Two goals underlie most discussions of crop yield gaps (Van Ittersum et al., 2013). The first is to measure the size of the yield gap, defined as the difference between yield potential (Yp) and average yields, in order to identify the potential scope for raising average yields via management changes. The second is to identify the key causes of the yield gap, in order to prioritize efforts in extension, research, and policy to raise land and labor productivity.
A fundamental challenge in pursuit of either of these goals is the considerable spatial and temporal heterogeneity of agricultural landscapes. In the measurement of yield gaps, for example, actual yields are often reported for administrative units that span hundreds or thousands of fields. Yield potential, meanwhile, is most readily estimated for individual fields, using either agronomic trials or well-tested crop simulation models (Lobell et al., 2009). How should the measurements at these two different spatial scales be compared when computing a yield gap? Some studies ignore the scale mismatch, implicitly assuming that point-level estimates of Yp are a good proxy for average Yp across the spatial domain of the reported average yield. Other studies attempt to estimate Yp at multiple points within the domain and then take an average, a sensible approach provided that data of sufficient quality exist to estimate Yp at multiple points.
Similarly, studies to understand causes of the yield gap may reasonably start by evaluating yield responses to different