Date of Completion


Embargo Period



Downscaling, Error Modelling, Hydrologic Modelling, Precipitation, Remote Sensing, QPE-Quantitative Precipitation Estimation, Soil Moisture

Major Advisor

Emmanouil N. Anagnostou

Associate Advisor

R. Edward Beighley

Associate Advisor

Guiling Wang

Associate Advisor

Amvrossios C. Bagtzoglou

Associate Advisor

Jeffrey McCollum

Field of Study

Environmental Engineering


Doctor of Philosophy

Open Access

Open Access


The overarching goal of the research described in this study is to improve uses of satellite rainfall in hydrological applications. To address this goal, the study followed the following three research dimensions. First, we improved the characterization of rainfall retrieval uncertainty from instantaneous earth-orbiting platform passive microwave observations, which form the basis of the merged satellite products, by investigating the significance of conditioning an error model for NASA’s Tropical Rainfall Measurement Mission (TRMM) Microwave Imager (TMI) rainfall algorithm (2A12) to near-surface soil moisture data derived from a land surface model. A two-dimensional satellite rainfall error model, SREM2D, was used to provide ensemble error representation of 2A12 rainfall using two different parameter calibration approaches: conditioning vs. not the SREM2D parameters to the surface soil wetness categories. The statistical analysis of model generated ensembles and associated error metrics showed better performance when surface wetness information is used in SREM2D.

The second research dimension was to investigate the hydrologic applicability of a quasi-global high-resolution satellite precipitation product relative to a global reanalysis product through comparison with rain gauge adjusted radar-rainfall estimates. The study presents error metrics for moderate and extreme precipitation events over the Susquehanna River Basin in the Northeast United States. Results show improvements of the statistical scores up to 50% with increasing basin size, particularly for the satellite product. Overall, the satellite product exhibits better error statistics compared to the coarse resolution re-analysis product. For the simulated streamflow, the re-analysis precipitation product is shown to have up to seven times higher mean relative error compared to the corresponding high resolution satellite product for moderate streamflow values; three times higher for extreme streamflows. This significant divergence in the runoff simulation error statistics is attributed to differences between the two precipitation products in terms of the propagation of their error properties from precipitation to simulated streamflow.

The last research dimension was motivated by the above findings to develop an error correction and downscaling scheme to advance the use of global reanalysis precipitation products in hydrologic modeling. The error correction and spatial downscaling was based on the SREM2D model and the high-resolution satellite precipitation data of the previous studies. This study focused on 437 significant storm events from the past 10 years that occurred over the Susquehanna River Basin in the Northeast United States. We quantify improvements on the reanalysis datasets through the satellite-driven downscaling scheme by presenting error metrics in (1) rainfall, and (2) generated runoff as function of basin size and storm severity. Results show that the generated rainfall ensembles of the downscaled reanalysis products could encapsulate well the reference rainfall. The statistical analysis including frequency and quantile plots, mean relative error and root mean square error statistics demonstrated significant improvements on downscaled data relative to the original re-analysis data in terms of both precipitation and runoff simulations. For instance in fall the re-analysis precipitation product is shown to have up to three times higher mean relative error compared to corresponding high resolution satellite data, this ratio increases up to four times for the simulated streamflow. The proposed downscaling scheme is modular in design- it can be applied for any gridded dataset in any region in the world.