Close Window

Univ. of Alaska: "An integrated system to improve short-term weather forecasting using numerical modeling"

Final Report

Principal Investigators:
Jeffrey S. Tilley, Geophysical Institute, University of Alaska Fairbanks, Fairbanks, AK
Kraig B. Gilkey, Science/Operations Officer, National Weather Service Forecast Office, Fairbanks, AK


1.1 Objectives

The original goal of this project was to improve short-range forecasts for the Alaska Region. Specific objectives towards achieving this goal were as follows:

Implementation of a high resolution atmospheric model (the Penn State/NCAR MM5) in the Alaska Region of the NWS through the Alaska Region Operations Network (ARONET) environment
Evaluation of the MM5's performance in an Arctic and sub-Arctic environment via a set of case study experiments.
Evaluation of the utility of products derived from MM5 forecasts as an integral part of the forecast process, including comparison of MM5-derived products with those derived from other sources

1.2 Participants

In addition to the UAF and NWS Principal Investigators listed above, the following additional personnel have been involved in the project:

David Bailey, Research Programmer, University of Alaska-Fairbanks (6/95-3/96 only)
David Covey, Programmer/Analyst, University of Alaska-Fairbanks (8/95-5/96 only)
Dennis Fielding, Graduate Student and research assistant, University of Alaska-Fairbanks (9/98-present)
Jan Julian, Electronic Systems Analyst, NWSFO/Fairbanks, AK
Debi-Lee Wilkinson, Research Programmer, University of Alaska-Fairbanks (5/97-7/98 only)
Wanli Wu, Research Programmer , University of Alaska-Fairbanks (3/96-2/97 only)

Also, the following individuals from the Alaska Region of the National Weather Service have been involved in the project on mostly an advisory capacity:

Carl Dierking, SOO, NWSFO/Juneau, AK
Gary Hufford, Region Scientist, Alaska Region HQ, Anchorage , AK
H. Lee Kelley, Aviation/Fire Wx Coordinator, Alaska Region HQ, Anchorage, AK (currently MIC at NWSFO/Fairbanks)
Carven Scott, SOO, NWSFO/Anchorage, AK

1.4 Division of Responsibilities

For this project, the NWS provided initialization and verification data from the Alaska synoptic station network and operational NWS gateway servers via the Alaska Region's ARONET environment, information on NGM/ETA model grids and conventions, and routines (gridRead and gridWrite) for conversion of input/output data over an Alaska Region domain to/from ARONET file structuring conventions and formats. The NWS also provided expertise on all facets of the ARONET environment and HP workstations constituting that environment. Finally, the NWS provided suggestions for appropriate case study experiments and assisted in evaluating the model outputs when available. This assistance also included aid in debugging various aspects of the ported codes.

For this project, UAF/GI provided access to the Penn State/NCAR MM5 mesoscale modeling system as well as documentation and consultation and training in the use of MM5 to the NWS PI. UAF/GI also provided expertise on regional numerical weather prediction models and took the lead in development and implementation of ARONET-compatible pre- and post-processing modules for the MM5 modeling system. UAF/GI personnel also took the lead in implementation of the primary numerical model within the MM5 system on the ARONET HP-UX workstations as well as testing via historical mode simulations.

1.5 Description of Research/Development Accomplishments

MM5 Implementation on ARONET, including pre-and post-processing

The development of a version of MM5 suitable for use on the ARONET environment ended up to be a more problematic proposition than expected, especially given the proliferation of MM5 on other workstation platforms simultaneously during the time period of this project. The primary tasks requiring extensively more time and effort than originally expected involved streamlining and conversion of the standard MM5 preprocessors, originally hardwired for CRAY platforms, to the ARONET HP-UX environment. Prime among the conversion tasks was conversion of the preprocessing routines, which assume data is in one large file, to accept data from a large amount of individual files, each containing one field at a given level and time, in NETCDF format. Other issues complicating these tasks are elaborated on in section 4.

As of early 1998, a modified version of the MM5V1 modeling system (including pre- and post-processors) was nominally operating, under HP-UX 9, on a single ARONET workstation dedicated to the NWS SOO. "Nominally operating" means the following: 1) No automated job scripting had been set up to execute MM5 on a routine basis, including quality control checks on incoming datastreams; 2) no automated graphics capability had been yet established on ARONET, though manual graphics capability was available via the ARONET xmMap utility; 3) problems with boundary values at the edge of the model domain, set the same as the standard Alaska grid domain received for NCEP model data; and 4) two test cases covering short periods in April 1997 and May 1998 have been successfully run.

