For a more thorough discussion on methodology and on SPAM’s mathmatical model, download fulltext PDF, or browse documentation for more.

Overview

1. We start with the administrative (geopolitical) units for which we have been able to obtain production statistics. These may typically be national or sub-national administrative regions such as states, districts, or counties

2. We reinterpret the already classified land cover imagery into crop land and non crop land.

3. We integrate crop-specific suitability information based on local climate and soil conditions, which provides how MUCH cropland exists at the pixel level.

4, The model utilizes all these input data and applies a cross entropy approach to obtain the final estimation of crop distribution.

Compare the starting point (upper left) with model results (lower left) and see the tremendous improvement in revealing spatial heterogeneity within the administrative units

Spam Reveals

Drawing from a collection of spatially relevant input data and crop production statistics, the SPAM model aggregates data to produce desegregated results, uniform at the pixel level.  The data reveals differences in yield based on technology practices and emergent patterns between geography and agriculture.

SPAM Inputs

SPAM relies on a collection of relevant spatially explicit input data, including crop production statistics, land cover and land use data, biophysical crop “suitability” assessments, population density, distance to urban (i.e. market) centers, as well as any prior knowledge about the spatial distribution of specific crops or crop systems.

Crop production statistics

While crop production data at the national level are reported by Food and Agriculture Organization of United Nations (FAO), similar data within sub-national boundaries are rarely available on a global scale. To satisfy an increasing necessity to have better crop production and land use data to support their respective programs, FAO, IFPRI (International Food Policy Research Institute) and SAGE (Center for Sustainability and the Global Environment, University of Wisconsin-Madison) started, in 2002, an informal collaborative consortium titled Agro-MAPS (Mapping of Agricultural Production Systems).

The goal of Agro-MAPS is to compile a consistent global spatial database based upon selected sub-national agricultural statistics. Agro-MAPS holds not only tabular statistical data but also links to maps of administrative districts (http://www.fao.org/landandwater/agll/agromaps/interactive/index.jsp). As input into SPAM, we started with Agro-MAPS data, and made a great effort to add more sub-national data, paying particular attention to developing countries in Africa, Latin America, and Asia. We established a network of data resources from various local subnational offices in many countries throughout the world. Currently most of the data used are from World Food Programme (WFP) crop and food supply assessment mission surveys, agricultural performance surveys, national bureaus of statistics, regional agricultural centers, ministries of agriculture, rural and extension services, regional NGOs, agricultural censuses, ministries of the environment, and water resource groups.

Taking advantage of these national partners and the institutes of the CGIAR (http://www.cgiar.org), we were able to compile a robust database with crop production data for more crops, longer time series, and smaller administrative units than any single collection of subnational production data currently available. These data were compiled in HarvestChoice’s new information database system which allows for the collection of time series data as well as data at varying administrative levels for all countries of the world. This database now serves as the feeder of crop production data into SPAM.

Below is a table showing a regional overview of data available in SPAM 2000 by administrative level:

Land cover / Land use

Satellite-based land cover datasets serve to provide detailed spatial information on cropland extent – distinguishing cropland from other forms of land cover such as forest, grassland, and water bodies and, therefore, delineating the geographical extents within which crop production must be allocated. The reliability of the land cover data in terms of measuring cropland can have significant implications for the overall reliability of the allocation.

At the time that the model was finalized for SPAM v3 there were two global land cover datasets for the year 2000, BU-MODIS Land Cover, and JRC’s GLC2000 and one for 1992/93 (USGS’s GLCC) that were used as input into the model. Each dataset has its own pros and cons so based on an evaluation of the three for Africa, we chose to use aggregates of all three. The merger of three different satellite derived products allows the model to identify, and prioritize, areas that were classified as cultivated in any of the three input datasets. The shortcomings of individual datasets in certain areas can thus potentially be overcome by the strengths of the others. The source data were all at a resolution of 30 arc seconds (approx 1x1km at the equator) but were aggregated to a 5 minute (approximately 10x10km at the equator) resolution for input to the SPAM allocation. More information on the various land cover datasets used for SPAM and other HarvestChoice projects can be found on the HarvestChoice site.

