Extraction Method Of Forest Ground Control Points Based On ICESat-2 Data

For estimating the carbon sink capacity of the whole forest ecosystem,the extraction and analysis of forest structural parameters are the basis for studying global changes in forest biomass and carbon cycle.Lidar technology has developed rapidly as a remote sensing tool in recent years,and has considerable advantages in forest structure parameter inversion because of its ability to overcome the effects of signal saturation,cloud cover and observation time period,and to extract and process forest vertical structure parameter information quickly and accurately.ICESat-2 is a photon-counting laser altimetry satellite with high sensitivity not found in other LIDARs,which can respond to signals at the single photon level and determine the location of the photons by calculating the start and end times of the signals,however,ICESat-2 does not classify the data in detail.SRTM remote sensing data covers most of the land area due to its ability to provide a digital elevation model with global coverage,and as a mainstream DEM product for global-scale terrain elevation measurements.Due to the vegetation occlusion effect of SRTM in forested areas,i.e.,the occlusion effect of vegetation on ground elevation,there are many height errors in SRTM data.In contrast,ICESat-2 data can be extracted to obtain more accurate ground elevation points in the forest understory,which is important for the study of forest understory topography with different forest types and cover.In this study,the study area of Everglades National Park in the United States with flat topography and low vegetation cover and the study area of Aiken County in the United States with complex topography and high forest cover were selected for comparison,and the accuracy of extracted understory ground control points was analyzed.The main studies are as follows:（1）The impact of the results on the ground point elevation is analyzed for different ATL08data parameters.The ICESat-2 data are first denoised using the DRAGANN algorithm.The process includes differential,regression and Gaussian adaptive nearest neighbor photon cloud denoising algorithms,which can effectively retain signal photons and delineate noise photons.The results show that the summed mean F values of recall and correctness of the denoising results of the algorithms in the two study areas exceed 0.8.The denoised photon cloud data are then used to extract the segment elevation information,brightness flag information,cloud confidence flag,surface coverage information,photon number parameter,accuracy description parameter,topographic photon information and photon height information from the ATL08data,based on the G-Li HT airborne validation data to study The change of elevation accuracy of each parameter screened under different thresholds,the accuracy result is more than 30%higher compared with the accuracy of photon point before screening,and its P-value keeps 10^-5level on average,the sample result is very significant.（2）A method of extracting forest ground control points based on ATL08 data parameters is proposed,and the required forest ground control points are extracted by an eight-step screening method considering the influence of topography and vegetation on the elevation accuracy of forest ground points.The results show that the RMSE of this study’s ATL08 data attribute parameter based extraction control point method is 0.30m and 2.18m in two study areas,and the obtained ground control points are compared with the elevation data of forest understory ground points extracted by DEM products of SRTM remote sensing data,and the accuracy results are significantly improved by about 14m compared with SRTM data.（3）The forest understory ground control point extraction method based on the attribute parameters of photonic point cloud data proposed in this study is applied to the forest ground control point elevation extraction in the northeastern region of China.In view of the lack of airborne validation data in China,the ground point results obtained by the forest ground control point extraction method of this study are relied on as the control point elevation datum of forest terrain,and the analysis corresponds to SRTM remote sensing data in different forest types and vegetation cover cases,and the elevation accuracy of the forest ground points obtained by it is improved by 28%,and the P-value is kept at 10^-3 level on average to present significant.