In this era of big data, sophisticated computer models and near ubiquitous satellite remote sensing estimates, it’s amazing how limited we are in our ability to map cropping systems at a global scale. My colleagues at IFPRI and I investigated this issue in a recent study released in Global Ecology and Biogeography. Also see the IFPRI blog post for a short take on the issue and some additional info on IFPRI’s SPAM model.
The problem is that while developed countries have begun to map their crops, most developing countries simply don’t have the resources to do so. Data is often only available through national or sub-national statistics, which provide limited spatial information (none beyond the administrative boundaries associated with the statistic). This data constraint makes mapping crops in a spatially explicit manner difficult to say the least.
Our analysis looks at differences between the four major models used to map global cropping systems (M3, MIRCA, SPAM and GAEZ) as well as the implications of those differences for subsequent food security analyses. The differences between the models is significant for not only the crop-specific information, but even the extent of cropland delineated by each model. Below is a figure from the analysis showing the differences in the cropland extent:
While it may seem unsurprising that these differences have implications for food security analyses, such as calculating the global yield gap, this has yet to be widely recognized. For example, having calculated the global yield gap using each of the four data sets, we found that the differences in the results were even larger than the estimated yield gap itself (i.e. the average from the four models).
Improving spatial understanding of crop production systems is a vital part of providing accurate analyses of global food security. Recognizing that the differences between models is significant is only the first step. Moving forward, we need to begin to reconcile these differences by evaluating external information against each of the four models to provide an indication of confidence in particular regions for each model, or more generally for each methodology.