yield threshold level as 3 t/ha, following the discussion by the Hunger Task Force of the UN Millennium Project and the African Green Revolution that set the cereal yield need to increase from the current level of 1 t/ha to 3 t/ha to achieve the Millennium Development Goal (MDG) on halving the number of people suffering from hunger in Africa.

Results and Discussion

Average potential yield under low-input and high-input systems were simulated as 2,271 and 4,270 kg/ha, respectively. In reality, however, actual yields achieved by farmer can be much lower than the potential level as the simulation does not take into account all biotic and abiotic constraints that may exist in the field.

Result at the lowest yield threshold level, 1 t/ha, indicated that there were 74% of maize area that reliably (more than 3 out of 5 years; hereafter) produce more than 1 t/ha. In other words, there are about a quarter (26%) of maize area produce less than 1 t/ha in most of the years. Map 1 shows the distribution of such low-potential areas across the region (red colored areas); they are distributed in much of the Horn of Africa countries (Ethiopia, Somalia, and Kenya), as well as Southern Africa countries (e.g., South Africa and Zimbabwe) and arid and semi-arid areas in West Africa (e.g., Ghana, Burkina Faso, and Nigeria). If we shift our focus to the target yield level of 3 t/ha, there are only about 15% of areas that can reliably produce yield higher than the level. Map 2 shows such areas are sparsely distributed in mostly humid and highland agro-climatic areas, such as Western Ethiopia, South-Eastern Ghana, Rwanda, and Burundi. With intensification, however, such areas could vastly expand to about 83% of maize areas, as shown in Map 3, except arid and semi-arid regions in Eastern Ethiopia, Northern Kenya, and Chad. That is, the simulated intensification methods in this study, inorganic fertilizer and hybrid variety, not only increased the yield levels but also contributed to improve the reliability of maize production over time.

This quick study assessed the status of (un)reliability of maize production in SSA region under two management practice scenarios of low-input/subsistence and high-input/intensified systems on the current maize growing areas. However, the improvement of production at regional or country level could also come from the expansion of crop land to new areas, whose land quality for maize cultivation could be better (in newly cleared land) or worse (in marginal land under arid climate). As we did not consider such changes in the land use, our estimation of land areas maybe underestimating, especially for the case of high-input/intensified systems. In addition, we only focused on the output side of food production system (i.e., yield), not considering the input (e.g., fertilizer and seed cost, labor charges, transportation cost to the market), such that the reliable production does not necessarily translate into the economic profitability. There maybe areas where fertilizer is cost-prohibitively expensive that the marginal yield improvement does not justify the upfront investment.

Despite the caveat, the reliable potential yield improvement under the intensification scenario in 83% of maize growing area is highly encouraging. This finding also supports the interim outcome from the Millennium Villages Projects that reported the achievements of average maize yield of more than 3 t/ha in all villages across SSA where maize is the major crop, through subsidized fertilizers, improved germplasm, and intensive training on appropriate agronomic practices.

Screen-capture from NHK World

We’re not normally focused on this part of the world, but something like the Japanese tsunami tends to divert your attention. There was no shortage of dramatic aerial footage; some of the most striking, to me, showed the wave chewing through what appeared to be acres upon acres of greenhouses along the eastern coast of Japan.  Along with destroying towns, the disaster clearly ravaged some agriculture-intensive areas, and watching the greenhouses crumple as the water touched them was a memorable display of the tsunami’s power.

A post from Big Picture Agriculture takes a … well, a “big picture” look at the possible ag impact of the tsunami.  Her bottom line seems to be that this is not an event that will devastate or even significantly harm the country’s agricultural output, though some grain storage facilities may have been affected. (Worth noting: Her post predates today’s revelations that radioactivity could be affecting certain agricultural products.)

Nonetheless, I got curious just what might have been going on in those greenhouses and the surrounding farmland.  So I called up our Droppr tool, which is a nifty interface combining Google Maps with our many layers of HarvestChoice crop and other data. HarvestChoice is mostly focused on sub-Saharan Africa and parts of south Asia, but the data behind this Droppr, Spatial Production Allocation Model (SPAM; http://mapspam.info) is world-wide:

Droppr pointing the tsunami-damaged areas in Japan

While mountains rise up a fairly short distance in from the shore, the darker splotches indicate the coastal areas north and south of Sendai are indeed heavily farmed. While this region might not represent a large portion of Japan’s agricultural output, clearly there are many farmers whose livelihoods will be affected.  And just what are they farming?

Sub-national crop production statistics for the Droppr-point site, retrieved from the Spatial Production Statistics Model (SPAM)

Not surprisingly, lots of rice.  I recall from my time living in Japan that Japanese have a real cultural pride in growing and consuming their own rice (backed up by government agricultural policy) which means much of the country’s agricultural acreage is turned over to (mostly small-holder) rice farms.  Some possible good news is that rice fields are relatively resilient to encursions of salt water, and the planting season was not yet underway. But one has to imagine the damage done by a disaster of this scale is something on a larger scale than your run-of-the-mill salt water encursion.

Interestingly, it’s not all rice.  Soybeans, sugarbeet, potato, wheat, even sugarcane are all grown in that region.  The usual patterns of Japanese agriculture suggest a portrait of many small or even part-time farmers wiped out by the surge.  I wonder if they’ll return to this lifestyle, or if the land post-disaster will have a different future, with agribusiness playing a bigger role or perhaps with the land pulled out of production altogether.

Still curious about just what, if anything, was in those poor greenhouses this time of year — perhaps some rice farmers getting an early start?  My web searching has come up empty on the question; with so many other life-and-death concerns in Japan, that’s not surprising. Any notions?

Questions

More often than not, we get same types of questions in this format: “What’s the (A) of (B) in (C)? Quick! I need to use it in the report I’m writing right away.”

• A: Choose from [Area, Production, Value of Production, or Yield]
• B: Choose from [Banana/Plantain, Barley, Dry Beans, Cassava, , Cowpea, Groundnut, Maize, Millet, Pigeonpea, Potato, Rice, Cotton, Sorghum, Soybean, Sugarbeet, Sugarcane, Sweet Potato, Vegetables, or Wheat]
• C: Choose from [ (all the countries in Sub-Saharan Africa) or (country aggregates - like ASARECA, COMESA, etc) ]

Yes you can go to FAOSTAT yourself (if you have access and your login ID and password handy) and start querying around, but shouldn’t there be an easy way to get this simple piece of information quick?

Answer

Yes, sure it is – enter the HarverstChoice HappyStat (Harvest Area, Production, Value of Production, and Yield Statistics) at http://www.harvestchoice.org/lab/index.php/stat. It’s a simple tiny web widget that dynamically/super-quickly answers to your questions on country-level agricultural statistics data.

In fact, the HappyStat has been always available at the HarvestChoice main website under the Market Data section since 2008, but became one of the best-kept secrets. So, we felt it’s maybe worth posting here to start collecting your feedback. If you find it useful (or not), let us know (use the comment form below, or use the feedback form). We’ll use your suggestions/comments when we update it with newly available statistics data.

Notes

• Make your choices in the top part (Show, Display, Where, Commodity), and you’ll see the results at the bottom automatically.
• All the underlying statistics data was retrieved from FAOSTAT 2006. We plan to update with FAOSTAT 2010 in the first half of 2011.
• Negative value (-9999) means no data available for the given year/commodity.
• No, of course we don’t mind answering to your questions – like the one above – but wouldn’t it be nice if you can find the answers *yourself* even quicker? Besides, by exploring the database yourself, you may find something interesting for your research work.

Despite a common (trust us, it’s not just you) misconception, sweet potato (Ipomoea batas L. Lam) is not at all related to potato (Solanum tuberosum), nor to the true yam (Dioscorea batatas). In fact, they are all quite different. Sweet potato is a root crop, and potato and yams are tuber crops. Even researchers sometimes get confused that their publications mistakenly relate the crop species and data. For example, some countries reported the production of sweet potato as yam, and vice versa. Some countries still combine sweet potato and yam in the same category as an aggregate.

Useful Facts about Sweet Potato in Sub-Saharan Africa

1. 

   Sweet potato harvest in Uganda

   Sweet potato belongs to the morning-glory family, and it was originated in Latin America and is thought to have been brought to Africa by slave traders.

2. The growth period for sweet potato is 3-7 months. In countries with two rainy seasons (e.g., Rwanda, Burundi, Uganda), sweet potato is available in the market 11 months of the year and is a primary staple. Elsewhere in Africa, it is available 4-8 months of the year.
3. Sweet potato is the 3rd most important food crop in seven East and Central African countries (6.8 million tons), after cassava (28.4 million tons) and maize (13.2 million tons). It ranks 4th in six Southern African countries, and 8th among four in West Africa.
4. Sweet potato is high in carbohydrates and can produce more edible energy per hectare per day than wheat, rice, or cassava.
5. Sweet potato requires fewer inputs and less labor than other crops such as maize, and tolerates marginal growing conditions (e.g., dry spells, poor soil).
6. 

   Farmers planting sweet potato vines between the cassava in Tanzania

   Many parts are edible. The leaves and tips of the sweet potato plant are widely consumed by people in Africa. Sweet potato vines provide a high-protein animal feed.

7. Sweet potato is also a valuable source of vitamins B, C, and E, and it contains moderate levels of iron and zinc.
8. The flesh color of sweetpotato ranges from white, cream, and yellow to orange and purple.
9. Sweet potato in Sub-Saharan Africa is primarily grown on small plots by poor farmers, mainly women.
10. Better agronomic practices, such as site selection, planting techniques, spacing, weed control, soil fertility and water management could more than double sweet potato yields in Sub-Saharan Africa.
11. One of the greatest threats to sweet potato production is sweet potato weevil, which often causes losses of 60-100% especially during droughts.
12. Sweet potato is bulky and perishable, yet promising pilot efforts are expanding market opportunities through the use of sweet potato flour, dried chips, juice, and bread as well as its use as animal feed. Investments in improved infrastructure and value chain efficiency could expand sweetpotato markets, including into growing urban markets.
13. Most commonly used varieties in Sub-Saharan Africa are white or yellow-fleshed.  There is a strong emphasis on introducing and promoting orange-fleshed varieties because just one small root (100-125 gms) supplies the daily vitamin A needs for a young child.

We <3 Sweet Potato

Given the importance of sweet potato in the food consumption in many countries and its nutrition value as well as potential market opportunities, HarvestChoice is closely working together with International Potato Center (CIP; one of the CGIAR Centers) to improved the quality of sweet potato production statistics data, modeling growth and yield of sweet potato in Sub-Saharan Africa, and assessing the potential benefits of introducing new management technologies to smallholder farmers.

More information on sweet potato on the Sweetpotato Knowledge Portal: www.sweetpotatoknowledge.org

• “Does anybody have experience (good or bad) using DSSAT with WINE on Linux. WINE is supposed to allow Windows applications to run under Linux but does not guarantee full Windows emulation.” – JW
• “If it is possible in Linux I think it can also be tried in MAC OS. Will you please share the procedure for compiling it linux, so that I can try in MAC.” – JT
• “I have been looking to run DSSAT models in UNIX platform. I certainly need your help in doing this.” – GA

Answer

This post assumes:

• You have a valid license of DSSAT and access to the source code of v4.5.
• You have good knowledge of DSSAT and CSM in general.
• You have good knowledge of running CSM from the command line.
• You have a valid license of Intel Fortran Compiler for Linux (free for non-commercial use)
• You feel comfortable handling files and executing programs from the command line on Linux.

On Windows, even with the DSSAT shell program, the actual model behind the interface runs on the command line, reading input files in plain ASCII format, writing output files again in plain ASCII format. You might have thought this seemed to be an old-school way (“What is DOS?”, “Who uses Fortran?”), this way in fact provides great flexibility and scalability – including the possibility of porting the program into other operating systems (as long as there is a modern Fortran compiler, that is). Besides, as computer scientists put it, there is only few things in the Earth faster than well-compiled Fortran programs.

The idea is simple; once you can compile the core executable program of DSSAT (dscsm045.exe in this case), you can run it on Linux (or any other platform). Here are few minor tricks that you’d need to know.

Put all the input files into one directory

In case you didn’t know, you can run the model from one directory when you have all the input data files there (weather, soil, cultivar, ecotype, species, standard input, etc). Of course you can maintain the original directory structure as in Windows, but it is a lot easier to manage your set of files in one place per project.

DSSAT Profile (Filename: DSSATPRO.v45)

This file indicates the paths to find various model executables and input file directory. Once you put everything in one directory, it gets easier in this step: simply put a period (.) in the place of C:\DSSAT\, indicating “Hey, it’s in the same directory.” For example, from [MBA C: \DSSAT45 DSCSM045.EXE CSCER045] to [MBA .   DSCSM045.EXE CSCER045]. Make this change throughout the file (or, you can take a look at mine here as an example; may or may not work on your system).

Settings for Forward Slash

There are two places in the source to take care of the slash direction difference between Windows and Linux. Just switch the comment (! symbol) between [DOS, Windows] and [Linux, UNIX].

ModuleDefs.for

!     CHARACTER(LEN=5), PARAMETER :: OPSYS = 'WINDO'   !DOS, Windows
CHARACTER(LEN=5), PARAMETER :: OPSYS = 'LINUX'   !Linux, UNIX

CRSIMDEF.FOR

!     CHARACTER(LEN=1),PARAMETER::SLASH = '\' !DOS, Windows
CHARACTER(LEN=1),PARAMETER::SLASH = '/' !Linux, Unix

Weather file path

When you have all the files in the same directory, you don’t need to have PATHWT attached to the weather file name (in the INP file). It’s not an official fix, but I found it was causing problem finding the weather file. Find the appropriate line (line 154 in my case; may be different depending on your build number) and take [PATHWT] out.

optempy2k.for (L154)

!      WRITE (LUNIO,2800,IOSTAT=ERRNUM) FILEW,PATHWT
WRITE (LUNIO,2800,IOSTAT=ERRNUM) FILEW

Keep file extensions in lower case

The Fortran files have inconsistence extensions (.FOR or .for), which is fine on Windows but not on Linux.  So, let’s keep them all in lowercase .for not to confuse the compiler (or ourselves).

Compiling order matters (or not?)

By definition, modules are better to be compiled first. In fact the compiler is smart enough to take care of this, but if you get annoyed by warning messages popping up..

Debug or Release?

Debug-mode compiled program runs slower but gives you more detailed error messages when things go wrong (which file and what line). Release-mode compiled program runs fast. Bash scripts to do this work included here.

Few final steps

Now, remaining steps should be straightforward – rename the executable as dscsm045.exe, change permission to executable, copy it to your workspace, and run it from the command line using the batch mode. Of course, you’ll also have to manipulate the DSSBatch.v45 file manually: For example, change [C:\DSSAT45\maize\BRPI0202.MZX to] just [BRPI0202.MZX]. Outside of shell, you’re now on your own!

Worth it?

I think there will be many motivations to do this work – but in our case, at IFPRI, this was needed to run DSSAT crop models on a Linux cluster (80-CPU, as of early 2011). With this, we were able to crank out global-scale simulation results on 10 km grids. It was certainly worth the effort!

Let’s talk

Now that we all knew that there are many of our colleagues interested in this issue, let’s talk openly. Got stuck? Know a better way (sure there will be!)? Feel free to leave your comment/question below.

Special Thanks

I must say that figuring out all this was not possible without insightful helps and advices from frontiers like Cheryl Porter and Guillermo Baigorria at University of Florida. Also, many thanks always to my colleague/friend Ricky Robertson at IFPRI.

Arc-minute?

Many of spatial datasets that HarvestChoice is generating are based on grids (we sometimes call it “meso-scale”) at three different resolutions – 60 arc-minute (1 degree), 30 arc-minute (0.5 degree), and 5 arc-minute (0.08333333333… degree). We will spare the definition of arc-minute here. For convenience we often call the 5 and 30 arc-minute grids as 10 km and 50 km grids, respectively (although that’s not exactly/technically true – since the size of grid cell changes as latitude). However, how exactly big is one cell at 5 arc-minute resolution?

