Forest inventories require up-to-date data on dominant tree height and stand top height from forest sample plots. These data are used to characterize the vertical structure of forests, providing a baseline for volume and yield tables as well as many other biomass studies. Obtaining height information through ground measurement is laborious, costly, and time-consuming. The aim of this study is to estimate stand top heights of the Artvin-Hatila Valley's forests using freely available laser scanning (LiDAR) data from the ICESat-2 satellite for the first time in Turkey. For this purpose, the dominant tree heights, traditionally measured by digital hypsometer in 52 sample plots, were evaluated by stand types and compared with the ICESat-2 canopy data. Then, two data sets were modeled using the Convolutional Neural Network (CNN) and simple regression methods. The model accuracies were evaluated with correlation (Pearson's R), coefficient of determination (R-2), and root mean squared error (RMSE) using ground-based data. The results showed that the CNN-based model performed better than the linear regression model in height estimation. Its R, R-2, and RMSE values were.82,.68, and 4.2 m, respectively. As for stand types, broadleaves- dominated, mature, and fully covered stands seem more appropriate for top height modeling with spaceborne LiDAR data. Degraded, coniferous, and young stands, as well as non-forest areas, barely allow accurate top height estimations due to their complex canopy surfaces and small openings among trees. Given the promising results, we conclude that satellite-based LiDAR systems provide opportunities to forest professionals as a free auxiliary data source for operational forest management in Turkey.