Multilayer perceptron neural network for downscaling rainfall in arid region: A case study of Baluchistan, Pakistan

Abstract

Downscaling rainfall in an arid region is much challenging compared to wet region due to erratic and infrequent behaviour of rainfall in the arid region. The complexity is further aggregated due to scarcity of data in such regions. A multilayer perceptron (MLP) neural network has been proposed in the present study for the downscaling of rainfall in the data scarce arid region of Baluchistan province of Pakistan, which is considered as one of the most vulnerable areas of Pakistan to climate change. The National Center for Environmental Prediction (NCEP) reanalysis datasets from 20 grid points surrounding the study area were used to select the predictors using principal component analysis. Monthly rainfall data for the time periods 1961-1990 and 1991-2001 were used for the calibration and validation of the MLP model, respectively. The performance of the model was assessed using various statistics including mean, variance, quartiles, root mean square error (RMSE), mean bias error (MBE), coefficient of determination (R 2)and Nash-Sutcliffe efficiency (NSE). Comparisons of mean monthly time series of observed and downscaled rainfall showed good agreement during both calibration and validation periods, while the downscaling model was found to underpredict rainfall variance in both periods. Other statistical parameters also revealed good agreement between observed and downscaled rainfall during both calibration and validation p e r i o d si nm o s to ft h es t a t i o n s .

Statistical downscaling of global climate model outputs to monthly precipitation via extreme learning machine: A case study

“…, Ahmed et al . , Okkan and Kirdemir , utilized several artificial intelligence models (e.g., support vector regression, ANN, genetic programing) for downscaling GCM outputs. Hashmi et al .…”

“…Ahmed et al . applied ANN model for downscaling rainfall and they found the best R 2 value as 0.63. It is clear from the Table that the ELM generally provided accurate and reliable results in predicting precipitation with respect to R 2 criteria.…”

“…Sachindra et al [50] applied SVM for predicting catchment streamflow and they found R 2 of 0.65 in the validation phase. Ahmed et al [48] applied ANN model for downscaling rainfall and they found the best R 2 value as 0.63. It is clear from the Table 4 that the ELM generally provided accurate and reliable results in predicting precipitation with respect to R 2 criteria.…”

Absolute homogeneity assessment of precipitation time series in an arid region of Pakistan

“…The arid region is considered the most vulnerable area of Pakistan to climate change. It has been projected that droughts and water scarcity will continuously increase in the area throughout the 21st century, which will severely affect the economy and livelihood of people if adaptation measures are not taken (Ahmed et al, 2015(Ahmed et al, , 2016. It is necessary to model and analyze meteorological variables (i.e., rainfall, temperature, evaporation, etc.)…”

Seasonal prediction of high‐resolution temperature at 2‐m height over Mongolia during boreal winter using both coupled general circulation model and artificial neural network

“…Details of the MLR analysis of TI and SHI on the BKI and ESI, referred to as TI MLR and SHI MLR , respectively, are shown in Table 1 variables into local-scale variables in some studies (e.g., Chadwick et al, 2011), is also used in this study. This is one of the most popular methods and known as a flexible type of neural network (Ahmed et al, 2015). This is one of the most popular methods and known as a flexible type of neural network (Ahmed et al, 2015).…”