Comparison of corrected surface reflectance, principal component analysis, and kernel principal component analysis for seagrass percent cover mapping using Sentinel-2 based on various regression models

Variations in seagrass percent cover is one of the key indicators of the health of a seagrass ecosystem. Given that seagrass is regarded as one of the most effective and efficient nature-based solutions for the mitigation and adaptation to climate change, there is a clear need for spatially explicit and extensive information on seagrass percent cover in order to inform management, restoration and conservation efforts. This paper presents a comparison of the accuracy of seagrass percent cover mapping derived from the sunglint and water column corrected surface reflectance (SR), principal component analysis (PCA), and kernel principal component analysis (KPCA) bands of Sentinel-2 L2A. The comparison is based on the application of various regression models, including stepwise regression (SWR), automated partial least squares regression (AutoPLSR), random forest regression (RFR), and support vector machine regression (SVR). The metrics for evaluating the accuracy of different input-model combinations are R², RMSE, and plot 1:1 between reference and estimated percent cover. The findings indicate that incorporating PCA or KPCA into deglint bands and DII within the SWR, SVR, RFR, and AutoPLSR models has varying effects on accuracy. Notably, accuracy consistently improved when deglint bands or DII were combined with both PCA and KPCA, suggesting that this combination is preferable for enhancing model performance and mapping accuracy. © 2025 SPIE.