Feel It Yourself

To help you understand how big (or small) a grid cell is at each spatial resolution, use the Droppr app below – showing the three resolutions of grids over global land area. The marker (Droppr) will show some information on the ground as well as the Cell ID’s (we call this HCID; this is the key to retrieve spatially explicit data from our database) at each resolution. Feel free to move the Droppr around; even try it for your backyard!

• Black cells: 60 arc-minute (1 degree) grids.
• Blue cells: 30 arc-minute (0.5 degree) grids. Note that black lines overlaid on top of blue lines.
• Pink cells: 5 arc-minute (0.083333…. degree) grids. Often called as 10 km grids.

Implication

For the raster-based data layers (counterpart of raster is vectors, such as country-level datasets), the resolution indicates to the unit area of same value. For climate dataset as an example, 5 arc-minute resolution means you will get a same value within the cell; no more finer information. Would that be detail enough? It often depends on the research question and analysts’ capacity as much as the data availability. For regional scale studies too-high resolution data might overwhelm analysis and interpretation of results, while too-coarse resolution data would disregard spatial variability of biophysical information. There is no magic single resolution that serves all purposes; however HarvestChoice found a sweet spot at 5 arc-minute (which is also one of the most commonly used resolutions in datasets in similar disciplines). Whenever possible, upcoming spatial data products from HarvestChoice will be standardized at 5 arc-minute resolution, including the HarvestChoice DataTiles.

Word cloud of the name of crop models retrieved from the survey result

Crop models help researchers to simulate the future of food security under climate change scenarios. Many crop models exist around the world, but they are often developed independently and not widely known or used. Some models have been developed to represent a single crop or part of a crop production process in detail, whilst others have the capability to model multiple crops in complex rotations and under various management, weather and soil conditions.

To better understand the global extent of crop model development and to identify gaps in capabilities, and to determine the geographical coverage and range of crops represented, the Macaulay Land Use Research Institute and IFPRI/HarvestChoice were commissioned by the CGIAR Agriculture and Food Security Challenge Program (CCAFS) to conduct a rapid meta-analysis of crop models for climate change and food security researches using on-line survey to the crop modeling community in the world. For about 1-month of time (September 2010), 141 respondents from 74 countries completed the survey, covering more than 100 crop models.

Key Findings

• 

  An improvement in the quality of data used for calibration and testing purposes and as input to the models was seen as one of the most important ways of improving models.

  An improvement in the quality of data used for calibration and testing purposes and as input to the models was seen as one of the most important ways of improving models.
• This is associated with a high requirement for improved availability of, and ease of access to shared data sets for calibration and model input.
• Use of models to improve understanding of processes was seen to be the best outcome, but policy development and climate change mitigation were not seen as key outcomes of model use.
• There is a paradox in that the main strengths of models were seen to be the detail of process representation, but not the skill in representing observed phenomena.
• The main strengths of the models were the representation of detailed processes, whereas the robustness in the quality of outputs was rated much lower.
• For improved modelling of climate change impacts, the best developments in process representation were seen to arise from better understanding and model representation of crop responses to extremes (particularly temperature and water limitations) and to elevated atmospheric carbon dioxide.
• The main food crops are represented by models, but the focus of application is cereals, maize and rice.
• 

  Word cloud of crops that respondents modeled

  Models were seen as being easily transferable to new locations, but limited by the availability of location specific data (e.g. soils, management, and weather).
• About half of respondents said their models had not been calibrated against elevated CO₂ experiments.
• Model evaluation and testing would be improved by availability of better quality data.
• Models need to be tested more against extremes of rainfall and temperature.
• Some models incorporate damage by insect pests, pathogens and physical damage (lodging, frosts, flooding), but there is a need for closer dynamic linking between weather, soil conditions and crop status with the characteristics of the individual form of damage in order to better represent observations.
• Modelling has been applied in most parts of the world, but the results indicate that the Middle East, Central Asia, African and Russian Federation countries have been under represented by modelling efforts.
• The quality and level of detail of documentation varies considerably between models, with clear potential for improvement.
• Funding was seen as the main factor limiting further model development.

Full report can be downloaded at here.

What’s Next

When researchers design a new project with crop modeling components, they frequently ask around questions of “Which model simulates this crop?”, “Which model works best in this region?”, or “Who is the modeling expert on this crop in this region?”. The co-authors of the report are planning to work on publishing the survey results as a publicly available/editable online database (think Wikipedia for crop modeling community) that researchers can find answers to the common scoping questions.

Download

• Questionnaire used in the survey (PDF)
• Report – Crop Modeling for Climate Change and Food Security Survey (PDF)

It is now grown from Senegal to Kenya, and from Sudan to Mozambique, at elevations from sea level to 2,200 metres, and in environments ranging from moist savanna to humid forest.

What’s So Unusual?

Cassava is a very unusual staple crops in many ways.

Peeling cassava skin

Firstly, it is propagated only from cuttings, not from seeds or tubers. This is a serious constraint in trying to introduce new varieties, since cuttings are relatively bulky and have a severely limited storage life (3-5 days in most cases), and cannot therefore be sold through seed merchants or similar commercial channels. This propagation mechanism is also problematic in breeding new varieties, and in reducing the spread of disease. New laboratory techniques allow in vitro germination and the use of micro-cuttings, but these techniques are currently out of the reach of almost all peasant farmers.

A second unusual feature is the post-harvest processing and storage. Cassava tubers, although relatively unaffected by ageing or decay while still in the ground, decay rapidly after harvest, and must be eaten or treated rapidly. Cassava tubers have a fairly thick skin which is high in cyanide-type compounds which protect them from insect attack but are poisonous to humans. The skin must be removed, usually manually after softening in water, and the tuber must then be treated by chipping and drying, by grinding or in some countries by fermentation to ensure preservation. The cyanide toxicity can be a problem for new farmers inexperienced in the post-harvest requirements, and tubers of “bitter” cassavas still have trace amounts of cyanide after removal of the skin, but this is tolerated and the flavour is even preferred by many consumers.

Drying cassava

The third characteristic which distinguishes cassava from all other starchy staples, and which leads to much confusion and inaccuracy in production statistics is that cassava has no clear harvest season. Tubers can be edible, although small, after as little as six months after planting, but very little is harvested until nine months or one year after planting. Even then, the whole crop is not harvested at a single time. Rather, individual plants are harvested as required for consumption in the household or for sale, and the field may not be completely harvested until as much as two years after planting. Cassava is often treated as a reserve crop. If the main starchy staple, often maize, fails due to poor rainfall, then cassava in  the ground provides interim nutrition and income, and can also keep the household fed between harvests of other more seasonal crops. In some parts of Africa, notably in northern Zambia, cassava was introduced by the colonial government early in the 20th century to provide improved food security during a time of regular locust plagues. The African Red Locust, breeding in swamps in northern Zambia and south-western Tanzania, attacked all above-ground vegetation, devastating the maize crop, but underground cassava tubers were unaffected and the tough stems were not eaten and soon regenerated.

Although yields of cassava are improved by good husbandry, particularly regular weeding, it is still  possible to simply plant a crop and then to return after a year to reap some kind of harvest. In areas where land is scarce, households can plant cassava in remote fields far from the village, visiting the fields only for planting and harvest. In most African farming cultures, land preparation (felling forest, clearing fallow) is done by men, who may also do the actual planting of cassava cuttings. All subsequent work, especially weeding and harvesting, is done by women.

Subsistence Crop with Commercial Potential

Cassava (Manihot esculenta)

Much cassava in Africa is grown as a subsistence crop, although it is rarely grown only for household consumption. Tubers are sold in nearby markets to raise essential cash in times of need. The increasing urbanisation of some major cassava producing countries, for example Nigeria, has resulted in an important market for cassava tubers and cassava-derived products in the growing towns. This in turn has led to the growth of a significant cassava processing industry, using simple locally manufactured machinery for chipping, grinding and drying cassava, and also to commercial-scale farming, with growers planting tens or even hundreds of hectares of cassava instead of the more traditional fractional hectare plots.

Cassava has wider commercial potential. It is an important source of starch for industrial use and for stock-feed, and industrial processing is starting in Nigeria and other important cassava-growing countries in Africa, following the example of Thailand which is currently the largest producer of industrial cassava. The Nigerian government has recently introduced a law to ensure that at least ten percent of the flour used in making bread and pizza in Nigeria is derived from cassava, in order to reduce wheat imports and to provide support for local farmers. Other countries, including Mozambique, may follow this example.

Breeding Strategy

Breeding of improved cassava varieties has been directed by two main pressures.

One has been to increase yield. Traditional cassava varieties, grown by peasant smallholders, provide yields of 2-5 tonnes per hectare, depending on climate and agronomic techniques. New varieties have been developed which can yield more than 40 tonnes per hectare, although very high levels of fertiliser input are necessary in most African soils to achieve this level. Without inputs, which are beyond the means and experience of most smallholders, these high-yielding varieties will rapidly impoverish the soil.

The second main pressure has been from a range of diseases, many originating in the original home of cassava, in Brazil, but mutating in new African environments. Cassava varieties tolerant to new virus diseases may not provide higher yields in disease-free situations than the traditional varieties, but at least yield is maintained under disease pressure.

Much smallholder cassava in Africa is grown inter-cropped with other food crops. In rainforest areas of Cameroon, for example, cassava is inter-cropped with maize and groundnuts, and sometimes also with plantains and beans. Groundnuts are harvested first, then maize, and finally the cassava. These traditional inter-cropping patterns have probably evolved to provide a more or less continuous supply of food, but also to maintain soil fertility and even to mitigate the effects of disease.

Cassava tubers provide substantial starch for the farming household, but the tubers are notably deficient in protein, minerals and vitamins. The cassava must therefore be eaten in combination with other more nutritious foods. Fortunately, cassava leaves are high in iron and some vitamins, and are eaten as a vegetable in many African cultures. Attempts are being made, for example in the HarvestPlus Project, to develop cassava varieties naturally more nutritious than traditional varieties, in much the same way that orange-fleshed sweet potato is more nutritious than white-fleshed. The new biofortifying cassava with provitamin A are to be released in D.R. Congo and Nigeria in 2011.

I have been spending time the last few days trying to understand the kind of inputs that DSSAT requires to run it from the command line.  You two have already figured these things out, so I hope you can coach me at the places I get stuck.  I have discovered that the DSSAT documentation lags years behind the program.  So in looking at the 3.5 manuals, they look at the inputs for FileX.  But I have not been able to see how FileX knows what climate variables to use in a simulation.  Could you please let me know where that occurs when using the command line?  Thanks!!!

Answer

Sorry to hear you spent days to figure this out.. Next time when you feel stuck, please don’t hesitate to ask us first!

Crop models in DSSAT operate on daily basis, thus daily weather data is needed. However, for future climate conditions, we only have monthly mean of climate variables (solar radiation, min/max temperature, rainfall, and rainy days) – so we are using a weather generator program to generate daily weather data. DSSAT conveniently comes with two weather generators, WGEN and SIMMETEO, and we’ve been using SIMMETEO, which uses monthly summaries to estimate parameters. Descriptions of the two weather generators can be found in the second volume of DSSAT v3.5 (yeah…) documentation. Soltani and Hoogenboom have published a number of articles on the comparison of two methodologies.

Opting to use SIMMETEO instead of recorded daily weather data is quite simple:

1. Name your climate profile with four-character code (e.g., CLIM.CLI)
2. In your X-file, in the FIELDS block, put the four-character code as the weather station ID [WSTA]
3. In your X-file, in the SIMULATION CONTROLS block, set the option of WTHER to S (simulated)
4. In your X-file, in the SIMULATION CONTROLS block, set the random seed value to your favorite 5-digit number

By the way, at this point I presume you constructed the climate profile for each grid cell. Just in case, the climate profile is a small text file containing monthly summary of climate variables (which are all included in the FutureClim database), and it should look like this:

*CLIMATE:Gainesville,Florida,USA
@ INSI      LAT     LONG  ELEV   TAV   AMP  SRAY  TMXY  TMNY  RAIY
  UFGA   29.630  -82.370    10  20.9   7.3  16.6  27.4  14.4  1310
@START  DURN  ANGA  ANGB REFHT WNDHT
  1958    49  0.25  0.50   2.0   3.0
@ GSST  GSDU
     1   365
*MONTHLY AVERAGES
@  MTH  SAMN  XAMN  NAMN  RTOT  RNUM
     1  10.9  19.6   6.0  86.4   8.1
     2  13.5  21.4   7.3 107.9   7.6
     3  17.3  24.5  10.1 101.7   7.8
     4  21.4  27.9  13.3  75.1   5.7
     5  22.3  31.0  17.0  95.3   7.6
     6  20.8  32.6  20.6 164.1  11.9
     7  20.5  33.1  21.8 166.9  15.8
     8  19.1  32.9  21.7 195.2  15.4
     9  16.7  31.6  20.5 134.5  11.1
    10  14.6  28.4  16.2  56.7   6.3
    11  11.9  24.4  11.1  57.4   6.0
    12   9.8  21.1   7.6  68.9   7.4

See the third volume of DSSAT v3.5 (I know, I know..) documentation to learn more about this file format.

Couple of things to note:

• Due to the nature of stochastic method, you won’t get any spatial correlation on daily weather. If you look at the generated daily data, you can easily spot that one grid cell has flood and its neighbor cell has drought in a same year.
• Crop growth/water stress and yield are very sensitive to the rainfall distribution during the season, not necessarily to the total amount of rainfall. To (partly) overcome this, we run a number of realizations (about >100).
• Especially, not having spatial correlation of rainfall is particularly problematic if you need to analyze yearly production from region to region and conduct trade/policy analysis, for example. One common method to address this is using a delta method. Which means, shift historic climatology (observed daily weather, if available) to the future condition (e.g., In January in this grid cell, temperature increases by 1 C and rainfall decreases by 5%). You can implement this by re-generating daily weather data (WTH file) or embedding the “shifters” in the ENVIRONMENTAL MODIFICATIONS block of X-file (see the second volume and find examples). For HarvestChoice studies, I pre-generated 100-year daily weather data for SSA countries (see here for more information) and use this method when I need to run under future climate conditions.

Hope this helps your understanding. I’ll be happy to discuss what’d be best for your particular study.

Cheers,
Jawoo

Long-term maize trial plots at the Hatfield Experimental Farm, University of Pretoria

Long-term crop trials provide valuable research data to help understand complex interactions between field management practices and environment conditions. Such datasets are often used as a basis for developing analytical biophysical models that are used to assess the sustainability of crop production, strategies for improving the status of food security, and enhanced resilience to the climate variability/change impacts. Two historic example sites are still managed by Rothamsted Research in Harpenden, UK, and University of Illinois in Morrow Plots, Illinois, USA for generations (see NASA Global Change Master Directory for the list of long-term trials globally), and many of their outcomes are being used even in outside of their original purposes of the experiments. For example, the Rothamstead’s long term experiments were set up to study soil fertility and soil properties change and became the dataset to study and model soil organic matter dynamics. However, they are now also addressing other issues including “… the incidence of pests, diseases, and weeds; soil pollution; ecology of farmland; carbon sequestration; factors influencing the sustainability of arable agriculture.”

Experiments

In Sub-Saharan Africa, however, only a handful long-term trial sites have been established, and most of their datasets are left in grey literatures and not readily available to a broader research community. Among those, the Hatfield Experimental Farm Long-Term Maize Trial in Pretoria, South Africa, is an exceptional case. Being one of the oldest long-term experiments in the world, 32 treatments of the combination of water, fertilizer (N/P/K) and green manure applications, have been continuously applied on 128 experimental plots of maize (summer; planting in October/November) in rotation with legume crop (field pea during winter season) since 1939.

