Modèlisation de la variabilité spatiale de l'humidité du sol à partir d'imagerie radar à synthèse d'ouverture

Abstract

Soil moisture content is a key variable in many environmental sciences. Even though it represents a small proportion of the liquid freshwater on Earth, it modulates interactions between the land surface and the atmosphere, thereby influencing climate and weather. Accurate modeling of the above processes depends on the ability to provide proper spatial characterization of soil moisture. In this dissertation we report on an investigation of the possibility of that the spatial variability of soil moisture can be evaluated using Radarsat-1 C-band SAR data. The data were acquired during five different periods over the Roseau River watershed in southern Manitoba, Canada. For validation purposes, ground measurements were carried out at 62 locations simultaneously with the satellite data acquisitions. The first step in this analysis was to assess the potential of wide mode acquisitions from the Radarsat1 SAR to provide reliable measurements of the rms height surface roughness parameter. It was found that the best estimation occurred when a constraint on local incidence angle was introduced. Without this constraint, none of the used models (OM: Oh Model and IEM: Integral Equation Model) provided satisfactory predictions. In order to reduce the surface roughness effect on radar backscatter behavior, the semiempirical calibration technique of IEM as proposed by Baghdadi et al. (2004) was implemented. Overall, a better agreement was found between the calibrated model results and SAR-based backscatter coefficients compared to simulation results based on the original IEM version. Soil moisture maps were then computed by inversion of the calibrated IEM based on a simplex algorithm routine. Derived spatial patterns of near-surface moisture content were then examined using an exponential semivariogram model for spatial extents ranging from tens of meters to kilometers. Despite quite different soil proprieties, similarities were found between the semivariogram range results (~ 100 m) and those reported in the literature. Our scaling analysis of near-surface moisture images confirms the existence of power law decay between the variance and the increasing scale over the range of log scale of 3 to 4.75 (R > 0.94). The slopes of the corresponding fitted lines were found to be in the range [-0.27, -0.51], which indicates the existence of a clustering effect. Combining these results with those obtained by passive microwaves, one could infer that the power law decay of the soil moisture moments could be valid over an extended log scale ranging from 3 to 7.5. To investigate the multi-scaling properties of the 5Ai?-based soil moisture images, moment analyses were performed. A consistent linear log-log dependency of the higher statistical moments on cell resolution was found over different moisture regimes, yielding good coefficients of determination (R > 0.90). The relationship between the regression slope of this power law decay and the moment order exhibits a concave shape, clearly showing a multi-fractal pattern in soil moisture during all acquisition periods.