Multi-Feature-Based Identification of Subtropical Evergreen Tree Species Using Gaofen-2 Imagery and Algorithm Comparison

The species and distribution of trees in a forest are critical to the understanding ecosystem processes development management strategies. Subtropical landscapes feature complex canopy structure high stand density. Studies on effects classification algorithms remote sensing-based identification tree few. GF-2 is first satellite China with sub-meter accuracy which has resolution short replay cycle. Here, we considered three representative types (Masson pine, Chinese fir, broadleaved evergreen trees) southern subtropical region as research objects. We quantitatively compared five machine learning algorithms, including backpropagation neural network, k-nearest neighbour, polytomous logistic regression, random (RF) support vector (SVM), four features (vegetation index, band reflectance, textural features, topographic factors) using Gaofen-2 panchromatic multispectral sensing images field survey data. All could effectively identify major areas (overall [OA] > 87.40%, kappa coefficient 81.08%). SVM model exhibited best ability (OA = 90.27%, 85.37%), followed by RF 88.90%, Kappa 83.30%). combination vegetation factor performed best, feature, factor. In addition, find that classifier constructed single not effective multiple factors. addition factors can significantly improve identification. According results classifiers, separability was good. producer’s user’s Masson pine were more than 90%, broad-leaved fir 80%. commission errors omission pine. variable importance assessment showed normalized difference greenness altitude, modified soil-adjusted index key variables. this study used accurately main forests China, help managers regularly monitor composition provide theoretical for formulate policies, sustainable plans wood mining, conservation measures.