Study On Canopy Water Content Inversion Of Six Typical Desert Plants Based On Hyperspectral Remote Sensing

As a symbolic factor of desert retrograde succession and progressive succession,desert plants play an indispensable role in maintaining the stability of desert ecosystem structure and function.As a decisive factor in the life activities of desert plants,water is also a basic factor to promote the flow of material and energy in desert ecosystems.Using hyperspectral remote sensing information to realize the identification and division of typical desert plants and the efficient,non-destructive and accurate monitoring of canopy water content can provide scientific basis for the accurate control of regional desert plants.In this study,six typical desert plants(Artemisia ordosica,Bassia dasyphylla,Corethrodendron fruticosum,Astragalus adsurgens,Amorpha fruticosal,and Thalictrum aquilegifolium)in the North and South Margins of Kubuqi Desert were selected as the research objects.Based on the field spectral detection experiment and the measured canopy water content data,the hyperspectral response characteristics of six typical desert plants were explored,and the sensitive characteristic bands,characteristic parameters and vegetation indexes of typical desert plants canopy water content were extracted.Combined with linear,exponential,logarithmic,polynomial,power function regression analysis and BP neural network algorithm.The inversion models of canopy water content of typical desert plants were constructed respectively,and the inversion performance of each model was comprehensively evaluated,and the optimal inversion models of canopy water content of six typical desert plants were established.The main findings are as follows:(1)Based on multi-type spectral transformation,the hyperspectral characteristic bands of typical desert plant canopy were selected.A total of 26 canopy hyperspectral response characteristic parameters were obtained from six typical desert plant canopy original spectra and nine transformation spectra.The canopy hyperspectral differences of six typical desert plants under the original canopy spectrum and nine spectral treatments methods were obtained.Among them,the original reflectance of the 1100 nm band spectrum can significantly identify six typical desert plants.(2)Extraction of canopy water content sensitivity parameters for typical desert plants.Based on the original canopy spectra and eight kinds of transformation spectra of typical desert plants,44,55,53,40,43 and 45 canopy water content sensitivity parameters were selected from Artemisia ordosica,Bassia dasyphylla,Corethrodendron fruticosum,Astragalus adsurgens,Amorpha fruticosal,and Thalictrum aquilegifolium,according to the selection criteria of correlation maximum value.Among them,the m NDVI705 vegetation index(r=-0.868)of Artemisia ordosica,the second derivative of the spectra at 1135 nm(r=-0.815),1609 nm(r=0.819),1501 nm(r=-0.759),and 1414 nm(r=0.787)bands of Bassia dasyphylla,Astragalus adsurgens,Amorpha fruticosal,and Thalictrum aquilegifolium,the first-order derivatives of the spectrum at 1687 nm(r=-0.856)band in Corethrodendron fruticosum were the most sensitive parameters of canopy water content.(3)Based on the spectral sensitivity parameters,the optimal inversion model of canopy water content of six typical desert plants was constructed.Based on the sensitivity parameters of canopy water content,five common regression and BP neural network models were used to construct the inversion model of canopy water content of six typical desert plants.The results show that the derivative transform spectrum and vegetation index can significantly improve the inversion accuracy of canopy water content of typical desert plants.Polynomial function and BP neural network inversion model can provide higher fitting accuracy.The optimal inversion models for canopy water content of Artemisia ordosica(R~2=0.887),Astragalus adsurgens(R~2=0.845),and Thalictrum aquilegifolium(R~2=0.809)were BP neural network inversion models based on spectral first-order derivatives;The BP neural network inversion model based on the second derivative of the spectrum was used as the optimal inversion model for the canopy water content of Corethrodendron fruticosum(R~2=0.768),Amorpha fruticosal(R~2=0.777);The optimal inversion model for canopy water content of Bassia dasyphylla(R~2=0.757)is a polynomial function inversion model based on the second derivative of spectrum.