Towards centimeter-scale habitat mapping using satellite hyperspectral and LiDAR UAV fusion: linking ENMAP, SuperDove, and Zenmuse L2 over Grand-Barachois (Miquelon, France)

Climate change is intensifying coastal vulnerability worldwide through more frequent extreme events and rising sea levels. Miquelon, a French island, is particularly exposed, motivating fine-scale monitoring of its coastal habitats. UAV surveys of the Grand-Barachois lagoon provide RGB and LiDAR data at ultra-high spatial resolution (10 cm), but their four spectral bands may limit land–sea use-cover classification accuracy. Conversely, the ENMAP satellite delivers 224 spectral bands at a coarser 30 m resolution with limited acquisition frequency, while SuperDove satellites acquire 8 bands daily at a 3 m resolution. This study investigates whether combining these datasets can generate imagery with both high spectral (224 bands) and high spatial (down to 0.10 m) resolution. ENMAP bands (30 m) are first predicted at 3 m using SuperDove bands as predictors via least-squares regression (LS-Fit, ENVI Classic); results are then used to predict the same bands at 0.10 m using UAV RGB and LiDAR-derived DSM. Reconstruction reliability is assessed from per-band residual images at each step. Minimum distance classifications are performed on nine image products (ENMAP and SuperDove at native and predicted resolutions, with and without DSM; UAV with and without DSM), each tested with progressively restricted spectral subsets (MIR, PIR, VIS for ENMAP; PIR, VIS for SuperDove; VIS, DSM for UAV), yielding 20 classifications, with sample sizes scaled to resolution (60 pixels at 30 m, 600 at 3 m, 18,000 at 0.10 m; 50/50 training/validation split). Preliminary results show kappa scores of 0.2730 for ENMAP at 30 m (224 bands), 0.8017 for ENMAP at 3 m with PIR+VIS (91 bands) versus 0.7617 for SuperDove at 0.10 m (8 bands), and 0.6121 for ENMAP at 0.10 m (224 bands, DSM-predicted) versus 0.5401 for UAV at 0.10 m (4 bands), illustrating the influence of spatial resolution, spectral band number, and DSM inclusion on classification performance. These outcomes support the feasibility of more precise land–sea use-cover mapping in coastal habitats such as the Grand-Barachois lagoon.