In 1996, Nel et al. published an article describing the experiment and findings based on decadal (5 periods: 1940-1949, 1950-1959, 1960-1969, 1970-1979, and 1980-1990) mean yield impacts by combinational effects of treatments and discussed the importance of balanced management of N, P, and K, and the critical roles of leguminous rotation crop and manure application. No follow-up publication has been made in public since 1996, but the trial is still being carried out by University of Pretoria. Hatfield is now preparing for its 71^st harvest in 2011.

Beyond the focus of Nel’s discussion on the nutrient management on decadal time-scale, we retrieved the underlying annual yield data to illustrate the temporal pattern of yield and its variability for the extended period of 1939-2005¹. To simplify the analysis and reflect typical² management choices of farmers in SSA, we focused on following four treatments under no-water condition³:

• Control
• N, P, and K fertilizer
• Manure
• N, P, and K fertilizer + Manure

The amount of fertilizer and manure and the used cultivar varies from period to period; but they can be generally considered as moving toward a high-input system. Unlike Nel’s study, no statistical adjustment was made on the raw yield data yet, primarily to illustrate the extent of potential yield variability as-is.

Rainfall

Rainfall data for the growing season covering the same time period was retrieved from the University of East Anglia CRU-TS v3.0 database at monthly-basis. The sum of rainfall for the first three-month of growing season, assumed to be from October to December, was computed and analyzed in relation to the yield, based on the hypothesis that this period roughly coincides with the period between planting and flowering, and the water stress of this period is critical to yield.

Figure 1. Annual early-season (Oct-Dec) rainfall amount (mm). Shaded areas indicate plus/minus 1 and 2 standard deviations.

When the amount of the early-season rainfall was plotted annually (Figure 1), overall there were 13 (19%) and 9 (13%) years with rainfall amount larger and smaller than the 1 standard deviation of rainfall over time, respectively. It was also shown that some increasing frequency of extreme events over the years. There were three incidents of rainfall amount larger than 2 standard deviation (i.e., wet years), and these all occurred after 1990. Between 1980 and 2005, there were 64% of years with rainfall less than long-term average. Overall there were about 60% of chance that the given year’s rainfall is less than the long-term average.

Figure 2. Cumulative distribution of rainfall by time segment (1939-1960, 1961-1980, and 1981-2005)

This was also shown at the cumulative probability chart of rainfall shown by three time periods (1939-1960, 1961-1980, and 1981-2005) (Figure 2). Especially, the probability line for the last time period, 1981-2005, indicated that more chances of receiving high amount of rainfall than previous periods. However the impact of rainfall variability was not clearly correlated with the variability of yield, possibly due to the initial water management practiced for all treatment plots.

Yield

In general, overall historical changes of yield show increasing trend over time (except for the control), but with noticeably high degrees of variability (Figure 3). Across treatments (excluding the control), it was shown that given year’s yield was >50% more or less (red dotted lines) than the previous year in about 23% of cases (Figure 4). That is, in about 1 in 5 years, given year’s yield – no matter how it was managed – can be unpredictably vary, either double of halve, from the previous year. Some extent of the variability may be due to the changing treatment conditions over time (e.g., different varieties) or undocumented biotic (e.g., pest infestation) or abiotic (e.g., acute drought) damages, yet the exact cause of such high variability is not yet identified in this quick study, but it will be the main subject of matters in the follow up study. The thickness of lines in Figure 4 correspond with the average yield level of each treatment each year, and they are showing the increasing trends over time with thickening lines, compared to the early years.

Figure 3. Annual yield & rainfall

Figure 4. Annual yield changes from the previous year (%)

To take into account the changing technology over time, the whole period was segmented into six (1939-1950, 1951-1960, 1961-1970, 1971-1980, 1981-1990, and 1991-2005), and the cumulative probability of yield during each period was plotted (Figure 5). Over time, the probability of getting higher yield level noticeably increased (i.e., moving toward right hand side), yet the variability persisted (i.e., wide spread of each line) even for the high input case of NPK fertilizer plus manure application (red line).

Figure 5. Cumulative probability of yield by decade

Remark

This post revealed the highly variable yield trends over time from the long-term maize trial in Hatfield Experiment Farm in Pretoria, South Africa. Since the trial started in 1939, the average yield levels across treatment significantly increased, partly due to the changing varieties and amount of nutrient inputs, yet so does their variabilities. About one in five years, average yield has been doubled or halved from the previous year. Not exact causes of the yield variability was identified yet, but will be further studied in the near future as a follow-up study.

It is worth reiterating that the long-term agronomic trials can give unprecedented insightful study materials for various study fields involving sustainability, ecology, food security, climate change, as well as conventional agriculture. Although the management of such trials demand nontrivial resources and dedications, we believe the value of well-maintained and carefully documented long-term trials easily outweigh the necessary cost and will provide valuable guidance for generations to come.

Download

We plan to make the annual maize yield dataset used in this post available in Excel format. Please fill out the data request form (not yet available); download instructions will be sent to your email.

Reference

Nel et al. 1996. Trends in maize grain yields in a long-term fertilizer trial. Field Crops Research 47 (1996) 53-64.

Acknowledgement

HarvestChoice deeply thanks University of Pretoria for the great efforts being made to lead the unprecedented long-term trial in Sub-Saharan Africa, and we sincerely wish this effort will continue for generations to come. We especially thank Johan de Beer, Johan van der Watt and Burger Cille for their data collecting efforts over the decades, and our colleagues/collaborators in Pretoria who helped to track down and compile the datasets, including Frikkie Liebenberg, John Annandale, and Martin Steyn.

Footnote

1. The annual yield dataset expands until 2008, but this study limited its focus until 2005 to coincide with the rainfall data.
2. We assumed it is not common to only apply potassium fertilizers.
3. Prior to planting in each season, sufficient water was applied to attain field capacity in all treatment plots. However, irrigation is not commonly practiced in the area. As a reference, Gauteng province (where Pretoria district belongs to) was reported to have irrigation in maize field was practiced in about 8% of area (Statistics South Africa, 2002).

How to find a right soil profile for this grid cell?

The Harmonized World Soil Database (FAO/IIASA/ISRIC/ISSCAS/JRC, 2009) (HWSD) delineates Sub-Saharan Africa (SSA) with 6,870 soil mapping unit (SMU) polygons. In each mapping unit, the share and detailed properties of up to 9 soils are presented in the database. The HWSD is a great data source that harmonizes regional/country-basis soil datasets in a globally consistent method, yet some aspects of the database make the properties not directly usable for complex crop systems models such as DSSAT and APSIM (e.g., no rooting depth defined; soil reference depth is harmonized for 30 cm for topsoil and 100 cm for subsoil). Thus, for researchers who uses the complex models, HWSD can provide information on some key soil variables at regional scale, but not site-specific soil profiles that crop models use.

On the other hand, the global soil profile databases, such as the ISRIC WISE Global Soil Profile Dataset, provides site-specific soil profile information at very detailed level (and also converted to the crop model-compatible format), yet their distribution is not uniform.

By mashing-up the mapping unit-based HWSD and site-specific WISE on 5 arc-minute grids, using 5 key soil variables (soil type, texture, organic carbon content, pH, and water availability), we created a simple look-up table that shows the best-matching soil profiles and their shared in each grid cell in SSA. When no appropriate soil profile is found from the WISE 1.1 Soil Profile Dataset, the HC27 Generic Soil Profiles was used to find a proxy soil profile. Following maps showing (A) the boundaries of 6,870 soil mapping units of HWSD and 588 units after the mash-up/matching process. (More text to be added later..)

Download

• Boundary of the 588 units in SSA and the soil profile ID for its predominant soil: https://hc.box.net/shared/609191jd1y (shapefile format)
• Table of soils and their shares on 5 arc-minute grids in SSA: https://hc.box.net/shared/66nskxlqla (tab-delimited ASCII text)
• WISE 1.1 Soil Profiles in DSSAT-compatible format: https://harvestchoice.wufoo.com/forms/download-wisol/
• HC27 Generic Soil Profiles for DSSAT and APSIM format: https://hc.box.net/shared/b5uy8vrrg1

1. Center for International Earth Science Information Network, The Earth Institute at Columbia University, Palisades, NY, USA
2. Tropical Agriculture Program, The Earth Institute at Columbia University, Palisades, NY, USA

Mapping Soil Constraints

Mapping the global extent of soil constraints to crop growth can play an important role in developing strategies for agricultural production, environmental protection, and sustainable development at regional and global scales. The soil fertility capability classification system (FCC) is a widely used technical system for interpreting soil taxonomy and additional soil attributes in ways directly relevant to plant growth (Buol et al.1975). In 2003, FCC4 was released with eight top- and sub-soil texture types and 17 condition modifiers defined to quantitatively delimit the soil’s capacity to provide ecosystem functions and services (Sanchez et al. 2003). As an update to the FCC4, HarvestChoice facilitated the development of an updated version using the Harmonized World Soil Database v1.1 (FAO/IIASA/ISRIC/ISSCAS/JRC 2009) as the underlying input data for analyzing 13 key modifiers that are the most relevant to achieving the Millennium Development Goals. This new dataset will play a key role in HarvestChoice’s forthcoming flagship study on the biophysical/economic impact assessment of agricultural production constraints in Sub-Saharan Africa (planned to be carried out in 2011-2012).

List of 13 Layers

Please refer to Sanchez et al. 2003 for more details on each layer.

• Clay in topsoil
• Loam in topsoil
• Sand in topsoil
• Aluminum toxic (a modifier)
• Calcareous (b modifier)
• High leaching potential (e modifier)
• Waterlogged (g modifier)
• High P fixation (i modifier)
• Low nutrient capital reserves (k modifier)
• Sodic (n modifier)
• Saline (s modifier)
• Cracking clays (v modifier)
• Volcanic (x modifier)

Download

Please fill out the data request form (Download instructions will be sent to your email)

References

• FAO/IIASA/ISRIC/ISSCAS/JRC, 2009. Harmonized World Soil Database (version 1.1). FAO, Rome, Italy and IIASA, Laxenburg, Austria.
• Sanchez, P.A., Palm, C.A., Buol, S.W. 2003. Fertility capability soil classification system: A tool to assess soil quality in the tropics. Geoderma 114:157-185.

Initially we’re looking at the relationship between poverty (as measured by per capita expenditure from recent national surveys) and agro-ecological variables.

Our goal is extend this type of analysis to include other physical variables (rainfall variability, soil quality and crop yield potentials) with a view to better inform targeted interventions. The spatial – and not just social and economic –  marginalization of poverty has long been recognized, with poorer rural households typically leaving in less-favorable environment, poorer soils and under harsher climatic conditions, and poorer urban households occupying fringes of cities beyond the boundaries of basic public services. More empirical evidence is needed however to better understand the interplay between spatial and economic exclusion.

Ghana - Interplay between agro-ecology and poverty prevalence amongst rural households

Kenya - Interplay between agro-ecology and poverty prevalence amongst rural households

Maps of Ghana and Kenya above show agro-ecological zones next to poverty prevalence. In both cases the spatial concentration of poor rural households (under $1.25/day per capita) and “ultra-poor” rural households (under $.75/day per capita) seems to be highly linked to overall climatic conditions, with higher poverty prevalence rates recorded in arid and semi-arid zones.

[Note: the data and maps in this article are being updating].

We have just recently embarked on producing fine-grained characterizations of small farm-holder households for a – yet limited – set of Sub-Saharan countries. This particular work attempts to produce household characteristics at “level-2″ administrative boundaries (typically district-level, “level-0″ corresponding to countries).

Kenya - Land productivity by district.

Malawi - Land productivity by district.

Nigeria - Land productivity by district.

Though estimates provided here are very rough and preliminary in many ways, we are exploring various methods to produce reliable Level 2 estimates in a more systematic and consistent format. Given HarvestChoice’s focus on measuring agricultural production technology adoption and potentials, our goal is to assemble household attributes with direct impact on – or generally believed to influence – farmer’s technology uptake and overall factor productivity (e.g. household composition, literacy and educational attainment, land tenure, field and parcel sizes, availability and use of inputs, access to agricultural extension services, access to water and energy sources).

Another related line of work is the production of level-2 sub-national poverty maps of Sub-Saharan Africa, so we will also be looking at the determinants of farmer’s income.

Initially, district-level estimates allow us to relate household characteristics to other spatial information, in particular agro-ecological zones (AEZ) and market shed areas – but spatially referenced household attributes may also be used to better inform crop production and economic evaluation models, or to segment households for targeted development interventions.

Four countries were selected here covering five AEZs. The countries were selected based on data availability and to maximize AEZ coverage. Data sources by country are as follows:

• Ghana: Ghana Living Standards Survey 5 (GLSS5), 2004/05
• Kenya: Kenya Integrated Household Budget Survey (KIHBS), 2004/05 — RIGA aggregates
• Nigeria: Nigeria Living Standards Survey (NLSS), 2003-2004 — RIGA aggregates
• Malawi: Malawi Integrated Household Survey (IHS-2), 2004 — RIGA aggregates

With the exception of Ghana we relied primarily on derived data sets from the RIGA project “components of income aggregates”. These pre-processed aggregates greatly facilitate cross-country comparisons, as data treatment, units and definitions are clearly documented and harmonized. However they do not provide sufficient coverage of agricultural activities, production, land and input uses, and we intend to fill in that gap in subsequent analysis.

Data

For each national survey average values and standard errors were statistically estimated using survey design specifications (stratified  samples and sampling weights) across multiple sub-domains (urban/rural population, poor/ultra-poor households).

In instances where no data is available for a particular district and/or sub-domain (either because of non-treated missing values or because there is just no data point available) we generally used regional (level-1) estimates instead.

Even with the “pre-treated” RIGA datasets many oddities remain, especially on key variables such as land ownership. The decisions we make in the treatment of missing values and outliers is particularly problematic when deriving level-2  estimates. We are very much aware that more refined calibration techniques (or a more judicious use of secondary data) are needed to ensure that no external distortion is added to the survey sample.

Standard errors on key variable estimates are often in the order of 10 to 20%, which in many instances make estimates fairly unreliable. That was clearly expected as nationally representative surveys are not designed to be significant at higher spatial resolution.

Farm sizes by country averaged across districts with average standard errors (19.7% across all countries).

Value of crop production by country averaged across districts with average standard errors (18.6% across all countries).

Share of female-headed households by country averaged across districts with average standard errors (22.1% across all countries).

Years of education by country averaged across districts with average standard errors (11.9% across all countries).

High standard errors on selected attributes (farm sizes, value of crop production, share of female-headed households, years of education by head of household) are shown on the above plots (means and standard errors are averaged across all districts for each country).

One way to increase the robustness of the results at higher resolutions might be to correlate household attributes with other known spatial characteristics (soil type and quality, agricultural production, agro-ecological zones and rainfall patterns, market shed areas, poverty prevalence rates, etc..) with a view to identify recurring patterns across the entire SSA sub-continent.

Sources

Household Surveys

• Ghana Living Standards Survey 5 (GLSS5), 2004/05 at http://www.statsghana.gov.gh/nada/?page=catalog
• Malawi Second Integrated Household Survey (IHS-2), 2004-2005 at http://www.nso.malawi.net/
• Rural Income Generating Activities (RIGA) Project at http://www.fao.org/es/ESA/riga/
• Suri, Tavneet, David Tschirley, Charity Irungu, Raphael Gitau and Daniel Kariuki (2008) Rural Incomes, Inequality and Poverty Dynamics in Kenya, Tegemeo Institute of Agricultural Policy and Development, WPS 30/2008 at http://www.tegemeo.org/documents/work/Tegemeo-WP30-Rural-incomes-inequality-poverty-dynamics-Kenya.pdf
• Mose, Lawrence O. (1997) Factors Affecting the Distribution and Use of Fertilizer in Kenya: Preliminary Assessment, Kenya Agricultural Marketing and Policy Analysis Project, Tegemeo Institute of Agricultural Policy and Development, Egerton University at http://www.tegemeo.org/documents/conference/factor.pdf
• Nigeria Bureau of Statistics (2006) Agriculture – Filling Data Gaps 1995-2006 at http://www.nigerianstat.gov.ng/nbsapps/agric/agric_gap.zip

Survey Estimation Techniques

A number of on-line resources were consulted to validate our choice of estimation techniques, as well as to help select and use appropriate R libraries.

