Quantitative Inversion Of Crop Parameters Coupled With Multi-class Radiative Transfer Models

Known as the“Lungs of the earth”,forest is one of the most important components of the terrestrial biosphere and an important foundation of the whole terrestrial natural ecosystem.Forests make up about 30%of the Earth’s land surface and are important for conserving water,regulating t he atmosphere,resisting wind and fixing sand,and maintaining biodiversity and the balance of ecosystems.Forests play an important role in protecting the regional ecological environment and maintaining the carbon balance of the world.The estimation of forest canopy height and forest aboveground biomass is essential to enhance understanding of the world’s carbon cycle and to develop more rational carbon reduction strategies.At present,optical remote sensing data,microwave remote sensing data and lidar data are three kinds of commonly used remote sensing data sources in estimating forest aboveground biomass.How to effectively combine the advantages of various sensors to improve the accuracy of forest aboveground biomass estimation is the current researc h focus and trend of quantitative remote sensing of forest land.Based on the Autokeras deep learning platform,this paper combines GEDI satellite-borne full-waveform lidar data,Sentinel-1 C-band Polarimetric interferometer data and Sentinel-2 multi-spectral remote sensing data in Nanchong,Sichuan Province,to construct the Autokeras depth learning framework and combine the advantages of the three types of data,so as to achieve the goal of retrieving the forest canopy height of Nanchong.Nanchong from th e multi-modal data,and then combine the allometric growth equation to obtain the aboveground biomass of the forest,a 30m spatial resolution forest canopy height map and biomass map of Nanchong were also drawn.In this paper,the following three aspects are studied:1)Estimation of forest aboveground biomass based on Autokeras deep learning algorithm;2)Estimation of forest aboveground biomass based on compressed sensing CLSANSAL algorithm;3)Forest canopy height and biomass retrieval based on multi-source remote sensing data.The results of this study are as follows:(1)The estimation of forest biomass is critical to improving our understanding of the global carbon cycle and achieving effective carbon reduction strategies,however,current studies usual ly select RH85,RH95,RH100and other limited relative height percentiles as model input variables,and few studies consider all the impact of RH on the model factors,based on the Autokeras deep learning algorithm,this paper estimated the aboveground biomass of forest,using the selected non-negative relative height Rh as the data source,the regression model of relative height percentile RH value and aboveground biomass in GEDI Point was established,and the precision of the model was verified.The final simulation accuracy of training set and test set is0.93 and 0.89,respectively.(2)In the process of data acquisition,there will be a lot of redundant data,only a small amount of data can finally be used,greatly reducing the efficiency of data use.In this paper,a method based on sparse regression is used to calculate the contribution of data,find the best subset in a large database,and calculate its contribution to the whole.The calculated contribution of RH88 to aboveground biomass was found to be up to 19%and that of RH66 to aboveground biomass to be up to 14%,with a similar trend of continuous contribution of the remaining Rh values distributed between 0 and 10%.(3)The spectral information of GEDI L2A forest canopy height data,Sentinel-1 data,Sentinel-2 data and relevant vegetation index were fused,and the advantages of all kinds of remote sensing data were fully combined.The canopy height of Nanchong Forest was retrieved using Autokeras deep learning framework.The aboveground biomass of Nanchong Forest was obtained by allometric growth equation.The results showed that the predicted values of forest canopy height were mainly distributed between 10 m and 20 m,and the spatial resolution of forest canopy height of 30 m in Nanchong,the accuracy of airborne lidar data validation(R~2=0.62,RMSE=2.40 m)and real measurement data validation(R~2=0.70,RMSE=1.82 m)is higher.The predicted values of aboveground biomass were mainly distributed between 10-30 Mg/ha in Nanchong,and the 30-M Resolution Forest AGB in Nanchong was validated with R~2 of 0.53and RMSE of 6.83 Mg/ha.The results show that GEDI has great application potential in high resolution forest height mapping of large area.