An Object-Based Approach to Model Verification in Tropical Cyclone Forecasting

An intense tropical cyclone (TC) presents a dramatic picture from space. TC convection is multi-scalar, spanning the large-scale synoptic (~2000 km) environment over which TCs gather moisture, the characteristic meso-α scale (~200-2000 km) spiral rainbands seen in radar and satellite imagery, and the meso-β scale (~20-200 km) eye and eyewall(s) in the TC inner core. As a TC intensifies, it produces structures in these predictable patterns, evolving from a disorganized cluster of clouds and precipitation into a compact and symmetric storm. These consistent patterns throughout the TC lifecycle facilitate the use of object-based metrics to quantify spatial attributes of the overall storm and individual rainbands. Further, object-based verification techniques are useful for comparing model predictions of precipitation versus observations. Here, I present a proof of concept study of Hurricane Harvey (2017) to demonstrate an objective method for measuring TC precipitation forecast skill. Model forecast precipitation from the operational High Resolution Rapid Refresh (HRRR), North American Mesoscale (NAM) and Global Forecast System (GFS) models is compared against the National Centers for Environmental Information (NCEI) Stage IV precipitation analysis. I focus on short- range (+12 to +72 hours) forecasts from the 00 and 12 UTC model cycles. All verification measures are evaluated at 6-hourly intervals to analyze how forecast skill varies with lead time. Due to the nonlinear nature of atmospheric convection, it is important to assess precipitation forecast skill at multiple thresholds. In this study, I delineate a binary precipitation field using rain rate thresholds of 0.01, 0.1, 0.25, 0.5, 1, 2, and 5 inches per hour. Next, I calculate spatial metrics of area, dispersion, closure, solidity, elongation, and fragmentation on a common 15-km horizontal resolution Cartesian grid, as well as centroid position errors for matched objects, based on the forecast and observed fields. I present a model intercomparison at each threshold, contextualized for the Hurricane Harvey case study, to demonstrate the capabilities and limitations of operational model precipitation forecasts in an extreme TC rainfall event. Results demonstrate the value of object-based, quantitative measures for assessing TC model forecasts rather than relying on subjective visual comparisons of forecasts and observations.