• Chromy, James R. and Savitri Abeyasekera (2005) Household Surveys in Developing and Transition Countries:  Design, Implementation and Analysis, Chapter 19 – Statistical analysis of survey data, Studies in Methods, Series F No. 96 at http://millenniumindicators.un.org/unsd/HHsurveys/pdf/Household_surveys.pdf
• Lumley, Thomas (2010) Estimates in Subpopulations, August 7, 2010 at http://cran.us.r-project.org/web/packages/survey/vignettes/domain.pdf
• Lumley, Thomas (2004) Analysis of Complex Survey Samples, Journal of Statistical software, Vol. 9, Issue 8, Apr 2004 at http://www.jstatsoft.org/v09/i08/paper
• Oksanen, Jari (2010) Cluster Analysis: Tutorial with R, January 20, 2010 at http://cc.oulu.fi/~jarioksa/opetus/metodi/sessio3.pdf

Layered information on a grid cell

Starting with the Spatial Production Allocation Model (SPAM), many HarvestChoice data products are organized at 5 arc-minute (also known as 10 km) grids. We often align the grids with other additional layers and aggregate them in different ways. For example, by overlaying market accessibility and crop growing areas or livestock population, we assess the proportion of crop and/or livestock production in the  high (or low) market access zones. Much of these spatial analyses are being done in GIS, but also can be quickly done through simple database operations. 

As a by-product of recent exercise that overlaid agricultural production, agro-ecological characteristics, and market accessibility on the rural population in Sub-Saharan Africa (SSA), we created a simple table that lists about 300K grid cells (i.e., tiles) in SSA with related attributes, including: 

• Cell ID (CELL5M; see the HCID post for more information)
• Country name
• Administrative unit name and level (source: HarvestChoice HCADMIN v1.0)
• Total cell area in ha (including water body; rough arithmetic estimate)
• Total crop land area in ha (source: SPAM v3.02)
• Agro-ecological zone (source: HarvestChoice AEZ)
• Population of 2000 in headcount – Total, rural, and urban (source: CIESIN Gridded Population of the World)
• Market accessibility in hours – Travel time to the nearest town with 20K+ population (source: HarvestChoice Travel Times)
• Livestock population density of 2005 in headcount – Cattle, goat, and sheep (source: FAO Global Livestock Production and Health Atlas)
• Elevation in meter (source: CGIAR-CSI SRTM 90m v4.1)
• Annual total rainfall in mm (source: WorldClim)
• Crop areas for major staple crops of 2000 in ha (wheat, rice, maize, barley, millet, sorghum, potato, sweet potato and yam, cassava, banana and plantain, soybean, beans, sugarcane, coffee, cotton, and groundnuts) (source: SPAM v3.02)

Nothing in this list of variables is new – yet we found it useful to put all the variables in one place and quickly summarize/aggregate different dimensions of characteristics; we thought that it is ideal material for the Labs! This table can be also used for multiple purposes creatively. Please suggest any significant yet missing variables – we’ll try to add them for next release. 

Download

• Cell ID (CELL5M) grids in shapefile: https://hc.box.net/shared/hdg0j8esj6
• Attributes table in tab-delimited ASCII text file: https://hc.box.net/shared/n8f6cxu6ue

What’s Next

• Add/update crop production and area values to 2005
• Add suitability of crop production
• Market accessibility (travel time) for other town population classes (20K, 100K, 250K, and 500K)
• Update population to 2005
• …? (Feel free to make suggestions!)

In August, we posted a new collection of more than 3,400 soil profiles that are converted/formatted for crop model applications based on the WISE 1.1 Soil Profile Database. Utilizing these new soil profiles, as described in the post, we anticipate crop modeling studies to expand their coverage areas to even those locations where  no soil measurement data was previously available. To answer the questions of exactly how, here is a quick example application that can help you find the one(s).

Unless there is soil information collected on site exactly for the modeling purpose, your best bet on soils would be to find the closest match from the pool of available soil profiles. Depending on the scope of study, what aspects of the soils make the best/closest match can vary. Here we set the criteria with four variables: organic carbon content, texture, available soil water content, and soil type (but this could be easily expanded to whichever variables are significant for the specific scope of study at hand). Starting from more than 3,400 soil profiles, you can narrow the list down to the profile that comes closest to fitting the specific conditions of your study site by changing the values of these four variables. Once you’ve found a reasonable subset of soil profiles, you can export their list, find the particular soil profile from the database file (downloadable here) and use them in your crop model application.

This quick example implies that some information about the soil properties should be available.  We assume that the minimum information available includes soil texture, some measures of fertility, and the water availability. These are also particularly sensitive properties for crop models thus they are used as the basis of the HC27 Generic Soil Profiles. When this minimum information is not available, global soil databases, such as the Harmonized World Soil Database, can provide rough estimates; however, some properties are highly dependent on the land use and management history so that they change over time.

In addition, the more soil profiles become available to us, the better the chances are of finding the closest matches in the pool of soil profiles for particular conditions. There is a good chance that the team behind this work (the universities of Georgia and Florida) continues their work using the latest version of WISE database (v3.1) to expand the pool to more than 10,000 soil profiles in a near future; we will keep you posted on the progress.

100 Years of Daily Weather Data in SSA

Mashup of UEA CRU-TS with NASA-POWER (SLATE was previously known as CRU-Mashup)

Analyzing crop yield variabilities across large area (e.g., across Sub-Saharan Africa) requires long-term historical weather data for extensive coverage areas, but measurement datasets at the appropriate spatial and temporal scales (e.g., SSA-wide coverage with daily measurements for at least 30-year period) do not exist. Instead, there are following two commonly used climate/weather global data sources that are complementing each other, to some extent:

• University of East Anglia CRU-TS is a historic time-series climate database with monthly mean of six climate elements (cloud cover, diurnal temperature range, frost day frequency, precipitation, daily mean temperature, monthly average daily minimum and maximum temperatures, vapor pressure, and wet day frequency) over the global land area. Version 3.1 of the CRU-TS covers the time period from 1901 to 2009 at 0.5 degree spatial resolution. The database uses available station records and interpolation methods. See Mitchell and Jones (2005) for more details.
• NASA POWER, the Prediction of Worldwide Energy Resource, is a NASA database provides satellite-based estimates on surface meteorology and solar energy since 1997 at 1 degree spatial resolution. POWER provides all the climate elements that crop models typically requires, including solar radiation, daily temperature minimum and maximum, and rainfall.

Loosely combining the two databases, HarvestChoice synthesized a plausible historic daily weather database, SLATE (Synthesized Long-term Weather), that covers 100 year period from 1910 to 2009, for sub-Saharan Africa at 0.5 degree spatial resolution.

Methodology

Describing how the mashup process is being done between the CRU-TS and POWER databases

The synthesizing methodology is simple, and the process is around the finding of the closest match of total rainfall amount in a given month. For example, one POWER grid cell A (1 degree) covers four CRU-TS grid cell A1, A2, A3, and A4.  From the CRU-TS grid cell A1, let’s focus on a particular year and month – say July 1980, whose total rainfall amount was 100 mm. POWER does not have records of 1980. However, if we boldly assume the weather pattern of A is equality applicable to A1, A2, A3, and A4, POWER can provide 12 realizations (1997-2008) of July’s daily weather pattern for the cell A. From the daily weather, one can quickly get monthly total rainfall of July in those years. If one of those years, say 2000, has the exactly same amount of total monthly rainfall of 100 mm (or the closest to 100 mm out of the 12 sets), then we pull the daily weather record, including all the other elements as well as rainfall, and plug into the July 1980 of the new database, SLATE.

This loose method is built upon a series of assumptions, thus this can not be regarded as a real data or replacement of measurement data. Synthesized daily rainfall patterns could be still far off from what really happened, even their monthly total match (or very close). However, we assume the outcome of this process is plausible enough to be used in a quick modeling exercise to examine the impact of rainfall-induced crop yield variability.

One advantage of this method over stochastic weather generators is the maintaining spatial correlation of rainfall. For example, due to their stochastic nature, weather generators often result in unlikely weather patterns at short distance (e.g., on a given season, drought and flood could occur in neighboring grid cells). By using CRU-TS as a reference,  SLATE maintains the occurrences of regional climate events as they were recorded in the CRU-TS.

Download

The SLATE v1.1 data files can be downloaded at:
https://hc.box.net/shared/2nr28vapjrb3dglpeydh (updated: 24 August 2011)

• Geographic coverage is sub-Sahara Africa.
• A GIS data layer of the cell boundary can be also downloaded for mapping purposes.
• File name indicates the ID of grid cell at 0.5 degree (30 arc-minute), described at HarvestChoice Grid Cell Databases (HCID).
• The weather data file format follows the standard weather file format used in DSSAT-CSM.
• Due to the limitation of the two-digit year, only 100 years has been processed from 1910 to 2009.
• Information in the header columns
  • LAT and LONG columns indicate the centroid coordinates of the 0.5 degree grid cell from HCID.
  • ELEV column represents mean elevation (m) within the cell boundary.
  • TAV column is the average annual temperature (C), computed from the CRU-TS v3.1.
  • AMP column was computed by the historical mean temperature of the warmest month minus the coldest month, computed from the CRU-TS v3.1.
• Weather elements
  • SRAD: Solar radiation (mj/m²)
  • TMAX: Daily maximum temperature (C)
  • TMIN: Daily minimum temperature (C)
  • RAIN: Daily total rainfall (mm)

What’s Next

• Selective validation with nearby weather station data
• Assessment of spatial structure of rainfall over time

Acknowledgements

The POWER agroclimatic datasets were obtained from the NASA Langley Research Center POWER Project funded through the NASA Earth Science Directorate Applied Science Program.

Scale-Up Crop Modeling Applications

Complex crop system models can be useful decision/discussion-support tools at both large areas and small-scale plot-levels, for which the models were originally developed and designed. This idea is seemingly straightforward. As the simulation runs on a homogeneous management unit (HMU) area, one can treat a reasonably large area (e.g., a grid cell) as an HMU and compile crop model input datasets (e.g., soil, weather, crop, management practices) for each HMU in the target area. One can also design/develop a driver program that batch-processes the runs one by one. However, this process can be a big hurdle for non-skilled users with limited programming skills and, more importantly, a lack of access to the right scale of model input datasets.

For crop model users in developing countries who would like to jump start the process of using crop models at the regional context, HarvestChoice teamed up with IFDC to develop a customized interface for DSSAT v4.5. The Harvest Toolkit enables users to run scaled-up crop growth simulations at regional scale. Simulations can be run either on a grid cell-basis at 30 arc-minute resolution (also known as 0.5 degree or 50 km grids) or the sub-national administrative units for crop land areas in Sub-Saharan Africa countries. Harvest Toolkit comes with HarvestChoice-developed and pre-compiled standard crop model input datasets so that users can immediately start running simulations. These model input datasets can also be easily replaced with user-provided ones.

Features

• Any experiments that run on DSSAT can be run across space and time on the Harvest Toolkit, including sensitivity analysis crop cultivar selection, optimizing fertilizer application, irrigation scheduling, and cultivating alternate crops.
• Two types of simulation units are available (model can be run for all coverage areas or user-selected subset areas):
  1. Grid cells at 30 arc-minute resolution (n=6.085)
  2. Sub-national administrative boundaries (n=642)
• Pre-configured model input datasets for each simulation unit
  1. Rainfed planting month based on long-term climate variability analysis using MarkSim and a simple soil water balance model was provided by P. Jones and P. Thornton
  2. Soil profiles (predominant in the unit or multiple soils and their area weights)
  3. Historical weather record for a 40-year period, generated by mash-up NASA -POWER and CRU-TS v2.1
  4. Profile of mean climate variables for 1950-2000 at the centroid of each unit
     • Rainfall and temperature data from WorldClim
     • Solar radiation and rainy days estimated by P. Jones and P. Thornton
• Simulation results can be analyzed for the seasonal variability of each model output variable, including yield, biomass, and the nutrient balances of the system.
• When user-provided price information on specific management practices is available, a simple economic profitability analysis can be also performed.
• The output can be also displayed on the map and tabular format.

Technology

The Harvest Toolkit currently runs on Windows environments. The GIS component that lets users choose simulation coverage areas and display outputs uses ESRI MapObject 2.1 (Due to its licensing limitation, this component will be replaced with an open source alternative in the next release).

Where to Get the Software?

We’re currently trying to finalize the documentation and the detailed installation processes on various environments. We’re also sorting out the licensing of the various components used in the software.

In the meantime, the Harvest Toolkit can be executed on a pre-built virtual machine on the Amazon Elastic Compute Clout (EC2) platform (Amazon Machine Image ID: ami-98b853f1). If you’re  interested, please contact us via email at j.koo@cgiar.org to arrange a test drive.

Credits

The Harvest Toolkit was developed by a consortium of like-minded institutes, including the International Center for Soil Fertility and Agricultural Development (IFDC), the Southeast Climate Consortium (SECC), the International Consortium for Agricultural Systems Applications (ICASA), and HarvestChoice. The precursor of the Harvest Toolkit was the Climate Information Analysis Tool (CIA Tool), developed at IFDC from support from SECC and the International Food Policy Research Institute (IFPRI). The individuals involved in the consortium were Paul Wilkens and Upendra Singh, IFDC; Jawoo Koo and Stanley Wood, IFPRI; and Gerrit Hoogenboom, University of Georgia. The CSM model, and other model inputs, is a product of DSSAT v4.5 and countless individuals at many institutions over many years of work.

This surface divides Sub-Saharan Africa into market sheds by measuring the nearest city or ‘market’ with a population of 20,000; 50,000; 100,000; 250,000; and 500,000 respectively (year 2000 estimates from GRUMP alpha data). Nearness to a city or market is determined by measuring the least accumulated ‘cost’ or travel time to each market center. The market shed is the total area surrounding each market for which that market has the lowest cost in terms of travel time. Travel time was estimated based on the combination of different spatial data layers, or variables, which affect the time required to travel across to the given points (i.e. cities). These variables include: elevation, slope, landcover, roads, road type, rivers, borders, and major water bodies. For more details on the travel time see the Travel time to cities dataset.

Market shed data can be used to determine the number of people or households that are more than likely dependent on a given market center (assuming that most people would travel to the closest market for their needs). It is a good indicator of the value of a city in terms of market accessibility as well as the nearness of that market center to other market centers in terms of total area and people served.

The data is presented in a geodatabase format and besides the marketshed locations, the individual crop productions, population and livestock are also summarized into each market shed.

Data

Download

https://hc.box.net/shared/q1cq59c24g (1.8 mb)

Map

This surface divides Sub-Saharan Africa into port sheds by measuring the nearest port considering the port type and port size besides the least costs. Nearness to a port is determined by measuring the least accumulated ‘cost’ or travel time to each port location. The port shed is the total area surrounding each port for which that port has the lowest cost in terms of travel time. Travel time was estimated based on the combination of different spatial data layers, or variables, which affect the time required to travel across to the given points (i.e. cities). These variables include: elevation, slope, landcover, roads, road type, rivers, borders, and major water bodies. For more details on the travel time see the Travel time to cities dataset. It is a good indicator of the value of a port in terms of port accessibility. The data is presented in a geodatabase format and besides the port shed locations, the individual crop productions, population and livestock are also summarized into each port shed. 

Data

Download

https://hc.box.net/shared/zqmijcdtik (230 kb) 

Map

Major ports in sub-Saharan Africa and their sheds

Need a Market?

