Linear spectral unmixing algorithm for modelling suspended sediment concentration of flash floods, upper Tekeze River, Ethiopia

• Published in: International journal of sediment research 2020-02, Vol.35 (1), p.79-90
• Main Authors: Gebreslassie, Hagos G., Melesse, Assefa M., Bishop, Kevin, Gebremariam, Azage G.
• Format: Article
• Language: English
• Subjects: Empirical remote sensing; Fjärranalysteknik; Flash floods; Linear spectral unmixing; Oceanografi, hydrologi, vattenresurser; Oceanography, Hydrology, Water Resources; Remote Sensing; Suspended sediment concentration; Tekeze River
• ISSN: 1001-6279
• DOI: 10.1016/j.ijsrc.2019.07.007

Flash floods are the highest sediment transporting agent, but are inaccessible for in-situ sampling and have rarely been analyzed by remote sensing technology. Laboratory and field experiments were done to develop linear spectral unmixing (LSU) remote sensing model and evaluate its performance in simulating the suspended sediment concentration (SSC) in flash floods. The models were developed from continuous monitoring in the laboratory and the onsite spectral signature of river bed sediment deposits and flash floods in the Tekeze River and in its tributary, the Tsirare River. The Pearson correlation coefficient was used to determine the variability of correlations between reflectance and SSCs. The coefficient of determination (R2) and root mean square of error (RMSE) were used to evaluate the performance of the generated models. The results found that the Pearson correlation coefficient between SSCs and reflectance varied based on the level of the SSCs, geological colors, and grain sizes. The performance of the LSU model and empirical remote sensing approaches were computed to be R2 = 0.92, and RMSE = ±0.76 g/l in the Tsirare River and R2 = 0.91, and RMSE = ±0.73 g/l in the Tekeze River and R2 = 0.81, RMSE = ±2.65 g/l in the Tsirare river and R2 = 0.76, RMSE = ±10.87 g/l in the Tekeze River, respectively. Hence, the LSU approach of remote sensing was found to be relatively accurate in monitoring and modeling the variability of SSCs that could be applied to the upper Tekeze River basin. •Sedimentation is one of the outermost challenge for water and energy.•Remote sensing is applicable for suspended sediment modelling in data scarce rivers.•Spectral un-mixing analysis improves remote sensing accuracy to model sediment loads.•Linear spectral un-mixing analysis is effective to model sediment load of flood.