A multi-stage framework for classifying complex coastal environments using Sentinel-2 data and the DEVWE method

Technical University of Varna, Department of Communication Engineering and Technologies, 9000, Varna, Bulgaria

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Accurate and detailed shoreline classification remains challenging due to complex transition zones, mixed pixels, and high spatial heterogeneity. The main objective of this study is to develop and evaluate a hierarchical, multi-stage methodology for accurately mapping coastal zones into seven classes. The framework integrates the Dynamic Ensemble Voting for Water Extraction (DEVWE) algorithm with strict spatial constraints. It involves delineating an initial shoreline using a global binary DEVWE classification, which is subsequently refined within a localized 200-meter buffer zone. This buffer is divided into separate water and land subpolygons. Localized One-vs-All ensemble classifiers are deployed independently to separate shallow from deep water in the water segment, and to classify five land classes (wet sand, dry sand, vegetation, buildings, and rocks) in the land segment. These independent results are synthesized into a unified thematic map with seven classes. The performance of the proposed model was evaluated using 5-fold cross-validation, yielding an overall accuracy of 0.7996 and a macro F1-score of 0.5656. A key advantage of this spatial hierarchy is the accurate delineation of transition zones during the DEVWE stages with limited buffers, which effectively mitigates the confusion between classes. By overcoming the limitations of conventional single-stage approaches, the modular methodology allows for independent optimization of each stage and demonstrates strong potential for operational regional coastal monitoring.