In almost any type of business, location is often the key to the success (yes, location, location, location!). If you’re in agricultural sector and wanting to open a new business for selling, say, fertilizer and seeds, that targets farmers with particular needs, you’d need to know about the status of agriculture in the area as well as how many potential crop/livestock and/or population you can potentially cover within certain travel time.

AgMarketFinder (http://marketfinder.info) is a web-based geo-processing GIS application that provides spatially-explicit agricultural statistics data (crop, livestock, and rural/urban human population) provided by HarvestChoice. Three types of spatial analysis can be conducted on-the-fly:

• from a user-selected point, four travel time zones (i.e., 0-2, 2-4, 4-6, and 6-8 hours of travel time),
• by sub-national administrative units, or
• by user-drawn polygon boundary.

What’s New in v2.0

• Processing speed improvement (no more (or at least much less) time-out)
• Livestock population data layer added (cattle, chickens, goats, pigs, and sheep)
• Two more geo-processing tools are added (by administrative unit and by user-drawn polygon boundary)

How to Use It?

Credit

• Data layers developed and provided by HarvestChoice
• Prototype developed by ESRI Applications Prototype Lab (APL) in March 2009
• Version 2.0 perfected by Spatial Development International (SpatialDev) in August 2010

Application

This application helps you identify a potential market location by analysing the travel time distances for 0-8 hour in 2 hour increments. The results are shown both on the map and as tabular data which can be downloaded.

Technology

The web application is built using the ArcGIS Server JavaScript API, the geoprocessing service to analyse the travel time is powered by ArcGIS Server and the base maps are being powered by ArcGIS Online.

Disclaimer

Data used in these analyses are from very different scales, and results are to be treated as indicative. We are continuously working to improve the harmonization and reliability of the underlying data layers.

Links

• AgMarketFinder v1.0 featured at Where2.0 (2009)
• AgMarketFinder v1.9 Screencast by ESRI APL
• So, what do you think? Give us your feedback!

Why?

A big part of our experimentations at the HarvestChoice Labs is focused on creating web-based application development. Many of these innovations are already being embedded in posts/articles throughout the HarvestChoice Main and Labs websites. Sharing documents, even if they are big, over the internet is increasingly easy; but figuring the best platform for sharing spatial dataset has been a challenge due to size, skill, required software, and so on. To benefit a broader user-base, we need a quick way of publishing/sharing any format (vector and/or raster) of spatial dataset, and, we believe, users need a quick way of browsing the data without having to download and use highly specialized GIS software. We’ve been experimenting multiple platforms for recent years–currently we’re experimenting with the ESRI ArcGIS Server platform. Yes, the server software is commercial, but it’s on us; users do not need to have anything other than a modern web browser and a reliable internet connection.

Example: Port Sheds

In this example, we quickly showcase what the data look like, let users choose a particular port shed and see its attributes, and execute a simple query function (e.g., which port sheds have the most maize production?). Try it out yourself!

HarvestChoice is assembling large sets of sub-national data at the household, farm and plot levels.

HarvestChoice relies heavily on large sets of household survey data to evaluate the economic impact of biophysical productivity responses at the farm level and on target populations.

At the farm level, the impact of technology change is largely driven by holders’ attitude towards risk and innovation. In economic impact analysis such attitudes are typically connected to observable household and community attributes — household size and characteristics of the head of household (age, health, gender, and education), consumption patterns, level of participation in local and regional markets, current use of production inputs, and access to extension services.

HarvestChoice is progressively assembling a collection of farm household attributes (with particular relevance to production technology), and using these attributes to characterize and segment populations at the sub-national level. The purpose is to facilitate on-going analysis, replicate research results, and disseminate often hard-to-acquire micro data.

Primary Data Sources

There are two main sources of harmonized micro data in addition to national household surveys:

• RIGA income estimates [details]
• FAO food security statistics [details]
• AFINS food security statistics [details]

We currently provide five categories of attributes:

1. household attributes (size, urban/rural, male/female-headed, age, education)
2. income/expenditure levels
3. market participation
4. consumption patterns
5. production systems (input use intensity)

Estimates (mean and standard deviation) at the regional level are disaggregated across the following classes:

• urban vs. rural households
• male-headed vs. female-headed households
• 5 quintiles of total household expenditure (used as a proxy for household income)

Sub-National Aggregates

The table below shows a list of countries and regions currently covered and a sample list of household attributes available for download.

You may need to wait a few seconds for the table to load. Use the filters to the left to switch between countries and to show/hide different economic attributes.

Acknowledgement

These preliminary harmonized estimates were obtained from Eduardo Magalhães, Datalyze Consulting Corp.

Crop systems models are often used to assess the potential yield of crops with or without certain types of constraints. As a quick and rough demonstration, this simulation result describes potential yields of rainfed maize with and without soil nitrogen constraint simulated for 50 years. The long-term average yield difference due to the soil nitrogen constraint was considered as the rainfed maize yield gap, assuming the potential and actual yield levels is defined by nitrogen availability in soils (In reality, there are many more types of constraints to be considered, such as soil nutrients other than nitrogen, pest/disease/weed, and variability of water; This post describes a simple proof-of-concept). The simulation was conducted at 5 arc-minute grids and aggregated over time (50-year average) and space (HCADMIN v1.0).

Simulation Settings

• Model: CERES-Maize 4.5-beta
• Crop: Maize
• Cultivar: Medium maturity generic (990002)
• Weather: CRU-Mashup v1.0.2
• Soil: HC.SOL v1.0
• Years: 1955-2004
• Reporting unit: HCADMIN v1.0

Download

https://hc.box.net/shared/nklxay3bsx (Shapefile format)

HCADMIN v1.0 - HarvestChoice Standard Reporting Unit

Many of HarvestChoice’s spatial analyses are done using raster datasets at various spatial resolutions (e.g., 1 km, 10 km, or 50 km grids). However, it is often more convenient and easily conceivable to aggregate the granular results at a more easily recognizable level using commonly-used administrative units. To standardize the aggregation and reporting process across the project, HCADMIN v1.0 was developed as a set of sub-national administrative unit boundaries in sub-Saharan Africa countries – at levels 1 or 2, depending on the number and size of units and the population within per country. Version 1.0 of the dataset contains 684 inland units.

Download

https://hc.box.net/shared/bg8orhlp85 (Shapefile format)

As an appendix to the Harmonized World Soil Database (HWSD) v1.1, seven soil indicators have been published to global extent at 5 arc-minute (10 km) resolution. From the website:

“On the basis of soil parameters provided by HWSD seven key soil qualities important for crop production have been derived, namely: nutrient availability, nutrient retention capacity, rooting conditions, oxygen availability to roots, excess salts, toxicities, and workability.

Soil qualities are related to the agricultural use of the soil and more specifically to specific crop requirements and tolerances. For the illustration of soil qualities, maize was selected as reference crop because of its global importance and wide geographical distribution.”

Indicators

The seven indicators are:

Each of the seven layers qualitatively represents the status of soil quality for a given grid cell:

1. no or slight constraints
2. moderate constraints
3. severe constraints
4. very severe constraints
5. mainly non-soil
6. permafrost area
7. waterbody

Although not quantitative, these indicators can be used as a starting point to estimate the soil-inherent abiotic constraints.

Data

The data can be downloaded from the HWSD website and the widget in this post shows the location-specific preview of the dataset. For sub-Saharan Africa region, the boundary of Soil Mapping Unit was overlaid to help understand the spatial extent of areas under the same constraint.

Source

• Fischer, G., F. Nachtergaele, S. Prieler, H.T. van Velthuizen, L. Verelst, D. Wiberg, 2008. Global Agro-ecological Zones Assessment for Agriculture (GAEZ 2008). IIASA, Laxenburg, Austria and FAO, Rome, Italy. (http://www.iiasa.ac.at/Research/LUC/External-World-soil-database/HTML/SoilQualityData.html?sb=11)

In this simulation, all major maize growing sites were assumed to have started cultivation in 1960 on previously grasslands. Forty years of continuous maize farming was then simulated with no nutrient or water management practices. Severely declining yield trends can be seen in most areas along with declining soil organic carbon stock.

Simulation Settings

• Model: CERES-Maize 4.5-beta with CENTURY
• Crop: Maize
• Run mode: Sequential
• Cultivar: Long maturity generic (990001)
• Nutrient management: None
• Water management: None
• Residue management: None (removed after harvest)
• Soil: HC27 v1.12
• Weather: CRU-Mashup v2.1
• Planting window: HC STWK-A v1.0-beta
• Simulated years: 1961-2000

It is important to farming households and other producers to have access to markets in order to trade/sell their goods. The more accessible markets are to the given population, the greater the population’s ability to remain economically self sufficient and maintain food security. These travel time maps indicate the degree of accessibility from a pixel to a populated place. The patterns shown here describe the geographic accessibility between places in Sub-Saharan Africa (SSA). We define accessibility as the time in hours required to travel from a given single point (1x1km pixel) to the nearest market center. Travel time to market centers is used as a proxy for market accessibility and shows the likely extent to which farming households are physically integrated with or isolated from markets.  Our travel time approach is estimated based on the combination of different global spatial data layers which represent the time required to cross each single point.

Accessibility is determined using a cost distance function to measure the ‘cost’ in time (hours) to the nearest market for each 1km pixel. The ‘friction’ or adjusted speed is based on a number of input variables, including: road locations, road type, elevation, slope, country boundaries, water bodies, & land cover. Each input variable was converted to a value representing the time it takes to travel 1 km. For example,  paved roads were given a value of 60km/hour whereas gravel roads may be given a value of 15km/hour. Water, landcover, slope, country boundaries, and elevation were used to modify the speed of travel (e.g. steeper areas were assigned slower speeds). The results are not intended to be accurate travel times between places but serve only as estimates of accessibility. To be accurate, we would need to have much better road data and more accurate assessments of average travel times over each component of the friction surface. Even then, it is doubtful that this could be achieved on a global scale. However, we would be very interested to know how the current travel times in the access surface relate to known travel times between locations.

Data

Download

https://hc.box.net/shared/6yps25u82b (229 mb)

Maps

References

• Nelson, A. 2000. Accessibility, transport and travel time information. CIAT Hillsides Project Report, CIAT, Cali, Colombia. pp 16.
• Nelson, A. 2008. Travel time to major cities: A global map of Accessibility. Global Environment Monitoring Unit – Joint Research Centre of the European Commission, Ispra Italy. Available at http://gem.jrc.ec.europa.eu/
• Nelson, A. and Leclerc, G. 2007. A spatial model of accessibility: Linking population and infrastructure to land use patterns in the Honduran Hillsides. In Hall, C. and Leclerc, G. (eds), Making World Development Work: Scientific Alternatives to Neoclassical Economic Theory. University of New Mexico Press, Alburquerque, NM. USA.

The global human settlements database comprises a global dataset of about 55,000 cities and towns with populations of 1,000 or more. The data are originally derived from CIESIN and further developments and improvements have been made using supplement datasets (e.g., gazetteers). Meanwhile, some of the cities’ points have been adjusted based on Google Maps and duplicated cities have been removed. There are five sub-settlements datasets that are derived using different population cutoffs: population > 20,000; population > 50,000; population > 100,000; population > 250,000; and population > 500,000 respectively. The dataset’s timeline is 2000 and it is in geo-database format.

The population figures were based on country-level census data, FAO data, and other international sources. In an effort to better understand the accessibility of populations to markets and services, HarvestChoice uses the location of human settlements as a proxy of market centers. These data serve as the target points when estimating and mapping travel time to cities as well as overall market accessibility.

Data

Download

(Coming..)

Canned Maps

References

• Balk, D., Pozzi, F., Yetman, G., Deichmann, U. and Nelson, A. 2004. The distribution of people and the dimension of place: methodologies to improve the global estimation of urban extents. Working Paper, CIESIN, Columbia University. Palisades, NY. pp 31.
• Pandey, D., Wheeler, D., Ostro, B., Deichmann, U., Hamilton, K. and Bolt, K. 2006. Ambient Particulate Matter Concentrations in Residential and Pollution Hotspot areas of World Cities: New Estimates based on the Global Model of Ambient Particulates (GMAPS), The World Bank Development Economics Research Group and the Environment Department Working Paper, The World Bank, Washington DC.

World Poverty at $1.25 (%)

The poverty maps (and related study) were commissioned by the CGIAR Strategy Results and Framework Team and produced through contributions from the International Center for Tropical Agriculture (CIAT), the Center for International Earth Science Information Network (CIESIN), the International Food Policy Research Institute (IFPRI), and the World Bank. The principal aim of the project was to produce a new poverty map of the developing world that goes beyond simply mapping national average poverty rates (e.g. around 150 data points in a global map of developing world poverty) to provide a more nuanced depiction of poverty.

Measuring Poverty

Poverty is defined as an economic condition in which one lacks both the money and basic necessities, such as food, water, education, healthcare, and shelter, necessary to thrive. (It is also characterized by less easily quantifiable characteristics, such as lack of participation in decision-making and civil, social, and cultural life.) Commonly measured by the average daily amount of money a person lives on, poverty is currently set at less than US$2 (PPP) per day (also called the $ 2 poverty line) for poverty and less than US$1.25 (PPP) per day (also called the $ 1.25 poverty line) for extreme poverty. The most common poverty metric is head count ratio (HCR), the percent of the population living below the established poverty line. These numbers are used to calculate absolute poverty numbers.

Mapping Poverty

These maps, constructed using more than 24,000 sub-national data points for the developing world, are groundbreaking in two ways. First, the spatial team was able to negotiate access to many previously unavailable national poverty maps based on national poverty lines. Second, while acknowledging that significant conceptual and methodological issues remain,  the team was able to construct the first-ever sub-national poverty map of the developing world (head count ratios and absolute numbers of poor people at $1.25 and $2.00 PPP 2005 poverty lines).

Current work in progress (scheduled for 2011 completion) focuses on improvements to the Sub-Saharan Africa (SSA) maps and the inclusion of more recent or higher spatial resolution data for another 10 countries in West and Central Africa.

The Maps

Below is a selection of national and sub-national maps based on the above-mentioned national and sub-national poverty line data. Please consider the following issues and acknowledgements before using these maps.

• These are provisional results and should be interpreted with caution.
• The spatial resolution of mapping, the poverty measures, and (where relevant) the consumption baskets to which they are applied vary widely among countries.
• Where 2005 sub-national estimates are based on re-scaling of existing national poverty line headcount index (p0) results, the reliability of that re-scaling depends, among other things, on the year of the national survey, the change in local consumer prices between 2005 and the survey year, and the gap in the national and the internationally-comparable ($PPP) poverty lines (when all expressed in 2005 local currency).

National/Sub-National Extreme Poverty ($1.25/day) Head Count Ratio (HCR; %) in Africa

Poverty in Africa at $1.25 (% of poor)

Delineation of sub-national boundaries at level 1 and level 2 where available, showing the percentage of population within the boundary which lives below the $1.25 (PPP 2005) poverty line.

• Table

National/Sub-National Number of Extreme Poor ($1.25/day)

Poor in Africa at $1.25 (nr of poor)

Absolute number of poor circa 2005 at the $ 1.25 poverty line for countries and sub-national units where available.  Representation units are pixels of approximately 10×10 kilometers.

• Table

Poverty Head Count Ratio (%) of Extreme Poverty ($1.25/day) and Percentage of Poverty within each Country in Africa

Delineation where available of sub-national boundaries at level 1 and level 2, showing the percentage of the population within the boundary as compared to the total national population living below the $1.25 (PPP 2005) poverty line. (The map is coming soon).

Poverty Table at $1.25

Column headings mean:

• Cntr_nr – number of poor in country
• Admin_nr – number of poor in administrative sub-unit
• Cntr_% – percentage of poor in country, measured on total country population
• Admin_% – percentage of poor in sub-unit, measured on total sub-unit population
• AdminCntr_% – percentag of poor in sub-unit, measured on total country population

Set filter to desired country for a country overview.

