What type of study is this?

This is a Quantitative Study study.

October 8, 2025Open Access

A Novel Method for Salinization Estimation in the Yellow River Delta Based on XGBoost and Three-Dimensional Feature Space Models

Key Points

The use of a three-dimensional feature space model significantly improves soil salinization estimation accuracy.
The highest accuracy for the model was achieved with a coefficient of determination of 0.921.
Salinized areas were primarily located in northern regions, with spatial differences observed in levels of salinization.
Sentinel-2 data and Bayesian optimization enhance the model's effectiveness for monitoring soil conditions.

Abstract

Abstract Soil salinization in the Yellow River Delta (YRD) is becoming more and more serious, and has become one of the key factors restricting the sustainable development of the region. Compared with two-dimensional feature space, three-dimensional feature space is superior in monitoring effectiveness and can more effectively utilize the multispectral information of remote sensing images. At present, the monitoring of soil salinization in the Yellow River Delta based on 3D feature space is not sufficiently reported. In this study, based on the Sentinel-2 remote sensing image data, 24 indices (vegetation index, salinity index and other indices) of different categories were calculated, and the Bayesian optimized XGBoost model was used to optimize the features of the indices for each category, construct the three-dimensional feature space model, and carry out the one-dimensional linear regression analysis to determine the optimal model for salinity estimation. The results showed that: (1) among the vegetation feature coefficients, Normalized Difference Red Edge Index, Extended Ratio Vegetation Index and Normalized Difference Vegetation Index were the most important. Among the salinity feature covariates, the salinity indices SI5, S3 and SI7 had the largest contribution; (2) the salinity monitoring model constructed based on the NDRE-Albedo-SI7 feature space had the highest accuracy, with the coefficient of determination was 0.921; (3) the extremely salinized and heavily salinized areas were mainly distributed in the northern coastal area of Hekou District, Kenli County and the eastern part of Dongying District, while the mildly salinized and non-salinized areas were mainly distributed in Lijin County, the western part of Kenli District, the western part of Dongying District and the southwestern part of Guangrao County. In this study, a three-dimensional feature space model was innovatively applied to monitor soil salinization in the Yellow River Delta by introducing the Sentinel-2-specific red-edge index and using the Bayesian-optimized XGBoost model for feature selection, which significantly improved the accuracy and reliability of the model. The results of the study provide a scientific basis for the monitoring and management of soil salinization in the Yellow River Delta.

Ask AI

Helpful

Bookmark

View Full Paper