Forecasting residential water demand in Kuwait to 2050 and 2100 using the cellular automata Markov chain model

ABSTRACT Accurate forecasting of residential water demand is essential for sustainable infrastructure planning in arid regions facing water scarcity. Kuwait, among the world's most water-stressed nations, depends almost entirely on desalinated water for domestic supply, yet spatially explicit, long-horizon demand forecasts remain scarce. This study applies the cellular automata Markov chain model (CA-MCM) to forecast residential water demand across Kuwait for 2050 and 2100. Three population density raster maps at 90-metre resolution (1975, 2000, 2025) were reclassified into five ordinal classes and used as inputs for CA-MCM transition analysis in TerrSet v20 (LiberaGIS). Model accuracy was validated by comparing the predicted 2025 class map to actual data, yielding an area under the curve of 0.980 via receiver operating characteristic (ROC) analysis, and a Kstandard of 0.978 and Klocation of 0.989 through the multi-resolution VALIDATE procedure. Residential water demand was then computed at the pixel level using a per-capita consumption rate of 500 litres per person per day. Results show marked spatial intensification and expansion of high-demand zones, with total forecasted demand reaching approximately 3.25 million m3/day by 2050 and 4.53 million m3/day by 2100. These findings offer spatially explicit guidance for long-term water security investment planning in Kuwait.