Comparing Machine and Deep Learning Methods for the Phenology-Based Classification of Land Cover Types in the Amazon Biome Using Sentinel-1 Time Series

The state of Amapá within the Amazon biome has a high complexity ecosystems formed by forests, savannas, seasonally flooded vegetation, mangroves, and different land uses. present research aimed to map vegetation from phenological behavior Sentinel-1 time series, which advantage not having atmospheric interference cloud cover. Furthermore, study compared three sets images (vertical–vertical co-polarization (VV) only, vertical–horizontal cross-polarization (VH) both VV VH) classifiers based on deep learning (long short-term memory (LSTM), Bidirectional LSTM (Bi-LSTM), Gated Recurrent Units (GRU), GRU (Bi-GRU)) machine (Random Forest, Extreme Gradient Boosting (XGBoost), k-Nearest Neighbors, Support Vector Machines (SVMs), Multilayer Perceptron). series englobed four years (2017–2020) with 12-day revisit, totaling 122 for each VH polarization. methodology presented following steps: image pre-processing, temporal filtering using Savitsky–Golay smoothing method, collection samples considering 17 classes, classification methods polarization datasets, accuracy analysis. combinations pooled dataset Neuron Networks led greatest F1 scores, Bi-GRU (93.53) Bi-LSTM (93.29), followed other methods, (93.30) (93.15). Among learning, two highest F1-score values were SVM (92.18) XGBoost (91.98). Therefore, variations long Synthetic Aperture Radar (SAR) allow detailed representation cover/land use water dynamics.