Page	Explanation (self-explanatory pages omitted)
2	While this training is geared specifically for fire weather forecasting, it will be useful for many other forecast problems too.
4	The graphic shows the various small domains that can be selected. A consequence of the small domain size is that a mean wind of 40 knots will advect conditions at the boundary into the center of the domain in just 10 hours. The one shown in solid dark green was used for the case we will be examining in detail.
5	With initial and boundary conditions coming from the Eta model, the NMM forecast is heavily constrained to resemble the Eta forecast. Differences will result from topography differences better effective resolution and amplitude of small-scale features due to less numerical damping and the slight improvement from 12 to 8 km grid spacing better vertical resolution in the boundary layer over elevated terrain different handling of flow around and over topography in Eta and terrain-following coordinate Physics are mostly inherited from the Eta and thus are not tuned to the NMM. Additionally, the Eta analysis is not consistent with the NMM dynamics and topography. This inconsistency is alleviated in a sort of spin-up method utilizing a time filtering technique which will be explained later. The topography differences will be shown and discussed in detail later.
6	Model changes in Eta and NMM from just before the case example used here up to 2004 fire weather season. Note that different versions of the land surface model are used starting within a week after this case example, while the NMM had two bug fixes at later dates. In fall 2004, NMM will catch up to the physics changes implemented in the Eta in July 2003 and few minor tweaks added. The notable of those small modifications are 1) to discontinue boosting the surface roughness values by terrain variance, meaning near-surface winds in the NMM will increase in regions of steep mountains and valleys, and 2) convection triggering will be somewhat reduced, so that resolved motions leading to grid-scale precipitation will play a slightly larger role. Meanwhile, Eta model changes include decreasing the surface roughness in all locations, which will increase near-surface winds in the Eta. After the fall 2004 Eta change package, the Eta will be frozen and all development will switch to the NMM.
7	Large-scale flow constrained by Eta solution Temperature plummets under calm, clear nighttime conditions in the lowest few model layers. Sometimes this verifies well, but often it is too cold or much too cold. The cold bias is shallow but it produces somewhat too large a net lower tropospheric cooling, so the mixed layer is slow to warm up enough to deepen the next afternoon. Flow follows terrain far more than in the Eta model, scouring out valleys and blowing up mountains Gravity wave propagation and tilt and consequent flow acceleration and mountain wave temperature structure appears to be physically realistic, unlike in the Eta model Detail is enhanced compared to Eta model, partly because of downscaling from 12 to 8 km, but also because the numerical scheme provides less damping, nonhydrostatic solutions require less damping, and less artificial diffusion is added for numerical stability. Thus features will have more amplitude and more structure. Isolated/scattered mountain convection often missed in Eta is captured in NMM, but perhaps too much? Orographic heavy precipitation episodes tend to be overdone in NMM. Quality of forecast may not be fairly assessed by standard point RMS error statistics because a timing or placement error of a weather feature whose structure is otherwise correct will produce a larger RMS error for a model that forecasts it at full amplitude than another model that has the same timing and placement but has the feature at only half its actual amplitude
8	The output parameters sent from NCEP for use in the field are shown. All are on an 8 km grid every 3 hours to 48 hours. They do not include turbulent kinetic energy (TKE) which could be a useful parameter to determine the depth of mixing. FXNET is only pushing out a fraction of these available output parameters - all those shown in black font. Those in burnt orange font are available in the same file but are not being distributed by FXNET. BUFR soundings are available from the NCEP server for use with BUFKIT, along with much larger grib files containing many more variables including many 3-d variables.
9	Moisture, temperature, and winds aloft are practically identical at the INITIAL time in NMM and Eta, as would be expected and as shown in these plots. But they are not quite identical, as will be explained in the next slide. Also, due to the different topography, there can be profound differences at the surface. The surface CAPE (not shown in this sequence) shows differences not seen for parcels using 30-hPa average conditions (shown here). The NMM fields were all plotted using the 8-km grid while the Eta upper air data were plotted using a 32-km grid and Eta CAPE was plotted using a 12-km grid. Winds are not plotted at every grid point.
