Monitoring Hydrological Regime And Classifying The Wetland Based On Time Series Radarsat-2 Images

In recent years,the development and utilization of water and soil resources on climate change and the human unreasonable exploitation(such as the increasing of industrial and agricultural water and construction of water diversion project in the upper reaches of the river)has led to significant changes in the in hydrological processs in the downstream wetland.It has seriously affected and destroyed the structure and function of the wetland ecosystem,endangering the regional ecological security,water security sustainable economic and social development,it is the serious challenge to the protection and restoration of wetland.At the same time,due to the lack of wetland management experiences in developing countries,wetlands is becoming increasing serious,and it aggravates the situation because of lack of basic data.In order to protect wetland resources,reduce the destruction of wetlands and formulate the scientific management measures,it is important to get the wetland hydrological regime accurately.Remote sensing monitoring of hydrological regime can not only provide data for establishment of regional hydrological model,but also has an important significance to improve the classification accuracy of wetlands.In this paper,two paraments of hydrological regime(flood inundation area and inundation frequency)are selected and analyzed.However,due to the imaging principle of the traditional optical image,it can not distinguish the inundation of the wetland vegetation,and the radar image can penetrate the vegetation through the microwave,and then detect the flooding situation under the vegetation.This paper uses eight Radarsat-2 time series images in 2015 to extract the hydrology on Zhalong Wetland.Because of radar images exist the ‘salt and pepper' noise in radar images,traditional classification such as supervised classification and decision tree is unsuited on the classification of the radar image,the eight time series Radarsat-2 images were classified to obtain the result of flood inundation area based on the object-oriented classification.The inundation frequency was calculated by the results of flood inundation area.The Zhalong Wetland was classified based on random forest classification algorithm by selecting eighteen predictive variables including Landsat 1-7 band;seven texture features(angular second moment,contrast,correlation,dissimilarity,energy,homogeneity,mean),normalized difference vegetation index(NDVI);HH and HV polarization backscatter coefficient,flood inundation area and inundation frequency,the influence of inundation frequency on the classification accuracy is analyzed.The result indicate: 1)The backscattering coefficient of the flood inundation area and the backscattering coefficient of non-inundation area have the obviours difference by comparing the mean and standard deviation of backscattering coefficients among the Radarsat-2 images.The flood inundation area can be easily extracted.The polarization image of HV is more obviours than the polarization image of HH to compare between the flooded and non-flooded arear of reed wetland.2)The eight flooded inundation area images are mapped through 8 Radarsat-2 images classified by using 2/3 training data based on the object-oriented classification.The accuracy is above 80% by making the accuracy verification with 1/3 training samples.In September 10 th,the accuracy and kappa coefficient are 98.24% and 0.97,respectively.In April 29 th,the accuracy and kappa coefficient are 82.86% and 0.48,respectively.Inundation frequency is obtained by superposition of 8 stages,the frequency of temperature changes recorded by temperature recorder is used to verify inundation frequency.The accuracy of inundation frequency is 84.62%,the area of high inundation frequency concentrates in the core area of Zhalong Wetland.3)The accuracy of the using RF algorithm accuracy is higher than using MLC algorithm and CART algorithm by comparing the accuracy of classification results.In the transition zone of grassland and marsh wetland,the wetland was misclassified to the non wetland area by using the MLC and CART.The accuracy with inundation frequency was 91.73%,the accuracy without inundation frequency was 76.49%.The importance score of the inundation frequency is 98.69,the main errors come from the commission of grassland and omission of wetland.However,the inundation frequency did not affect the accuracy of the water.