Testing the GAEZ agronomic model in the fields:Evidence from Uganda

The Global Agroecological Zones (GAEZ) model developed by FAO and IIASA provides granular crop-specific expected yields worldwide. An increasing number of papers in economics are using this dataset for a number of different purposes, such as deriving an exogenous source of variations in institutions or to parametrize computable general equilibrium models of trade in agricultural goods. However, as pointed out by the GAEZ creators, its results †should be treated in a conservative manner and at appropriate aggregation levels, which are commensurate with the resolution of the basic data†. In this paper, I †test†the GAEZ predicted yields by comparing them with the ones measured through agricultural surveys †in the fields†. In particular, I use data from two different agricultural surveys in Uganda, where farmers typically grow a multitude of crops. In both instances, I find that GAEZ predictions often correlate negatively with yields, and that farmers’ crop choices are more responsive to survey-based yields than to GAEZ’s predictions. These findings suggest that, at least in the context of Uganda, predicted GAEZ yields fail to make reliable predictions at the granular level.