10	Near-surface fields are considerably different in places. This might be expected from using different topography, but there is another cause too, and it has a slight impact on all fields at all levels. If the model were started from the Eta fields plus vertical interpolation to the NMM topography, the model would sputter along, spewing out lots of gravity waves. Its forecast of any parameter at any grid point might look like the green curve in the lower graph. To create an initial condition consistent with the NMM, the Lynch digital filter is applied, meaning the model is run with all physics (radiation, surface fluxes, moist processes, etc.) turned off, just running dry dynamics. It is run for 110 time steps (36 minutes) forward and also backwards in time. The resulting solution might look like the green curve in the top graph. This solution is filtered in time. The resulting filtered solution might look like the blue curve in the top graph. This process is repeated a few times. The value of this solution at the initial time is the initial condition used in the NMM. Note that it is different than the value from the Eta. However, when the model is actually run, the solution is now much smoother but within 12 hours converges to the same solution the NMM would have had with the Eta initial condition, just without the noise. The result is that initial winds and temperatures reflect an adjustment to the NMM topography, not merely an interpolation. We will next see some of these differences. In case you were wondering, the 12-km Eta also performs a similar procedure to its own 3d-var analysis, because the corrections the 3d-var analysis procedure makes to the first guess were found to create a noisy forecast in the 12-km Eta. A few years ago when the Eta was running at coarser resolution, this procedure was not performed.
11	The initial NMM winds (shown in knots) over the lowest 30 hPa are weaker than in the Eta, even over the plains. (data are plotted from 8-km grid for NMM and 12-km grid for Eta, as will be the case for all surface data shown in this lesson)
12	The background shades of gray are the topography, lighter shades for lower elevation, of each model over a small section of the domain. More about the topography will be shown later. Be sure to click on the check box to display the 2-meter temperatures. Note that the temperatures are warmer in valleys in the NMM and colder on mountain tops, and that the more detailed terrain has a corresponding temperature pattern.
13	Be sure to click on the check box to display the 10-meter winds. Note that the NMM has stronger winds in the valley at the north end of the plot and over the peaks in northern Wyoming towards the middle of the plot, while flow is channeled more through the valley in eastern Idaho.
14	The remainder of this lesson will mostly focus on the Eta and NMM forecast from 6 UTC July 1, 2003 and verifying observations. All NMM data are from the same 8-km grid distributed to IMETs. Eta data are from the 12-km grid for surface and boundary layer fields, some of which are also available in AWIPS on this grid and others of which are available in AWIPS on the 20-km grid. Eta upper air data are from the 32-km grid, and the same fields are available in AWIPS on the 40-km grid. Soundings all are from BUFR sounding data from each model, which are raw model data at the nearest model grid column.
16	Large scale conditions are seen to be the same in NMM and Eta out to 48 hours except for a high height bias problem in the NMM caused by an inconsistent lateral boundary condition formulation that was corrected in October 2003.
17	Same as previous except winds instead of heights. Note differences evolve over the middle of the domain, especially over North Dakota.
19	Shown here is the 700-500 hPa temperature difference, which would be used in the Haines index over high terrain to measure how steep the low-level lapse rate is. Haines index would add 1 for values less than 18, 2 for 18-22, and 3 for above 22, so the key contours to focus on are 18 and 22. Only contours from 16 to 24 are plotted. Note that differences in the two forecasts do occur, both over the plains and the mountains, but overall the same areas are highlighted for high values. The Eta contours are from a 32-km output grid while the NMM contours are from an 8-km output grid.
20	The other part of the Haines for high elevation locations is the 700 hPa dewpoint depression, a measure of low-level dryness. The critical values here are 15 and 20. Contours range from 12 to 24, with smaller and larger values not contoured. A lot more terrain-related detail shows up in the NMM forecast, especially during the daytime but even some at night. Some of the greater detail in the NMM is a result of plotting on its 8-km grid while the Eta is plotted on a 32-km grid. You are accustomed to viewing the "meso-Eta" on a 40-km output grid in AWIPS, but this is what moisture fields at 8-km look like, so don't be surprised or alarmed. Do these differences have operational impact on your fire weather assessment?
