Mangrove taxon mapping from Sentinel-1/2 using a compact one-dimensional network with probability calibration, predictive entropy, and frozen transfer evaluation
4 affiliations
- 1College of Artificial Intelligence, Hainan Normal University
- 2Key Laboratory of Data Science and Smart Education, Hainan Normal University, Ministry of Education
- 3Key Laboratory of the Ministry of Education for Coastal and Wetland Ecosystems, College of the Environment and Ecology, Xiamen University
- 4College of Life Sciences, Hainan Normal University
Abstract
Introduction: Mangrove extent maps lack the taxonomic information needed for carbon assessment, restoration evaluation, and management. Taxon mapping at 10 m remains constrained by spectral similarity, mixed pixels, limited reference coverage, and spatial dependence.
Methods: We developed the Inverted-Residual Spectral One-Dimensional Convolutional Neural Network (IR-Spec1D) as a compact classifier for a 14-feature Sentinel-1/2 optical and radar vector. To capture nonlinear optical–radar relationships compactly, inverted-residual and depthwise-separable transformations learn feature interactions, while Efficient Channel Attention emphasizes informative learned channels. Spectral-Domain Dropout reduces dependence on individual inputs through training-time feature masking. The workflow combines this configuration with spatial evaluation, post hoc temperature scaling, predictive-entropy analysis, and frozen transfer assessment. The six-category source taxonomy includes the B. sexangula / B. gymnorhiza complex. Six classifiers were evaluated using April 2024 observations from the selected Dongzhai Harbor study area.
Results: IR-Spec1D achieved the highest mean overall accuracy (0.9720) and macro-F1 (0.9406) under five spatially grouped cross-validation folds. Performance remained comparatively high with a 100-m exclusion buffer. Held-out source-domain diagnostics showed that temperature scaling reduced mean expected calibration error from 0.0346 to 0.0125 without changing categorical predictions. Predictive entropy ranked cross-validated source-domain errors with an AUROC of 0.8808. The frozen ensemble retained an overall accuracy of 0.9450 in both same-site temporal evaluations, whereas four-category closed-set geographic transfer to Gaoqiao yielded 0.7118.
Discussion: The combined evidence characterizes taxon discrimination, held-out probability behavior, and performance retention under specified transfer conditions within a common mangrove-mapping workflow.
Related
- Beyond Discrimination: Evaluating Probability Calibration in Radiomics-Image Fusion for Renal Cell Carcinoma Classification
- Reliability-Centered Evaluation of Sparse Longitudinal CT Lesion-Size Forecasting with Conformal Interval Calibration and Gompertz-Inspired Regularization
- DCAF: Unifying Data Validity, Evaluation Validity and Predictive Modeling for Single-cell Perturbation Simulation
- Panvascular-JEPA: Unseen-Structure Evaluation of Factorized Predictive States for Hemodynamic Networks
- Predictive comparison and evaluation of ANN and ANFIS as effective tools in modeling prodigiosin production by Serratia marcescens ND93
BibTeX
@article{liu2026mangrove,
title = {{Mangrove taxon mapping from Sentinel-1/2 using a compact one-dimensional network with probability calibration, predictive entropy, and frozen transfer evaluation}},
author = {Guohao Liu and Shulei Wu and Yaoru Wang and Yihui Zhang and Huandong Chen and Yukai Chen and Zhongqiang Wu and Nan Wang},
journal = {Frontiers in Plant Science},
year = {2026},
month = oct,
doi = {10.3389/fpls.2026.1967679},
url = {https://www.frontiersin.org/articles/10.3389/fpls.2026.1967679/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Mangrove taxon mapping from Sentinel-1/2 using a compact one-dimensional network with probability calibration, predictive entropy, and frozen transfer evaluation
AU - Guohao Liu
AU - Shulei Wu
AU - Yaoru Wang
AU - Yihui Zhang
AU - Huandong Chen
AU - Yukai Chen
AU - Zhongqiang Wu
AU - Nan Wang
PY - 2026
DA - 2026/10/16
JO - Frontiers in Plant Science
DO - 10.3389/fpls.2026.1967679
UR - https://www.frontiersin.org/articles/10.3389/fpls.2026.1967679/full
AB - Introduction: Mangrove extent maps lack the taxonomic information needed for carbon assessment, restoration evaluation, and management. Taxon mapping at 10 m remains constrained by spectral similarity, mixed pixels, limited reference coverage, and spatial dependence. Methods: We developed the Inverted-Residual Spectral One-Dimensional Convolutional Neural Network (IR-Spec1D) as a compact classifier for a 14-feature Sentinel-1/2 optical and radar vector. To capture nonlinear optical–radar relationships compactly, inverted-residual and depthwise-separable transformations learn feature interactions, while Efficient Channel Attention emphasizes informative learned channels. Spectral-Domain Dropout reduces dependence on individual inputs through training-time feature masking. The workflow combines this configuration with spatial evaluation, post hoc temperature scaling, predictive-entropy analysis, and frozen transfer assessment. The six-category source taxonomy includes the B. sexangula / B. gymnorhiza complex. Six classifiers were evaluated using April 2024 observations from the selected Dongzhai Harbor study area. Results: IR-Spec1D achieved the highest mean overall accuracy (0.9720) and macro-F1 (0.9406) under five spatially grouped cross-validation folds. Performance remained comparatively high with a 100-m exclusion buffer. Held-out source-domain diagnostics showed that temperature scaling reduced mean expected calibration error from 0.0346 to 0.0125 without changing categorical predictions. Predictive entropy ranked cross-validated source-domain errors with an AUROC of 0.8808. The frozen ensemble retained an overall accuracy of 0.9450 in both same-site temporal evaluations, whereas four-category closed-set geographic transfer to Gaoqiao yielded 0.7118. Discussion: The combined evidence characterizes taxon discrimination, held-out probability behavior, and performance retention under specified transfer conditions within a common mangrove-mapping workflow.
SN - 1664-462X
ER -