Quantifying scattering characteristics of mangrove species from Optuna-based optimal machine learning classification using multi-scale feature selection and SAR image time series

• DOI: 10.1016/j.jag.2023.103446
• Publication Date: 2023-08-07T10:22:43Z
• Authors: Bolin Fu; Yiyin Liang; Zhinan Lao; Xidong Sun; Sunzhe Li; Hongchang He; Weiwei Sun; Donglin Fan
• Keywords: Hyperparameter

Mangroves play a significant role in carbon sequestration and storage. Mapping mangrove species monitoring their conditions have been crucial issue for achieving sustainable development goals. Currently combing multidimensional optical SAR images with machine learning become an important approach classification, but there are still some challenges feature selection hyperparameter optimizations. In this study, we proposed novel classification framework by multi-scale variable algorithm (MUVR) state-of-the-art optimization method (Optuna) mapping the Beilun Estuary Maowei Sea nature reserves using dual-polarization images, further quantified scattering characteristics of image time series. We found that: (1) The MUVR could determine optimal scale features different scenarios species, improve performance overall accuracy (OA) improvement 12.85%; (2) Optuna-based CatBoost outperforms LightGBM NGBoost algorithms which achieved highest OA (93.18%). This study demonstrated that was suitable identifying Aegiceras corniculatum, while discriminating Avicennia marina, Bruguiera gymnorrhiza, Cyperus malaccensis, Kandelia candel Sonneratia apetala; (3) its derivatives improved identification ability collaboration multispectral SAR-derived produced better classification; (4) From 2018 to 2020, backscattering coefficients VV VH polarization focused on 0.053–0.327 0.015–0.062, respectively. coherence mangroves displayed seasonal change trend large variations summer small winter. range Entropy Alpha from 0.65 0.88 17–33, indicated main mechanism moderate random surface scattering.