Mapping tree species composition in a Caspian temperate mixed forest based on spectral-temporal metrics and machine learning

The tree species composition (TSC) reflects a forest's diversity and is relevant for forest planning, biodiversity conservation, resources management. Yet, accurate information on at landscape scale largely missing, especially mixed forests remote areas. One reason being that mapping time-consuming, costly, in Here we develop robust method TSC temperate forest. Based inventory plots considering the frequency of dominant dataset, five groups were defined: pure oriental beech, common hornbeam, deciduous. classification based three-year time series data Landsat-8 (L8) Sentinel-2 (S2) derived spectral-temporal features (STMs) vegetation indices within long-term, seasonal, monthly scales. Model performances three Machine Learning (ML) algorithms, Random Forest (RF), Support Vector (SVM), Classification Regression Tree (CART) compared revealed different accuracies (overall (OAs) between ∼ 70 % 86 %). Highest OA was obtained using SVM regardless dataset (STMs satellite series). comparisons scales indicated with both L8 S2 seasonal STMs produced higher than long-term outperforming across all tested ML algorithms. We conclude freely available series, features, algorithms are favourable mapping.