Research Efforts in the US and Collaborations with Asia toward Storm-Scale Numerical Weather Prediction

Kelvin K. Droegemeier and David Jahn
Center for Analysis and Prediction of Storms, University of Oklahoma
Norman, OK 73019, USA

With recent enhancements in weather observation technology, such as the national US network of Doppler radars, satellite remote sensing and highly resolved surface observation networks, the availability of the world's most powerful parallel-processing supercomputers, and a much improved understanding of the small-scale atmosphere through intensive research across all sectors of the meteorological community, the ability to predict explicitly the initiation and evolution of individual convective storms and their winter counterparts appears to be within reach.

In light of these advances, there has evolved a rapidly growing interest in the application and operational implementation of storm-scale numerical weather prediction not only by the federal weather agencies of various countries, but also within private industry around the world. In the US, the National Centers for Environmenal Prediction hosted a Workshop on Centralized and Local Modeling December 1995 as a first step to evaluate nationally the current level of research in the area of real-time weather modeling at the national labs and university research centers and to address what kind of relationship and cooperation should exist between the federal and academic components to affect and enable the implementation of storm-scale numerical weather prediction (NWP) nation-wide. It was noted that, at the present time, 22 universities across the nation are collaborating with a total of 25 NWS Forecast Offices (WFOs) on local modeling efforts. This activity ranges in scope from informal exchanges of PC-based forecast products to extensive evaluations of locally-generated predictions that make use of realtime WSR-88D and other high-resolution data. In most cases the associated numerical models are initiated from a cold start, i.e., without data assimilation, and use resolutions of 10 to 25 km -- slightly higher than current operational models.

In particular, the National Science Foundation has invested a portion of its resources to establish a national research center, the Center for Analysis and Prediction of Storms, as a means of addressing the challenges involved in successfully predicting storm-scale weather phenomena. Founded in 1989, the mission of CAPS is to demonstrate the practicability of storm-scale numerical weather prediction and to develop, test, and validate an end-to-end regional forecast system appropriate for operational, commercial, and research applications.

The challenge of storm-scale NWP

The need to develop more accurate forecasts is demonstrated by the recognized limitations of the information currently afforded the National Weather Service in predicting weather for a specified time and location. Although the models used operationally by the National Weather Service are able to forecast the environmental components required for thunderstorm development, such as moist air near the surface, dry unstable air aloft, and the development of features, such as cold fronts and dry lines, that provide concentrated regions of vertical forcing required to trigger storms, these models do not have adequate spatial resolution to pinpoint exactly where and when an individual storm will develop or even the storms themselves once they do form.

The fact that these models are run over the entire U.S. puts a practical limit on the grid resolution that can be achieved. Currently the national operational models use a grid spacing of no less than 29 km. (Although it should be noted that NCEP is currently conducting tests with the Eta model using a grid resolution of 10 km.) A storm-scale model, such as the Advanced Regional Prediction System (ARPS), developed by CAPS, is typically run with a grid spacing of 1 to 5 km to explicitly resolve specific storms. Because thunderstorms have relatively short lifetimes (a few hours) as compared to larger-scale weather systems (a few days), their associated numerical forecasts must be generated and disseminated with sufficient lead time as to be of practical value. The computational requirements of storm-scale prediction poses an extreme challenge that demands the availability of supercomputers having sustained teraflop performance, gigabyte capacity memories, and parallel input/output coupled with the formulation of a very efficient numerical code suited for such computer architectures.

It is necessary that the model account for all processes that are dynamically pertinent on the storm-scale even though they may be neglected for the large-scale. For example, neglecting the effect of vertical wind accelerations is adequate when forecasting larger-scale systems, such as low-pressure systems and fronts, but inappropriate for thunderstorms that can have updrafts as large as the environmental horizontal wind (> 40 m/s). Also, storm-scale models can explicitly resolve, i.e. not just parameterize, certain processes such as the development of cumulus clouds, which alters the overall radiation budget and greatly impacts the forecast result.

Besides the development of an accurate and robust numerical model, an effective prediction system also requires the means of accurately sensing and analyzing the observational data needed to initialize the model. There are currently several data sources available that provide wind, temperature, and dewpoint quantities both at the surface and aloft. The temporal and spatial increments of the data, however, vary considerably depending on the type of sampling network and the resolution of the instruments. For example, rawinsondes give the wind and temperature profiles of the atmosphere but are spaced 300 km apart on the average and are launched only twice daily. Surface observations, however, from relatively highly resolved "mesonets" provide data every 15 minutes. To incorporate these data into the model solution, data analysis techniques are needed that account for data density disparities and insure that the analyzed fields are dynamically consistent. CAPS has developed the ARPS Data Analysis System (ADAS) to objectively analyze observational data and create a suite of "nowcast" products and provide an initialized state for the ARPS.