It should be noted that the aforementioned problem related to the boundary conditions did not substantially degrade the model solution in the interior of the domain. This problem has also been seen by other groups running MM5 in real time and is believed related to the fact that in our implementation a) the model is initialized on the exact same grid it is executed on, which may lead to an exacerbation of b) mismatches between solution of the initializing analysis (Eta, NGM or AVN models) and the MM5 solutions. Given that the interior domain solution did not seem to be adversely affected, this problem was deemed of a lower priority than an upgrade in the basic model version, described in the next paragraph.

Given the general MM5 community move to MM5V2 (and soon to be MM5V3), the PIs decided that upgrading the functioning MM5V1 system to MM5V2 on an AWIPS compatible workstation running HP-UX10.2 was the best area on which to focus efforts during the latter stages of the projects. Unfortunately, this change proved to be more problematic than expected given preprocessor code incompatibilities between the two versions of HP-UX as well as physical resource constraints associated with the NWSFO's move to a new facility in the last quarter of 1998. This move effectively disrupted work on the project from August through November of 1998, other than work on the post processing visualization modules, to be discussed next.

Collaboration with a M.S. student in Computer Science at the University of Alaska-Fairbanks was initiated with respect to visualization of ARONET MM5 output. In the original implementation of ARONET (and AWIPS as well), MM5 output can only be represented in 1-or 2-dimensional (planar) formats. This limits to some extent the utility of the MM5 model on ARONET since important atmospheric processes can sometimes be discerned more clearly through a 3- or 4-D visualization of atmospheric fields. Discussion amongst the PIs concluded that a viable approach to higher-dimensionality visualization would be a fusion of the Vis5D package originally developed at the University of Wisconsin with the ARONET graphical user interface and visualization utilities. Mr. Fielding, who is now employed as a staff programmer, has made excellent progress and a functioning prototype has been developed, with a May 1999 anticipated porting/implementation date on NWSFO systems.

Evaluation of MM5 Utility Based on Case Study Experiments

Tests to evaluate the MM5 performance over the Alaska Region have, with the exception of the April 1997 and May 1998 events discussed above, been conducted solely with the use of the Arctic Region Supercomputing Center Cray systems. A brief summary of test cases follows below.

1) 12-15 September 1992 Interior Alaska heavy snow event
During the period of 12-15 September 1992, substantial (30 cm or greater) snow fell over portions of interior Alaska in association with a quasi stationary trough, the development of which was not captured by the operational NGM model, but was captured well by the MM5.

MM5 simulations were conducted using a hierarchy of three nested grids at 75, 25 and 8.33 km resolutions. The model captures the development of the trough over the period, even at the coarsest resolution (75-km grid) and predicts precipitation over approximately the correct regions. Forecasts of precipitation type (liquid vs. frozen) and timing of the precipitation also align surprisingly well with the observations, though the amount of precipitation is underforecast.

Simulations on the 25-km grid showed general improvement over the 75-km grid resolutions with respect to precipitation. Precipitation type again appeared to be forecast relatively well. The simulations on this grid also depicted fairly well the eastward development of the mesoscale quasistationary trough from the Chukotsk Peninsula into the Tanana Valley of Alaska, as well as a weak cyclone associated with this trough.

Simulations on the 8.33-km grid did not continue to show an improvement, largely due to problems in the location of the precipitation systems relative to the grid boundaries and an attendant problem relating to the use of the cumulus schemes at this boundary.

2) 31 March 1993 Turbulence event
On the afternoon of 31 March 1993, an Evergreen Airlines cargo plane dropped an engine into a vacant parking lot near Anchorage International Airport shortly after takeoff. The airline claimed that extreme boundary layer turbulence contributed to the incident. Operational NWS forecasts indicated a possibility for up to moderate turbulence south and west of Anchorage, but not within the Anchorage bowl area proper.

MM5 simulations were conducted using a hierarchy of four nested grids at resolutions from 63 km down to 2.33 km, with the finest grid centered at Anchorage International Airport and covering the entire Anchorage bowl and surrounding terrain. Turbulence diagnostics, including Richardson number, derived from simulations at both 7 and 2.2 km initialized 24 hours prior to the event, strongly suggested a likelihood for strong turbulence in the boundary layer in the vicinity of the airport during the afternoon of 31 March. A potential for strong turbulence was also indicated at numerous locations around the Anchorage bowl, associated with wavelike structures in the vertical motion field. All of these features could be linked to strong downslope lee side flow associated with a strong synoptic cyclone translating through the Gulf of Alaska to the south of Anchorage. In all, the simulations indicated that the MM5 could have provided useful guidance to aviation forecasters for this event and gave some credence to the claims of Evergreen Airlines regarding the nature of this incident.

