Artificial Neural Networks with various Transfer Functions for Modeling Rainfall Patterns in Sokoto, Nigeria

Abstract

Accurate rainfall prediction is crucial for agricultural practices and water resource management  in semi-arid regions. Hence, the objective of this paper is to employ Artificial Neural Networks (ANNs) with various transfer functions for modeling rainfall patterns  in Sokoto, Nigeria. Rainfall data from 1990 to 2019 alongside average temperature, relative humidity, and year were utilized.  Three candidate transfer functions (logsig, purelin, tansig) were compared within a multi-layered ANNs architecture. The performance of each model was evaluated using correlation coefficient  (R) and root mean square error (RMSE). The results revealed that the ANN with the tansig  transfer function achieved the highest R (0.8789) and the lowest RMSE (0.0125), demonstrating  a strong positive relationship between predictions and actual data with minimal errors. This  performance surpassed previously reported ANNs models for rainfall prediction in some  Nigerian northwestern regions. The study concludes that tansig is the most effective transfer function for modeling Sokoto's rainfall patterns using ANNs. This model can be a valuable tool for stakeholders in agriculture and water management to make informed decisions based on predicted rainfall patterns.