An Adaptive Model to Monitor Chlorophyll-a in Inland Waters in Southern Quebec Using Downscaled MODIS Imagery

Abstract

The purpose of this study is to assess the performance of an adaptive model (AM) in estimating chlorophyll‑a concentration (Chl‑a) in optically complex inland waters. Chl‑a modeling using remote sensing data is usually based on a single model that generally follows an exponential function. The estimates produced by such models are relatively accurate at high Chl‑a concentrations, but accuracy drops at low concentrations. Our objective was to develop an approach combining spectral response classification and three semi-empirical algorithms. The AM discriminates between three blooming classes (waters poorly, moderately, and highly loaded in Chl‑a), with discrimination thresholds set using the classification and regression tree (CART) technique. The calibration of three specific estimators for each class was achieved using a multivariate stepwise regression. Compared to published models (Floating Algae Index, Kahru model, and APProach by ELimination) using the same data set, the AM provided better Chl‑a concentration estimates (R2 of 0.96, relative RMSE of 23%, relative Bias of −2%, and a relative NASH criterion of 0.9). Moreover, the AM achieved an overall success rate of 67% in the estimation of blooming classes (corresponding to low, moderate, and high Chl‑a concentration classes). This was done using an independent data set collected from 22 inland water bodies for the period 2007–2010 and for which the only information available was the blooming class.

Keywords

• remote sensing
• MODIS
• inland waters
• HABs
• Chl‑a
• classification
• CART
• multivariate regression
• stepwise