ESTIMATION OF MANGROVE FRACTIONAL COVER FROM MULTISPECTRAL AND HYPERSPECTRAL DATA USING MIXTURE TUNED MATCHED FILTERING

Abstract

Mangroves provide various ecosystem services and contribute to climate change adaptation being one of the blue carbon ecosystems. The extents of mangroves are mapped and monitored using commonly available multispectral images, such as Landsat and Sentinel-2, to detect and assess gains and losses. However, this presence or absence per pixel based on crisp classification or index thresholding offers limited information on the dynamics of mangrove growth or decline. In this paper, we evaluated the use of Mixture Tuned Matched Filtering (MTMF) in estimating mangrove fractional cover (MFC) from multispectral (Landsat-8) and hyperspectral (PRISMA) satellite images. We also examined the utility of the mangrove vegetation index (MVI) and enhanced MVI (EMVI) for this purpose. The images were first denoised using Minimum Noise Fraction (MNF). MTMF was then separately applied to the sets of MNF bands, excluding noise bands, to generate Matched Filtering (MF) Score and Infeasibility layers. The endmember (mangrove) spectrum was extracted from a pixel identified using pixel purity index (PPI) and examination of high-resolution Google Earth base image, from which detailed mangrove extents were also delineated. A 30-m vector grid file was created and populated with MFC, MF Score, and Infeasibility values using zonal analysis. Correlation analysis, exploratory regression, and ordinary least squares (OLS) regression were performed. MF Score is moderately and positively correlated with MFC. In contrast, Infeasibility, MVI, and EMVI are uncorrelated or very weakly correlated with MFC. MFC can be estimated using an OLS model with MF Score and Infeasibility as explanatory variables. The performance of the PRISMA-based model (R2Adj=0.30, AIC=98643.20) was found to be better than the Landsat-8-based model (R2Adj=0.36, AIC=97428.89).