National/Sub-National Poverty ($2/day) Head Count Ratio (HCR; %) in Africa

Poor in Africa at $2 (% of poor)

Delineation of sub-national boundaries at level 1 and level 2 where available, showing the percentage of population within the boundary which lives below the $2 (PPP 2005) poverty line.

• Table

National/Sub-National Number of Poor ($2/day) in Africa

Poor in Africa at $2 (number of poor)

Absolute number of poor circa 2005 at $2 poverty line for countries and sub-national units where available.  Representation units are pixels of approximate 10×10 kilometers.

• Table

Poverty ($2/day) Head Count Ratio (HRC; %) and percentage of poverty within each country in Africa

Delineation of sub-national boundaries at level 1 and level 2 where available, showing the percentage of the population within the boundary as compared to the total national population which lives below the $1.25 (PPP 2005) poverty line. (map coming soon)

Poverty Table at $2

Column headings mean:

• Cntr_nr – number of poor in country
• Admin_nr - number of poor in administrative sub-unit
• Cntr_% – percentage of poor in country, measured on total country population
• Admin_% - percentage of poor in sub-unit, measured on total sub-unit population
• AdminCntr_% - percentag of poor in sub-unit, measured on total country population

Set filter to desired country for a country overview.  

Rankings

Ranking of Countries in Sub-Saharan Africa (SSA) according to Extreme Poverty ($1.25; left) and Poverty ($2; right) Head Count Ratios (%)

Ranking of poor countries in Sub-Saharan Africa (%)

SSA countries with poverty (at both $1.25 and $2 levels) and for which poverty data is available are ranked from the least to the most  poor.

On both charts, three percentages are displayed: those living on $1.25/day or less, those living between $1.25 and $2/day or less, and those living on $2/day or less.

Data compiled from poverty tables at $1.25 and $2 levels.

Table for Total Population at National and Sub-National Levels

Total population for each country and it’s sub-national units (approximately 2005)  shown for reference purposes.

Set filter to desired country for a country overview. (Do not select Admin_Units since this choice is misleading. )

Poverty and Agro Ecological Zones (AEZ)

Agro-ecological zones in Africa

An Agro-ecological zone (AEZ) is a land resource mapping unit that is defined by climate, landform, soils, and possibly landcover. It has a specific range of potentials and constraints for land use, especially for crop and animal production. Parameters used in the definition of an AEZ pay special attention to the climatic and edaphic requirements of crops, and on the management systems under which these crops can be grown. By overlaying the AEZs and the poverty numbers, we generated poverty information by AEZs and administrative sub-units. The corresponding map is not yet available, but the numbers are.  The actual poverty headcount data was calculated by determining the average poverty rate for each adminxaez unit and then calculating the headcount values using these rates and the total headcount for 2005.  These values will be different than the actual poverty headcount data; therefore, you may see some differences when comparing totals from the administrative units.

• Table

Table of poverty at $1.25 and $2 levels in AEZs

Set filter to desired country and variable for a country overview. Do not select Admin_Units since the choice is misleading. Variables are:

Only numbers and not percentages have been aggreggated. 

What’s Next

The following features will be added soon:

Reference

Wood, S., G. Hyman, U. Deichmann, E. Barona, R. Tenorio, Z. Guo, S. Castano, O. Rivera, E. Diaz, and J. Marin. 2010. Sub-national poverty maps for the developing world using international poverty lines: Preliminary data release. Available from http://povertymap.info (password protected).

This dataset was developed as a part of the “Geographic Domain Analysis to Support the Targeting, Prioritization, and Design of a CGIAR Mega-Project (MP) Portfolio” in 2009, and taken from the following authorship on the poverty mapping:

Stanley Wood^a, Glenn Hyman^b, Uwe Deichmann^c, Elizabeth Barona^b, Ria Tenorio^a, Zhe Guo^a, Silvia Castano^b, Ovidio Rivera^b, Enna Diaz^b, and John Alexander Marin^b

Additional data and technical contributions from:

Harold Coulombe^c and Quentin Wodon^c, Maria Muniz^d, Sam Benin^a, Todd Benson^a, Julio Berdegue^e, Jesus Gonzalez^e, Ana Maria Barufi^e, Peter Lanjouw^c and Ken Simler^c, Brian Blankespoor^c and Siobhan Murray^c

^a International Food Policy Research Institute, Washington DC, USA
^b CIAT-International Center for Tropical Agriculture, Cali, Colombia
^c World Bank, Washington DC, USA
^d Center for International Earth Science Information Network, Earth Institute at Columbia University, New York, USA
^e Rimisp-Centro Latinoamericano para el Desarrollo Rural, Santiago, Chile

Disclaimer

(text)

Links

Poverty

• PovCalNet
• Poverty.com
• WB Poverty Reduction & Equity
• Portfolios of the Poor (2009)

Poverty and Agriculture

• DFID Growth and Poverty Reduction (2005)
• Down to Earth: Agriculture and Poverty Reduction in Africa (2007)
• The State of Food and Agriculture (2005)
• UNU The (Evolving) Role of Agriculture in Poverty Reduction (2010)
• World Development Report (2008)

Poverty and Environment

• Poverty Environment Net
• UNDP-UNEP Poverty-Environment Initiative

Inequality

• UC Atlas of Global Inequality

General

• FAO
• Human Development Reports (All)
• The Millennium Development Goals Report (2009)
• WB Topics in Development
• Wold Development Reports (all)

Statistics

• FAOSTAT
• OECD Statistics
• UN Data
• UN World Population Prospects (2008)

Hunger

• Bread for the World Hunger Report (2010)
• FAO Hunger
• Free Rice
• IFPRI Global Hunger Index (2009)
• World Food Programme
• World Hunger

Food Security

• ADB and Food Security
• FAO Food Security Statistics
• The State of Food Insecurity in the World (2009)

Growing Seasons - Map 1

Growing seasons define the period of time when temperature and moisture conditions are suitable for crop growth. Understanding when these periods of growth occur helps researchers, policymakers, and farmers to better manage their land and water resources and to better understand how variability in climate affects the ability of farmers to plant, grow, and harvest specific crops. The concept of growing seasons takes into account the seasonality and length of potential growing periods during the year. The growing periods are determined based on the start of the rainy season, potential evapotranspiration, and temperature. Some areas of the world are not suitable for crop growth at any time, whereas others are suitable year round; still others are defined by multiple growing seasons. The HarvestChoice growing seasons surfaces were derived specifically for Sub-Saharan Africa (SSA). Season A refers to the primary growing season. If a region is bi-modal (defined by two growing seasons), it is identified as growing season B.

Defining Growing Seasons using Satellite Data

Measuring Growing Seasons has consistently been a challenge to farmers, agricultural researchers, and those concerned with the effects of climate change. Historically LGP has been calculated based on time series climate data from existing rainfall stations. The data are collected from the stations in the study area for a minimum of 20 years (FAO 1996), and then extrapolated from and analyzed across space to determine the start and end dates, and ultimately create a continuous LGP surface.

Determining growing seasons using satellite data provides a more accurate surface since it is based on actual reflectance values. This method also relies on the statistical analysis of time series data but since the satellite data are available continuously across space there is no need to estimate values using sparsely available point data. There is, thus, added confidence that the growing period data for each individual cell is a measure of truth. With the increased availability of satellite data and the growing periods can be more easily measured for multiple points in time.

Use of the Enhanced Vegetation Index

In order to improve our definition of growing seasons for input into the crop growth model and ultimately to redefine the Agroecological Zones surface, we explored the use of 1 kilometer (km) resolution greening up/down data derived from MODIS satellite images. These data are available for 4 years (2001-2004) and provide a comprehensive picture of the start and end days of the growing season for each year based on the Enhanced Vegetation Index (EVI). The EVI is a refined vegetation index that ‘de-couples’ the canopy background signal and reduces atmospheric influences (Huete et al., 1999). This index is considered a compliment to, but also an improvement over, NDVI since it uses both the red, NIR and blue reflectance values. The greening up/down data are presented in days based on a value of 1 for Jan 1, 2000 up to a value of 1826 for December 31, 2004. The value for each cell represents the date of greenup and the date of senescence (greening down). The processing of the satellite data was done at UMD. For most years there are four surfaces of start days and four surfaces of end days (two each forJan.-Dec.and two each for July-June of the following year). For HarvestChoice the data sets were analyzed together to determine the start and end dates for each calendar year and whether the pixel represents a bimodal area. The annual values were then compared to determine a ‘best guess’ representation of the start and end dates for a given pixel. The source data are 1x1km resolution. The analysis was done at 10x10km.

The graph to the right illustrates how the onset, senescence (decline), and thus length of the growing season are determined based on the satellite derived EVI. The bottom axis represents time (in days or weeks) and the side axis represents reflectance values. The growing season begins at the point that the reflectance value indicates a ‘greening up’ of the given cell.
A detailed description of how these data were used to determine the start, end, length, and modality of the growing season is available in Creating Growing Period Start and End Dates using satellite derived data – notes on methodology v1.2.

Data Gaps

The greening up/down images only provide data for cells that have a change in phenology throughout the year. Thus cells that are never green (desert, ice, urban, and other bare areas) or always green (evergreen forests) are left null. This is problematic for a number of reasons: 1) Because the documentation is no longer readily available, we do not know how these cells were identified. Did they eliminate certain cells based on their classification in ancillary land cover data set(s) or did they eliminate these areas based strictly on the reported EVI? 2) These cells do not naturally coincide with areas that we (IFPRI/HC) define as non-vegetated or evergreen forests based on GLCCD, GLC2000, MODIS so we need to decide how to best ‘fill-in’ values for these cells.

To test the extent of the problem we summarized the GLC2000 & MODIS (2001 data) land cover classes into six summary classes: Evergreen forest; deciduous or mosaic forest; dhrub or grassland; croplands or cropland mosaic; non-vegetated areas; and water. The above map shows the deciduous forest, shrub/grassland, and cropland classes as ‘other’ in order to better isolate the evergreen and non-vegetated classes. This is overlayed with a Boolean surface showing whether a cell has a growing season or not (yes = 1; no = 0). The GLC2000 and MODIS aggregations were overlayed, and the extent of growing season (for SSA) into one surface (ssa-grseasonxlndcov) and created a map showing areas that had growing season data and if not, how the cells were classified in the GLC2000 and MODIS data. In the map, the evergreen and non-vegetated classes took precedence over the other classes, ie. if either of the datasets had a cell classified as evergreen, we classified it as evergreen and did the same for non-veg & water (in that order). The rest fell into ‘other’.

DATA

Start of Growing Season A (Version 1.0)

The start weeks for the primary growing season (A) are based on values dating from 2001-2004. The source data contains a value in days (0-365), indicating the greening up day for the given year. These values were reclassified to weeks (values 1-52) in order to decrease file size and processing time. The ‘greening up’ data for all years were combined and the minimum and maximum start dates were identified for each year. The growing season start date was determined based on the median of the minimum start dates for the four years of data. Medians were used as a quick means of avoiding the assignment of anomalous values for the start date.

• Frequency distribution by start week A by country
• Growing season reporting tool by country & HarvestChoice administrative unit
• Download data

Start of Growing Season B (Version 1.0)

The start weeks for the secondary growing season (B) are based on values dating from 2001-2004. The source data contains a value in days (0-365) indicating the greening up day for the given year. These values were reclassified to weeks (values 1-52) in order to decrease file size and processing time. The ‘greening up’ data for all years were combined and the minimum and maximum start dates were identified for each year. The growing season start date for the primary growing season was determined based on the median of the minimum start dates for the four years of data. The growing season start date was determined based on the median of the minimum start dates for the four years of data. If there was a gap of 10 weeks or greater between the earliest and latest start dates within a calendar year for 2 or more years then it was assumed that the cell has a bi-modal growing season. This cell was then flagged and the start date for the second growing season was determined using the median of the later dates for each year. Medians were used as a quick means of avoiding the assignment of anomalous values for the start date.

• Growing season reporting tool by country & HarvestChoice administrative unit
• Download data

Bi-modal Regions (Version 1.0)

Bimodal areas are those with two or more distinct growing seasons. For this analysis, regions were defined as bimodal if there was a gap of 10 weeks or greater between the earliest and latest start dates within a calendar year for two or more years. This measure was taken on a cell by cell basis and the qualifying cells were then flagged as bimodal.

• Growing season reporting tool by country & HarvestChoice administrative unit
• Download data

End of Growing Season A (Version 1.0.beta)

The end weeks for the primary growing season (A) are based on values dating from 2001-2004. The source data contains a value in days (0-365) indicating the end of scenesence (greening down) for the given year. These values were reclassified to weeks (values 1-52) in order to decrease file size and processing time. The ‘greening down’ data for all years were combined and the end date was determined based on an analysis of the latest dates for each year. Cells identified as bi-modal were analyzed separately since the latest date would more than likely correspond to the second growing season so this had to be accounted for. The growing season start date was determined based on the median of the minimum start dates for the four years of data. Medians were used as a quick means of avoiding the assignment of anomalous values for the start date.

• Growing season reporting tool by country & HarvestChoice administrative unit
• Download data

Length of Growing Season A (Version 1.0.beta)

The length of the primary growing season (A) was determined by subtracting the calculated start week from the calculated end week (see start and end date sections for a more detailed description of these surfaces and how they were derived).

• Frequency distribution by start week by country
• Growing Season reporting tool by country & HarvestChoice administrative unit
• Download Data

Download Data

Reporting Tools

Location-specific Start Week of Seasons A and B

• Drag and drop the marker to retrieve the site-specific growing seasons information
• Overlaid map indicates the start week of the growing season A (see the Map 1 for legend)
• The map looks better on Firefox/Chrome/Safari than on Internet Explorer

Frequency distribution by start week for select countries in Sub-Saharan Africa

Frequency distribution by start week for the agroecological zones within the cool, Tropics in Sub-Saharan Africa

Frequency distribution by start week for the agroecological zones within the cool, Subtropics in Sub-Saharan Africa

Frequency distribution by start week for the agroecological zones within the warm, Subtropics in Sub-Saharan Africa

Earliest start week – Season A (col 1) & Season B (col 2)

Coming Soon

• Comparison to cropping calendar by country
• Updated and revised end weeks and LGP data
• Improved Agroecological Zones dataset using satellite derived LGP surface

References

• Huete, A, C Justice, W van Leeuwen. April 1999. MODIS VEGETATION INDEX (MOD 13). Algorithm Theoretical Basis Document. Version 3. Available at: http://modis.gsfc.nasa.gov/data/atbd/atbd_mod13.pdf

Soil property variables in the WISE v1.1 and the DSSAT requirements

Biophysical models are commonly used in ecosystem impact assessment studies at different temporal and spatial scales, and their results implicate decisions on the overall system at multiple levels—from micro-scale natural resource management to country- or regional-scale policy decisions. In agricultural systems, soil data is often key for understanding the current status of the ecosystem and its constraints, and strategizing proper interventions to sustainably improve its productivity. 

Increased demand for comprehensive, large-scale assessments of ecosystems increase, especially in the context of enhancing food security and improving resilience to the climate change impacts, leads to a corresponding increase in global/regional-scale soil databases. However, it is difficult to find extensive, quantitative, and geo-referenced soil property databases that are readily available for the areas (or regions) of interest, especially in developing countries.  

The International Soil Reference and Information Centre (ISRIC) developed a detailed geo-referenced global soil profile database, entitled World Inventory of Soil Emission Potentials (WISE). It is designed to provide scientists and ecosystem modelers with a homogenized set of soil property data that is useful for multiple purposes, from assessing soil vulnerability to determining crop production (Batjes, 2002). However, the WISE database does not provide all the information required by simulation models; the raw database is not immediately usable in all ecosystems models. Using the DSSAT (Decision Support Systems for Agrotechnology Transfer) Crop Systems Model (Hoogenboom et al., 2009; Jones et al., 2003) as an example, Figure 1 shows which variables are directly provided by WISE (solid arrows) and which ones can be estimated from other variables using pedotransfer functions  (dotted arrows).  