The national network of NEXRAD radars is another vital source of data for storm-scale prediction. A major difficulty in using radar data, however, is that the radar can only sense precipitation intensity and the radial component of the wind, i.e., the wind parallel to the radar beam. Yet, model initiation requires the full 3-D wind as well as the temperature, pressure, and water vapor profiles. CAPS has developed several methods of retrieving those fields not directly measured by the radar via a numerical solution, which nets a set of fields that is both dynamically consistent with the physical equations and constraints that govern atmospheric flow and is also in agreement with observed fields.

Is the small-scale atmosphere predictable?

In spring 1995 and 1996, CAPS conducted major tests involving a full real-time run of its forecast system, from ingest and quality control of observational data (including WSR-88D data) to forecast generation. These tests have provided an opportunity to evaluate the feasibility of the prediction of storm-scale weather phenomena in real-time and in an operational setting. During each testing period, the ARPS was run daily over areas of anticipated severe weather in order to produce forecasts up to 6 hours in advance. For days with significant weather, forecast results were transferred to the local National Weather Service Forecast Office as well as NCEP's Storm Prediction Center, who were asked not only to comment on forecast accuracy but also to provide suggestions regarding products and product format most useful for the forecasting of storms. Although both years evidenced a number of encouraging cases for which the forecasts were very similar in timing and location to storms that later developed, there is evidence that the improvements incorporated in 1996, such as enhanced model numerics and the ingest of radar data, resulted in an overall success in storm prediction which was a step up from that achieved the previous year. In particular during the 1996 operations, with a success rate of 8 out of 10 days for which convection occurred, the ARPS correctly predicted up to 6 hours in advance the location and timing of storm development.

Based on its operational tests to date, CAPS believes that the small-scale atmosphere is indeed predictable, i.e., that the initiation, morphology, and decay of individual storms can be foreseen with definable skill, in some cases for 6 to 9 hours. Experience to date also suggests that the timing of some events can be captured to within 15-30 minutes, with spatial accuracies up to about 30 miles over a 6 to 9 hour period. Of course, these numbers are problematic and likely represent the best-case scenarios where storm morphology is associated with strong forcing, e.g., terrain, fronts, drylines. Nonetheless, these levels of accuracy have in fact been achieved in several realtime forecasts during the past 2 years, and continued testing is underway by CAPS scientists to broaden the spectrum of cases in the dataset.

Collaborations in Asia

It should be noted again that CAPS is not the only center that is contributing toward the science of numerical prediction, although they remain unique in their quest to demonstrate the practicability of storm-scale NWP on an operational level. As stated earlier, up to 22 university and federal research groups in the US are working toward real-time small-scale NWP in some form or another. Furthermore, several groups overseas including Asia also have a significant interest in this branch of NWP and are interested in cooperating with counterpart US research entities.

In particular, the Korea Meteorological Administration (KMA) would like to implement a mesoscale model operationally to improve their forecasts of smaller-scale phenomena, e.g. the prediction of localized heavy rain to improve flooding forecasts or prediction of boundary layer winds over a restricted vicinity for purposes of air quality management. Toward this goal, they are collaborating with CAPS and will be conducting an operational test this summer using both ADAS and ARPS to generate fine-scale weather forecasts over the Korean peninsula.

Such collaborations between the US and the Republic of Korea is a natural consequence of their common research interests. Specifically, both countries exhibit state-of-the art observation networks that include spatially dense automatic surface observation networks, which are unique to any other in the world, as well as a nation-wide network of Doppler radars. Both countries have also undertaken extensive research interests in the effects of terrain on the dynamics of small-scale weather processes. Korea exhibits a wide-range of orographic features in a relatively small area with a narrow transition region between sea coast and mountainous terrain.

In order to identify beneficial means of collaboration between the US and Korea, a joint meteorological workshop will be held at Seoul National University next October. The purpose of the workshop is to create a working forum by which scientists from the US and Korea can share and seek to significantly improve current techniques used in the observation, analysis, and prediction of localized weather effects such as severe storms, fog, flooding, or strong damaging winds.

US collaborations with other countries in Asia are also promising. For example, a group in Thailand will be installing ARPS at their facilities in Bangkok in preparation for an operational test similar to what will be conducted in Korea. Their interests are in improving their forecasts over Southeast Asia with regard to typhoon development as well as prediction of wave height for the Gulf of Siam and the Andaman Sea.

Kelvin K. Droegemeier
Center for Analysis and Prediction of Storms and School of Meteorology
University of Oklahoma, Norman, OK 73019, USA
E-mail: kdroege@ou.edu
URL: http://wwwcaps.uoknor.edu/CAPS/kdroege.html

David Jahn
Center for Analysis and Prediction of Storms
University of Oklahoma, Norman, OK 73019, USA
E-mail: djahn@ou.edu