Preprint version - Not peer reviewed Accurate tree species mapping is critical for effective forest management and ecosystem monitoring, especially under climate change. Mapping species composition is essential for understanding ecological dynamics, which supports conservation and sustainable resource management. However, traditional field-based monitoring is time-consuming and resource-intensive, particularly in mountainous regions. This study explores the use of remote sensing and machine learning techniques to improve the efficiency and accuracy of individual tree crown (ITC)-level species mapping. We utilize imaging spectroscopy data from NASA’s AVIRIS-NG sensor covering the 400 to 2400 nm spectral range with 425 bands and 2-meter spatial resolution, offering a great opportunity for high-resolution vegetation analysis. Our study area covers 630 km² within the Swiss Pre-Alps. A Random Forest classifier was trained to identify five dominant tree species. Model performance was evaluated using ground reference points and compared with classifications derived from Sentinel-2 multispectral imagery and structural features extracted from airborne LiDAR. The spectral model achieved strong predictive performance (Overall accuracy = 0.74). LiDAR-derived structural features provided marginal improvements when incorporated into the spectral model but performed poorly when used independently. Multispectral data produced limited classification accuracy. To address spatial gaps from incomplete flight coverage, we applied the tile-based gap-filling algorithm chessQS to generate a continuous, high-resolution species map. This study presents the first application of AVIRIS-NG data for tree species mapping in the Swiss Pre-Alps and demonstrates its potential for large-scale, high-precision forest monitoring in complex mountainous environments.
Lambiel et al. (Fri,) studied this question.