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

Due to its geographical location and its poor economic conditions, Africa  is the continent most exposed to the adverse consequences of climate change, particularly on agriculture. The very low percentage of land equipped for irrigation, 3.5% in Sub-Saharan Africa, is another element of concern, sensibly reducing the ability to mitigate the likely productivity losses caused by increasing climate variability and extreme events. Fostering irrigation in Africa is therefore a priority, but due to a limited amount of resources, both in economic and physical (e.g. harvestable water) terms, irrigation projects have to be planned carefully and appropriate locations should be prioritized. The present paper tries to assess the potentials of irrigation in Sub-Saharan Africa and to individuate the locations to be prioritized. The analysis focuses on four cereals, maize, millet, sorghum and wheat, among the most common staples in the region, and relies on a mix of crop modelling (DSSAT) and machine learning (XGBoost) to draw its conclusions. Specifically, for all four crops, crop simulations under rain-fed conditions and optimal irrigation, with DSSAT adding water every time a need for it is observed, are performed on a sample of all Sub-Saharan agricultural plots. Yields differentials and water requirements for optimal irrigation are then computed. Subsequently, yields and water requirements are predicted for all remaining agricultural locations through machine learning, using as explanatory variables the same inputs, soil characteristics, management practices and weather variables, required by DSSAT. Water productivity, defined as the ratio of yields differentials over water requirements for irrigation, is finally computed to individuate the locations where irrigation projects would be most beneficial. By further relying on a continental map of run-off values, we individuate two types of priority locations: areas where simple water capture and storage devices are viable and areas where more complex systems are necessary. The paper points out the importance of irrigation in Sub-Saharan Africa, showing significant gains in yields, up to 100% compared to rain-fed conditions. It also finds high potentials for water capture and storage devices in the south-eastern part of the continent and in South Africa, while the western part and the stripe bordering the Sahara desert would have to rely on more complex irrigation systems.