Rice Irrigation Water Use Efficiency under Drivers of Climate Change: Identifying the Key Drivers Using Least Absolute Shrinkage and Selection Operator (LASSO) Regression

This paper asks which weather and soil-water conditions make rice irrigation at the Rwangingo marshland (Gatsibo, Rwanda) more or less efficient from day to day. It uses six years (2017–2022) of daily NASA POWER data — rainfall, max/min temperature, humidity, wind, UV radiation, surface wetness and profile soil moisture — and a daily irrigation efficiency index built from the scheme's crop water requirement and canal releases. LASSO regression ranks the drivers; ridge, elastic-net, bootstrap and a hold-out on 2021–2022 check the result. Findings: soil-water status (profile soil moisture and surface wetness) dominates, explaining most of the variation; heavy rainfall is the main atmospheric cause of lost efficiency; temperature, wind and UV have small effects; humidity adds nothing once soil water is accounted for. The daily index is not related to seasonal yield-per-water (IWUE, 0.59–0.84 kg m⁻³), which is governed by how much water the scheme diverts — Season A diverts more and is less efficient. Recommendations: trigger releases on field wetness rather than a fixed 7-day rotation, skip rotations when rain keeps the marshland wet, close intakes ahead of heavy rain, and keep a simple daily water and yield log so the analysis can be repeated.