Research On Forest Canopy Height Retrieval Based On ICESat-2 Spaceborne LiDAR Data

Forest canopy height refers to the distance from the top of the canopy to the ground,which is an important indicator for forest ecosystem evaluation and can reflect the level of forest ecological health and vegetation growth.High precision measurement of forest canopy height in large areas can reduce the uncertainty of forest carbon cycle,and plays a vital role in the estimation of forest biomass,carbon storage and dynamic change research.Traditional measurement methods are time-consuming and laborious,making it difficult to obtain large-scale data.Synthetic aperture radar has a certain penetration,but it is prone to signal saturation,which leads to a decrease in its accuracy in forest canopy height extraction.As an active remote sensing detection technology,lidar can quickly and accurately acquire three-dimensional surface data by calculating the round-trip time of laser energy pulses between the sensor and the target.Among them,spaceborne Li DAR has unique advantages in obtaining three-dimensional data of large-scale forests due to its high orbit and wide observation range.Therefore,based on the new generation of spaceborne photon counting lidar ICESat-2/ATLAS data,this study explores the denoising and classification algorithms of photon cloud data,and then realizes the forest canopy height mapping in the research area based on the proposed algorithm.Finally,a comparative study of the surface elevation information retrieved from ICESat-2/ATLAS data and GF-7 stereo images was carried out.The specific research content and conclusions of this paper are as follows:1)A multi-level filtering algorithm was constructed,using different combination algorithms for the distribution characteristics of photon cloud data under different observation conditions.The search domain parameters were adaptively improved,and the differences in denoising results between different filtering directions were explored.The research results show that under daytime strong beam observation conditions,the denoising accuracy(R_sand F values are better than 0.97 and 0.67,respectively)of our algorithm is superior to ATL08algorithm(R_s=0.85,F=0.65),and the filtering direction has no significant impact on the denoising results;Under daytime weak beam observation conditions,our research institute adopts multi-level filtering algorithms in both horizontal and intra group unified directions with the same denoising accuracy(R_s=0.92,F=0.69),which is better than the denoising results using each photon adaptive direction and ATL08 algorithm(R_sand F values are 0.94and 0.88,0.65 and 0.67,respectively).Although each photon adaptive direction can retain signal photons to a greater extent,the overall denoising accuracy is relatively low;Under strong and weak beam observation conditions at night,the denoising accuracy of this research algorithm(R_sand F values are better than 0.98 and 0.92,respectively)is superior to ATL08algorithm(R_sand F values are better than 0.90 and 0.88,respectively),and the filtering direction has no significant impact on the denoising accuracy of the algorithm.2)Construct a multi-level progressive photon cloud classification algorithm that achieves the classification of photon cloud data through the process of potential photons,initial photons,precise photons,encrypted photons,and contour fitting.The classification accuracy was evaluated under different observation conditions.The research results show that:compared with the ATL08 algorithm,under daytime observation conditions,the research algorithm can obtain higher classification accuracy(strong and weak beam ground/crown classification accuracy R~2and RMSE are 0.99/0.69,0.99/0.65 and 1.52m/9.47m,3.16m/9.58m,respectively).For nighttime conditions,the ground classification accuracy of the two algorithms is similar(both strong and weak beam data R~2is better than 0.99,and RMSE is lower than 2.46m).However,the classification accuracy of the algorithm in this study is significantly better than the ATL08 algorithm for the classification accuracy of the canopy(the classification accuracy R~2and RMSE of strong/weak beam crowns are 0.84/0.82and 6.84m/6.85m,respectively).Regardless of day or night,the classification accuracy of strong beam data is better than that of weak beam.Comprehensive analysis shows that strong beam data at night can provide relatively higher data quality.3)Based on the ICESat-2/ATLAS data of Liuzhou City in 2019-2021,the 30m resolution forest canopy height mapping in Liuzhou City was realized.Based on the Liuzhou discrete CHM data set obtained by the photon cloud data processing algorithm proposed in this research,Landsat-8,SRTM DEM and climate data are used as characteristic parameters.And the canopy height extrapolation model was constructed by random forest algorithm,and the canopy height mapping of Liuzhou City was realized.The results show that:based on the test set data,the verification accuracy is R~2=0.58,RMSE=3.95m.Compared with the measured data,the forest canopy height mapping accuracy is R~2=0.51,RMSE=2.24m.4)The differences in the results of surface elevation retrieval from ICESat-2 and GF-7satellite stereo images were explored.This paper conducts a comparative study on the surface elevation results generated by the two types of data under different vegetation coverage conditions.The research results show that the correlation R~2and RMSE of DSM generated by ICESat-2 strong beam and GF-7 satellite are 0.98/6.56m,0.99/8.76m,and 0.95/35.17m respectively under the coverage of cultivated land,grassland and forest.The R~2and RMSE of weak beams are 1.00/4.45m,0.99/7.09m,and 0.96/29.36m,respectively.The results of the two types of data in the forest area are relatively different.By re-filtering the ICESat-2 data in the forest area,the R~2of the strong/weak beam and the DSM generated by the GF-7 satellite are both increased to 0.99,and the RMSE and MAE are also greatly reduced.And as the statistical scale increases,the two gradually decrease.