Analysis of Detailed Wetland Land Mapping Using Landsat Imagery: A Hybrid Segmented PCA and Machine Learning Approach Integrated with Spectral Indices

Lakes serve as vital ecological assets, sustaining biodiversity, supporting human activities, and maintaining environmental balance. Effective mapping of land cover types is essential for assessing ecological health and monitoring biodiversity dynamics in such ecosystems. This study focuses on Pulicat Lake, the second-largest brackish water lagoon in India, which represents a complex and dynamic environment influenced by both riverine and marine inputs. A methodological framework integrating Segmented Principal Component Analysis (SPCA), Base maps for Water, Vegetation, Soil created using SVM algorithm and spectral indices was analyzed in enhanced land cover mapping accuracy using the RF algorithm in multispectral Landsat 8 imagery. The SPCA method selectively applies PCA to targeted spectral band combinations for specific land cover types, thereby improving spectral separability of the classes. The SPCA/RF/spectral indices approach yielded a significantly improved classification accuracy of 97.52% and a Kappa coefficient of 0.96. These results highlight the effectiveness of the SPCA framework with an RF classifier and integration of spectral indices in land cover mapping and ecosystem monitoring in complex and dynamic environments like Pulicat Lake.