Towards Improved Hydrologic Forecasting and Hydropower Planning In Data-Scarce Regions Using Remote Sensing

Akash Koppa

Civil and Environmental Engineering Department, UCLA

Tuesday, October 23, 12pm -12:50pm
Boelter Hall 5264

Water resources management in regions such as East Africa and South Asia is hindered by acute scarcity of hydrologic information. For example, most existing hydropower reservoirs in sub-Saharan Africa exhibit suboptimal performance (as low as 30% utilization in many cases). The lack of hydrologic measurements prevents the implementation of reliable Decision Support Systems (DSS) to optimize these reservoirs. In this regard, remote sensing and satellite-based observations of hydrologic fluxes combined with ensemble meteorological forecasts and distributed hydrologic models have the potential to alleviate the issue of data scarcity. We make novel contributions to the following stages of streamflow forecasting and hydropower planning leveraging satellite-based hydrologic measurements: 1) Validation of remote sensing data, 2) Calibration of hydrologic models using remote sensing, 3) Generating reservoir inflow scenarios and stochastic optimization of hydropower.

For validating remote sensing data in regions where ground-based measurements are not available, we propose a new validation framework that characterizes the error in precipitation (P) and evapotranspiration (ET) measurements using the parametrically efficient Budyko hypothesis. Results show that the validation framework performs similar to traditional validation methods that use ground-based measurements. For calibrating hydrologic models without streamflow measurements, we explore the utility of other water balance components such as ET and soil moisture (SM). We propose a new multivariate calibration methodology for large scale hydrologic models. We show that ET and SM can potentially replace streamflow measurements for calibration. Finally, we address the issue of generating reservoir inflow scenarios without streamflow measurements. We propose aN ET-based Bayesian model averaging (BMA) framework for assigning scenario probabilities.