21	What do you notice about the vertical levels? How do you expect this to affect the forecast in the boundary layer? Vertical structure in the boundary layer is affected by vertical resolution, and topographic flow is affected by the vertical coordinate system. Both models have 60 layers, with similar layer thickness at the bottom at sea level. However, the Eta levels are quasi-horizontal, and some disappear underground over topography, while all the NMM levels are above ground, retaining small vertical grid spacing over elevated terrain. NMM uses a sigma coordinate between the surface and 400 hPa and uses isobaric surfaces above 400 hPa.
23	Hourly BUFR soundings like you would see in BUFKIT - raw model data - from NMM (cyan) and Eta (blue) at Riverton, WY with raob (yellow) overlayed at raob times. Note that the elevation is lower in the NMM than in the Eta, the initial moisture profile is shifted downward slightly and initial temperatures are colder at the bottom. The temperature immediately plummets near the ground in the NMM, with the verifying 12 UTC raob splitting the difference between the Eta and NMM at the surface. This difference is probably related to the much thinner layers near the ground in the NMM, concentrating the influence of the ground and promoting stronger/faster decoupling of the surface winds and resolving drainage flows better. However, both models have too much net cooling through the boundary layer and too much moisture trapped below the residual layer, probably due to the initial moisture profile. Looking at the 12 UTC soundings each day, what do you expect the afternoon dewpoint to be and when during the day do you expect the dewpoint to plummet and the winds to pick up? See what happens... The first morning into afternoon, the boundary layer development is nearly the same in both models, though the NMM lags the Eta slightly due to being slightly cooler. The NMM is drier in the upper part of the boundary layer during the middle of the day. The second night, the NMM temperature plummets quickly again near the ground and is cools a little more than Eta through the boundary layer. The Eta maintains a mixed moisture profile from the surface upwards for around 150 hPa while NMM does not - it has more moisture trapped at the bottom and less in the residual layer. Morning verification shows again that the surface temperature falls between NMM and Eta, with a sharp nocturnal inversion in the bottom 50 hPa, below which moisture is trapped and above which the moisture is well mixed but much drier than in the Eta. The second day, the NMM is playing catch-up with the Eta in boundary layer development, finally catching up by 2100 UTC (1400 local standard time) with nearly identical temperature profiles thereafter. NMM is moister than Eta throughout the day; verification is moister at the surface but drier through the boundary layer above the surface.
24	The Eta uses silhouette topography, which is accounts for the blocking effect of higher terrain but isn't as high as envelope orography, which goes over top of the peaks. Additionally, the Eta step topography has the step edges at the points where velocity is defined (rather than halfway between velocity grid points), so the velocity is assumed to be zero at these edges. To avoid a valley with no flow, there has to be at least two adjacent grid squares at the valley floor (allowing wind through the middle of the valley), thus the valley cannot be narrower than two grid boxes, which is 24 km. This limits how deep valleys can be cut below surrounding terrain. The NMM uses mean topography, which is simply an average of the subgrid-scale topography data inside each model grid square.
25	Topography comparison. Note that peaks are not higher in NMM but there is a lot more detail. The box on the left panel is the zoomed area shown next and is the area that will be shown in close-up data plots overlayed over topography.
26	Same as previous plot, including same color scale, but zoomed on area in box on previous plot. What do you notice about the NMM topography compared to Eta, and how would you expect this to affect the forecast? At 8 km grid spacing, the NMM still cannot come close to capturing most individual peaks or canyons, though the variability it shows is a marked improvement from that in the Eta model. The second frame shows the topography underlay to be used in upcoming plots. It is the exact same map, with NMM topography on the left and Eta topography on the right. The color scale is changed to gray shades, varying from lightest at lowest elevation to darkest at highest elevation, with the same scale for both models.
28	Be sure to click on the check box to view the 2-meter temperatures in each model overlayed on its topography. The numbers are observations from METARs and MesoWest sites, including RAWS stations. Model contours are color coded on the same scale for easy comparison. At 12 UTC on the first day, early morning, what do you notice about the temperature patterns? During the afternoon near peak burn time, at 21 UTC, what do you notice? Note how much more cooling the NMM has than Eta in just the first 3 hours at all locations, ridge, valley, and plains. The NMM fits the observations better in the higher elevations but cools off too quickly in the valleys and plains. During the afternoon, the two forecasts are about the same for valley regions but cooler over ridges in the NMM - perhaps from more upslope adiabatic cooling, as the ridge elevations are not higher. Also there is much more detail in the NMM, which corresponds to the topographic detail. The second night the differences are again large. With the help of the meso-west observations, we see the NMM is cold temperatures on the slopes verified well but NMM is too cold in the valleys. Overall, NMM looks closer to most of the observations than the Eta.