3) 22 December 1996 turbulence event
During the evening of 22 December 1996, Alaska Airlines flight 67 was traversing a typical flight path parallel to the coastline over the northernmost Gulf of Alaska. After experiencing some light to moderate turbulence on ascent out of Juneau, the plane briefly encountered turbulence near Yakutat, AK at 35000 feet which was strong enough to result in injuries to two members of the flight crew. The aircraft was able to continue to Anchorage where the injured crewmembers were taken to hospital for treatment. Some turbulence is often common along this flight path, but turbulence of this severity was not well forecast by the operational models for this incident.

MM5 simulations were again conducted with a hierarchy of three nested grids at resolutions of 63, 21 and 7 km, with 36 vertical levels used in the simulations. Simulations were initialized both at 16 hours and 4 hours prior to the event in order to determine the degree that MM5 could simulate such an event at the short range. Turbulence diagnostics computed from both simulations strongly suggested the likelihood of turbulence in the vicinity of Yakutat at the proper flight level and within approximately the correct time frame. Further, the simulations gave support to the pilot's contention that the turbulence was wave-based. The MM5 simulations clearly suggested that upward propagating gravity waves were associated with a periodic strengthening and weakening of the vertical and horizontal shears at the aircraft's flight level during the evening of 22 December. Such behavior is consistent with pilot reports suggesting that turbulence in the Yakutat area was intermittent during the entire day on 22 December. The role of a southward propagating jet streak with respect to this event is still unclear, though we hope to be able to investigate that further in the future.

Most encouraging was the fact that the MM5 simulations initialized only 4 hours prior to the event could have been used in real -time (were the model actually run in real time for this case) and provide valuable guidance. This suggests further that the MM5 could be robust enough to meet operational needs in Alaska if properly implemented.

4) 1-2 December 1992 icing/turbulence event
A large synoptic system in the southern Bering Sea translated slowly westward during the period 30 November- 2 December 1992. Numerous reports of inflight icing in southwestern and south central Alaska were associated with this system, which brought a broad moist southerly airflow over much of the region. Inflight icing was also prevalent in southeast Alaska in association with a meso-alpha scale cyclone in the eastern Gulf of Alaska which eventually merged with the advancing synoptic low. The airflow around the meso-alpha scale low also appeared to be associated with strong winds and turbulence throughout Turnagain Arm east of Anchorage, which was not well forecast operationally.

MM5 simulations of this event on a hierarchy of four nested grids (60, 20, 6.67 and 2.23 km) were undertaken as part of a cooperative effort during the 1997 COMAP training course that all of the Alaska NWS SOOs participated in. The main objective was to determine how well MM5 might forecast inflight icing conditions through its cloud liquid water fields. This objective and other objectives, including the application of current icing algorithms to the MM5 output, are being examined in more detail in another COMET project currently entering its second year. Here we will simply note that the MM5 cloud water (and other hydrometeor) fields contained useful information that could have benefited an inflight icing forecast.

The MM5 provided a good forecast of the large synoptic system, including the merging with the aforementioned meso-alpha scale low. The main deficiency in the model involved the speed at which moisture was advected into southwestern and south central Alaska. The simulations were slow to bring in adequate moisture during the first 12 hours but rapidly compensated during the second 12 hours so that agreement was reached with observations by 24 hours. This behavior can likely be attributable to model spin up of the cloud condensate fields. Further, it was able to produce an environment on the 6.67 and 2.23 km grids that was consistent with the reported strong winds and turbulence (i.e., strong mesoscale vertical motions and gradients in vertical motion, flow deformations, etc.) in Turnagain Arm during much of the proper timeframe.

5) 1 Nov 1993 icing event
As for the preceding case, these simulations were originally undertaken as a collaborative effort with the 1997 COMAP training course, and again the focus of the simulations was an inflight icing period that occurred during 1 November 1993. A similar type of scenario was involved with this case, except that the synoptic low involved tracked more northerly into the Bering Sea during the period in question. As a result, the strong moist southerly flow was confined more to southwestern Alaska and the Aleutian Islands than in the December 1992 event.

