Estimation of secondary dune density in the hinterland of the Taklimakan Desert

The secondary dune density is a crucial parameter for studying desert surface characteristics and monitoring the development of sand ridges. It provides an effective means to describe the distribution of regional secondary dunes. However, there is currently no effective method for estimating dune density, particularly for secondary dunes that are attached to large sand ridges, which are often overlooked due to their smaller scale. Therefore, this study presents a more effective method for estimating secondary dune density by utilizing Google Earth images and the YOLOv8s algorithm. This study utilizes Google Earth images from the hinterland of the Taklimakan Desert to construct a secondary dune identification dataset. Based on the dataset, we train and validate the secondary dune identification model to estimate secondary dune density. The research results indicate that the model achieved an average precision (AP50) of 63.58 % for secondary dune identification, outperforming other baseline algorithms. Furthermore, the dune identification model successfully identified a total of 18,208 secondary dunes within the verification area. The model demonstrated a higher predictive capability for secondary dune densities in the Taklimakan Desert hinterland, with a coefficient of determination (R²) of 0.89 between estimated and observed values. The mean absolute error (MAE) was 20.94 km⁻², and the root mean square error (RMSE) was 25.04 km⁻². Crucially, the accuracy of the secondary density estimation is highly dependent on the precise delineation of dune outlines, crest lines, dune arcs, and ridge lines from the imagery. The method for estimating secondary dune density proposed in this study overcomes the limitations of existing research and provides new insights into the evolutionary processes of aeolian dunes.