Analysis of multitemporal MODIS and landsat 7 images acquired over amazonian floodplains lakes for suspended sediment concentrations retrieval

Abstract

The objective of this paper is to analyze the spatial and temporal variabilities of the water signatures as a function of sediment concentration at the surface. Images acquired simultaneously with 3 campaigns of water measurements, conducted from high water stage to low water stages, are analyzed and discussed. The potential of ETM+ and MODIS for sediment retrieval appears clearly, especially in the red bands, for concentrations in the range of 0 to 100 mg/l, commonly encountered during most of the hydrological cycle. First retrieval attempts give RMSE errors of about 10 mg/l. We show also that phytoplancton blooms, occurring over large areas during decreasing water levels, alter especially ETM+ and should be first detected to separate them from the rest of the retrieval. These results show that Landsat and especially MODIS, because of his higher revisit frequency, are very promising for surface sediment retrieval over the large lakes of the Amazon floodplain. If accurate estimates of the Amazon mean annual liquid flows exist, intra and inter annual variations of the solid flows as well as the mechanisms driving the sediment transport from the Andes to the Atlantic are only partially characterized. Sediment transport is followed in the Amazon by a sparse network of stations, regularly collecting water samples along the river mainstreams. The case of the floodplain is of special interest, covering 300000 km² at high water, it stocks large volumes of waters, and catch temporally or definitely great amounts of sediments. The floodplain are considered as net sinks of sediments but the rates of sedimentation are not yet assessed, and thus the influence of the floodplain under the sediment transport into the Amazon mainstream remain almost unquantified. A dense network of stations measuring sediment concentrations in the floodplain would be required, but the spatial and temporal heterogeneity of the floodplain, constituted of thousands of lakes, makes it clearly unaffordable. The solution is to develop new techniques, among which earth observation satellite is very promising. Use of satellite data to interpret water color in terms of chlorophyll concentration and infer primary production is routinely used over ocean waters. Recent development aimed at develop algorithms to retrieve in turbid coastal waters both chlorophyll content and total suspended sediment (TSS) concentrations. However, very few works dealt already with the use of satellite data to monitor water quality in the Amazon basin, and these few works (1), lacked dramatically of in situ measurement to validate their conclusions. The objective of this study is to develop a robust tool to monitor TSS for the large inland freshwater systems of the Amazon floodplain. This article presents, in particular, the analysis of remote sensing data using a large dataset of water quality measurements, to assess the spatial and temporal variations of water signatures. Those data were acquired within the HYBAM program (Hydrology and Geochemistry of the Amazon basin), developed by the French Institute for the Development (IRD) for 20 years in the whole Amazon basin.