Constraining Satellite Precipitation Biases Through Geophysical Arguments
July 15, 2025
Eric Goldenstern
Committee: Chris Kummerow (Advisor); Christine Chiu; Kristen Rasmussen; Imme Ebert-Uphoff (Electrical and Computer Engineering, CIRA)
Abstract
Passive microwave satellite precipitation products provide crucial information on global rainfall and the only source of such information over much of the world. These products operate under the assumption that microwave brightness temperature (TB) is sufficient to constrain rainfall. This information, however, often represents multiple rain states, resulting in substantial and highly variable errors. Traditional error analyses use a reference precipitation dataset, often from rain gauges or radars, which are only available in limited portions of the world and can only represent the areas in which they operate. As such, much of the world is left without an objective quantification of satellite precipitation errors. The work of this dissertation aims to address this gap in error coverage by determine alternate methods by which they can be estimated.Because satellite precipitation products utilize radiative information to attribute rainfall via a mathematical framework, the resultant errors can be expected to stem from assumptions within the framework itself and the radiative theory it exploits. To characterize these potential error sources and their effects, coincident GPROF Version 7, Global Precipitation Measurement (GPM) Microwave Imager (GMI), and GPM combined observations were examined over three tropical land regions known to exhibit distinct biases relative to one another. The interrelation between rain rate and ice-rain ratio (IRR) was identified as the primary descriptor of regional bias, which accounts for roughly 50% of the observed GPROF biases. Accounting for second-order effects further improves this, accounting for roughly 70% of the observed biases when considered with rain rate and IRR. Comparing the effects of these three parameters between GPROF Version 7 and the 1-D version of GPROF-NN showed similar improvements, indicating the utility of this error attribution across precipitation products.
From the three error sources identified, IRR is the least available. While ice content information can be gathered by ground-based or spaceborne radars, these platforms have limited spatiotemporal coverage, complicating their use in operational products. Since the formation of atmospheric ice is driven by the incident environment, an investigation on this linkage was undertaken. Seven years of spaceborne radar and precipitation measurements from GPM taken over three tropical land regions were obtained to investigate this hypothesis. Five ice content regimes, as defined by ice rain ratio (IRR), were identified and found to relate to distinct precipitation regimes. These regimes were coupled with CAPE, TCWV and Tavg information from ERA5, which proved to both reproduce the spatiotemporal characteristics of the regimes and their effects on regional bias constraint. These effects appear largely unaffected by shifts in the input data and time series, showing that the environments in which certain precipitation regimes occur are generalizable to other locations and times.
Having now established the sources of error in passive microwave precipitation products and the ability of the large-scale environment to diagnose these sources, a method for quantifying these errors using ERA5 information was developed. This framework, termed the Satellite Precipitation Errors from Ancillary Data (SPREAD) model, utilizes support vector regression to directly predict product errors from the retrieval rain rate and ERA5 thermodynamic and kinematic information. Evaluations of SPREAD show promise in developing error estimates in the Tropics, where it was trained, while further work is needed to improve its ability to predict errors in regions outside of the Tropics. Investigations of the resolution sensitivity on the model indicates that SPREAD is most stable when assessing precipitation errors at large scales, reflecting the resolution capabilities of the ERA5 data. Though a prototype, the SPREAD model shows the capability of attributing satellite precipitation retrieval errors in cases where traditional validation analyses are not possible by leveraging information provided by the large-scale meteorological conditions.