Locations of the WISE 1.1 soil profiles

In 2007, using the 1.0 version of WISE, Gijsman et al corrected and converted 836 out of 1,125 soil profiles to a format suitable for the DSSAT. A new, improved methodology implementing a rigorous quality control to detect and estimate errors and missing values was developed by Romero et al (2010), who successfully converted 3,404 out of 4,382 soil profiles into the same format (Figure 2). In this new version of the database, about 38% of original soil profiles were improved by correcting erroneous values and estimating key missing values. By utilizing this large sample of soil profiles, it would be possible to extend crop modeling studies to the areas where no soil profile data was previously available, by, for example, finding the closest matching soil profile from the database based on key soil parameters (e.g., texture, organic carbon, rooting depth, nitrogen). 

Preview

• Click the soil profile cluster (number indicates the number of soil profiles in each cluster) and zoom into it until the locations of individual soil profiles is shown.
• Some soil profiles are reported on the same location; their profile information will be shown together.

Download

Please fill out the data request form. Download instructions for the WI.SOL and documentation will be sent to your email.
(NOTE: The WI.SOL file contains soil profiles data formatted specifically for the DSSAT Crop Systems Models. The original WISE database can be downloaded from the World Soil Information (ISRIC) website.)

References

• Batjes, N.H. 2002. A Homogenized Soil Profile Data Set for Global and Regional Environmental Research (WISE, version 1.1) 2002/01. International Soil Reference and Information Centre, Wageningen, The Netherlands.
• Gijsman, A.J., P.K. Thornton, and G. Hoogenboom. 2007. Using the WISE database to parameterize soil inputs for crop simulation models. Computers and Electronics in Agriculture 56:85-100.
• Hoogenboom, G., J.W. Jones, P.W. Wilkens, C.H. Porter, K.J. Boote, L.A. Hunt, U. Singh, J.L. Lizaso, J.W. White, O. Uryasev, F.S. Royce, R. Ogoshi, A.J. Gijsman, and G.Y. Tsuji. 2009. Decision Support System for Agrotechnology Transfer (DSSAT) Version 4.5 [CD-ROM]. University of Hawaii, Honolulu, Hawaii.
• Jones, J.W., G. Hoogenboom, C.H. Porter, K.J. Boote, W.D. Batchelor, L.A. Hunt, P.W. Wilkens, U. Singh, A.J. Gijsman, and J.T. Ritchie. 2003. The DSSAT cropping system model. European Journal of Agronomy 18:235-265.
• Romero, C.C., G. Hoogenboom, A. Gijsman, G.A. Baigorria, and J. Koo. 2010. Strengthening soil quality databases: A new global soil database for crop and environmental modeling. Environmental Modeling & Software (Submitted).

Grid datasets typically have a continental or global extent and are stored and processed in different formats. Some files cover the world; others cover a particular region (e.g., Africa). The  global grids available here provide cell number identifiers, country identifiers, the fraction area that is land, and values representing the area covered by each cell. There are grids for a number of different but consistently-generated resolutions (30 seconds, 5, 10, 30 minutes, and 1 degree).

Grids and Resolution

We define five grids with a global extent using a “geographic” projection. Thus the corner coordinates are (in decimal degree):

• Upper left corner is at lon = -180.0, lat = 90.0
• Lower-right corner is at lon = 180.0, lat = -90.0

Different “resolutions” have different cell sizes. The one degree global grid has 360 columns and 180 rows, thus (360 x 180 =) 64800 cells (0, 1, 2, …, 64799).

To avoid confusion between grid cell numbers for grids with a different resolution, we refer to the cell numbers of these grids as: cell1d, cell30m, cell10m, cell5m, and cell30s. For example cell5m = 720 refers to the column 720 and row 0 while cell1d = 720 refers to column 0 and row 3 on their respective grids.

Countries

In the country grids, cell values are numeric codes that identify a country. The link between the identifier and the country name can be made via an access database or with this text file. The country grids were created by converting the GADM (Global Administrative Areas, http://gadm.org) version 0.9 polygons to a 30 second global grid, and aggregating using the mode (most common value).

Area

As we deal with un-projected grids (latitude/longitude) spatial units are in degrees, and cell resolution is constant in degrees, but not in m2. This is because one degree longitude is about 0.83 km at the equator, but 0 at the poles. The area grids provide an estimate of the size of each cell (in km2).

Fraction Land Area

Identifies the fraction of the grid cell that is land area. Derived from the 30 seconds country data.

Unique Cell Identifiers

For a grid of a certain resolution, irrespective of its extent, it is possible to always use the same identifiers for a specific grid cell, even if you are using only a subset of the data (e.g. Africa). Such a consistent numbering system assures a smooth data exchange and analysis. Grid (or raster) data consists of a rectangular area divided into rectangular (typically square) cells. In the example grid drawn [1], the green area represents the grid, with in red the cell numbers starting at zero. There are 10 columns and 5 rows, hence 10 x 5 = 50 cells.

The sequential numbering starts in the upper-left corner, moves to the right, and then to the next line, ending in the lower-right corner. For computational reasons is easier to start with 0 than with 1 (because in most computer languages arrays are indexed from 0…n and not from 1…n). Therefore the identifier of the last cell will be the total number of cells minus one, which is  10 x 5 – 1 = 49  in this example. Row and column numbers also start with zero. Cell numbers only have meaning for a specific grid (computationally only the number of columns must be the same; but semantically the grid also has the same spatial extent and resolution).

While a cell could be referred to with a row and column number, it is in most cases much easier to have a single unique identifier (in the case of simpler queries for example), also because for necessary cases, computing the row or column number from a cell number is relatively trivial. The next page shows a number of example functions in R language that are useful in this context. Such functions can also be easily implemented in other programming languages.

Download

• Raster: Grid cells are in the ESRI ASCII raster type format. For each resolution there are three files. One file with cell numbers (filename = “hc_seq_*”), i.e. the unique identifier for each cell. There is a also a file indicating the country to which (the majority of) that cell belongs (filename = “hc_cnt_*”). Countries are identified with a numeric code that is linked to country names in the access database and also here. The country grids were created by aggregating from a 30 second grid, using the mode (most common value). Finally, there is a file in which the value represents the area of that cell in km2 (filename = “hc_area_*”).
• Vector: Same as above for the raster data, but the data are stored in a shapefile format: hc_grid_shp.zip. That is, each raster cell is a rectangular polygon with the cell number as an attribute for easy (albeit perhaps inefficient) linking and displaying of data. Grid cells that only cover oceans or seas are not included.
• Cell database: The database file (hc_grid_mdb.zip) is in the Microsoft Access format, and it has three tables: “cells”, “countries”, and “gridspecs”. Table cells links the cell numbers of the different resolutions and also links these to the country numbers. This can be used to (dis) aggregate data, for example to distribute data at a 1 degree resolution to different countries. Table “countries” provides the link between the country codes “CID” and country names. The “gridspecs” table provides some essential parameters for each grid such as number of rows.
• Country boundaries: The country boundaries used to make the country grids is from the GADM version 0.9.
• Gridded data is included for all resolutions except 30 seconds resolution because of the large file sizes. If you require cell numbers of country identifiers at this resolution you can use these  grids: 30 second cell numbers (hc_seq30s_asc.zip) this one is a very big download) and for countries (hc_cnt30s_asc.zip).
• All of these files are provide for reference; in many cases using a raster type file format and a programmatic approach to calculating cell numbers may be more efficient then linking to these files.

Farm army worm damage on corn

As plant pathologists say, “Plants get sick too (Esnard, 2003)”. This is particularly so in highly-stressed smallholder farming systems in developing countries. As with humans, crops are susceptible to damage by many organisms, such as bacteria, viruses, and insects. Crops are particularly susceptible when they are under water and nutrient stress and thus not healthy enough to sustain immunity and recover their usual vigor. Such stress conditions are commonly found in subsistence-oriented, low-input cropping systems where farmers often have limited capacity to mitigate crop damage or even to prevent total crop loss. Some types of damage can be minimized by proper prevention or intervention measures (e.g., residue management, applying pesticides or planting resistant crop varieties), but such measures are either knowledge-intensive, costly, or both and are therefore often improperly utilized or underutilized by resource poor farmers. Providing assessments of the vulnerability of different cropping systems to pest and disease damage over space and time represents a potentially major contribution to the design, targeting and support of crop loss mitigation efforts by both the public and private sectors. 

To assess the potential impacts of pests and diseases on crop production, some crop models include coupling points, special model variables whose changing values can be used to represent pest and disease damage to crop organs or growth processes. Examples of coupling point variables include leaf mass or area, stem mass, root mass, root length and seed mass or number, all of which might be negatively impacted by pests and diseases. By identifying specific damage pathways and rates of damage using these variables (e.g., an army-worm infestation might be defined as reducing leaf area by 10% by the 40th day after planting), growth models can quantify how crop development would be affected and, ultimately, what would be the impact on crop yield. The coupling point concept was first introduced in 1983 (Boote et al.), and later formally implemented in the DSSAT crop modeling platform (Hoogenboom et al., 2009; Jones, et at., 2003) as a Pest Module (Batchelor et al., 1993). The Pest Module allows users to input field observations and scouting data on insect damage, disease severity, and physical damage to plants or plant components (e.g., grains or leaves) and to simulate the likely effects of those pest and diseases on crop growth and economic yield. 

For damage to be simulated, information must be provided on the specific pathways by which individual pests or diseases impact crop organs or crop growth processes. These pathways can define, for example, how much damage occurs on which parts of a plant on a daily basis as either a relative (e.g., percentage leaf area destruction per day) or an absolute (e.g., 10g seed destruction per day) value. The DSSAT Pest Module supports definition of the following pathways of pest and disease impact on crop growth: 

• Leaf mass destruction (%/day or g/m2/day)
• Leaf area destruction (%/day or g/m2/day)
• Stem mass destruction (%/day or g/m2/day)
• Root mass destruction (%/day or g/m2/day)
• Number of plants destroyed (#/m2/day)
• Share of plants destroyed (%/day)
• Reduction in assimilation of biomass (%/day)
• Seed destruction (%/day , #/m2/day, or g/m2/day)

Depending on the characteristics of the pest or disease, damage can be expressed in a single or in multiple pathways. For example, one corn earworm larva is known to damage susceptible maize in following ways: 

• Reduce small seed numbers by 10/m2/day
• Reduce large seed numbers by 2.5/m2/day
• Reduce leaf area by 0.5 %/m2/day

Example: Evaluating Acute Disease Infestation Impacts at Site and Regional Scales

Figure 1. Simulated impacts of leaf-damaging pest infestation on maize canopy development throughout the growing season at a single site. A range of “single shot” leaf damage events (at day 60) was implemented through damage pathway “coupling points” in the DSSAT model.

We illustrate the coupling point (or damage pathway) approach using the scenario of an acute “single shot” pest infestation on maize occurring two months (60 days) after planting. 

Figure 1 shows the simulated leaf area damage according to a range of assumptions about the severity of the infestation on canopy development throughout one growing season. Although the maize plant continues to grow (except in the case of complete leaf loss), the plant’s initial production potential cannot be recovered, resulting in a permanent loss of overall leaf development and, hence, compromised photosynthesis capacity and biomass production.  As more detailed ecological information about pest infestation becomes available, more realistic scenarios (e.g., damage with or without crop stress, extreme weather events, tolerant cultivars, or flexible planting windows) can be developed and tested to assess impacts on crop productivity, as well as to design efficient integrated pest management strategies (e.g., Willocquet et al., 2002). 

Figure 2. Simulated seasonal maize yield losses due to the impacts of recurring leaf-damaging pest infestation on maize canopy development throughout each growing season at (A) single site and (B) averaged over multiple major maize growing sites in Tanzania.

Beyond the single plot for a single season, such simulation experiments can be scaled up across larger areas for multiple seasons using the HarvestChoice Crop Systems Simulation Platform (Simplr) and help understand the spatial and temporal variability of crop production and yield under the impacts of disease infestation. For all major maize growing sites in Tanzania, represented as 10-km grid cells, three levels of hypothetical single-shot pest/disease infestation scenarios that are repetitively occurring two months after planting every season were simulated on rainfed/low-input maize production system (i.e., 20 kg[N] of urea application on traditional variety) for 20 years. 

Figure 3. Simulated maize yield (A) averaged for 20-year period in major maize growing areas in Tanzania and (B) average yield loss (%) from 50% of leaf area damage compared with the no-damage case.

Figure 2 shows the maize yield trends at one example site (Kwimba) and averaged over all sites. Seasonal yield loss was calculated by comparing the yield with controls with no damage. For the one site example, the highest level of yield loss coincided with the highest temporal variability, compared to the other two levels (Figure 2A). When averaged over space, however, such variability of yield loss was reduced and shown to be relatively stable over time. These suggest that there may be a threshold of leaf damage on maize, beyond which the damage is so devastating that the yield loss highly exacerbates, and the threshold level can be site-specific. 

Figure 4. Histogram showing the area for a range of average yield loss over the simulated time-period of 20 years.

Spatially, on average, yield loss increased as the damage became more severe–but at different rates (Figure 3). This is likely due to the different soil and climate conditions at each site influencing the stresses and development of maize thus the different extents of susceptibility. A histogram drawn for the range of yield loss also showed that there were distinctive spatial patterns of yield losses for the severity of leaf damage (Figure 4). 

With more detailed information about potential pest prevalence over time and space (e.g., HarvestChoice’s pest mapping outputs), more reliable estimate of crop yield losses can be made. HarvestChoice is evaluating the feasibility and reliability of such approaches for predicting the changing vulnerability of crops to pests and diseases at local, national and regional scale. 

References

• Batchelor, W.D., J.W. Jones, K.J. Boote, and H.O. Pinnschmidt. 1993. Extending the Use of Crop Models to Study Pest Damage. Transactions of the Asae 36:551-558.
• Boote, K.J., J.W. Jones, J.W. Mishoe, and R.D. Berger. 1983. Coupling Pests to Crop Growth Simulators to Predict Yield Reductions. Phytopathology 73:1581-1587.
• Esnard, J. 2003. Plants Get Sick Too!, pp. Poster. The American Phytopathological Society, St. Paul, MN.
• Hoogenboom, G., J.W. Jones, P.W. Wilkens, C.H. Porter, K.J. Boote, L.A. Hunt, U. Singh, J.L. Lizaso, J.W. White, O. Uryasev, F.S. Royce, R. Ogoshi, A.J. Gijsman, and G.Y. Tsuji. 2009. Decision Support System for Agrotechnology Transfer (DSSAT) Version 4.5 [CD-ROM]. University of Hawaii, Honolulu, Hawaii.
• Jones, J.W., G. Hoogenboom, C.H. Porter, K.J. Boote, W.D. Batchelor, L.A. Hunt, P.W. Wilkens, U. Singh, A.J. Gijsman, and J.T. Ritchie. 2003. The DSSAT cropping system model. European Journal of Agronomy 18:235-265.
• Willocquet, L., S. Savary, L. Fernandez, F.A. Elazegui, N. Castilla, D. Zhu, Q. Tang, S. Huang, X. Lin, H.M. Singh, and R.K. Srivastava. 2002. Structure and validation of RICEPEST, a production situation-driven, crop growth model simulating rice yield response to multiple pest injuries for tropical Asia. Ecological Modelling 153:247-268.

Simulated maize yield level changes after 35 years of continuous extractive cultivations

Farming entails a great deal of choices and uncertainties. From season to season, weather varies, price fluctuates, soil degrades, pest damages, and climate changes. Farmers everywhere must cope with these uncertainties. Throughout the history of agriculture, many options have been developed to help manage these risks, increase yields, increase efficiency, and, more recently, promote the sustainability of the overall system.

