Assessment of BUFKIT Methodologies to Forecast

Assessment of BUFKIT Methodologies to Forecast
Wind and Wind Gust Speed

Kenneth R. Cook and L. David Williams
National Weather Service, Wichita Kansas

September 11, 2007

1.0 Introduction

Wind and wind gust forecasts remain a day to day challenge for forecasters at the National Weather Service (NWS). Winds produced by synoptic scale pressure gradient forces, commonly referred to as gradient winds, consist of a large part of this challenge, especially in the fall, winter, and spring months on the Plains. BUFKIT (WTDB, 2007) is one of the primary software packages used to assess the environment and produce wind forecasts by forecasters. It would only be prudent to investigate the accuracy of using such software and determine what "best practices" can be utilized to improve said forecasts.

2.0 Methodology

Hourly observation data from the National Climatic Data Center were gathered for 6 sites across the Weather Forecast Office (WFO) Wichita's area of responsibility. Specifically this included Salina (SLN), Russell (RSL), Hutchinson (HUT), Wichita (ICT), Chanute (CNU), and Winfield (WLD). The dates examined consisted of the 6 month period from January through June of 2006. This corresponded to the locally available BUFKIT data archive.

Cases were identified by looking at Wind Advisory and High Wind Warning Cases for the WFO ICT service area. This amounted to in excess of 20 cases for the said time period as the spring of 2006 was quite windy. In fact, 19 wind advisories were issued during the spring months of 2006 alone.

Once cases were identified, hourly wind and wind gust observations were matched with the hourly forecasts of mixed layer winds (referred to in the images below as the model name -X (e.g. NAM-X)), or in BUFKIT "momentum transport winds", and wind speed at the top of the mixed layer (referred to in the images below as the model name -T (e.g. NAM-T)). For this study, these fields were used as a proxy for a surface wind forecast in a well mixed atmosphere as this can be common practice by operational forecasters. The goal is to ascertain the validity in making such an assumption and to determine the most accurate way to forecast a sustained wind and a wind gust. The NAM-Eta, GFS, and RUC models were used for this evaluation as the model forecasts in BUFKIT. Only the 0-6 hour forecast time period was assessed as this would, in theory, produce the most accurate forecasts of any model element (Zhu, 2007).

Once the data were assessed, graphs were produced showing a plethora of statistical analysis. Additionally, errors were placed into "bins" in order to further assess the level of accuracy.

3.0 Analysis and Results

Using the mixed layer wind (hereafter MLWIND) or wind speed at the top of the mixed layer (hereafter TMLWIND) directly as a proxy for a surface wind forecast would result in speeds much stronger than observed. This is evident in the following graphical representation of the output:

At right is a graph of the total wind bias using both MLWIND and TMLWIND as the forecast surface sustained wind speed. All models are compared and are included in the results for the 6 stations shown.
At right are the same comparisons stratifying the data by station.
Clearly, the NAM-X has the least amount of bias. Stratifying the NAM-X forecasts by station using the same scale of error identifies likely boundary layer influences that should be considered in the forecast process. The RSL site is the most exposed of the sites and therefore should have and does exhibit wind forecasts closest to what is contained in the mixed layer. On the contrary, ICT and HUT both have effects (friction from trees and depression from elevation respectively) reducing the surface wind speed causing a systematic bias in the forecasts.
Finally, by placing the errors into a "bin", an attempt was made to determine where much of the error was located in the dataset. This is shown at right.

Until this point, it would seem difficult at best to use this method to forecast a sustained wind without performing some significant "on the fly" adjustment. The next assessment performed examined the utility in using MLWIND or TMLWIND directly as a proxy for a surface wind gust forecast. This examination shows some promise predicting wind gust via BUFKIT. The following graphs illustrate this:

At right, the MLWIND and TMLWIND variables have been examined. The GFS and RUC MKWIND performed remarkably well for these cases.
At right is the same comparison as above only the data is stratified by station.
Further investigation of the RUC MLWIND forecast showed good clustering of error near zero.
The GFE MLWIND forecast showed even better clustering near the zero error. Given the low bias and these results, this would be the best choice to predict a wind gust.

4.0 Summary and Conclusions

Forecasters use BUFKIT as an integral part of the forecast process. High wind cases were examined to ascertain the validity of using said software, developing best practices from this research to forecast sustained wind and wind gust speed. MLWIND and TMLWIND were used as a proxy for a surface wind speed forecast in a well mixed atmosphere.

Sustained wind speed was not predicted very well overall as a significant bias to over forecast surface wind speed was noted. However, MLWIND showed the most promise as a method using the NAM forecast where a bias of nearly 3 knots was found to exist. This bias must be taken into consideration while making this forecast.

Secondly, the ability to forecast wind gust speed using this same methodology proved fruitful. Both the GFS and RUC MLWIND forecasts were quite accurate for the cases studied. Examination of the RUC showed a bias and mean absolute error (MAE) of 1.43 knots. Evaluation of the GFS produced a bias of 0.40 knots and a MAE of 1.51 knots.

From these data, forecasters could deduce that using the GFS or RUC MLWIND forecast as an approximation of an observed surface wind gust speed in a gradient wind environment would be deemed appropriate.

Finally, it can be concluded that using the momentum transport winds in BUFKIT to predict a surface wind gust speed is a great asset to the forecasters. This software should be utilized as much as possible to facilitate an improved surface wind gust speed forecast.

5.0 References

Warning Decision and Training Branch (WDTB), 2007: BUFKIT information available at the following web site: http://wdtb.noaa.gov/tools/BUFKIT/index.html.

AvnFPS, 2007: Wind Rose Calculations from data 1973-2004.

Zhu, Yuejian, 2007: Verification data from the Environmental Modeling Center (EMC) available online at http://www.emc.ncep.noaa.gov.

National Weather Service Weather Forecast Office