The MM5 simulations used the same grid hierarchy as for the December 1992 case, although results from the 2.23-km grid, centered over Anchorage and Turnagain Arm, were viewed from an icing standpoint for this event given that little orographically related turbulence was reported. The simulations adequately simulated the overall system evolution on the synoptic and meso-alpha scales, although, as was the case for the December 1992 event, the cloud condensate fields developed initially a bit more slowly than observed.

Further analysis of this event is continuing in conjunction with the other COMET project referred to above. To summarize, it would appear that MM5 will have some significant utility, either alone or in conjunction with other icing diagnosis/forecast algorithms, to provide for improved depiction and forecasts of inflight icing in the Alaska Region.

6), 7), April 25 1997 and May 27, 1998 events
MM5 simulations on a single grid with ~ 80 km spacing were conducted on the NWS ARONET platforms for these two events. The prime motivation for these simulations was merely to test the system functionality on the ARONET platform. Thus, extensive evaluation of model performance has yet to occur for these cases.

The first event constituted a quiescent period dominated by a large scale anticyclone over the majority of the state. Preliminary analysis suggested that MM5 adequately captured the anticyclone behavior although differed with operational models on some of the mesoscale details as might be expected.

The second event constituted a convectively driven heavy rain event over Interior Alaska, which culminated in flash flooding over the southern foothills of the Brooks Range. Preliminary analysis suggests that the MM5 simulation, if available in real time, would have provided additional quality guidance for forecasters in predicting a widespread heavy rain scenario with flash flooding. Further analysis is needed to quantify the performance of the model in both of these cases.


In addition to participation in the COMET Outreach Workshops noted in Section 3, the following exchanges between the University, NWS, Air Force and other entities associated in some way with this project are briefly listed below:

1) August 1995: the UAF PI traveled to the Alaska Region HQ (ARHQ) in Anchorage for meetings with the Alaska Region Chief Scientist (Dr. Gary Hufford) and the Region Director (Mr. Richard Hutcheon) to discuss future collaborations and lay the foundation for an upcoming COMET Symposium and Alaska Region workshop on Mesoscale Prediction.

2) September 1995: meetings took place involving the Alaska Region Chief Scientist, the Fairbanks SOO, the directors of COMET and UNIDATA, the UAF PI and other Geophysical Institute personnel at the UAF campus to discuss potential projects (COMET-funded or otherwise) and the application of UNIDATA to those projects. The NWS/Fairbanks assisted with organization of this visit.

3) September 1995: the UAF PI participated in joint meetings with the Fairbanks SOO, Mr. Stan Keefe (NWS/Anchorage) and Dr. Warren Blier (UCLA) in Fairbanks to discuss this project and the efforts of Blier and Keefe related to polar lows in the Gulf of Alaska and storm surges in Norton Sound.

4) Fall 1995: a joint proposal was submitted to the NOAA Cooperative Research Program by the UAF PI and the Alaska Region Chief Scientist for additional mesoscale simulation studies. An additional $30,000 was awarded for this work, some of which is also reported above. (i.e., the two projects leveraged each other)

5) March 1996: both PIs met in Fairbanks with the Region Scientist, Dr. Ken Dean (UAF) and forecaster Douglas Christopherson (NWSFO-Fairbanks) to discuss plans for the COMET symposium that was held in June 1996. (A separate report submitted to COMET in 1996 details this meeting).

6-8) 1997: Three visits of the UAF PI to ARHQ and the NWSFO/Anchorage SOO occurred. Ideas for future projects and collaborative efforts were discussed during these visits, as were potential ideas for project proposals. The last visit in 1997 (October) also associated with a presentation to the Anchorage local chapter of the AMS on mesoscale modeling. Ground work was also laid at the October meetings, which also included representatives from Elmendorf Air Force Base and the Head of the University of Alaska Anchorage (UAA) Aviation Technology Program for what is now the Alaska experimental forecast facility at UAA.

9) October 1997: Presentation of mesoscale modeling activities at the CSTAR Science Workshop, Silver Spring
MD and the FAA's "Alliance for Safety" meeting, Anchorage AK by the UAF PI.

10) February 1998: At the request of the NWS/Juneau SOO and ARHQ, the UAF PI conducted a case study simulation of an event that took place off the coast of Yakutat, AK in early December 1997 was conducted. The event featured unusually strong southeasterly winds offshore of Yakutat in association with an approaching complex, occluding cyclone system and attendant mesoscale structures. A preliminary analysis was conducted which resulted in a paper presented by the Juneau SOO at the 1998 Northwest Weather Conference in Seattle.