With these techniques and tools in mind, each farmer must assess their local context and analyze the costs and benefits of adopting them, such as the additional labor and/or investment required. National and international donors and policymakers share the farmers’ goal of improving food security as cost-effectively as possible. They, like farmers, must evaluate the feasibility and profitability of available strategies and policy options and decide which ones to promote and where. If reliable estimates could be made of the current and potential patterns of crop productivity under different scenarios, many agricultural development investment and policy decisions would be significantly improved or made with greater confidence.

Simplr, the crop systems simulation platform

Crop systems models mathematically describe the growth and yield of crop and its interaction with soils, climate, and management practices. As a decision-support tool, crop systems models can be particularly useful when combined with a sound understanding of the farming systems, their location-specific traits, and quality data.

Most modern crop models can quantify on a daily basis a crop’s various biological processes (e.g., the amount of solar energy transformed into biomass, water and nutrient requirements, supply, stresses, and growth stages), as well as physical processes related to crops (e.g., water runoff, soil carbon sequestration, and nitrogen leaching). Since the early 1970s, agricultural scientists developed various crop models based on improved knowledge of plant photosynthesis and respiration processes. Models range from the generic and simple to the very specific and complex. Some models use response functions at their core (e.g., yield is a function of rainfall and nutrients), while others use  intricate sets of differential equations to describe the complexity of different processes and their interactions. There is no final and universal crop model—rather, crop models are selected based on the type of research question. The performance of different crop models is openly evaluated in different conditions.

Modeling the gridded world

Conceptual schematic of the SChEF, the HarvestChoice Spatial Characterization and Evaluation Framework

Conventionally, agronomic researchers often use crop systems models to better understand the current status of farming systems at small/micro-scales and systematically test possible scenarios and estimate efficient uses of given resources before (or instead of) conducting real experiments. For example, modeling soil water and nutrient status can help make the irrigation and fertilizer application on crops more efficient.

At the meso-scale (i.e., pixelated view of the world on 10-km grids), given our best understanding of the current status of the smallholders’ farming systems in Sub-Saharan Africa (SSA), we use crop systems models to understand and assess the biophysical impacts of multiple abiotic/biotic constraints and evaluate the potential benefits of adopting a range of potential intervention scenarios. Ultimately, HarvestChoice aims to better inform economic and policy analysis of the overall agriculture sector in the region through its analytical research platform, within which crop systems models play a role as the biophysical estimator of the cropping systems’ productivity responses under scenarios of change, including a broad range of technology, crop management assumptions, and historic/contemporary/projected climate conditions. The overarching platform, collectively branded as Simplr (simulation platform), is a component of the overall HarvestChoice Spatial Characterization and Evaluation Framework (SChEF). The productivity assessments are made at a detailed spatial resolution across the entire SSA region.

Cumulative probability distribution of simulated maize yield under the different variety and fertilizer application practices for 30 years in major maize growing areas in Ethiopia

What is the spatial unit of simulation in the regional context?

Strategic questions often cover large areas where significant heterogeneity in crop growth conditions is found. However, crop systems models are developed for use in small unit areas where crop growth, environmental conditions, and management practices are considered homogeneous. To overcome this geographic scale limitation, Simplr implements a meso-scale analysis platform that assumes a uniformly generated grid cell as a unit area, which is currently designed as a 10×10 km grid cell (There are about 300,000 grid cells in SSA at this resolution). For each cell, a set of data required to run crop models has been compiled, including soil properties, weather/climate, and typical management practices. As different types of research questions require different levels of input data, various spatial and temporal resolutions of model input and evaluation data from multiple sources are being compiled. Input datasets introduce implicit spatial correlations across landscapes, but biophysical processes in each grid cell are independently simulated from neighboring cells. As necessary, the simulation results can be aggregated and reported to the 1st or 2nd level of administrative units in each country.

What are the productivity changes that Simplr can assess?

The most widely-used measure of cropping system productivity is crop yield; thus a primary use of Simplr is to assess crop yield variation in both space and time under different scenarios of changes in the cropping systems components. Additionally, Simplr also provides other measures of the productivity and performance of cropping systems. At the core of Simplr is a crop systems model built around the best scientific understanding of the biological, physical, and chemical processes supporting crop growth and its dynamic interaction with environment. Drawing on these hard-wired process models, Simplr generates seasonal estimates of crop growth and yield components (above and below ground biomass, yield, and residue) and tracks stocks and flows of nutrient (carbon, nitrogen, phosphorus) and water in a crop and soil profile. By comparing estimates of these indicators across different intervention scenarios, Simplr supports in-depth analysis of relevance to a range of topical agricultural development issues. Recently focused issues include crop improvement, integrated soil fertility management, small scale water management and irrigation, conservation agriculture, enhanced carbon sequestration, payment for ecosystem services, and farm scale adaptation to climate change.

Estimated value:cost ratio of applying urea on rainfed maize fields in five East African countries

What data and results has Simplr already generated?

The development of the Simplr has involved progress on several fronts:

• Climate data: The creation of a consistent historic, contemporary, and projected set of six key climate variables of sufficient spatial (10 km) and temporal (daily) resolution for landscape scale analysis of crop growth.
• Soils data: Assembly of a range of options for representing the heterogeneity of soil conditions on both a point and area basis.
• Contemporary production systems: Creation and collection of survey and expert-based evidence on key attributes of crop production systems in SSA; land holdings, cropping patterns, planting dates, input use, and market orientation.

Using evolving versions of these databases, several systematic assessments have been undertaken,  including: maize yield response to changes in planting date, varietal maturity periods and genetic potential, nitrogen application, and supplementary irrigation, potential changes in the spatial and multi-year patterns of yields and soil fertility status for maize, sorghum and cotton production systems in Mali under a range of improved soil and water management technologies, variability cereal yields based on simulation of long term (more than 40 years) of historic climate data, and projections of SSA cereal yields under a range of climate change scenarios. These results are accessible through the HarvestChoice website. On-going applications of Simplr include an evaluation of conservation agriculture technologies, an assessment of the economic cost of various biotic constraints on food staples, and a more extensive spatial comparison of the maize varietal performance.

What crop systems simulation tools are employed?

To date, HarvestChoice has employed the DSSAT (Version 4.5) and APSIM (Version 7.1) suite of models, but has also applied the ORYZA2000 rice model (lead by IRRI) and is engaged in harmonizing input datasets across different crop modeling platforms. HarvestChoice sees the further development of crop model platforms to support strategic investment and policy analysis as a core area of its on-going work. Partners in that process include the Universities of Georgia and Florida, IFDC, and CSIRO as well as CGIAR centers. Two key areas of focus will be tighter coupling of pest and disease and crop models and the assessing the potential impacts of improved technologies and practices on profitability and resource sustainability over time.

How will HarvestChoice provide outreach for its cropping system modeling capacity?

There are currently three main strategies by which we are sharing Simplr technologies and striving to enhance user capacity to leverage these outputs. First, we are providing access to input Simplr datasets and analysis outputs via the HarvestChoice website.  Second is we are providing access to a purpose built crop modeling tool Harvest Toolkit in partnership with IFDC and ICASA and, of increasing emphasis, through the development and delivery of training materials and courses with specialized partners. A particular goal is to promote the adoption and application of cropping system modeling tools by African scientists and analysts involved in relevant decision support processes in the region.

Map of the distribution of 27 generic soils in sub-Saharan Africa at 5 arc-minute grid cells. Each cell shows the predominant soil.

Ecosystems models are currently used in various types of impact assessment studies at different temporal and spatial scales, and their results often implicate policy and management decisions at multiple levels (e.g., from micro-level farm management to macro-level natural resource management). Soil information is often a key input to the models, yet it is difficult to obtain extensive, quantitative, and geo-referenced soil property data for the areas (or regions) of interest. Global soil databases do exist (e.g., Harmonized World Soil Database (HWSD) by IIASA/FAO/ISSCAS/ISRIC/JRC, 2009), but they do not provide all the required information for the models at a specific site. Whereas existing global soil profile databases  (e.g., WISE by ISRIC, 2002) do not extensively cover large areas in the developing world.

To overcome the limitation of location-specific soil profile data for crop modeling applications, we generated a set of generic soil profiles based on three criteria that crop models are most responsive to: texture, rooting depth, and organic carbon content. By classifying three levels for each category and setting their boundary conditions (Box 1), 27 soil profiles, HC27, were generated in formats compatible with DSSAT and APSIM. The boundary conditions were defined based on soil profiles recorded in Sub-Saharan Africa (SSA) and are thus subject to further adjustments in other regions where extensive soil profiles are available.

How to Use the 27 Soil Profiles

There are multiple ways toutilize HC27 in crop modeling applications. First, for a given site, users can choose which one best matches the soils found in the area. It would be difficult to estimate values of all soil properties that crop models require, but selecting one out of 27 by answering three multiple-choice questions would be relatively straightforward to users with some level of agronomic knowledge. Secondly, a model can be run with all 27 soil profiles for a given site to create a set of simulation results, then narrowed down to the most relevant one later as more site-specific information becomes available. Finally, based on additional information from other databases, a new kind of soil map that locates 27 soils can be generated and used in large-scale applications.

For example, Figure 1 is an example of a soil map indicating which one of 27 soils is predominantly distributed where. This data layer was generated by 1) overlaying 10-km grids on HWSD v1.1; 2) computing zonal statistics of soils on grids; 3) determining the predominant soil in each grid cell; and 4) matching the soil with one of 27 soils based on the predominant soil’s texture, organic carbon content of top soil, and available water content classification (as the proxy of rooting depth) from the HWSD.

Is HC27 the Ultimate Soil Database?

The HC27 does not replace existing high resolution soil mapping efforts nor duplicate site-specific soil measurements. Instead, this approach tries to address the need for a reasonably representative meso-scale soil profile database to be used in certain types of spatial crop systems modeling applications. For example, in 2009, HC27 was used in regional/global scale climate change impact assessment studies by IFPRI. However, due to the nature of “generic” characteristics, there will be applications for which the use of HC27 is not desirable, namely where detailed soil property dynamics beyond the three criteria are emphasized.

Preview

Download

The latest version of HC27 can be downloaded at:
https://hc.box.net/shared/b5uy8vrrg1

References

• FAO/IIASA/ISRIC/ISSCAS/JRC, 2009. Harmonized World Soil Database (version 1.1) http://bit.ly/cDZLyq
• ADB, 2009. Building Climate Resilience in the Agricultural Sector of Asia and the Pacific. http://bit.ly/bV6JtV
• IFPRI, 2009. Climate Change Impact on Agriculture and Costs of Adaptation. http://bit.ly/aAL4tI

Assessing the Impact of Rainfall Variability on Maize Yield Potential in Sub-Saharan Africa

Water availability is the most critical factor for sustaining crop productivity in rainfed agriculture. Even if a drought-tolerant trait is introduced, water isn’t available to crops when there is no water in the soil. Rainfall variability from season to season greatly affects soil water availability to crops, and thus pose crop production risks. Ideally, crop cultivations should be situated in areas with high rainfall with low variability; however, subsistence farming can be found in a wide range of environmental conditions—from very suitable to marginal lands. Variability in seasonal rainfall (i.e., the accumulated amount of rainfall from the planting to the harvest of a crop) is higher in the areas with smaller amount of rainfall (Figure 1). Overall, about 37% of maize-growing areas in the Sub-Saharan Africa (SSA) region are located in the areas with the coefficient of variation (C.V.) of seasonal rainfall higher than 0.2 (Figure 2).

Figure 2. Map of sub-Saharan Africa showing the coefficient of variation (C.V.) of seasonal rainfall at major maize growing areas during 1955-2004.

Simulated Crop Yield Potential

One of the common crop modeling applications is the estimation of potential productivity under the different assumptions of biophysical constraints. To assess the seasonal rainfall variability impact on crop production and its variability, the site-specific yield potential of rainfed maize (with no nitrogen constraint) was simulated at grid-cells (10 km resolution) for 1955-2004.

When aggregated over time, the overall trend showed that more rainfall correlates with higher yield and lower yield variability (Figure 3). However, it was also noted that there is a wide range of yield even from the same amount of seasonal rainfall. For example, about 500 mm of rainfall during the crop growing season yielded about 1.5 t/ha in East Shewa, Ethiopia, whereas it was almost 10 t/ha in Kroonstad, South Africa. This is partly due to the different soil characteristics (e.g., clayey soils hold water more and longer than sandy ones), but mostly the influence of location-specific rainfall patterns throughout the season on a daily basis. That is, how the same 500 mm of rainfall is distributed is an important factor. Different crop growth stages have different sensitivity levels of development to water stress; low availability of water during a critical stage can have a higher impact on yield than others. Hence, to efficiently use water, it is important to understand “when” crop needs water the most.

Water Productivity

A simple measure of estimating a crop’s water use efficiency is water productivity, which is the crop yield divided by the amount of water applied (i.e., crop-per-drop). If the crop is highly water-efficient, there would be a positive relationship between water (rainfall, for the rainfed systems) and water productivity. Figure 5 shows the water productivity trend increases when the seasonal rainfall amount is limited (e.g., below 400 mm), suggesting that investment in water management will be relatively more efficient in those areas. However, as discussed in the previous section, the total amount of water input does not always correlate with higher yield due to soil characteristics and rainfall patterns. Intensive rain runs the water off from the field. Figure 5 also shows the slightly decreasing water productivity trend as the seasonal rainfall amount increases.

Figure 6. Map of sub-Saharan Africa showing the C.V. of maize yield potential at major growing areas during 1955-2004.

Suitability Assessment

The location-specific yield potential and variability can be used to compare the suitability of cultivation across the region. By scatter-plotting the yield potential against its variability for each grid-cell of a major maize growing area, Figure 4 provides a quick measure of which maize growing areas are more suitable for rainfed cultivation. Most high rainfall areas (those with a larger circle) have relatively high yield with low variability–and are thus well suited for rainfed cultivation. For the unsuitable areas with high variability of yield, yield potential is almost always low; Figure 6 shows where the marginal areas are located on the map.

Figure 7 shows the cumulative distribution of the C.V. of yield potential when compared across countries  so it is clear which country is better off overall in terms of the proportional areas with the given range of yield variability. For example, comparing Ghana with Kenya at the C.V. of 0.1, the figure shows about 95% of Ghana’s major maize growing areas have the C.V. of potential yield less than 0.1, whereas it’s only about 15% in Kenya. Comparatively, of all the countries in SSA, Somalia is in the most marginal condition.

When It’s Not Just Water

Beyond water, crop systems models can assess the impact of multiple biotic/abiotic constraints and management practices on crop growth and yield in a systemic way. For example, Figure 8 shows location-specific changes in the relationship between yield potential and variability under a set of scenarios on crop, cultivar, fertilizer, and irrigation. Starting from the point A, the medium-maturity maize without fertilizer or irrigation yielded about 2t/ha with 0.75 of C.V. in this area. Adding fertilizer (B), irrigation (C), or fertilizer and irrigation (D) can bring various changes to the average yield and variability. Especially of note, the  fertilizer and irrigation combination can more than double the average yield with 50% less variability. In addition, as the rainfall is limited, the cultivar can be changed to a shorter maturity (E); a slight increase in yield (from D) with 14% less variability. Similarly, adding fertilizer (F), irrigation (G), or both (H) can change different degrees of impact on yield and variability. Another strategy could be switching to a more drought-tolerant crops, such as sorghum (I). Compared to the same conditions of maize (A and E), sorghum can produce a yield with about 25% lower variability, although the yield level is different. When irrigated (J), sorghum’s yield variability was even lower than any other simulated cases of maize.

Conclusion

Rainfall variability is an important characteristic of climate in SSA that imposes crop production risks, especially on rainfed subsistence cultivation systems on marginal land. The crop systems model can help by providing information to assess the extent of cross-regional risk and potential impact, and better target interventions.