Wind Verification | Mixing Height | Local Inversion Potential | Ventilation Index

Wind Verification

Click on a point to view the characteristics of the observation location and comparison location and comparison of modeled and observed wind data.

The observations we selected for comparison came from NOAA National Weather Service (NWS) primary observing stations (National Renewable Laboratory 1992; National Climatic Data Center 1997). Data from local and regional networks, such as the interagency RAWS network (U.S. Department of Interior 1995), did not have adequate quality or consistency for model verification.

The Wind Development section describes the wind model and some of its known strengths and weaknesses. Basically the model implements only two drag coefficients, one for open water and one for forests. As a result, agreement is expected to be better for forested lands than grass or urban lands. Also, the model does not implement radiative heating and cooling, which is likely to influence performance in regions of large diurnal temperature fluctuations.

To compare modeled winds with observed winds, we generated polar diagrams of modeled, observed, and differences for each observation location (How to Interpret Wind Verification Plots). Plots were generated for each month and each synoptic time period (Time of Data). This allows for examination of seasonal and diurnal performance.

At each observation location, we include maps of model terrain (2.5' latitude-longitude in the 48-conterminous states and Hawaii, and 5km horizontal resolution in Alaska), and actual terrain and land use (90 m spatial resolution). These help determine whether the model is failing because of its simple physics or numerical schemes or because its terrain and land-use are not representative of higher-resolution values affecting the anemometer measurement. All verification plots are explained in How to Interpret Wind Verification Plots.

In general, the model performed very well with respect to both wind speed and wind direction. The two land-use categories (land and water) in the model, however, tend to bias rough terrain (forested and mountainous). This causes strong winds (> 8 m/s) to be underestimated over broad flat areas and grass lands. Also, the smoothed model terrain appears to cause a modest directional bias of less than 45° in either the clockwise or counterclockwise direction. Finally, restrictions on the model lapse rate (temperature difference with height) and lack of radiative heating appear to cause poor model performance in some months at western arid sites.

Because several domains were merged together to generate a national coverage of wind (Wind Development) we closely examined winds in the overlap regions to evaluate potential edge effects. In all cases, we found model-generated winds from one domain to be reasonably consistent with winds generated from an overlapping domain. For example, Medford, Oregon and Winnemucca, Nevada represent two stations in the overlap zone between the Northwest and Southwest domains. Winnemucca is near the center of the overlap zone while Medford is near the northern border of the Southwest domain. The two model runs agree more closely at the central site but do not differ grossly at the border site. Because border sites are weighted less, the edge effect becomes negligible in the merged data.

We used a standard method of deriving mixing height (Mixing Height Development). Because we used coincident surface observations instead of maximum and minimum, however, we chose to verify our methods against mixing heights derived by the Environmental Protection Agency (U.S. EPA 2001) for the years 1984 to 1990. At each radiosonde location we created scatter diagrams of mixing heights to help highlight differences. The following figure shows an example from Pittsburgh, Pennsylvania. Points falling on the diagonal indicate perfect agreement. Most values agree quite well. Significant differences arise for missing data (solid blue triangles). Our missing values were omitted and were different than EPA missing values, which were filled. Differences also occur for temperature change values (open green triangles). EPA adjusted values when the selected temperature values were less than the 12Z RAOB temperature while we made no adjustments.

Two sites, Omaha, Nebraska and Corpus Christi, Texas show significant differences in all values (see figures below). We are investigating the cause of these discrepancies. Until we isolate and fix the cause, mixing-height values in regions around these sites should be used cautiously.

Local Inversion Verification

There are very few observations of local inversion occurrence or location. At RAOB locations, we tested our inversion occurrence criteria and found that observed surface-based inversions occurred on nearly all days that the criteria of calm wind and clear skies were met. A surface-based inversion was determined to exist if the 12Z RAOB included two adjacent layers within 1000 meters of the surface that reported warmer air over cooler air. No distinction was made between inversion strength or depth. The resolution of the temperature data is to the nearest tenth of a degree centigrade.

A table shows "p values" associated with chi-squared tests for each month. A p-value less than 0.05 indicates a strong relationship or agreement between the RAOB observed inversion and the surface-based algorithm. The table is color coded to show areas of no relationship (white), significant relationship (p<0.05), and highly significant relationship (p<0.01). Table.

Inversion occurrence is determined from radiosonde observations (RAOBS), which are sparsely distributed over the United States. Additionally, these stations are typically located in flat areas or broad valleys where local inversions are less significant than in narrow valleys, small hollows, and basins that are typical of wildland areas. Therefore, most of our verification techniques for local inversion potential are qualitative in nature.

Another way to check the reasonableness of the terrain algorithm is to compare maps of local inversion potential with satellite observations. The following figures show a MODIS satellite image over the Salmon River in central Idaho during the 2000 wildfires and a map of local inversion potential for the same general area. Inversion potential is shaded with light blue being the strongest potential. From the satellite image, it appears that most smoke is concentrated in the Salmon River valley, where the strongest inversion potential is indicated. Darker blue colors indicate potential inversions in side valley and tributaries, just as appear in the satellite image.

Ventilation Index Verification

As an index, one only can judge its value from its measured components, which are wind and mixing height. Modeled winds were shown to be reasonably accurate in many cases, with randomly distributed errors within a range of observation accuracy, but there seems to be a relatively consistent slow bias. Mixing heights appear reasonably accurate in all cases, except within 10s of kilometers from Omaha, Nebraska and Corpus Christi, Texas. It is difficult to determine the accuracy of the local inversion potential, however, because there are so few observations. Also, the relatively coarse grid size (2.5' latitude longitude and 5km) does not capture many of the small hollows that can trap smoke at night. Together with the somewhat slow wind speed and inclusion of local valley inversions, we assume that the ventilation index errs conservatively, biasing toward potentially poor ventilation. The surface wind speed multiplied by the mixing height gives an estimate of ventilation potential. The resulting index in VCIS is lower than is typical because it is influenced by local inversions and derived from winds at the surface instead of higher in the mixed layer.