Analysis of spatially and temporally varying precipitation in Bangladesh

Abstract

Nowadays, time series analysis of rainfall data is a major tool to explore rainfall variability pattern. In this study, rainfall data at 28 stations of Bangladesh has been evaluated using trend test, spatial analysis and Artificial Neural Network (ANN) model. From the Mann-Kendall's (M-K) trend test, it was observed that the value of Sen's slope for the maximum increase of rainfall was 1.813 mm per month in Cox's Bazar and the maximum decrease of rainfall was-0.896 mm per month in Bhola. Rainfall time series were smoothed using wavelet transformation and then wavelet denoised signals were used for fitting Artificial Neural Network (ANN) model. The predictive capability of ANN model was assessed by PBIAS (Percent of bias), index of agreement (d) and R-squared approach. Finally, a forecast for five years (January 2010-December 2014) was carried out to evaluate the predictive capability. The values of normalized root mean square error (NRMSE) ranges between 0.07% and 0.17% for calibration and 0.07%-0.32% for validation which indicate the fitness of the model. The findings derived from the study are to understand the nature of possible rainfall variations in 28 stations of Bangladesh, which will help the hydrologists as well as the policy makers to make decisions.