Using Wave Models

Our developer team (laugh) has worked hard to augment real-time buoy telemetry with model data to provide users with an increase in spacial coverage and to provide wave forecasts. This has been delivered with two feature releases: virtual buoys on the homepage to cover more locations and the inclusion of 16 day forecasts from GFS-Wave. This blog hopefully sheds a little light on how models work, what they produce, and how Playbuoy uses this data.

Figure 1: Virtual Buoy Stations increasing coverage

Figure 2: 16 Day GFS-Wave Forecast

What are weather models?

Figure 3: HIRES and lower res ECMWF model geopotential height layer

Weather models are simulations of the earths atmosphere grounded in the governing physics (fluid dynamics, heat transfer, etc). Models operate generally by splitting the atmosphere into discrete pieces, setting the initial conditions (what the atmosphere is doing right now), then stepping forward in time in discrete steps to numerically solve for future conditions using finite volume methods. This is computationally expensive, so multiple models exist differentiating along spacial resolution, regional coverage, temporal coverage, and how they handle the underlying physics.

Non-exhaustive list of commonly discussed models
GFS- Global model updated 4 times a day
ECMWF AKA "Euro"- Global model updated 4 times a day
ICON Global model updated ~8 times a day
NAM Higher resolution regional model covering North America. Updated 4 times a day
HRRR High resolution model covering North America updated every hour
Note, higher resolution models can be useful in short-term forecasting however generally become less accurate out beyond a few days.

Wave models, initialization, and forecasts
While the typical consumer of weather forecasts and models is most concerned with the atmosphere, if you are reading this then you are interested in the ocean as well. As you can imagine, the physics of the ocean are different than that of the atmosphere. Therefore, modeling and simulating the ocean takes a separate but associated model to the atmospheric model most people refer too. The ocean and the atmosphere interact (wind creates waves), and therefore atmosphere and ocean models are linked in a process called coupling. As a result, the global models generally have an associated wave model. This is the GFS-Wave for GFS and ECWAM for the Euro.

Figure 4: Example of coupling between atmospheric and ocean models. Note IF3 is ECMWF atmospheric model.

Before simulating and modeling the atmosphere and ocean, the models must be initialized. Just as GoogleMaps needs to know your current location before it can give you direction, a weather or wave model must be loaded with the current conditions of the ocean or atmosphere in order to provide a decent simulation. This is process is not as simple a process as telling Google Maps your GPS location. Weather models must initialize hundreds of millions of parameters, relying as best possible on satellite data, weather stations, weather balloons, and other observations. The model run on the "best" initialized values is often called the operational run, but initialization is not perfect given the complexity of the atmosphere. Beyond the operational run, the model is run many more times with different input parameters to better understand the range of possible outcomes and the sensitivity to initialization error. This is a model ensemble.

Figure 5: Example of different GFS ensemble members (GEFS) showing different possible outcomes. Source: Colorado State University

Actually modeling the surface of the ocean (like you would see in a video game for example) is not possible with current computational resources, so wave models more simply use specific values to represent the state of the ocean. More specifically they use the wave energy at different frequencies, or the Energy Spectrum. The calculated wave energy spectrum is similar to that assessed by wave buoys. The wave model then simulates how this wave energy evolves over time based on interactions with the wind and the atmospheric model, swell propagation and refraction, and other interactions.

Playbuoy Virtual Buoys

Figure 6: Readout for Virtual buoy TPC55 offshore Costa Rica

While Playbuoy started with an emphasis on live data from ocean buoys, ocean buoys are only placed in select locations. Providing wave energy spectra from wave models in additional locations allows us to dramatically increase the coverage. While the accuracy of virtual buoys is not as good as physically buoys, it represents the best available current estimates based on the model initialization and forecast from GFS-Wave. This factors the latest from satellite readings, ship observations, and many more. Coverage is not limitless due to the computational intensity of the full spectra data, but adding virtual buoys more than triples our coverage!

GFS-Wave Forecasts
Playbuoy displays the GFS-Wave forecast for each buoy location (physical and virtual) out to 16 days, giving the latest wave projections. Note the forecast error increases the farther you go out in time. While results vary for different locations and seasons, forecast error for day 5 is typically twice that of one day forecasts. Forecasts beyond 5 days should generally be consumed with conditional caution for locally generated swells. Wave forecasts for locations receiving groundswell can be more accurate for a given time period, as these swells are generated well before they arrive at a location. Modeling the propagation of swells after they are generated is generally more certain than modeling the storms which generate the swells.

Figure 7: Forecast error at different time horizons. NCEP generally covers GFS-WAVE.

Hopefully this helps to increase your understanding how wave model data can be used in addition to live buoy data.