Information on actual land use is even more difficult to find than that on agricultural land cover. One key factor to successfully allocating production statistics is to know what areas are irrigated. The Land and Water Division of FAO and the University of Frankfurt are working together to develop the Global Map of Irrigated Areas (GMIA) which provides GIS coverage of areas equipped for irrigation at a 5 minute resolution. Using these data we were able to identify areas that were most likely irrigated and thus allocated the irrigated area and production to these locations.

Crop suitability

Different crops have different thermal, moisture, and soil requirements, particularly under rainfed conditions. FAO, in collaboration with the International Institute for Applied Systems Analysis (IIASA), has developed the agro-ecological zones (AEZ) methodology based on an evaluation of existing land resources and biophysical limitations and potentials for specific crops (FAO/IIASA). This methodology provides maximum potential and biophysically attainable crop yields and suitable crop areas. For SPAM we utilized three production system types from the FAO/IIASA suitability datasets: Irrigated; rainfed – high input/commercial; rainfed – low input/subsistence. For each crop by the three input levels, we define our suitable land as the sum of the four suitability classes in the AEZ model: very suitable, suitable, moderately suitable, and marginally suitable. These data were made available at a 5 minute resolution. Maps of crop suitability for the 20 crops allocated at the three input levels are available through theHarvestChoice data portal.

Input Systems– Three Levels

Each crop can grow in any of 3 input systems:

• Irrigated (I)
• Rainfed, high-input/commercial (H)
• Rainfed, low-input/subsistence (L)

The definition of these input systems (/management levels) more or less follows FAO/IIASA’s GAEZ project (http://www.iiasa.ac.at/Research/LUC/GAEZ/index.htm) since we use those suitability surfaces.

The rainfed, high input/commercial crop system is rainfed-based agriculture, but uses high-yield varieties and some animal traction and mechanization. It at least applies some fertilizer, chemical pest, disease or weed controls.

The rainfed, low-input/subsistence crop system refers to rainfed crop production which uses traditional varieties and mainly manual labor without (or with little) application of nutrients or chemicals for pest and disease control.

In contrast, irrigated crop system refers to the crop area equipped with either full or partial control irrigation. Normally the crop production on the irrigated fields uses high level of inputs such as modern varieties and fertilizer as well as advanced management such as soil/water conservation measures.

Example: Rice Harvested Area for Three Input Levels

Irrigated

Rainfed, high-input/commercial

Rainfed, low-input/subsistence

Area harvested is allocated to at least one of these systems according to further information available or expert judgement. Physical area, production and yield are also calculated for each input system and crop allocation results follow the same subdivision. Final results are presented in disaggregated (per system) and aggregated form (sum of all systems).

Additional Data sets:

Population density

Accessibility

SPAM Outputs

SPAM outputs are reported in .dbf tables referenced to 5 arc-minute (approximately 10 km x 10 km at equator) grid cells for the following variables for each crop:

• Harvested Area (hectares)
• Physical Area (hectares)
• Production (Tons)
• Yield (Kilograms / hectare)

Harvested Area

Amount of area where a specific crop in a given input system is being cultivated and harvested in the year 2000 (average 1999 – 2001). If a crop is being harvested more than once, the harvested area increases as well. Data for harvested area is collected at national and sub-national level for all SPAM crops. Harvested area is expressed in hectare.

Physical area

Area where a specific crop in a given input system is being cultivated in the year 2000. Multiple harvesting seasons in one year are not taken account of. Intercropping or successive planting of different crops on the same physical area do not increase the physical area. Cropping intensity factors convert harvested area into physical area.
Physical area is expressed in hectare.

Production

Production of a specific crop in the given input system. Production is calculated by multiplying harvested area by yield. Since harvested area and yield refer to the year 2000, so does production. Production is expressed in metric tonns.

Yield

The yield of a crop in the given input system is the amount of production per harvested area.Yields are given for the year 2000. Yield data is collected at national and sub-national level for all SPAM crops. Yield is expressed in kilogramme per hectare.

Note

Harvested area, yield and cropping intensity factors are also input parameters (at national and sometimes sub-national level). Physical area and production are calculated at those levels as well. SPAM processing further allocates physical area and calculates harvested area, yield and production for each cell.