Remote Sensing Inversion Of Regional Land Use-High Spatial Temporal Resolution Evapotranspiration And Soil Moisture Distribution

Land use information,evapotranspiration(ET)and soil moisture(SM)are the basis data for the efficient management of regional water resources.Remote sensing technology can be used to obtain related parameters which contribute to the dynamic management of regional water resources.In this study,the countries of Wugong,Fufeng and Yanglin in Guanzhong region of Shaanxi province were selected for the inversion of regional land use-high spatial temporal resolution ET and SM distribution based on the technology chart of theoretical analysis,numerical simulation and experimental research.The following main conclusions are drawn:(1)A new method of land use classification based on daily time series data was proposed,which overcomes the reduction of classification accuracy caused by missing data of time series data.Research indicates that S-G filtering algorithm is superior to the HANTS filtering algorithm for smoothly filtering unequal-interval time-series data.Daily-interpolated NDVI time-series data has a higher overall accuracy and Kappa coefficient in land classification,which could be used to classify similar NDVI trends during the period to improve the credibility of land use classification.The support vector machine(SVM)classifier was optimized via using the genetic algorithm(GA)and the classification accuracy of daily NDVI time series was improved.(2)An improved multi-classifier fusion algorithm was proposed based on the daily NDVI time series data.Research indicates that different single classifiers show different advantages in land cover classification.Mahalanobis distance classifier is unsuitable for extracting information from daily NDVI time series,since the required number of the sample due to its mechanism characteristicsis higher than the actual sample number dimension.K-means classifier is unsuitable for extracting land resource information with complex planting structure in irrigation districts.Extracting land use information can be achieved with the statistical index(mean,peak and skewness)of daily NDVI time series and it can reduce the tedious calculation process caused by the multi-dimensions of the time series feature space.The multi-classifier combination has higher accuracy than the original single algorithms.And the proposed improved multi-classifier combination was demonstrated to provide enhanced performance in different single classifier combinations,especially for sub-classifiers with greater differences.(3)A new inversion method of high-temporal resolution ET without thermal infrared data was proposed which based on artificial neural network(ANN)algorithm.Research indicates that land surface temperature(LST)and Digital Elevation Model(DEM)are the first and second dominant factors respectively,The LST–NDVI relationship is strong in summer and it can be used to obtain LST with the predictor variable NDVI in summer.The study found that ET models using LST based on thermal infrared data have higher accuracy,and ET models using LST based on regression model also have higher accuracy The ANN models for ET estimation without thermal infrared data can applied to long-term ET data acquisition,which increased the flexibility of ET research.(4)An ET-vegetation index(EVDI)model was proposed to inverse the regional SM,which overcomes the uncertainty problem of soil moisture inversion in the LST-vegetation index(TVDI)model.Research indicates that the EVDI feature space construction is feasible,and it can be used for SM inversion.At the growth time of corn,the SM inversion by the EVDI method is superior to the TVDI method.The measured SM content at 0–20 cm had a highest determination coefficient,then it declined with the increase of soil depth.At the growth time of wheat,TVDI and EVDI methods are both unsuitable for soil moisture monitoring during wheat seedling stage.Compared with TVDI method,EVDI method is not suitable for soil moisture inversion in winter with lower ET but performs better in other seasons with higher ET.The results of SM inversion in different crop periods indicated that the spatial characteristics of TVDI and EVDI and multi-point monitoring results can be used to inversed regional SM.The soil drought level monitoring can be achieved based on the EVDI and the different drought index represent different degrees of drought.This paper improves the land classification accuracy by improving the dimension of time series data,but it increases the burden of calculation at the same time.The further studies need to extract main features of the daily time series for land cover classification.Secondly,the stability of the high temporal resolution regional ET inversion model and SM inversion model affected by the regional meteorological conditions,surface cover types and soil properties.The affected factors were complexity and further research is needed.