Rainfall Small-Scale Variability: Certain Implications for Radar Rainfall Validation Problem

Despite recent technological advances in the area of radar hydrology, reliable quantification of radar-rainfall amounts on scales relevant to hydrological applications is not fully realized. A major problem is the lack of accurate estimation of uncertainty levels of radar-rainfall products. Discrepancies between radar and rain gauge rainfall quantities---traditionally considered as an approximation of the true ground rainfall---are mainly attributed to two main factors: rainfall small-scale natural variability and differences in sampling properties between radar and gauge. The present research addresses these issues in an attempt to develop sound validation methodologies
for radar-rainfall products. In the framework of this study, some fundamental questions are posed: How is rainfall variable over scales smaller than the radar-resolved scales? What role does such variability play in the evaluation of radar-rainfall estimates? Is there a radar-space/gauge-time equivalence where the observations of both sensors are possibly comparable?

This study makes use of extensive experimental rainfall multi-sensor observations collected during the field campaigns of the NASA's Tropical Rainfall Measuring Mission (TRMM) in addition to other experimental datasets. First, errors associated with gauge measurements---as the main independent data source for validation---are investigated and characterized. Then, extensive data analysis is performed to characterize rainfall variability over scales relevant to radar spatial/temporal resolutions. A statistical procedure is applied to quantify the contribution of rainfall natural variability to the significant discrepancy usually observed in radar-gauge comparisons. This eventually can be used in establishing an error budget for radar-rainfall products. Finally, a non-parametric statistical procedure is developed and applied in an effort to establish a comprehensive validation framework. A main element is a transformation model of point-to-area rainfall which is experimentally verified. The findings of this research have implications for the use of radar-rainfall data in applications such as hydrologic modeling, flood forecasting, numerical weather prediction models, and validation
of other remote sensing rainfall estimates, among others.

ehabib@tntech.edu