In the absence of suitable observations, the verification of tropical cyclone (TC) intensity predictions traditionally relies on estimates of wind speeds and mean sea level pressure (MSLP) at the centre of TCs. These estimates are usually derived from satellite observations of cloud structure using the Advanced Dvorak Technique (ADT). The ADT method has been assessed in dedicated studies by comparing the estimates with in-situ measurements by aircraft reconnaissance. A new way of assessing it uses simulated satellite images produced at ECMWF.
The ADT method
The ADT method, mainly developed at the University of Wisconsin, uses observations from the longwave infrared window channel to derive an intensity classification referred to as
the ‘T number’. The T number is calculated by exploiting empirical relationships between the image scene characteristics and TC intensity. However, uncertainties in the estimation arise from the empirical relationship between cloud patterns and the T number and from the translation of the T number into minimum MSLP and maximum wind speed. Traditionally ADT performance has been assessed by comparing ADT results with in-situ observations obtained by aircraft reconnaissance. Since such assessments are limited by the availability of in-situ observations, we here present an alternative based on simulated satellite images from ECMWF forecasts.
Assessing ADT estimates
We assume that there is a good level of consistency between simulated satellite images and the predicted intensity (expressed in terms of MSLP or maximum wind) since both are generated by the same model. Finding good consistency between the ADT results from simulated images on the one hand and the model surface fields on the other would thus provide additional confidence in the behaviour of the ADT algorithm as well as in the simulated satellite images.
Since 2016 ECMWF has been producing simulated satellite images with global coverage. These are generated using the operational high-resolution forecast model and the radiative transfer model used in the operational data assimilation (e.g. RTTOV 11). The output from the model is used as input to RTTOV to derive and simulate the brightness temperatures that geostationary satellites would observe given the relevant model profiles and surface parameters. Such images are routinely produced from the model in both the infrared atmospheric window and water vapour regions at three-hourly intervals. In this pilot study, images of TC Usagi (2013), Fitow (2013) and Neoguri (2014) have been used as input for the ADT. The resulting intensity estimates in terms of minimum MSLP can be compared with the minimum MSLP predicted by the model. The results, shown in the line charts (blue and green lines), agree well with each other for most of the time steps for these three cases. However, they differ for the most intensive phase of TC Neoguri. This is expected as the modelled cyclone is very deep and only few cyclones this deep have occurred in reality, making the statistical relations used in the ADT method uncertain. Particularly large differences are found for TC Neoguri during the intensification phase, where the estimated intensity from the simulated images using ADT is much less than the intensity predicted by the model. One explanation is that the cirrus canopy can obscure TC structure underneath as the vortex organization improves, thereby creating a false plateau in the intensity estimates derived from the IR scenes. In such conditions, the availability of polar-orbiting satellite microwave (MW) images can be used to aid in the detection of the developing eye. Unfortunately, model-simulated MW imagery was not available for use in the ADT method in this study.
The preliminary results of the study show that in most cases the ADT algorithm provides relatively good estimates of modelled intensities in terms of MSLP using the model-simulated satellite data as input. In terms of maximum wind, in nearly all cases the ADT intensities are stronger compared to the model forecast wind estimates (10-minute average winds – not shown). This is in line with previous findings, which show that ECMWF forecasts significantly underestimate time-averaged surface winds, especially for intense cyclones.
The pilot study results also show a good correlation between intensity estimates derived by the Hawaii-based Joint Typhoon Warning Center (JTWC), who use the Dvorak technique, and simulated ADT results. This is the case even during episodes of rapid intensification even though the time phasing may differ.
A potential future direction is to systematically compare model-simulated satellite images with real satellite images to verify tropical cyclone intensities. This would provide a more direct comparison between modelled and observed quantities than just comparing minimum pressure or maximum wind plots against estimates from JTWC and other tropical cyclone agencies. However, this would require model-simulated satellite images with hourly resolution and the inclusion of simulated MW imagery.