29	Thermal belt where burning may continue at night shows on some slopes in NMM (first frame), not at all in Eta (second frame). [The thermal belt is a zone along the slope where nighttime temperatures are a relative maximum compared to cold air draining below to lower elevations and cold air due to high elevation above.]
30	Be sure to click on the check box to view the 2-meter RH in each model overlayed on its topography. Observations and contours are color coded on the same scale. 70% is the highest value contoured because higher values are not of interest for fire weather. At just 3 hours into the forecast, what do you notice in comparing Eta to NMM? At 18 UTC on the first day? At 21 UTC on the first day? The initial RH in both models matches the observations poorly, but is improved by 3 hours. The NMM rapidly develops higher RH during the night than the Eta due to cooler temperatures, and these higher RH values verified better at most but not all stations. During the afternoon, the NMM is much moister over ridges and somewhat moister overall though is drier in a few valley spots. Most of the observations are drier than both models during the day. There also is rain in the northern part of this region in the NMM during the afternoon that may be adding some moisture to the soil, increasing surface evaporation. But since the rain is mostly from the convective parameterization, it isn't evaporating on the way down, thus not fully explaining the moist region in the northeast part of the plot. The NMM has northeasterly upslope in that region, drawing in some of the plains moisture from the east. The Eta has flow around the terrain rather than upslope.
31	Flow over terrain is different in models using the Eta step coordinate compared to models using terrain-following coordinates such as the NMM. Eta step-coordinates inhibit flow across valleys; the air tends to flow over top of the valley, even when lapse rates are steep. With terrain-following coordinates, air tends to flow down and then up the valley walls, too much under stable conditions when air should remain trapped in the valley. Downslope flow in Eta coordinates tends to flow one level above the model terrain, partly due to the no-slip (zero velocity) condition at the edge of each step and the horizontal nature of flow in Eta coordinates. Downslope flow in terrain-following coordinates tends to follow the coordinate surface, so it reaches the bottom and scours it out, perhaps too fast. In Eta coordinates, because the flow tends to be relatively flat, it flow around obstacles such as mountains instead of over top of them. In terrain-following coordinates, there will still be some deflection of the flow but it won't be as flat, more will rise up over the obstacle.
32	An example of a topographic flow with differences between the models is the Denver mesoscylone that occasionally forms with southerly winds across the Palmer divide. The first frame shows 10-meter winds over Colorado in the NMM at night in a situation with a strong southerly large-scale flow. Note the gyres south of the Palmer divide and south of the Cheyenne ridge (east-west ridges extending east from the front range over central and CO and extreme southern WY, respectively). The second frame shows the same for the Eta, which did not reproduce the gyres. Observations had weaker gyres than in the NMM but did have northerly winds at the north end of each gyre and westerly winds back along the front range.
33	Another topographic flow example is the Santa Ana wind in southern California. The figure shows 10-meter winds. Arrows are proportional to wind speed, with thick arrows for observations. The Eta blows the downslope winds above the lowest model layer offshore while the NMM has the winds reaching the coast. The NMM also has a more developed on-shore flow at the north end of the picture.
35	Comparison of 10-meter wind speed in each model with observations. The interpolation of 10-meter wind speeds in the postprocessor involves the surface roughness. There was a bug in the surface roughness fixed a few weeks after this case study date which caused the 10-meter winds to be far too light in three box-shaped regions in the far southern part of the plot (shows clearly at 21 UTC, 15 hour forecast) but had no forecast impact over most of the domain. Additional changes are coming to both the NMM and Eta surface roughness (see model physics changes, page 5). The winds in the NMM are generally too light at night, and over the mountains they clearly mix down too slowly during the daytime boundary layer development.
