Mapping land suitability for informal, small-scale irrigation development using spatial modelling and machine learning in the Upper East Region, Ghana

Small-scale irrigation has gained momentum in recent years as one of the development priorities in Sub-Saharan Africa. However, farmer-led irrigation is often informal with little support from extension services and a paucity of data on land suitability for irrigation. To map the spatial explicit suitability for dry season small-scale irrigation, we developed a method using an ensemble of boosted regression trees, random forest, and maximum entropy machine learning models for the Upper East Region of Ghana. Both biophysical predictors including surface and groundwater availability, climate, topography and soil properties, and socio-economic predictors which represent demography and infrastructure development such as accessibility to cities and proximity to roads were considered. We assessed that 179,584 ± 49,853 ha is suitable for dry-season small-scale irrigation development when only biophysical variables are considered, and 158,470 ± 27,222 ha when socio-economic variables are included alongside the biophysical predictors, representing 77-89% of the current rainfed-croplands. Travel time to cities, accessibility to small reservoirs, exchangeable sodium percentage, surface runoff that can be potentially stored in reservoirs, population density, proximity to roads, and elevation percentile were the top predictors of small-scale irrigation suitability. These results suggested that the availability of water alone is not a sufficient indicator for area suitability for small-scale irrigation. This calls for strategic road infrastructure development and an improvement in the support to farmers for market accessibility. The suitability for small-scale irrigation should be put in the local context of market availability, demographic indicators, and infrastructure development.