Study On The Spectral Characteristics Of Typical Halophytes In Ebinur Region

The Ebinur Wetland National Nature Reserve is temperate arid zone with high protection value desert ecosystem.There are many types of vegetation in the reserve.The soil has different degrees of salinization.It is of great significance to study the Halophytes in Ebinur Lake Wetland for alleviating salinization disaster,preventing sand dust storm and salt dust storm,and repairing desert ecosystem.In this study,the author selected the Ebinur Wetland National Nature Reserve as the study area,and took typical five types halophytes as the research object.The author used the hyperspectral technology to collect the spectral reflectance data of halophytes,and carried out a variety of mathematical transformations on spectral data and extracted characteristic parameters to analyze the spectral characteristics of halophytes.Then,the Mahalanobis Distance was choosed to select characteristic band for hyperspectral data in order to reduce the dimension of hyperspectral data.On this basis,the discrimination model of halophyte type was constructed to provide a reference for the investigation and monitoring of plant resources from aerial or aerospace remote sensing.The biochemical component data of plants is an important basic data of the ecosystem and global carbon cycle in arid areas.In terms of remote sensing to extract vegetation parameters,only one extraction algorithm is often studied for each parameter type of data product,and the results have large uncertainties.In this study,an estimation model of plant leaf equivalent water thickness was constructed by means of measured hyperspectral data and plant biochemical data,so as to provide reference for monitoring halophyte plant growth.The following conclusions are as follows:(1)The analysis of the spectral characteristics of Haloxylon ammodendron,Tamarix chinensis,Nitraria roborowskii Kom,Populus euphratica and Phragmites communis showed that the five types of halophytes have similar spectral characteristic curves and the spectral characteristics of general green vegetation.By using the mathematical transformation methods(first derivative,logarithm spectrum and the continuous removal spectrum)to process the spectral data of vegetation,the detailed information of plant spectrum can be amplified to a certain extent,and the spectral differences of different types of plants can be highlighted.By extracting the spectral absorption characteristic parameters of 540?760nm band,the absorption peak area of Tamarix chinensis was obviously different from that of other four types of plants.(2)The Mahalanobis distance method was used to select characteristic bands from the original spectrum and its transformed spectral,and the plant type discrimination model was constructed.The selected feature bands were mainly concentrated in the characteristic positions of the spectrum,such as the blue valley,red valley,red edge and near infrared water absorption valley.In the constructed plant type recognition model,the classification accuracy of original spectrum,continuum removal spectrum and logarithmic spectrum transformation was better,which were 0.88,0.90and 0.90 respectively.(3)The sensitivity and anti-saturation performance of the selected vegetation indices were evaluated by using the improved sobol sensitivity analysis method.NDWI₁₂₀₀,NDWI_1240,NDWI₁₆₄₀,WBI,GVMI,NDII and SRWI indices are all sensitive to the equivalent water thickness and weakly sensitive to other parameters,which are ideal indexes for estimating the equivalent water thickness of vegetation.(4)Three different models were used to estimate the equivalent water thickness of plants in Ebinur Wetland.The results showed that the prediction accuracy of the BP-PROSPECT model was the best(R² is 0.832,RMSE is 0.0028,RPD is 2.21).By using the PROSPECT model,the author discussed the use of classical iterative numerical optimization algorithm to retrieve the equivalent water thickness.The determination coefficient was 0.692 and RMSE was 0.0043cm.The estimation model slightly overestimates the water content.Multivariate empirical model inversion had a good prediction accuracy(R² is 0.709,RMSE is 0.0024,RPD is 1.76).This method can be considered when other model parameters are difficult to obtain.It is feasible and adaptive to invert EWT in Ebinur Wetland by combining PROSPECT model and BP neural network.