36	Comparison of wind speed averaged over the bottom 30 hPa in each model. The surface roughness is not used in calculating these winds during postprocessing, though it has an indirect affect since it is in the surface layer parameterization, thus affecting the model low-level winds. Note how much more sharply the topographic affects show up in the NMM compared to in the 10-meter wind speed.
37	Be sure to click on the check box to display the 10-meter wind barbs. The model winds are color-coded by speed and the observations are in purple. Barbs correspond to locations at the head. Barbs are a little longer for METARs than for meso-west observations. Remember, the model 10-meter wind speeds will generally be increasing after the upcoming changes later in 2004, in both models. However, the pattern of wind direction associated with topography, channeling, diurnal slope flows and nocturnal drainage should not be affected much by these model changes, so this case offers a good example. What differences do you notice at 12 UTC the first morning? At 21 UTC the first afternoon? The second night at 3 and 6 UTC? Next afternoon at 21 UTC?
38	Be sure to click on the check box to display the observed wind gusts. The model 10-meter wind speed contours and observed wind gusts are color-coded on the same scale. The RAWS/MesoWest data showed great gustiness during the afternoons, far exceeding the 10-meter wind predictions by either model. Even with the expected 10-meter speed increases in the models, they will still fall well short of these gusts.
39	Be sure to click on the check box to display the observed gusts. This is the same as the previous page except the model speeds are from the average over the bottom 30 hPa above ground. These winds are stronger than the 10-meter winds but still decrease somewhat at night with nocturnal decoupling. The NMM appears to be picking up some sort of mountain wave activity at 21 and 24 hours, showing high values over the high terrain and slopes which is matching observed gusts, while the Eta does not have this. At 27 hours, the NMM has subsided this activity except over the eastern most parts of the plot, while observations show gustiness has subsided everywhere. However, even these winds did not suggest the high gusts during the second afternoon, and the NMM again has strong winds on the eastern slope at the eastern side of the plot by late into the final night, when no strong gusts were observed.
41	The top row has CAPE based on the most unstable model layer within the lowest 70 hPa, which is the "surface" CAPE field created at NCEP. This is the CAPE distributed over FXNET. The lower panel refers to the CAPE taken by defining parcels averaged over the bottom 30 hPa, the next 30 hPa, etc., 6 of these averages, so the highest reaches to 180 hPa above the model terrain. The parcel among these six with the highest theta-e is used for the CAPE. How do the CAPEs in NMM compare to Eta over the plains? Over the mountains? Overall, the two types of CAPE ("surface" and most unstable of 6 layers) are very similar, but the surface CAPE is showing higher values earlier in the day over the mountains, suggesting that these values are for very thin layers near the ground, not representative of mixed-layer parcels.
42	Lightning observations are shown to verify where the lightning threat occurred, with model surface CAPE shown for reference. Before 19 UTC on the first day (corresponding to 13-h forecast), there was no lightning. Plots are shown every 2 hours thereafter to 06 UTC July 2, after which the complexes continued producing profilic lightning as they moved east across the plains. The next day, lightning started after 21 UTC (corresponding to 39-h forecast) and continued spreading east over the plains at times after the last shown. Is the CAPE in NMM useful for indicating when and where mountain lightning occurs?
44	The QPF plots show 3-hour accumulated total precipitation in the upper panels and the portion produced by the model convective parameterization in the lower panels. The color coding highlights danger for light amounts to focus attention on potential for dry thunderstorm activity. Note that almost all of the precipitation is produced by the convective parameterization in both models. An exception is the NMM at 21 hours has some small streaks of grid-scale precipitation in southern Montana extending downstream from the maxima in convective precipitation. Under moister conditions, and probably also after the NMM model change reducing convective triggering, these trails of heavier precipitation are more common extending downwind of convection. How does the precipitation in the two models compare, in timing, amount, and coverage, over the mountains? plains?
45, 46	1-hour lightning and RFC 6-hour precipitation estimates are shown for comparison against model QPF. Note that the color scale is the same as for the 3-hour QPF on the previous page except that the lowest two categories shown in red and orange on the previous page are combined into one category shown in dark orange in the RFC plots. How do areas of model precipitation compare with locations of lightning? How do areas of light precipitation verify? Heavy precipitation?