The potential of irrigation for cereals production in Sub–Saharan Africa: A machine learning application for emulating crop growth at large scale

The low percentage of land equipped for irrigation and the scarce land agricultural productivity render Africa an ideal target for irrigation projects. These have the potentials of increasing and stabilizing yields, thus contributing to food security and poverty reduction. The present paper investigated the potentials of irrigation in the whole Sub-Saharan region with the aim of individuating areas where intervention should be prioritized. The analysis is conducted via a mix of simulations through the crop model DSSAT and machine learning, namely XGBoost. Yield differentials for four cereals, millet, maize, sorghum and rice, are computed together with water requirements under a low fertilization scenario that reflects current agricultural practices in the region. By crossing the resulting water productivity levels and run-off values, most promising areas of intervention are individuated. The average increase in yields varies between roughly 14% and 17%, depending on crop, but these figures may be drastically improved if combined with an intensification of nutrient ans organic matter provision. • We quantify the potential of irrigation for cereals production in Sub-Saharan Africa. • A mix of crop modelling and machine learning is used to predict water potentials with a 1 km of resolution. • Water requirements for optimal irrigation and water productivity are computed. • Locations (clusters of points) to be prioritized for two types of irrigation projects are individuated.