Design and implementation of a hybrid MLP-FFA model for soil salinity prediction

Abstract

In this research, the ability of a hybrid model, which is integrating the firefly algorithm with the multilayer perceptron (MLP-FFA), is investigated for prediction of soil salinity (EC) using remote sensing and topography data in Miandoab city, northwest of Iran. Some important variables including the salinity ratio (SR), slope, ratio vegetation index (RVI), normalized differential vegetation index (NDVI), gypsum index, soil-adjusted vegetation index (SAVI), elevation and normalized difference salinity index (NDSI) are used as inputs of the model, while the EC is the output variable. The data consist of 80 soil samples. The estimates provided by the MLP-FFA model are compared with the standalone MLP model based on the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), Nash–Sutcliffe coefficient (ENS), Willmott’s Index of agreement (Wi) and percentage peak deviation (Pdv). The obtained results showed more precise estimations of the MLP-FFA than the MLP model with R2, MAE, RMSE, ENS, Wi and Pdv values of 0.641, 1.195, 1.672, 0.626, − 0.247 and 4.603, respectively, in training phase, while in the testing phase these values were equal to 0.662, 0.439, 0.538, 0.566, − 0.15 and − 32.013, respectively. Therefore, the results specified that the developed MLP-FFA model can be selected as an efficient technique over the MLP model for accurate prediction of EC.