Abstract: Aiming at the prevailing northwest arid zone of soil salinization, the paper selected the Ugan-Kuqa River Oasis as studied area and TM image as data source to establish BP neural network combined with Adaboost algorithm evaluation soil salinization warning model. First, according to the actual conditions of the area set up four warning indicators of the model: groundwater depth, elevation, salinity index, normalized Drought Index, and extracted its continuous surface information, made BP neural network as a weak predictor to predict, different training obtained results are combined into a strong predictor when using the model of training samples, according to the classification results of each evaluation factor in the share of the contribution rate to adjust their weights, the predicted outcome will be fair objectively reflect each evaluation factor's contribution for the soil salinization in the region, the experimental results show that the overall situation in the study area is more serious alarm, wasteland and water content inside a small area of arable land around the northern oasis, salinization have hazard higher degree. This paper from the point of view of soil salinization warning to conduct a preliminary study and lay a foundation for further study in the future salinization warning.

Key words: arid zone, salinization, warning, BP, Adaboost predictor