Development of a regional bio-optical model for water quality assessment in the US Virgin Islands

Loading...
Thumbnail Image
Date
Authors
Kerrigan, Kristi
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Previous research in the US Virgin Islands (USVI) has demonstrated that land-based sources of pollution associated with watershed development and climate change are local and global factors causing coral reef degradation. A good indicator that can be used to assess stress on these environments is the water quality. Conventional assessment methods based on in situ measurements are timely and costly. Satellite remote sensing techniques offer better spatial coverage and temporal resolution to accurately characterize the dynamic nature of water quality parameters by applying bio-optical models. Chlorophyll-a, suspended sediments (TSM), and colored-dissolved organic matter are color-producing agents (CPAs) that define the water quality and can be measured remotely. However, the interference of multiple optically active constituents that characterize the water column as well as reflectance from the bottom poses a challenge in shallow coastal environments in USVI. In this study, field and laboratory based data were collected from sites on St. Thomas and St. John to characterize the CPAs and bottom reflectance of substrates. Results indicate that the optical properties of these waters are a function of multiple CPAs with chlorophyll-a values ranging from 0.10 to 2.35 μg/L and TSM values from 8.97 to 15.7 mg/L. These data were combined with in situ hyperspectral radiometric and Landsat OLI satellite data to develop a regionally tiered model that can predict CPA concentrations using traditional band ratio and multivariate approaches. Band ratio models for the hyperspectral dataset (R2 = 0.35; RMSE = 0.10 μg/L) and Landsat OLI dataset (R2 = 0.35; RMSE = 0.12 μg/L) indicated promising accuracy. However, a stronger model was developed using a multivariate, partial least squares regression to identify wavelengths that are more sensitive to chlorophyll-a (R2 = 0.62, RMSE = 0.08 μg/L) and TSM (R2 = 0.55). This approach takes advantage of the full spectrum of hyperspectral data, thus providing a more robust predictive model. Models developed in this study will significantly improve near-real time and long-term water quality monitoring in USVI and will provide insight to factors contributing to coral reef decline.
Description
Keywords
Citation