Lessons Learned

Forecasting Weeks for Orlando, FL and Fresno, CA



For our second ePortfolio assignment, we were set the task of choosing a Lesson Learned from each forecast week, two per city, and elaborate on what we learned and how it affects our forecasting technique.  Orlando, Florida, KMCO, was our first forecasting city for the first two weeks of the WxChallenge Forecasting Contest.  This time of year, Orlando is still in the warm season, with tropical temperatures and solar heating dominating their weather.  We also used Orlando, Florida for our practice week the week prior to the beginning of the contest.  My first Lesson Learned actually came from that practice week, and it had a big effect on my forecasting technique - the effects and behavior of mid-latitude cyclones/low pressure systems/frontal boundaries between the Midwest and tropical Florida.  It changed my whole thinking on frontal boundaries when they reach Florida after traversing the Midwest and Southeast at the end of summer/beginning of fall.

The second Lesson Learned, from the second week of the contest, hinged around my use of the Delta Method, which was required to try at least once for this assignment.  My surprise came in the form of the accuracy of my forecast when I used this method - that was my best forecast for the week! 

Our second city was Fresno, California, tucked in a valley (San Joaquin Valley) just west of the Sierra Nevada Mountains.  The climate for Fresno at this time of year is basically sunny and dry with light winds.  However, there was a low pressure system traveling around the vicinity - it got knocked west by a large low pressure system over central Canada, where it moved to the north of Fresno then moved offshore, where it spun a day or two before moving east back onshore just south of Fresno.  Our first lesson learned was regarding the presence of showers nearby from this low, whereby evaporatively cooled air busted the high temperature prediction on MOS and by the class (it was first thought that a marine layer invasion had caused this temperature bust - details below). 

The second lesson learned for Fresno had to do with winds and MOS' handle on factoring in downward momentum mixing from 850 mb to the surface.  This mixing typically manifests as wind gusts, whereas we are forecasting for sustained winds here - a small but important point that caused some confusion and led to this lesson learned.  With the presence of the low pressure system nearby, a large pressure gradient showing up on the eWall progs, and the local NWS Area Forecast Discussion indicating that winds would be much higher than normal, we followed suit with predictions for much higher sustained winds and busted by several knots.

Below I will elaborate on each of these Lessons Learned and try to graphically show each occurrence.

Orlando, Florida KMCO

Forecast week 1 for Orlando, FL KMCO

Lesson Learned:  Though the sea breeze frontal issues with precipitation, clouds and wind was the biggest issue of the first week of forecasting for Orlando, my personal Lesson Learned was about tropical climatology of the southern states vs. mid-latitude cyclones and frontal boundaries that pack a large temperature change in the Midwest toward the end of the warm season.  I realize this occurred during the practice week, but it had a very important consequence to my forecasting mindset.

On September 17th, a frontal boundary packing a potent large temperature gradient passed through Kansas City, Missouri, dropping temperatures by 26 degrees between September 16th and September 17th for afternoon highs (though for some reason the local weather station determined it was 24 degrees).  You can see the frontal passage and its effect on these meteograms. 


Observations for KANSAS, MO (MCI)

1653Z 16 Sep 2006 to 1753Z 17 Sep 2006

Observations for KANSAS, MO (MCI)

1653Z 17 Sep 2006 to 1753Z 18 Sep 2006

This meteogram from the University of Wyoming site shows the high temperature at Kansas City Intl Airport of 92F prior to the frontal passage at 09Z September 17th.  By 15Z on the 17th, the temperature had cooled to 62F.  This was the first shot of cold air that came along with the front.  [Click on image to open in new window.] This meteogram from the University of Wyoming site shows the high reached on September 17th of 66F, then temperatures dropping all the way to 51F at 11Z.  [Click on image to open in new window.]

These surface analyses for 21Z September 16th and 21Z September 17th show the progression of the front and the changes in temperature and dew points behind it.  I admit, I was doing some upstream forecasting with this front, thinking since it was so potent here in Kansas City, that it would affect temperatures nearly as much when it reached Florida. 

