Abstract
This thesis addresses the application of atmospheric correction algorithms in the satellite remote sensing of reservoirs. The major aims of the project were, first to identify the problem of atmospheric effects on satellite images, second to review the available atmospheric correction methods, third to apply the image-based techniques and methods based on atmospheric modelling to time-series Landsat-5 TM images of large reservoirs in the Lower Thames Valley and, fourth, to evaluate the effectiveness of these methods in this application.The wide range of digital numbers observed in a time series of masked images of eutrophic reservoirs in the Lower Thames Valley demonstrated The important time-dependent contribution of atmospheric effects. Uncorrected at-satellite reflectance values for the waters in these reservoirs were found to be dominated by atmospheric effects, which accounted for between 63 % to 100 % of the at-satellite reflectance in the visible and near infrared bands. This confirms the importance of removing atmospheric effects from dark targets before attempting to correlate physical data with satellite measurements. Evaluation of the atmospheric correction techniques was achieved by comparing the atmospheric corrected reservoir reflectance data with the spectro-radiometric measurements acquired in-situ using a GER1500 field spectro-radiometer and with other reflectance values found in the literature. It was found that the darkest pixel method was the most suitable technique in the Landsat-5 Thematic Mapper (TM) bands 1, 2 and 3 for removing atmospheric effects from satellite images which include reseryoirs. The darkest pixel (DP) method is a fully image-based technique, and is simpler in adaptation and easier computationally than other more sophisticated atmospheric correction algorithms. Using data collected on the Lower Thames reservoirs, it has been shown that accounting for atmospheric effects in Landsat TM images enables the development of water quality predictive equations. The atmospherically corrected (using DP) satellite reflectance values were found to improve significantly the correlations with chlorophyll-a (chl-a) and particulate organic carbon (POC). The best correlations for predicting chl-a and POC were found in the TM 1 and TM 2 bands respectively. Concerning the use of eutrophic reservoirs as pseudo-invariant targets, it was found that (a) the low spectral reflectance values, (b) the negligible spatial variability within reservoirs, (c) their generally large size and distinctive shape and (d) their high turbidity make them suitable dark pseudo-invariant targets. Values of ground reflectance for eutrophic reservoirs are provided for the user to evaluate atmospheric effects in other studies. A new atmospheric correction method has been developed based on the use of a very bright object (concrete airport aprons) and a very dark target (eutrophic water storage reservoir). The new method has been applied to images of the Lower Thames Valley and sites in Greece and Cyprus. Two novel methods of assessing atmospheric pollution based on the use of eutrophic reservoirs as-dark targets are introduced and have been applied at two sites.