Spectral Unmixing -And Integrated Hydrologic Model for Sediment Estimation, Evaluation of Climate Change Impact and Management Types

No Thumbnail Available



Journal Title

Journal ISSN

Volume Title


Addis Ababa University


Remote sensing is a less costly and reasonably accurate technology for monitoring and modelling river systems. However, the coarseness of remote sensing data together with the dynamic inherent optical properties of the variables constrained its application. Thus, the overall purpose of this research was to propose the spectral unmixing approach for building remote sensing model and parametrizing the physically distributed hydrologic model, soil and water assessment tool (SWAT), for estimating sediment concentration and evaluating impact of climate and sediment management changes. Laboratory, time series in-situ and space remote sensing data analyses were triangulated to construct linear spectral unmixing analysis (LSUA) and compared with the conventional (empirical) remote sensing models. The models were constructed in laboratory experiments using sediment types sampled from the Tekeze and Tsirare Riverbeds deposited sediments and then tested to estimate the daily SSCs from in-situ and space remote sensing data and evaluated against the observed SSCs in the Rivers. To enhance the LSUA model accuracy into the mixed pixels of the moderate-resolution imaging spectroradiometer (MODIS) images, a new approach called Double-stage-LSUA (DLSUA) was proposed. In this case, LSUA was applied at two stages that LSUA in the first stage was used to unmix the pixels‘ reflectance into respective macro endmembers‘ (rock\bare-land and turbid water) reflectance and the LSUA in the next stage was used to determine spectral mixing coefficients (SMCs) of the constituents in the turbid water (micro components including pure water and sediment) was proposed. Finally, the SSCs of the Rivers were simulated by inserting the computed SMCs into the LSUA model generated in the laboratory. The LSUA approach was also tested to monitor the spatial variability of a vegetation parameter of soil erosion and sediment (C-factor) which is the required parameter in most sediment estimating hydrologic models. The spatial minimum C-factor of the upper Tekeze River basin was first mapped using the LSUA technique from the Landsat images and tested its accuracy using time series field monitoring. Average C-factor was integrated into hydrological response units (HRUs) of SWAT. This differs from the conventional approach where the C-factors have been integrated into land-use type units of SWAT. The LSUA integrated SWAT was demonstrated in evaluating climate and sediment management change scenarios on sediment yield. The goodness-of-fit indices including Nash-Sutcliffe coefficient (NSE), Coefficient of determination (R2 ), Root Mean Square of Error (RMSE), Root mean Square of error- observations standard deviation Ratio (RSR) and Percent Bias (PBIAS) were used to evaluate the performance of the model outputs. The application of LSUA approach to finer (ground) and coarser (MODIS) resolutions remote sensing data for modelling variability of SSCs of the Tekeze Rivers were performed averagely at R 2 = 0.92 with RMSE = ±0.75g/l and R 2 = 0.83 with RMSE ±9.96, respectively. These performances were relatively good compared to the simulations using the conventional empirical regression remote sensing model performed at R2 =0.78 with ix RMSE = ±6.76g/l and R2 = 0.74 with RMSE ±16.2, respectively. The success of applying the LSUA approach was not only for the direct estimation of SSCs, but it was also successful for determining the spatial variation of C-factor values within and among landuse types. The demonstration in the upper Tekeze basin showed that the use of the minimum C-factor map produced using LSUA and integrating it into HRUs of SWAT improved the fit between the predicted and the measured sediment yield. The coefficients including NSE, PBIAS, RSR and R2 for sediment yield were 0.72, 0.39, 34.2 & 0.68, respectively, when the C-factor values were for the land-use type units of SWAT. When the C-factor was for the HRUs in SWAT, the corresponding values were 0.84, 0.23, 10, & 0.89. The average rainfall and temperature over the basin experienced neither significant increasing nor decreasing trends in the time scales. In contrast, trend analyses of different variables on the simulated sediment yield from the upper Tekeze basin have shown a significant increasing trend. Moreover, the sediment concentration simulation using the LSUA-SWAT shows that applying filter strips, stone bunds, and reforestation or integration of these scenarios reduced the current sediment yields by different rates (Ave. 9-38%). The LSUA approach has found to be effective in generating relatively accurate and universal models working with both ground-based (finer-resolution), and space-based (coarser-resolution) remote sensing data from river systems. LSUA was also effective in determining the variability of C-factor among and within landuse types. The successful integration of the C-factor values into HRUs enhances the sensitivity of SWAT to the spatial variability of C-factor and then sediment yield. Therefore, the current study implied that prior studying and considering the inherent optical properties of endmembers during analysis is important to enhance remote sensing technology for modelling and monitoring sediment concentrations. The continuous and significant increasing trend of sediment concentration in the basin irrespective of the insignificant and non trending changes of climate variables has implied that the changes in catchment characteristics over time including changes in land use and/or land cover in the basin are the governing factors. Moreover, the sediment yield in the basin varies with the changes in sediment management type. Hence, though further calibration and validation are needed, the LUSA and its integration to hydrologic models (eg. SWAT) approach can support decision-making concerning the SSCs variability and impacts of climate change and management alternatives at the river basin scale better than the conventional approaches.



C-factor, Empirical remote sensing, Linear unmixing, Sediment monitoring, Sediment type, Spectral mixing coefficient, suspended sediment concentration, Tekeze River basin