The Yellow River Delta (YRD) has the world's highest land formation rate. However, soil salinization has caused severe land degradation in the region. Understanding the distribution of soil salinity and its variation is essential for saline soil management. This study combined soil salinity sampling data, remote sensing imagery, and layers of geographic environmental factors. Three different models, including Ordinary Kriging (OK), Geographically Weighted Regression (GWR), and Bayesian Maximum Entropy (BME), were used and compared to predict the soil salinity of two soil layers in 2006 and 2022. Then the spatial distribution characteristics and development trends of soil salinity in the YRD were analyzed. The results indicated that (1) The BME model is an optimal salinity prediction model that integrates soft data from multiple sources to perform nonlinear estimates. Compared to the OK and GWR models, the RMSE was reduced by up to 25% and 13%, respectively, and the greatest improvement in R² was increased from 0.0534 and 0.2718 to 0.5569, respectively. (2) Soil salinity in the YRD shows a spatially increasing trend from the southwestern inland to the northeastern coast. Over the past 16 years, the salinization pattern has become more complex: soil salinization has been mitigated in the central and southern regions, with the average salinity decreasing from 1.03% to 0.86% and the salinization rate decreasing from 99% to 88%; and it has significantly intensified in the northern part of the study area near the old Yellow River channel and the Gubei Reservoir, and in some scattered inland areas. Continuous water and sediment regulation in the Yellow River basin and ecological management of natural reserves can generally alleviate soil salinization, but the risk of soil salinization can be increased by seawater erosion, inappropriate land use, and resource exploitation.

中文翻译：

---

2006年与2022年黄河三角洲土壤盐分分布及其变化检测


黄河三角洲（YRD）拥有世界上最高的土地形成率。然而，土壤盐碱化导致该地区土地严重退化。了解土壤盐分的分布及其变化对于盐渍土管理至关重要。这项研究结合了土壤盐分采样数据、遥感图像和地理环境因素层。采用普通克里金法（OK）、地理加权回归（GWR）和贝叶斯最大熵（BME）三种不同模型对2006年和2022年两土层土壤盐分含量进行了预测，并进行了比较。分析了长三角地区土壤盐分的发展趋势。结果表明：（1）BME模型是一种集成多源软数据进行非线性估计的最优盐度预测模型。与OK和GWR模型相比，RMSE分别降低了25%和13%， R ²的最大改进分别从0.0534和0.2718增加到0.5569。 (2)长三角地区土壤盐分在空间上呈现出由西南内陆向东北沿海递增的趋势。 16年来，盐碱化格局更加复杂：中南部地区土壤盐碱化有所缓解，平均盐分由1.03%下降到0.86%，盐化率由99%下降到88%；研究区北部黄河故道、古北水库附近以及部分零星内陆地区有明显加剧。 黄河流域的持续调水调沙和自然保护区的生态管理总体上可以缓解土壤盐渍化，但海水侵蚀、土地利用不当和资源开发等也会增加土壤盐渍化的风险。