An ensemble based system for Chlorophyll-a estimation using MODIS imagery over Southern Quebec inland waters

Abstract

The purpose of this study was to enhance the performance of the Adaptive-model (AM) for Chlorophyll-a (Chl-a) concentration estimation using MODIS-imagery. The AM is based on the combination of spectral response classification and three semi-empirical estimators. Most critical step in Chl-a estimation when using the AM is the estimator selection. A wrong selection could lead to an over- or underestimation and induces a staircase effect in Chl-a modelization. Ensemble based systems (EBS) development can be in interesting solution. One of the most used techniques to develop an EBS is bagging-algorithm. However, its main weakness is time consumption. The Gaussian-quadrature (GQ) formula has the potential to handle this kind of problems. The objective of this study is to develop a mixed ensemble based system for Chl-a estimation in Quebec inland waters using GQ formula. Statistical indexes evaluation was satisfied, relative-RMSE=15% and R2=0.98 whereas matrix-confusion results were 0.92 and 0.8 for global-success and Kappa-index, respectively.