Affiliation:
1. Unidad Académica de Ingeniería, Universidad Autónoma de Zacatecas, Ramón López Velarde 801, Zacatecas 98000, Mexico
2. Unidad Académica de Ciencia y Tecnología de la Luz y la Materia, Universidad Autónoma de Zacatecas, Circuito Marie Curie S/N, Parque de Ciencia y Tecnología, Quantum, Zacatecas 98160, Mexico
3. Independent Researcher, Tabasco 86247, Mexico
Abstract
Drought is, among natural hazards, one of the most harmful to humanity. The forecasting of droughts is essential to reduce their impact on the economy, agriculture, tourism and water resource systems. In this study, drought forecast in the central region of the state of Zacatecas, a semi-arid region of Mexico, is explored by means of artificial neural networks (ANNs), forecasting numerical values of three drought indices—the standardized precipitation index (SPI), the standardized precipitation and evapotranspiration index (SPEI) and the reconnaissance drought index (RDI)—in an effort to establish the most suitable index for drought forecasting with ANNs in semi-arid regions. Records of 52 years of monthly precipitation and temperature were used. The indices were calculated in three different time scales: 3, 6 and 12 months. The analyzed models showed great capacity to forecast the values of the three drought indices, and it was found that for the trial set, the RDI was the drought index that was best fitted by the models, with the evaluation metrics R2 (determination coefficient), RMSE (root mean square error), MAE (mean absolute error) and MBE (Mean Bias Error) showing ranges of 0.834–0.988, 0.099–0.402, 0.072–0.343 and 0.017–0.095, respectively. For the validation set, the evaluation metrics were slightly better.