Three days later, on September 20th, this same frontal boundary was slated to reach the Orlando area.  Below are the surface analyses for 21Z September 19h and 21Z September 20th to show the progression of the front.  For all intents and purposes, the conditions pre-front and post-front look the same on these maps, showing that the power it had back in the Midwest weakened considerably by the time it reached Florida. 

This surface analysis from the HPC site shows the frontal boundary draped across the Florida Panhandle and Georgia, with not much temperature difference between areas ahead of the front and areas behind the front.  This should have been my first clue. This surface analysis from the HPC site shows the frontal boundary already having passed through the Orlando area and sitting on the tip of the peninsula.  Note the temperature of 83F in Orlando and 86F at the tip of Florida.  Only three degrees Fahrenheit difference!  Tropical climate, solar heating, and warm SSTs took their toll on this front that was so powerful only a few days before in the middle of the country.

This meteogram shows the frontal passage at Orlando - with solar heating in control, Orlando barely gave the front a nod of recognition.


Observations for ORLANDO, FL (MCO)

1953Z 19 Sep 2006 to 2053Z 20 Sep 2006

This meteogram from the University of Wyoming site shows the frontal passage at around 16Z on September 20th.  Temperatures actually kept warming up with the solar heating in spite of the frontal passage.


My forecasts for 06Z 091906 - 06Z 092006 and 06Z 092006 - 06Z 092106 were for high 88/low 74 and high 86/low 70, respectively.  I was fairly close on the lows.  However, I busted on the highs due to my thinking that the frontal boundary would significantly affect the air mass temperature over Florida as it had in the Midwest.  Actuals for both days were high 92/low 74 and high 90/low 71, respectively.  The high on the 20th of 2 degrees cooler has more to do with sea breeze frontal cloud cover than it does with the frontal air mass.  Just as a comparative note, the high on the 18th was 92/71 and on the 21st was 91/68 (cooler night with clear skies).

I have always considered frontal boundaries as a distinct dividing line between "warring" large temperature differences.  I had also always thought that all frontal passages involved storm activity, or at the very least, rain.  One of the most interesting, and sometimes confusing, thing I've learned during this program is that there are varied and sundry weather conditions, as well as varied and sundry cause-and-effect relationships, involved with frontal passages.  And that's just fronts - there are so many variables to weather and climate.  Though I knew it was warm in Florida year-round with no large temperature fluctuations usually, still I had it in my mind that fronts cause storms and cooler temperatures.  Even though I've learned that this isn't always so, it must have still been in the back of my mind when making this forecast. 

Forecast week 2 for Orlando, FL KMCO

Lesson Learned:  The Delta Method really works!  For the forecast period 06Z 10/4 - 06Z 10/5, I chose to use the Delta Method on high temp and max sustained winds both because the Orlando area was embedded in an air mass that was changing little day by day, and forecasted to be the same for the period.  This allowed for some persistence forecasting.  It turned out to be the WxChallenge day when I had the least amount of error!

I started out as it outlines in the lesson text, with monitoring the hourly METARS to determine the high for the current day.  On the METAR reading below on October 3rd, the high so far had come out to be 32.2C, or 89.96F - round it to 90F.  This turned out to be the actual high for the day, by the way.

KMCO 031753Z 07014KT 10SM FEW055 SCT070 SCT095 BKN250 30/18 A3013 RMK AO2 SLP202 T03000178 10322 20228 58014=

I used only the NGM Grid Interpolations at 12Z for October 3rd, though I realize in hindsight that I should have done a comparison between the three (ETA and AVN).  First we look at 850mb - since the high temp reading on the METAR was taken at 1753Z (18Z basically), I look at the 18Z 10/3 and 18Z 10/4 temps for comparison.  They are 14 and 13 Celsius, respectively.  That's a change of 1 degree Celsius (or 1.8 F to subtract). 

DAY / HOUR        03/12  03/18  04/00  04/06  04/12  04/18  05/00  05/06  05/12
------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------
1000 MB (C)            25    27    26    23    23    25    25    24    23
  850 MB (C)            16    14    13    14    13    13    13    14    14
From the NGM grid interpolation found at the Texas A&M University Atmospheric Sciences site.

And now for the math.  We've already determined that we're rounding 1.8 to 2 degrees F.  So you subtract 2 degrees from 90 degrees to arrive at 88 degrees F for the progged high for tomorrow.  I'm happy to say that the actual high for October 4th was ... 88F! 

