Partial least-squares regression for soil salinity mapping in Bangladesh

Abstract

Estimating the salinity of the soil along the coast of south-western Bangladesh is the focus of this study. Thirteen soil salinity indicators were computed using the Landsat OLI images, and 241 soil salinity samples were gathered from secondary sources. In this research, partial least-squares regression was used to estimate soil salinity. Soil salinity zones were then divided into five classes based on Jenks natural breaks. There is a strong relationship between soil salinity and several indices, including B2 (Blue band), B5 (Near-Infrared band), SI_1 (Salinity index 1), and SI_2 (Salinity index 2). The correlation between B2 and soil salinity is particularly high, hovering around 0.92. The majority of other factors, however, have a negative or barely positive correlation of −0.260 to 0.032. Very high and high levels of salinity are estimated to cover 29.02% and 13.45% of the study area, respectively, based on the soil salinity map. Finally, cross-validation has been used to see how well the model works to predict soil salinity for a specified weight. Iteration has been done three times in order to assess the model's performance on various subsets of data and produce a more accurate prediction. In the third iteration, PRESS, RSS, and Q2cum values were 1.844826, 3.354687, and 0.9705131, correspondingly, which indicates a higher predictive accuracy. The findings of the study will be useful as a tool for policymakers in the south-western coastal area of Bangladesh who are responsible for integrated coastal zone management.