11) April 1998: The UAF PI attended and presented an invited talk at an Alaskan Aviation Symposium on the subject of application of mesoscale modeling to aviation interests. This was followed by meeting with individuals from NCAR, the Anchorage NWSFO, ARHQ and the FAA to discuss possible future areas of research.


Tilley, J. S., A. H. Lynch, D. A. Bailey and K. Gilkey, 1995: Improving Short-term Weather Forecasting Using Numerical Modeling Techniques. Alaska COMET Outreach Symposium, University Corporation for Atmospheric Research, Fairbanks, AK June 1995.

Tilley, J. S., D. A. Bailey, K. Gilkey, J. Julian and W. Wu, 1996: Towards Improving Short Term Weather Forecasting Using Numerical Modeling Techniques: An Update. Alaska COMET Outreach Symposium and Workshop on Operational Mesoscale Prediction, University Corporation for Atmospheric Research, Fairbanks, AK, 10-13 June 1996.

Tilley, J. S, 1996: Modeling Meso-beta Scale Processes in Alaska. Alaska COMET Outreach Symposium and Workshop on Operational Mesoscale Prediction, University Corporation for Atmospheric Research, Fairbanks, AK, 10-13 June 1996. (invited)

Tilley, J. S. and K. Gilkey, 1997: The Use of High Resolution Mesoscale Models to Support Operational Forecast Problems. 13th Conference of Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography and Hydrology, American Meteorological Society, Long Beach, CA, 2-7 Feb 1997, 383-386.

Tilley, J. S., D. L. Wilkinson, J. Miller, K. Gilkey, C. Scott, H. L. Kelley, M. K. Politovich, T. T. Warner, W. L. Chapman, A. H. Lynch, D. A. Bailey, W. Wu, J. W. Bao and A. L. Gravier, 1997: Regional Modeling in the Western Arctic. ARSC User Forum, Arctic Region Supercomputing Center, Fairbanks AK, 12-14 Oct. (invited)

Tilley, J. S., K. B. Gilkey, C. A. Scott and D. L. Wilkinson, 1997: Using High Resolution Mesoscale Models to Support Operational Forecast Problems. CSTAR Science Workshop, NOAA/Office of Meteorology, Silver Spring MD, 6-9 Oct. (invited)

Tilley, J. S., D. L. Wilkinson, C. A. Scott, H. L. Kelley and K. B. Gilkey, 1997: Applications of Mesoscale Atmospheric Models for Improved Forecasting of Aviation Weather. FAA "Alliance for Safety" Meeting, FAA and NOAA/NWS, Anchorage, AK, 21-22 Oct. (Invited presentation; also presented at Anchorage Chapter of American Meteorological Society October meeting, 23 Oct).

Tilley, J. S., K. Gilkey, C. Scott and D. L. Wilkinson, 1998: Implementation of a High Resolution Mesoscale Model within an Operational NWS Regional Network. 14th Conference on Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography and Hydrology, American Meteorological Society, Phoenix, AZ, 11-16 January, J3-J7.

Tilley, J. S., D-L. Wilkinson, K. B. Gilkey and H. L. Kelley, 1998: Short Range Simulation of an Alaskan Clear Air Turbulence Event. 16th Conference on Weather Analysis and Forecasting, American Meteorological Society, Phoenix, AZ 11-16 January, 68-70.

Tilley, J. S., 1998: Towards Improved Forecasting of Icing and Turbulence in the Alaskan Region. 1998 Spring Air Fair Expo, Northern Alaskan Aviation Symposium, Fairbanks, AK, 6 March.

Tilley, J.S., 1998: Application of Mesoscale Numerical Weather Prediction Models to Aviation Weather at the University of Alaska: Past, Present and Future. National Weather Service/Alaska Region Aviation Weather Workshop, Anchorage, AK, 5-10 April.

Tilley, J. S. and K. Gilkey, 1998: Efforts to Implement MM5 within the Alaska Region Operations Network (ARONET) Second Workshop on Real-time Mesoscale Numerical Weather Prediction, National Center for Atmospheric Research, Boulder, CO, 17 June.

Tilley, J. S, K. G. Gilkey, D. L. Wilkinson, M. K. Politovich, C. A. Scott, H. L. Kelley, G. Hufford and C. Dierking, 1998: Mesoscale Modeling in Support of Observational Studies in the NWS Alaska Region. NOAA Review of Cooperative Institute for Arctic Research, Fairbanks, AK, 16 September 1998.

