Optimization of a Semi-Analytical Algorithm for Multi-Temporal Water Quality Monitoring in Inland Waters with Wide Natural Variability

• Published in: Remote sensing (Basel, Switzerland) Vol. 7; no. 12; pp. 16623 - 16646
• Main Authors: Bramante, James, Sin, Tsai
• Format: Journal Article
• Language: English
• Published: MDPI AG 01-12-2015
• Subjects: hyperspectral; inland waters; semi-analytical bio-optical algorithm; Singapore; water quality monitoring
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs71215845

Current spectrometer design and the increasingly affordable price of field hyperspectral sensors are making feasible their use for water quality monitoring. In this study, we parameterized a semi-analytical algorithm to derive constituent concentrations from field spectroradiometer measurements in ten freshwater reservoirs over two years. In contrast to algorithms parameterized for single airborne or satellite sensor deployments, we optimized the algorithm for robust performance across all reservoirs and for multi-temporal application. Our algorithm produced chlorophyll-a concentration estimates with a root mean squared error (RMSE) of 7.7 mg∙m−3 over a range of 4–135 mg∙m−3. The model also produced estimates of total suspended solids (TSS) concentration with an RMSE of 4.0 g∙m−3 over a range of 0–25 g∙m−3. Choosing a non-linear objective function during inversion reduced variance of residuals in chlorophyll-a and TSS estimates by 20 and 18 percentage points, respectively. Application of our algorithm to two years of data and over ten study sites allowed us to specify sources of suboptimal parameterization and measure the non-stationarity of algorithm performance, analyses difficult for short or single deployments. Suboptimal parameterization, especially of backscatter properties between reservoirs, was the greatest source of error in our algorithm, accounting for 17%–20% of all error. In only one reservoir was time-dependent error apparent. In this reservoir, decreases in TSS over time resulted in less TSS estimate error due to imperfect model parameterization. For future applications, especially with ground-based sensors, model performance can easily be improved by using non-linear inversion procedures and replicating spectral measurements.