Retrieval Of Forest Canopy Height In A Mountainous Region With ICESat-2 ATLAS And Sentinel-2 Data

Forest canopy height is one of the important forest structure parameters to accurately estimate forest biomass or carbon storage.Accurate and rapid estimation and mapping of regional forest canopy height is the key to understand vegetation growth process and ecosystem internal structure.In this study,ATLAS data and Sentinel-2 remote sensing images were used to retrieve forest canopy height in Jinzhai County,Anhui Province,which is located in subtropical mountainous area.The test of forest canopy height parameters of ICESat-2 ATLAS product found that the error of ICESat-2 ATLAS product was large in mountainous areas,so it is difficult to be directly used for forest canopy height estimation.ATL03 and ATL08 products were associated to obtain the spatial information and classification information of the photons,and then the altitude information of the associated photons was combined with the corresponding airborne Lidar ground elevation(DTM)to calculate the height of the space-borne canopy.The average space-borne canopy height of each unit was calculated by taking the segments with different sizes along the orbit as statistical units(20m,40m···· 200m).According to the distribution of statistical units with different number of canopy photons,a filtering method was developed to obtain forest height samples with high accuracy and reliability.In this study,the continuous forest canopy height with 20 m spatial resolution was retrived based on the ATLAS statistical unit with 20 m size and Sentinel-2 data.In this study,a forest canopy height estimation model was established based on multiple linear regression(MLR),support vector machine(SVM),k Nearest Neighbors(k NN)and stack of three algorithms,and the accuracy of the model was verified.The optimal model was selected to retrive the canopy height and analyze the spatial distribution characteristics.The results are as follows:(1)Analyze the relationship between the average height of the space-borne canopy and the average height of the corresponding unit airborne Lidar.The results show that the introduction of high-precision DTM can significantly improve the accuracy of ATLAS data in extracting the mountainous forest canopy height.The r between the mean height of the spaceborne canopy and the mean height of the airborne Lidar canopy increases from 0.30-0.56 to 0.70-0.94,and the RMSE decreases from 4.49-11.71 m to 1.32-3.23 m.(2)For mountainous areas with high canopy coverage,the average height data of low reliability space-borne canopy is filtered based on data retention rate,which can not only retain sufficient data,but also ensure that the average height of space-borne canopy after screening has a good correlation with the average height of the corresponding unit airborne Lidar and a small RMSE.The 20 m and 40 m statistical units filter the data retention rate based on 20%,the 60 m and 80 m statistical units filter the data retention rate based on 35%,and the 100 m or larger statistical units filter the data retention rate based on 75%.In this study area,all ATLAS strip data were processed based on the 20% retention rate method,and the correlation range of validation was 0.61-0.91,and RMSE was 1.79-4.35 m.(3)By analyzing the accuracy of multiple linear regression,random forest,k Nearest Neighbors and stacking algorithm of the three algorithms for broad-leaved,Chinese fir,Masson pine and undifferentiated forest types,it is found that the random forest algorithm without differentiated forest types has the highest estimation accuracy.The correlation coefficient,root mean square error and relative root mean square error were 0.724,2.44 m and 24.8%,respectively.Therefore,the forest canopy height of the study area is retrived based on the model.(4)The red-edge bands in Sentinel-2 remote sensing images are important for estimating forest canopy height.The red-edge band and its calculated vegetation indices(such as NDVIre2,RVIre,RECI,etc.)are selected in the modeling process.These vegetation indices can well reflect the situation of forest canopy height and are important independent variables of the modeling.(5)The results of forest canopy height in Jinzhai County range from 2.79 m to17.20 m.The spatial characteristics of forest canopy height were analyzed from three aspects of altitude,slope direction and slope.It was found that the forest canopy height was higher in the area with an altitude of less than 800 m and a slope of 15°-30°or 30°-45°,and the range of canopy height was mainly 10-15 m,while the slope direction had no significant influence on the spatial distribution of forest canopy height.(6)Compared with the forest canopy height products generated by space-borne lidar data,it was found that in the mountainous research area with complex and rugged terrain and high canopy coverage,the forest canopy parameters obtained by space-borne lidar could not be accurately described,leading to inaccurate and overestimated forest canopy parameters.