In forested watersheds, density, land cover, and its vertical structure are crucial factors for flood management, ecosystem monitoring, and biomass inventory. Nowadays, producing land cover maps with high accuracy has become a reality with the application of remote sensing techniques, but in some situations, it is not so easy to distinguish between the overstory and understory vegetation with only spectral information. The main goal of this study was to analyze the accuracy enhancement in overstory and understory land cover mapping at the watershed scale when using the data fusion of seasonal and annual time series of Sentinel-2 images complemented with low-density LiDAR and soil and vegetation indices. The study area was composed by two neighboring watersheds in Badajoz province (Spain). The accuracy of land cover classifications was trained in two ways: first, for each season and several soil-vegetation indices; and second, for the annual series and soil-vegetation indices. Next, LiDAR data were included in both analyses by means of a Boolean metric concerning the height. The obtained results showed that the overall accuracy was better with the annual evaluation when only spectral information was used for the classification. Additionally, if LiDAR data were included in the classification (data fusion), the overall accuracies were highly improved, especially in summer and autumn seasons. This improvement can be a significant issue in the analysis of vegetation structure and its spatial distribution as it is decisive for watershed ecosystem management.
Keywords: Forest land cover; LiDAR; Multitemporal analysis; Overstory; Random forest; Remote sensing; Sentinel-2A; Understory.