Forecasting Model Of Soil Selinization Cause In The Lower Reaches Of Kaidu River Basin

Soil salinization in arid, semi-arid poses a great threat to the regions of stableoasis ecological environment,and play an important hinder role to the sustainabledevelopment of oasis agriculture.With the importance of the study of soil salinizationhas become the consensus of the world, based on the GIS measure and remote sensingimage of soil salinization and remote sensing image analysis, extract informationsalinization,analyses soil salinization, inversion, simulation and prediction continuesto mature. Based on the platform of GIS comprehensive analysis of remote sensingand non-remote sensing data analysis mathematical model as a means of salinizationtechnology is constantly improving.Because Soil salinization area happen complexitysystem, the nonlinear,and current research a little bigger than the role of soilsalinization in natural factors,soil salinization together in the natural and humanfactors analysis and prediction is not mature enough.Therefore, analysis of the impactof natural and anthropogenic factors on soil salinization and soil salinization throughthe establishment based on BP neural network prediction model into a result can beachieved various factors make into understanding soil salinization maintaining aridzone oasis of stability and security and so has very important significance.Taking a typical arid region–the oasis in the lower reaches of Kaidu RiverBasin as the study area, and using1973,1990,2000and2010LANDSAT MSS, TM,ETM and ALOS/EVNIR-2remote sensing data and field survey data as the basic datasource, by SVM classification, soil salinization situation and change distribution inthirty-eight years are analyzed. By ArcGIS spatial analysis functions, using theshortest distance method, distance from the lakeshore, residential distance, thedistance from the irrigation canal and distance from drains, combined withgroundwater depth, groundwater salinity and elevation data, such as natural andman-made factors, using gray correlation analysis to determine of the relationshipbetween the various cause factors of soil salinity, soil salinity cause factors areexplored. Experimental data and spatial data, using BP neural network and Matlabplatform, cause factors of soil salinity are studied.The results show that:1. The soil salinization area increased122.83km2from1973to1990, and severesalinization and moderate salinization area continued to decline, severe salinizationdrop area between1990and2000was6.8km2and7.06km2from2000to2010, mild salinity area showed that an increasing trend and then an decreasing trend. The areareduced128.65km2from1990to2000and the area increased41.23km2from2000to2010. Soil salinization relatively intense period in the study area from1973~1990.2. Grey correlation degree of each factor from0.6651to0.7366in0~10cm,from0.4153to0.6138in10~30cm, from0.3585to0.5687in30~50cm. Variousfactors influence on soil salinity in0~10cm. Among them, the distance from theirrigation canal and groundwater depth as the most important factor, correlationcoefficient are the biggest and Again, the distance from the residential areas,drainage,elevation,distance from the lakeshore, groundwater salinity.3. Using additional momentum gradient descent method, LM optimizationalgorithm and Bayesian regularization algorithm for trained. Using7-16-1three layernetwork structure build the BP neural network prediction model.4. The seven main factors as input and soil salinity as output, BP neural networkmodel can be better simulate and predict the relationship between the genetic factorsand salinity.predicted value and the actual value of the residual from BP neuralnetwork is relatively small, the average of the residuals is0.1969, the predictionaccuracy of86.71%,it feasibly to using this method to predict the cause of soilsalinization.