Verification of satellite estimates and NWP model forecasts for tropical cyclone rainfall

© 2018 Dr. Yingjun ChenThis study evaluates the TRMM 3B42 version 7 (V7) estimates of daily rainfall in tropical cyclones (TCs) using the Comprehensive Pacific Rainfall Database (PACRAIN) of 24h rain gauge observations. The evaluation is performed on two different terrain types: low-lying atoll sites (assumed to represent open-ocean conditions) and coastal and island sites (over land). The results show that TRMM 3B42 has good skill at detecting intense TC rainfall, with good correlation and pattern matching with PACRAIN observations. However, it tends to overestimate heavy rain frequency on atoll sites, but tends to underestimate heavy rain frequency on coastal and island sites. Overall, TRMM 3B42 is better able to estimate the intensity of TC heavy rain over ocean than over land. It is least skillful at coastal and island sites with high elevation, where it significantly underestimates TC heavy rainfall, suggesting that TRMM 3B42 is unable to capture orographic enhancement during TC landfall. Finally, results from V7 were compared with results from its predecessor, Version 6, showing that Version 7 of TRMM 3B42 has higher values on average for TC rain. TRMM 3B42 rainfall estimates in landfalling TCs are then evaluated over Australia. A high quality gauge–based gridded rainfall product from the Australian Water Availability Project (AWAP) is utilized as reference data. The comparisons show good correspondence over space and time between TRMM 3B42 and AWAP analysis for rainfall at TC landfall over Australia. The results also show that TRMM 3B42 generally overestimates TC rain for low rain rate but underestimates TC rain at high rain rate. TC intensity, location, terrain, and TC seasons all have impacts on TRMM 3B42’s detection skill. For TC heavy rain, TRMM 3B42 shows better agreement with AWAP during more intense TCs (category 3–5), in the eyewall as opposed to the rain bands, in the tropics as op posed to the subtropics, and late in the TC season as opposed to early and peak of the TC season. The last part of this study is to demonstrate the useful information that can be derived from Contiguous Rain Area (CRA) evaluation, such as systematic errors in predicted tropical cyclone (TC) rainfall location and components of rainfall error due to incorrect predictions of location, rain volume, and rain pattern. CRA verification was applied to evaluate ACCESS-TC, the tropical cyclone version of the Australian Community Climate and Earth System Simulator (ACCESS), daily rainfall forecasts over 15 TCs in the north west Pacific ocean during 2012-2013, by comparing with TRMM 3B42 satellite estimates. The results showed that pattern error was the major contributor to the total TC rainfall forecast error, followed by displacement and volume. ACCESS-TC forecasts tended to predict more rainfall closer to the TC center compared to TRMM 3B42 estimates. Furthermore, rain event verification showed that for short lead time (24h) forecasts, overestimation of rain volume was a major problem for ACCESS-TC forecasts, while displacement error was more significant in longer lead time (72h) forecasts. Finally, probability distribution functions and radial probability distributions of rainfall in the forecasts and observations are used to further characterise the rain volume error. This confirmed that ACCESS-TC tended to produce more extreme rain in the locations closer to the TC center (eyewall)