Open source Global Digital Elevation Models (GDEMs) serve as an important base for studies in geosciences. However, these models contain vertical errors due to various reasons. In this study, data from two Satellite LiDAR altimetry systems, GEDI and ICESat-2, were used to improve the vertical accuracy of GDEMs. Three different machine learning methods, namely an Artificial Neural Network (ANN), Extreme Gradient Boosting (XGBoost), and a Convolutional Neural Network (CNN), were employed to improve existing DEM data with satellite LiDAR data. The methodology was tested in five areas with varying characteristics. Ground control data were selected from high accuracy DEMs generated from Airborne LiDAR and GNSS data. The use of ANN method improved the vertical accuracy of SRTM data from 6.45 to 3.72 m in Test area-4. Similarly, the CNN method demonstrated an improvement in the vertical accuracy of bare ground SRTM data increasing from 3.4 to 0.6 m in Test area-4. In Test area-5, the ANN method improved the vertical accuracy of SRTM data with slopes between 30 and 60%, increasing from 3.8 to 0.5 m. Notably, the results underscore the successful improvement of GDEMs across all test areas.@en

Key message: Despite showing a cost-effective potential for quantifying vertical forest structure, the GEDI and ICESat-2 satellite LiDAR missions fall short of the data accuracy standards required by tree- and stand-level forest inventories. Abstract: Tree and stand heights are key inventory variables in forestry, but measuring them manually is time-consuming for large forestlands. For that reason, researchers have traditionally used terrestrial and aerial remote sensing systems to retrieve forest height information. Recent developments in sensor technology have made it possible for spaceborne LiDAR systems to collect height data. However, there is still a knowledge gap regarding the utility and reliability of these data in varying forest structures. The present study aims to assess the accuracies of dominant stand heights retrieved by GEDI and ICESat-2 satellites. To that end, we used stand-type maps and field-measured inventory data from forest management plans as references. Additionally, we developed convolutional neural network (CNN) models to improve the data accuracy of raw LiDAR metrics. The results showed that GEDI generally underestimated dominant heights (RMSE = 3.06 m, %RMSE = 21.80%), whereas ICESat-2 overestimated them (RMSE = 4.02 m, %RMSE = 30.76%). Accuracy decreased further as the slope increased, particularly for ICESat-2 data. Nonetheless, using CNN models, we improved estimation accuracies to some extent (%RMSEs = 20.12% and 19.75% for GEDI and ICESat-2). In terms of forest structure, GEDI performed better in fully-covered stands than in sparsely-covered forests. This is attributable to the smaller height differences between canopy tops in dense forest conditions. ICESat-2, on the other hand, performed better in thin forests (DBH < 20 cm) than in large-girth and mature stands of Crimean pine. We conclude that GEDI and ICESat-2 missions, particularly in hilly landscapes, rarely achieve the standards needed in stand-level forest inventories when used alone.

@en