Spatial and temporal resolution effects on object-based mapping of termite mounds

• Single-date mapping: very high-res imagery (GeoEye-1, aerial) achieved >95% accuracy for heuweltjies. • Increasing temporal resolution (Sentinel-2) improved classification accuracy by >10% via phenology. • Segmentation quality directly affects classification; GEOBIA improves heuweltjie detectability. • High-resolution DEM-derived terrain variables contributed to improved classification accuracy. Biogenic mounds play a crucial role in shaping soil salinity patterns, carbon storage, nutrient redistribution, and rangeland functioning, making accurate information on their spatial distribution essential for effective ecosystem management. This study investigates the impact of spatial and temporal resolution on the mapping accuracy of termite mounds using remote sensing imagery. Mapping performance was evaluated using object-based image analysis combined with machine learning across multi-resolution datasets, including GeoEye-1, aerial imagery, and Sentinel-2. Two experimental designs were implemented to quantify resolution-driven differences in detection accuracy. The first set of experiments evaluated the effect of temporal resolution with 1) a seasonal Sentinel-2 image composite (June to August 2019), 2) a monthly Sentinel-2 image composite (June 2019), and 3) a single date Sentinel-2 image (June 2019). The second set of experiments assessed the impact of spatial resolution using 1) Geoeye-1 imagery, 2) aerial imagery, and 3) Sentinel-2 imagery. Classification results were analysed by comparing overall accuracies (OA) and kappa coefficients, with McNemar’s test used to assess the statistical significance of accuracy differences among experiments. Results indicated that very high spatial resolution images (Geoeye-1 and aerial) based on GEOBIA and SVM allow for the classification of termite mounds with accuracies exceeding 95%. Although lower spatial resolution evidently decreased classification accuracy, increasing temporal resolution can minimise these limitations. This is demonstrated by the 10.4% and 10.8% improvements in overall accuracy (OA) using seasonal (91.1%) and monthly (90.7%) Sentinel-2 composites compared to a single-date Sentinel-2 image (80.3%).