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

Rafael Magallanes Quintana; Carlos E. Galván Tejada; Jorge I. Galván Tejada; Santiago de Jesúa Méndez Gallegos; Fidel Blanco Macías; Ricardo David Valdez Cepeda

Ayuda

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

Rafael Magallanes-Quintanar ^[1] ; Carlos Eric Galván-Tejada ^[1] ; Jorge Isaac Galvan-Tejada ^[1] ; Santiago de Jesús Méndez-Gallegos ^[2] ; Fidel Blanco-Macías ^[3] ; Ricardo David Valdez-Cepeda ^[3]
1. [1] Universidad Autónoma de Zacatecas
  
  Universidad Autónoma de Zacatecas
  
  México
2. [2] Colegio de Postgraduados
  
  Colegio de Postgraduados
  
  México
3. [3] Universidad Autónoma Chapingo
  
  Universidad Autónoma Chapingo
  
  México
Mostrar afiliaciones +
Localización: Agrociencia, ISSN 2521-9766, ISSN-e 1405-3195, Vol. 57, Nº. 1, 2023, págs. 11-20
Idioma: inglés
Enlaces
- Texto completo
Resumen
- Globally, an increase in the occurrence of droughts may be due to the effects of climate change on rainfall patterns. Drought prediction based on historical rainfall behavior can be very useful in sectors where water is a critical element, such as rain-fed agriculture. Therefore, drought classification and characterization based on drought index prediction models could aid in mitigating their negative effects in such water-sensitive sectors. The primary goal of this paper was to test a Multilayer Perceptron Artificial Neural Network as a model to forecast the monthly Standardized Precipitation Index in north-central México using rainfall data from the 1964– 2014 period. The model was obtained using the Hyperbolic Tangent activation function and the optimization method from the Adaptive Moment Estimation algorithm. The model used a 26-12-1 network architecture with 4 weights and 365 trainable parameters. The scatter plot analysis between predicted and observed Standardized Precipitation Index values for the test dataset resulted in a Coefficient of Determination between 0.84 and 0.88. Based on quantitative statistics averaged across the test set, the Artificial Network Model performed substantially well in predicting the Standardized Precipitation Index at the four studied regions. This was confirmed by an all-region average value of the performance statistics Mean Absolute Error (0.081), Mean Square Error (0.014) and the Coefficient of Determination (0.867). We conclude that the Artificial Network models developed and tested in this research provided adequate monthly Standardized Precipitation Index skills for the analyzed stations in the studied territory.