Dépôt numérique

A Novel approach to monitor chlorophyll-a concentration using an adaptive model from MODIS data at 250 metres spatial resolution.

El Alem, Anas; Chokmani, Karem; Laurion, Isabelle et El Adlouni, Salah-Eddine (2013). A Novel approach to monitor chlorophyll-a concentration using an adaptive model from MODIS data at 250 metres spatial resolution. In: American Geophysical Union Fall Meeting - AGU 2013, 9-13 décembre 2013, San Francisco, États-Unis.

Ce document n'est pas hébergé sur EspaceINRS.


Occurrence and extent of Harmful Algal Bloom (HAB) has increased in inland water bodies around the world. The appearance of these blooms reflects the advanced state of eutrophication of several aquatic systems caused by urban, agricultural, and industrial development. Algal blooms, especially those cyanobacterial origins, are capable to produce and release toxins, threatening human and animal health, quality of drinking water, and recreational water bodies. Conventional monitoring networks, based on infrequent sampling in a few fixed monitoring stations, cannot provide the information needed as HABs are spatially and temporally heterogeneous. Remote sensing represents an interesting alternative to provide the required spatial and temporal coverage. The usefulness of air-borne and satellite remote sensing data to detect HABs was demonstrated since three decades ago, and since several empirical and semi-empirical models, using satellite imagery, were developed to estimate chlorophyll-a concentration [Chl-a] as a proxy to detect bloom proliferations. However, most of those models presented several weaknesses that are generally linked to the range of [Chl-a] to be estimated. Indeed, models originally calibrated for high [Chl-a] fail to estimate low concentrations and vice versa. In this study, an adaptive model to estimate [Chl-a], spread over a wide range of concentrations, is developed for optically complex inland water bodies based on combination of water spectral response classification and three developed semi-empirical algorithms using a multivariate regression. Three distinct water types (low, medium, and high [Chl-a]) are first identified using the Classification and Regression Tree (CART) method performed on remote sensing reflectance over a dataset of 44 [Chl-a] samples collected from Lakes over Quebec province. Based on the water classification, a specific multivariate model to each water type is developed using the same dataset and the MODIS data at 250-m spatial resolution. By pre-clustering inland water bodies, the results were very interesting as the determination coefficients as well as the relative RMSE of the cross-validation were of 0.99, 0.98 and 0.95 and of 0.5%, 8% and 17% for high, medium, and low [Chl-a], respectively. On the other hand, the adaptive model reached a global success rate of 92% using an independent, semi-qualitative, [Chl-a] samples collected over more than twenty inland water bodies for the years 2009 and 2010 over the Quebec province.

Type de document: Document issu d'une conférence ou d'un atelier
Mots-clés libres: concentration de chlorophylle; algues nuisibles; toxine; santé humaine; santé animale;
Centre: Centre Eau Terre Environnement
Date de dépôt: 19 nov. 2020 20:19
Dernière modification: 26 nov. 2020 16:19
URI: https://espace.inrs.ca/id/eprint/4400

Gestion Actions (Identification requise)

Modifier la notice Modifier la notice