Incorporating spatial autocorrelation into soil salinity models:Insights from the Minqin Oasis and its desert–oasis transition zone in Northwest ChinaOA

Remote sensing-based soil salinity inversion serves as a crucial approach for monitoring and assessment in arid regions.However,most existing models rarely account for the spatial autocorrelation(SAC)of soil salinity,which limits both their predictive accuracy and ability to capture spatial patterns.To address this gap,this study investigated the Minqin Oasis and its adjacent desert–oasis transition zone in Northwest China.Based on collected field soil samples and concurrently acquired Landsat-8 OLI remote sensing images in 2024,we incorporated characteristic bands reflecting SAC into conventional spectral indices.Through multi-band combination optimization and comparison of different models''predictive performance,we constructed an optimal soil salinity inversion model for the Minqin Oasis and its adjacent desert–oasis transition zone.The results demonstrated that incorporating SAC of soil salinity markedly improved model performance,with the Gradient Boosting Regression Trees(GBRT)model incorporating SAC(GBRT_SAC)achieving the best accuracy.Compared with the traditional spectral index-based GBRT model,the coefficient of determination(R2)increased by 7.320%,the root mean square error(RMSE)decreased by 20.230%,and the mean absolute percentage error(MAPE)decreased by 121.01%using the GBRT_SAC model.The soil salinity distribution derived from the GBRT_SAC model revealed pronounced spatial heterogeneity,with salinized areas covering approximately 1256.75 km^(2)(36.170%of the total area).Soil salinity was jointly influenced by natural and anthropogenic factors.At the regional scale,soil type and vegetation type emerged as the dominant drivers shaping soil salinity patterns.In contrast,within the oasis interior,soil salinity was primarily driven by groundwater table regulated by irrigation,leading to surface salt accumulation through capillary rise.In the 1000 m desert–oasis transition zone,the explanatory power(q-value)of all environmental factors for spatial variation of soil salinity significantly increased,indicating a sensitive interface where hydrological and aeolian processes interact.Notably,although soil salinity was relatively lower in sandy areas,sand content emerged as the most influential factor in this region(q-value=0.483),effectively serving as a key indicator of the transitional environment.By introducing SAC-based features into soil salinity inversion models,this study provides a robust methodological framework and valuable data to support understanding and management of soil salinization in arid desert–oasis ecotone systems.