Salinity intrusion prediction using remote sensing and machine learning in data-limited regions: A case study in Vietnam's Mekong Delta

Abstract

With population growth, the demand for land resources is expected to increase significantly in the coming decades. Maintaining the integrity of soil distribution requires a remarkable amount of work to deal with agricultural extension. Salinity intrusion monitoring is a crucial process, which directly affects sustainable development, especially in areas affected by global warming and in coastal zones. In recent years, various studies have used the soil-water salinity data to evaluate the spatiotemporal increase in salinity intrusion. This study aims to establish a novel framework for monitoring salinity intrusion using remote sensing and machine learning. It focuses on the salinity intrusion in soil, which affects water availability, food security, human health, etc. Numerous algorithms have been implemented to find the best solution for this issue, including Xgboost (XGR), Gaussian processes, support vector regression, deep neural networks, and the grasshopper optimization algorithm (GOA). A total of 143 samples collected from 2016 to 2020 at 39 measurement stations were divided into two sets: 70% training and 30% testing. Thirty-one independent variables were used to develop the model. Vietnam's Mekong Delta, where the salinity intrusion problem is becoming increasingly serious due to global warming and demographics, was selected as the study area. Each of the proposed models was compared and evaluated by applying various statistical indices such as the root mean square error, coefficient of determination (R2), and mean absolute error. The results show that the prediction model was built successfully by wielding data from the implemented salinity measurement stations, and the XGR-GOA model was better than the other models (R2 = 0.86, RMSE = 0.076, and MAE = 0.065). This finding demonstrates the feasibility of estimating and monitoring salinity intrusion in data-limited regions by integrating optical satellite images and machine learning, which are easily and cost-effectively obtainable. The proposed conceptual methodology in our study is novel and provides additional useful information for the monitoring and management of salinity intrusion not only in Vietnam's Mekong Delta, but also in other sites that have similar natural and anthropological conditions.