AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Probabilistic Uncertainty Estimation of Radar Rainfall: In a groundwater modelling context

Author

Term

4. term

Publication year

2012

Submitted on

Pages

119

Abstract

Denne afhandling undersøger, hvordan usikkerheden i radarestimeret nedbør kan kvantificeres og anvendes i grundvandsmodellering. Med udgangspunkt i Mt. Stapylton-radarens kvantitative nedbørsestimat (QPE) over Bribie Island, Queensland, udvikles en probabilistisk ensemblegenerator, der perturberer det deterministiske radarfelt ud fra spatiale fejlstrukturer og temporale fejlkorrelationer estimeret ved sammenligning med regnmålerdata. Ensemblevariationen tolkes som usikkerheden, og algoritmen indebærer en indirekte bias-korrektion, som giver anvendelige tidsserier til hydrologiske formål. En transiente MIKE SHE-grundvandsmodel for Bribie Island, der omfatter evapotranspiration, umættet og mættet zone, kalibreres til gennemsnitlige observerede grundvandsniveauer og anvendes til at undersøge følsomheden over for usikkerhed i radarnedbør. Resultaterne viser, at variabiliteten i de producerede nedbørstidsserier kun i begrænset omfang overføres til de simulerede grundvandsniveauer, hvilket peger på, at grundvandsmodellen opfører sig som et dæmpet system med lang responstid og primært reagerer på akkumulerede mængder frem for små variationer. Arbejdsindsatsen for at etablere ensemblegeneratoren fremstår derfor stor i forhold til den beskedne spredning i modeloutput for denne anvendelse, men metoden vurderes relevant i andre sammenhænge, såsom urban afvanding, realtidsusikkerhedsestimering og nowcasting. Bagvedliggende fejlkilder for radar og regnmålere (bl.a. dæmpning, vertikalt reflektivitetsprofil, dråbestørrelsesfordeling og vindrelateret underfang) gennemgås som grundlag for fejlspecifikationen.

This thesis examines how uncertainty in radar-based rainfall estimates can be quantified and used in groundwater modeling. Using quantitative precipitation estimates (QPE) from the Mt. Stapylton radar over Bribie Island, Queensland, a probabilistic ensemble generator is developed that perturbs the deterministic radar rainfall field based on spatial error structures and temporal error correlations derived from comparisons with rain gauge data. The ensemble spread represents the uncertainty, and the algorithm provides an implicit bias correction that yields time series suitable for hydrological applications. A transient MIKE SHE groundwater model for Bribie Island, comprising evapotranspiration, unsaturated and saturated zones, is calibrated to average observed groundwater levels and used to assess sensitivity to uncertainty in radar rainfall input. Results show that the variability in the probabilistic rainfall time series translates into only modest variability in simulated groundwater levels, indicating a damped system with a long response time that is more responsive to accumulated totals than to small fluctuations. Consequently, the effort required to implement the ensemble generator appears disproportionate to the limited spread in model outputs for this groundwater use case, although the approach is considered valuable for other applications such as urban drainage, real-time uncertainty estimation, and nowcasting. Underlying error sources for radar and gauges (including attenuation, vertical reflectivity profile, drop size distribution, and wind-related undercatch) are reviewed to inform the error specification.

[This summary has been generated with the help of AI directly from the project (PDF)]