Artificial neural network models for prediction of standardized precipitation index in central Mexico

Some of the effects climate change may be related to a in patterns rainfall intensity or scarcity. So, humanity is facing environmental challenges due an increase occurrence droughts. Forecasting droughts based on cumulative influence could greatly beneficial for mitigating adverse consequences water-sensitive sectors such as agriculture. Then, predictive models drought indices help assessing water scarcity situations, identification and their severity characterization. In this paper, we tested feasibility Artificial Neural Network data-driven model predicting monthly Standardized Precipitation Index 4 regions (Semi-desert, Highlands, Canyons Mountains) north-central México using variable data from 1965 2004 training simulated period 2005-2014. The best was found Hyperbolic Tangent activation function Adaptive Moment Estimation (Adam) algorithm optimization method. set following architecture: 26-12-1 network with weights 365 trainable parameters. Based analysis scatter plot between predicted observed precipitation values test dataset, Coefficient Determination 0.84 0.88. terms quantitative statistics averaged over set, Model performed very well at four analyzed regions. This verified by all-region average value performance Mean Absolute Error (0.0805), Square (0.0144) (0.8671). nutshell summarize that developed study had good prediction skills stations its drought-related properties region.