A Comparison of Machine Learning Models for Mapping Tree Species Using WorldView-2 Imagery in the Agroforestry Landscape of West Africa

Farmland trees are a vital part of the local economy as used by farmers for fuelwood well food, fodder, medicines, fibre, and building materials. As result, mapping tree species is important ecological, socio-economic, natural resource management. The study evaluates very high-resolution remotely sensed WorldView-2 (WV-2) imagery classification in agroforestry landscape Kano Close-Settled Zone (KCSZ), Northern Nigeria. Individual crowns extracted geographic object-based image analysis (GEOBIA) were to identify nine dominant (Faidherbia albida, Anogeissus leiocarpus, Azadirachta indica, Diospyros mespiliformis, Mangifera Parkia biglobosa, Piliostigma reticulatum, Tamarindus Vitellaria paradoxa) at object level. For every reference datasets, eight original spectral bands WV-2 image, their statistics (minimum, maximum, mean, standard deviation, etc.), spatial, textural, color-space (hue, saturation), different vegetation indices (VI) predictor variables species. Nine machine learning methods object-level classification. These Extra Gradient Boost (XGB), Gaussian Naïve Bayes (GNB), Boosting (GB), K-nearest neighbours (KNN), Light Machine (LGBM), Logistic Regression (LR), Multi-layered Perceptron (MLP), Random Forest (RF), Support Vector Machines (SVM). two top-performing models terms highest accuracies individual found be SVM (overall accuracy = 82.1% Cohen’s kappa 0.79) MLP 81.7% with lowest numbers misclassified compared other methods.