Ensemble-Based Systems to Monitor Algal Bloom With Remote Sensing

Abstract

In this paper, an ensemble-based system (EBS) for estimating chlorophyll-a concentrations (Chl-a) in inland water bodies using downscaled MODIS images was developed. Seeking additional opinions before making a decision is part of human nature, particularly for decisions involving health issues. The general concept behind EBS algorithms is based on this principle. Forty-six Chl-a measurements, collected over four water bodies between 2000 and 2008, were used to calibrate the EBS. Measurements ranged from 2.7 (oligotrophic waters) to 91 000 mg Chl-a $\text{m}^{{-3}}$ (hypertrophic waters). The EBS performance was evaluated by cross validation, and also using an independent database (Chl-a ranging from 1 to 14 mg $\text{m}^{{-3}}$ ). Cross-validation results were satisfactory both under high-blooming conditions ( $\text{R}^{2} = 0.98$ , relative RMSE (RMSEr) = 15%, relative Bias (BIASr) = −2%, and relative Nash (NASHr) = 0.95) and at the initialization of blooms ( $\text{R}^{2} = 0.77$ , RMSEr = 37%, BIASr = −8%, and NASHr = 0.70). The EBS also performed well on the independent database ( $\text{R}^{2} =0.93$ , RMSEr = 50%, BIASr = −27%, and NASHr = 0.70). A visual comparison mapping Chl-a on the Missisquoi Bay of Lake Champlain additionally illustrates the potential of the EBS to detect early phases of algal growth contrary to other models (adaptive model, Kharu, and APProach by Elimination). This approach is a proof of concept for a more efficient way to quantify algal biomass from remote sensing and could be exported to any types of satellite imagery in a context of water quality monitoring.