Evaluation Of Remote Sensing Based Lake Bathymetry For Sediment Yield Estimation: The Case Study Of Lake Hawassa.

Abstract

Sedimentation in Lake Hawassa is a major problem for lake level rise and reduction in storage capacity. To overcome this problem sediment yield estimation plays an important role. In order to estimate the amount of sediment yield, it needs an accurate and continuous bathymetry survey. Moreover, the lack of a continuous bathymetry survey of the lake is a major factor for managing and studying morphometries of the lake. Nowadays, Remote sensing technology had led to overcome this problem if used for depth estimation and applied to sediment yield estimation. However, remote sensing was not implemented before in retrieving the bathymetry of Lake Hawassa. So, the aim of this study was to evaluate the accuracy of depth estimated from remote sensing data and applying to sediment yield estimation. Landsat 7 ETM+ with a resolution of 30 m data was used for depth estimation. While Lyzenga or linear band model were used for depth estimation. Preprocessing activities were made on secondary data to obtain corrected surface reflectance. Corrected reflectance values were log-transformed and Multiple linear regression is carried out by using an input 70% of in-situ measurement and corresponding reflectance values to get depth estimating model parameters for construction. The model performance and accuracy in estimating depth were assessed and validated by quantifying the error between 30% of the in-situ field measured depth not used in model construction and corresponding estimated depths. The result indicates that the correlation coefficient (R = 0.7), Mean Absolute Error (MAE = 3.06) and Root Mean Square Error (RMSE = 3.69 m) for testing data set (30% of in-situ depth). The result showed Lyzenga or linear model has less performance in estimating depth less than 5 m and depth greater than 20 m. Moreover, the result was statistically unsatisfactory when the model is calibrated for the whole area at once. Relying on the estimated depths, 3D bed surface of 2011 and for the year 1999 based on generated bed elevation from bed surface topographic maps by three interpolation methods using 50% of bed elevation. The result of statistical indicators for Empirical Bayesian Kriging was (Mean Absolute Error =0.027 m and Root Mean Square Error = 0.096 m) for the year 1999 and (Mean Absolute Error =1.069 m and Root Mean Square Error = 1.403 m) for the year 2011 respectively lower than the Inverse Distance Weighting and Radial Basis Function method of interpolation. The result showed Empirical Bayesian Kriging has better performance in generating 3D bed surface of Lake Hawass. Using the two generated 3D bed surface by Empirical Bayesian Kriging, a map of bed changes were created to estimate sediment yield between the years 1999 and 2011. The result of sediment yield estimated between 1999-2011 were 94.1 million m3. The sediment yield by the present approach were compared with the in-situ based depth 3D bed surface of the lake. The result showed that the estimated sediment yield relying on the estimated depth by the present approach (from remote sensing data) overestimated the result by 24.1 million m3 compared to the result derived from the generated 3D bed surafce of the lake based on the field measurements. Therefore, the study concludes estimating depth from remote sensing data by applying a linear model for the whole area at once and applying the estimated depth for sediment yield estimation overestimates and gives unsatisfactory result. Finally, the study recommends Lyzenga or linear model could not be used for depth estimation and applied for sediment yield estimation. Additionally, Empirical Bayesian Kriging can yield better accuracy in the generation of the 3D bed surface of Lake Hawassa and can be used in the future.