Abstract: Soil salinization has caused soil degradation and the loss of arable land in modern agriculture. It is often required to accurately, timely, and spatially explicitly monitor the soil salinization at regional scales, particularly for crop production and the reclamation of saline-alkali lands in sustainable and high-quality agriculture. However, it is still lacking in the prediction of soil salinity using remote sensing imagery, due to the complex spectral features of the soils. In this study, the spatial distribution of the soil salinity was predicted with high accuracy, where the environmental variables were integrated with the multi-source Sentinel-2 remote sensing data. Two representative study areas were selected from Dingbian County in Yulin City, and Dali County in Weinan City, Shaanxi Province, China. Field sampling was conducted to collect the soil salinity data from multiple points over the study regions. Statistical analyses were performed to characterize the spatial distribution of the soil salinity at the sampling sites. Machine learning models were employed to establish quantitative relationships between environmental variables and total soil salt content. Model parameters were also optimized for the reliability of the prediction using cross validation. The total content of the soil salt was estimated using this framework. The saline-alkali lands were delineated with varying degrees of salinization. At the same time, Sentinel-2 multispectral imagery was utilized to extract the saline-alkali land information using specific spectral indices. The spatial severity of the soil salinization was mapped to fuse the environmental variables with the remote sensing-derived observations. Results indicate that the heavily saline soils were dominant in the saline-alkali lands of the Dingbian and Dali counties, thus accounting for approximately 20.00% and 22.45% of the total land area, respectively. Correlation analyses revealed that the soil pH, evapotranspiration, and precipitation also exhibited significant relationships with the total soil salt content, where the correlation coefficients were 0.70, 0.68, and -0.72, respectively, indicating the influence of the soil properties and climatic factors on the salinization. Saline-alkali lands were further extracted from the remote sensing imagery using threshold segmentation. The high classification accuracies of 85% were achieved for the normalized difference saline-alkali index and the soil permeability. A comparison was also made on the different prediction approaches—including spatial interpolation, point-to-area extrapolation, machine learning models, and the integrated machine learning plus remote sensing framework. The results also demonstrated that the integrated approach significantly enhanced the accuracy of the prediction in the area and spatial distribution of the soils with varying degrees of salinization. The mean absolute error was reduced to 0.82 g/kg after calibration, the root mean square error was 1.21 g/kg, and the coefficient of determination reached 0.85. The integrated model substantially outperformed the single-source prediction. Compared with the conventional soil salinity prediction, the high predictions of the saline-alkali land were closely aligned with the historical observations, indicating the reliability and practical applicability. Overall, environmental variables, field observations, and remote sensing data were integrated to support the evidence-based decision-making for the land reclamation and resource optimization, in order to mitigate the soil salinization impacts. These findings can provide a sound technical framework to real-time monitor the total soil salinity at the national scale. Valuable guidance can be offered for sustainable land planning in the salinization-prone regions.