Fielding, D. J., K. B. Gilkey and J. S. Tilley, 1999: Development of 4-D Visualization Capabilities Within the NWS Alaska Region Operational Network Environment. 15th International Conference on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, 10-15 January, Dallas, TX, 502-505/


4.1 University's perspective

Prime among the University benefits from this project has been the establishment and expansion of a positive relationship between UAF and NWS. New directions for collaborative efforts are being discussed by several UAF investigators and the NWS, including efforts with the Alaska Experimental Forecast Facility noted above. Particularly beneficial has been the extension of NWS/UAF interaction to include both the Anchorage and Juneau Forecast Offices and the Alaska Aviation Weather Unit. With the collaboration now informally extending to the other offices in the Alaska Region, a new perspective and appreciation for the types of forecast problems (and successes/failures of operational models) outside of interior and northern Alaska has been gained.

Much has also been learned along the way about mesoscale processes in Alaska and the relative ease or difficulty in simulating them to the level required for good operational application. Also a valuable lesson learned was that turning a research quality model into an operational product can often be a process full of unforeseen pitfalls, including hardware and software limitations. Given limited funding resources and time constraints that have been routinely faced by both investigators (not the least of which was the NWSFO office move noted above), this lesson has come at the price of not having a currently operational version of the entire MM5V2 system working at the NWSFO.

In general, progress on this project was made during relatively short, concentrated intervals in a start-stop fashion. This situation was not optimal and rather frustrating for both investigators, as was the "revolving door" of UAF programmers assisting the PI with this project. It has been another hard lesson to learn that it is difficult to recruit and keep qualified programmers in the University of Alaska system, partly because the salary scale is not competitive nationally. This has led to effectively an annual turnover in programmers as they find more lucrative employment elsewhere. This resulting new hire spinups have further slowed the project by limiting the UAF PI's time.

Strategies and planning for future projects will have to take this fact into consideration, as well as the fact that with NWS personnel stretched as thin as they are in the Alaska Region, it has equally difficult for NWS scientific staff (e.g. SOOs) to dedicate large blocks of time to any project that is not immediately related to present operations. This latter situation is starting to improve, so hopefully this issue will be less of an obstacle than in the past couple of years.

4.2 NWS Perspective

This project generated many positive results, which fall into roughly two categories: 1) demonstration of the utility of mesoscale models in NWS operations, and 2) the strengthening of interactions between academics and the NWS in the Alaska Region.

In every case study of MM5 in this project, information of direct usefulness to operational forecasts was seen. Had that information been available to forecasters, it would have altered forecasts sent to the public and industry in important ways. Although the MM5 system has yet to run operationally inside the NWS systems, it is clear that the physical resources (computers, connectivity, data, etc.) are either available or affordable. Also the MM5 runtines with useful resolution and physics on even low end Unix workstations is in the timeframe for useful forecast guidance.

The interaction between the NWS and academics due to this project has spread well beyond the dimension of the original project. This has been important for all three Forecast Offices (NWSFO), the Alaska Aviation Weather Unit (AAWU) and the developing Alaska Experimental Forecast Facility (AEFF). With the Forecast Office finally co-located with the International Arctic Research Center and Geophysical Institute on the UAF campus, interactions began quickly due to previous collaborative work such as this project.

There were some unforeseen dynamics that prevented this project from reaching the original goals. The loss of the Forecast Office's Meteorologist in Charge coincidentally with relocation of the Forecast Office removed the NWS SOO from participation effectively from May 1998 onward. The pressure to migrate to a new version of the operating system environment and a new version of MM5 also cost significantly more than expected.

An important lesson learned is that maintaining a complex modeling system like MM5 has fairly high costs in specialized manpower. What this means is that a modeling system can't be treated as an "appliance". To satisfy this requirement requires both a commitment by the NWS to support it, and access to the very specialized personnel available at universities or other modeling institutions. However, it is still believed that this is a viable and beneficial activity for NWS forecast offices.

When this project began, the technical infrastructure was provided by the Alaska Region Operations Network (ARONET). Over the course of the project the nationally supported Advanced Weather Information Processing System (AWIPS) has come from an unknown quantity to the mandated solution in NWS forecast offices. Thus ARONET as a legacy system will be replaced in the near future. This will be a significant change of operating environments and data formats. The migration of an MM5 system to AWIPS now, while cheaper than the original ARONET development due to our acquired experiences, will still be a substantial effort.