A Comparison of the Statistical Downscaling and Long-Short-Term-Memory Artificial Neural Network Models for Long-Term Temperature and Precipitations Forecasting

General circulation models (GCMs) run at regional resolution or a continental scale. Therefore, these results cannot be used directly for local temperatures and precipitation prediction. Downscaling techniques are required to calibrate GCMs. Statistical downscaling (SDSM) the most widely bias correction of However, few studies have compared SDSM with multi-layer perceptron artificial neural networks in studies, indicate that outperform other approaches. This paper investigates an alternative architecture networks, namely long-short-term memory (LSTM), forecast two critical climate variables, temperature precipitation, application five gauging stations Lake Chad Basin. is data scarce area which has been impacted by severe drought, where water resources influenced change recent agricultural expansion. was as benchmark this monthly time–scales weather prediction, using grid GCM output 5 degrees latitude × longitude global grid. Three performance indicators were study, root mean square error (RMSE), measure sensitivity model outliers, absolute percentage (MAPE), estimate overall predictions, well Nash Sutcliffe Efficiency (NSE), standard field forecasting. Results on validation set test LSTM indicated produced better accuracy average SDSM. For forecasting, RMSE MAPE 33.21 mm 24.82% respectively, while 53.32 34.62% respectively. In terms three year ahead minimum forecasts, presents 4.96 degree celsius 27.16%, 8.58 12.83%. maximum forecast, 4.27 11.09 percent, 9.93 12.07%. Given results, may suitable approach downscale simulation models’ output, improve management long-term precipitations forecasting level.