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, updated Septemer 19, 2012

 


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

Below are various graphs showing some of the results of this study, using the mixed layer wind (hereafter MLWIND) and wind speed at the top of the mixed layer (hereafter TMLWIND) directly as a proxy for a surface wind forecast.

At the top left is a graph of the total wind bias when forecasting surface wind speed. This is comparing the bias of both the MLWIND and TMLWIND. Data included here are all stations and all models.

At the top right is a graph of the total wind gust bias when forecasting surface wind speed. This is using the same comparison.

At the lower left is the same comparison as above for surface wind speed, stratifying by station.

At the lower right is the same comparison as above for surface wind gust speed, stratifying by station.

From these data, it is clear that for these cases, the NAM outperformed the other NWP models.

Clearly, the NAM-X has the least amount of bias when forecasting the surface sustained wind speed.

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, HUT, and CNU have effects (friction from trees while HUT is in a depression in elevation) reducing the surface wind speed causing a systematic bias in the forecasts.

 
 

For these images, the errors have been categorized into bins. This enables the reader to see where the bulk of the errors occurred for each model.

Although all of the models are skewed to the left, further identifying their high bias, the NAM mixed layer forecast showed the best result.

 
 
 The surface wind gust forecast is relatively the same. The top of the mixed layer winds of the NAM out perform the other models. That said, the mixed later winds of the GFS are the best for gust forecasting for these cases.  
 
 

For these images, the errors have been categorized into bins. This enables the reader to see where the bulk of the errors occurred for each model.

Although all of the models are skewed to the left, further identifying their high bias, the NAM mixed layer forecast showed the best result.

 
 
With these bins of the surface wind gust, the GFS and RUC forecasts of mixed layer winds and well as the NAM top of the ML winds showed the most promise.
 

 


 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 of -2.02 knots. Evaluation of the GFS produced a bias of -1.19 knots. Additionally, the NAM TMLWIND produced a bias of 2.13 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, taking into consideration the TMLWIND of the NAM would be deemed appropriate.

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

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. 
 


USA.gov is the U.S. government's official web portal to all federal, state and local government web resources and services.