Francois Gourand, Ecole Nationale de la Météorologie, Meteo-France
David Atkinson, International Arctic Research Center; UAF College of Natural Science and Mathematics, Atmospheric Sciences Program

This project is a joint ENM/IARC endeavour.

Alaska France


The sources of meteorological data have never been so numerous and diverse. They are grouped primarily in two distinct types : observation data and data coming from digital models of the atmosphere. The steady growth in computer power has allowed digital models to play an increasingly dominant role in the world of weather forecasting.

Digital models, as complex they can be, cannot account for all physical processes and environmental details. This means that small-scale physical processes must be parametrized, rather than explicitly modeled. Frequent qualitative improvements can be realized by increasing the models' resolution, but this comes at a cost of increased computer time.

The weather station network can never be dense enough to cover the totality of a given territory, and the measurements of a network in any region often reflect the presence of topoclimates that are not accounted for in large-scale models. This is particularly true for complex, sparsely inhabited regions, such as Alaska or the Yukon Territory, in northwestern North America.

There is therefore a real need to develop a tool that can fill the gap existing between the relatively coarse resolution of a fluid dynamic/spectral forecast model and the actual detail present in the landscape. Such a high-resolution "topoclimate" model must consequently rely as much on topoclimatic specificities as on the meterological situation : for this reason a high-resolution (less than 1 km) temperature model that blends landscape features with forecast model output has been developed for subarctic northwestern North America (Alaska and Yukon).

A high-resolution global digital elevation model was chosen for this project : the United States Geological Survey "GTOPO30". Its high-resolution 1/120° is then the basis for the model resolution, and allows a quite detailed description of the terrain. The selected weather forecast model is the Global Forecast System, developed and operated by the US National Weather Service (National Oceanic and Atmospheric Administration), because it is a global model, easy to access, reliable, and because it should be supported for some time. By combining these two models, one can account for detailed, topographically driven processes, by computing some derived parameters on the one hand, and accounting for the synoptic meteorological situation on the other hand.

Local characteristics include slope gradient and aspect, maximum gradient to an obstruction, which affects the exposure/shelter of a given location to a wind direction, as well as flow accumulation, which shows how and where the cold air drains during night, or during strong surface inversions. Various algorithms were developed or adapted to account for these effects. These locational elements are then linked with meteorology.

This is done by expressing the meteorological situation in terms of the site-specific parameters identified above. First, elevation plays a major role in the temperature estimate. Combining temperature vertical profiles from GFS and the elevation from GTOPO30 provides a first guess of temperatures at high resolution.

Another major process, particularly important in the northern regions concerned by this study, is the handling of drainage flow and nocturnal or wintertime thermal inversions. This process, which depends on the terrain and the meteorological situation, has been modeled carefully. The results, although in general agreement with observations, can be wrong when the inversions are especially strong, like in the Fairbanks area during wintertime.

The Foehn effect, or adiabatic compression, is another major process. It is modeled to account for the substantial temperature gain that affects some regions with a marked relief during particular synoptic situations. The results add important information, however an impediment to modeling this component was a poorly captured local scale wind field in GFS. This component should ideally be driven using a high-resolution wind map, but one was not available for this project. Nevertherless, some ideas were proposed to try to correct this fact.

The solar radiation effect is also modeled. It is driven by total cloudiness (modeled by GFS) and often represents a small correction to temperature, but is particularly useful when the other effects do not apply (except elevation). Finally, another effect was modeled : coastal cooling. For now, it is not activated in the model because preliminary tests indicated it was not necessary.

Once all these effects are defined, a temperature map is produced. It is then important to assess the error of the model : a good way to do so is to compare the model to a number of observations, as many as possible over as wide a region as possible. To do this the MADIS(Meteorological Assimilation Data Ingest System) observation network, available for North America, has been used. More than 100 weather stations belonging to this network, distributed as equitably as possible across the domain, were chosen and accurately located on the digital elevation model.

An automatic evaluation system was set up that compares model values and observed values for a given time period. The model biases are calculated by averaging each weather station over the given time period. After this, the bias field is applied to correct the model results to obtain the final temperature estimate. This is especially useful in some areas where the GFS model is far from reality, for instance in the northern part of the domain. Unfortunately, some major errors in GFS can not be corrected when they happen in areas without any weather station.

The whole system runs automatically : GFS data are downloaded every 6 hours, handled by the topoclimate model to produce a temperature map. This temperature map is compared to observations to generate a bias map for the last 15 days, and this bias is "inverse-applied" to correct the temperature map to get a unbiased temperature. Finally, the unbiased map is put online. This system thus provides a high-resolution operational surface air temperature climatology in real time, day after day, over sparsely inhabited regions.

Thanks to the use of standards and convenient tools (the topoclimate model was mostly written using the NCAR Command Language of the National Center for Atmospheric Research), everything is set and ready so that the model can also be initialized with models other than GFS, for climate research puroposes, for instance.