Abstract: [Objective] This study was intended to explore the spatiotemporal variation of soil salinization in the Weigan-Kuqa River Delta Oasis, through analysis of the importance of vegetation phenological characteristics and land surface parameters in the remote sensing-based monitoring of salinization, in order to provide scientific support for the management of soil salinization and desertification in arid oasis.[Method] Phenology metrics (Start of season, SOS; End of season, EOS; Length of season, LEN) were derived from MODIS-NDVI data, and then coupled with land surface parameters (vegetation and salinity index, terrain attributes, drought indexes, etc) and vegetation phenological metrics as input factors of Cubist regression tree model to predict soil salinity in Weigan-Kuqa River Delta Oasis from 2006 to 2016.[Result] The SOS ranges from March to June, and EOS ranges from mid-November to late-December. For inner oasis area, the LSI (Large seasonal integral) has a larger value concentrated on the range of 6.08-9.20. However, a lower range of LSI (3.64-6.08) was observed in the oasis-desert transition belt. The accuracy of estimation model based on phenological parameters is relatively low (RMSE=13.29,R²=0.12). The accuracy of model based on phenological parameters and land surface parameters is best (RMSE=9.02, R²=0.72), which is better than that of model based on land surface parameters (RMSE=12.66, R²=0.22). In the prediction model, TVDI (Temperature vegetation drought index), LSI, Salinity index (SIT and SI), MSAVI (Modified soil adjusted vegetation index), and surface reflectance have high relative importance, indicating that soil water content, vegetation growing situation, vegetation biomass and visible light reflectance are important indicators for monitoring soil salinization. Weigan-Kuqa River Delta Oasis is dominated by slightly saline and non-saline soils with the mean annual soil salt content of 7.08 g·kg^-1a^-1. The agriculture area inside the oasis is mostly covered by non-saline soil. The moderately salinization soil are mostly located in the east and south parts of the oasis. The soil salinization showed no significant decreasing trend, with a range of 0.00-0.764 g·kg^-1a^-1. Little area showed an increasing trend, and mostly located in the middle part of the oasis and near the southern part of Tarim river, with a range of 0.00-0.742 g·kg^-1a^-1. In 2007, the degree of salinization was the heaviest, and the average soil salt content was 12.68 g·kg^-1. In 2011, the degree of salinization was the lightest, and the average soil salt content was 4.61 g·kg^-1.[Conclusion] Regression tree model can establish a complete knowledge of soil-environment relationship, and effectively extract regional soil salinization information. The introduction of vegetation phenological parameters can significantly improve the accuracy of soil salinization prediction. The soil salinization in Weigan-Kuqa River Delta Oasis showed a mainly decreasing trend, and a small part of the area showed an increasing trend, and the salinization treatment effect was remarkable. There is a tight correspondence between different phenological characteristics and local soil salinity. In some parts of the study area, soil salinity is increasing because of human factors. Therefore, it is necessary to take effective measures for soil salinization control and ecological environmental protection in these specific areas.

Key words: soil salinization, vegetation phenology, land surface parameters, Cubist regression model, MODIS