In the pre-event stage, from five days out up until immediately prior to landfall, there is heavier reliance on model based data since the density and availability of observations is often sparser in areas farther offshore or more remote. While in-situ surface observations that come from a variety of sources (ships, buoys, automated and manned observation stations) can be limited for some storms, there is still a myriad of remotely sensed based products can be used to verify and improve model simulation performance. Other shortcomings with observations, such as instrument failure, calibration problems, inconsistent exposures and spatial density, often make it difficult for decision makers to aptly assess the extent and severity of an event. In many cases, remote observation of wind is valid only over open water. Remote sensing measurements such as those from NOAA’s Advanced Scatterometer (ASCAT) and NASA’s recently deployed Cyclone Global Navigation Satellite System (CYGNSS), will undoubtedly have the compound effect of assisting with the verification of model performance and subsequent model parameterization enhancements, and improving the underlying forecast models that are used to initialize some of our TC multi-model members to increase the accuracy of the hazards produced by the model.
In the immediate TC post-event timeframe (2hours to 2 days after) the response and recovery operations depend on accurate, readily available meteorological hazard intelligence to effectively deploy. Casualty insurers, emergency responders, property and asset managers rely on this vital information. Globally consistent, uniform model-based hazard datasets help decision makers create a process for responding in a systematic manner, addressing some of the limitations posed when confronted with inconsistent or sparse observational data. TC event characteristics in terms of spatial extent and severity can be easier to quantify when using the model-based data in concert with available observations.
A multi-model suite of hazard intelligence also informs the decision maker to how probable a particular outcome is by conveying uncertainty. When there is a wider spread, or divergence, in solutions it helps signal the need to remain more alert to changing conditions, but that the likelihood remains lower than when there is more consistency across the forecasted hazards. Observations can be leveraged to assess multi-model performance and ‘ground-truth’. The multi-model solution requires a level of understanding to interpret the model produced results knowing that they are guidance and put them into even better context using the best available observational data.
Our study undertakes a comparison of model outputs with a select sample of observational data sources at differing time scales for aiding in critical decision making. The evaluation framework identifies where there is appropriate overlap in temporal and geographic coverage between our internally produced model-generated TC multi-model hazard data and the external observation datasets for comparison. This will help highlight the strengths of each type of dataset, illustrating the benefit of combining them for the purposes of improving decision making capabilities in the pre, during and post event time scales. Model data has limitations tied to the quality of the forecast used to initialize the simulation (and inherent forecast uncertainty), while observations can be limiting in their capacity to fully capture the impacts of the event over a broader geographic region. To quantify / rank the quality of observations versus our model based data, we compare both spatially and at point locations, the wind fields from both remotely sensed and surface based observational sources at a few specific short-term time scales (pre, during and immediate post-event). This study compares the performance and utility of using each data source over at various time scales.
Comparing the application of multi-model results versus observations can be assessed on varying stages of the short term and immediate post-event time scale. The TC multi-model hazard results and different observational datasets are best used in conjunction with each other, by filling in the inherent gaps in consistency and coverage that may be present with the different data sources. In many cases, such as the case for Hurricane Patricia (2015) in the East Pacific, the in-situ land observations are too sparse to provide continuity whereas model data is always a consistent availability, even if not as accurate. Availability of high resolution (up to 1 km) TC model hazard data can instill confidence of TC impacts and reinforce confidence of the extent and severity of TC hazards in cases where there are inadequate surface observations limiting representation to a few points over a large area of exposure. Together, observational and model produced data provide decision makers with a more comprehensive picture for situational awareness, to formulate the specific location based assessments needed to respond to TC hazards from an imminent short-range forecast threat through the post-event assessment stage and knowing when to focus on each is paramount.