Then I went for the winds.  Knowing that this may be off due to the day not being done with, I'm taking the highest reading of winds of 14 knots at 1753Z as my starting value here (see METAR line above). On the NGM grid, we look at 850 mb winds:

1000 MB            07/012 06/009 08/014 07/017 06/011 05/011 06/013 06/013 07/011
  850 MB            08/015 08/011 08/008 07/015 08/012 07/017 07/012 05/015 05/017
From the NGM grid interpolation found at the Texas A&M University Atmospheric Sciences site.

The 18Z 10/3 wind is 11 knots and the 18Z 10/4 wind is 17 knots - a difference of 6 knots.  This is an addition instead of a subtraction, so you add the 6 knots to the 14 knots to arrive at 20 knots for the progged wind speed for tomorrow.  I was told on the discussion boards that my wind calculation was incorrect, but we never got around to delving into the logistics of that.  I forecast for 20 knots as a maximum sustained wind by taking the 14 knots at 18Z and adding the 6 knots change at 850 mb.  The actual wind was 19 knots. 

It was also pointed out that these calculations should be checked with surface/2m/1000 mb values, as well.  The value in this is if the 850 mb layer is not well mixed, those winds will not be filtering down to the surface and thus increasing wind speeds.  However, it was cautioned in the lesson text that stability must be similar between the comparison days in order to use this method.  I believe this was the case for this forecast period.  Here are the surface analyses for October 3rd and October 4th, both for 15Z.

This surface analysis for 15Z on October 3rd, from the home page of our class, shows the stable air mass under a high pressure area, with light northeast winds and temperatures in the mid-80s. Here is the surface analysis for 15Z on October 4th, from the HPC site.  You can see that the air mass has not significantly changed from October 3rd.


I think the biggest problem I have with studying the weather is relating the dynamics I can see and feel outside to the equations and formulas that come to play in the data.  There's still a disconnect for me between math and the great outdoors.  For the Delta Method to be so close to actual temperature forecast - which to me is a bunch of numbers trying to relate directly to the air outside - truly amazed and delighted me.  Not having any formal background, nor natural inclination toward, physics and math, it's been a struggle to put some of these concepts together in my head, but this experience helped to bridge that gap.  It also made the numbers more interesting, or rather have more meaning, so that I'm more comfortable using them.

Fresno, California KFAT

Forecast Week 1 for Fresno, CA KFAT

Lesson Learned:  Marine layer - to be or not to be.

Fresno, California was a more difficult city to forecast.  Fresno sits in a valley with both mountains and desert nearby, but with mountains between the city and the coast.  Climatology called for mostly sunny and warm conditions, but there was a low pressure system that affected the area during our forecast parameters.  During the forecast period for 06Z 10/13 - 06Z 10/14 (Friday the 13th), MOS indicated predicted high temps close to climatology, which was a high of 80F for October 13th.  The low pressure system was sitting offshore, and slated by the eWall progs to move inland just south of the Fresno area toward the end of the forecast period.  The two HPC Front maps below show the predicted location of the low at 12Z 10/13 and 00Z 10/14.

This forecast map for fronts and precipitation valid 12Z October 13th (from the HPC site) shows the low pressure center offshore southern California, with rain chances for the immediate coastal areas north of Baja.  This forecast map for fronts and precipitation valid for 00Z October 14th (from the HPC site) shows the area for rain chances widened considerably to include all of southern California and parts of Nevada and Arizona.  However, MOS indicated clear to partly cloudy conditions for Fresno through the forecast period.

In spite of the low coming onshore, precipitation chances were low on MOS, NWS Area Forecast, and the NWS Point Forecast.  There was a lot of dry air with this low, as can be seen on the lower left panel on the eWall progs linked above, and the visible and infrared satellite images below from 2030Z on October 13th confirm that the low brought a few clouds to the Fresno area, but the class consensus was that it was not enough to warrant lowering daytime highs much beyond MOS and climatology.

This visible satellite image valid 2030Z October 13th is from the PSU eWall and shows patchy partly cloudy conditions over southern California. This water vapor image from the PSU eWall shows the dry air coming into southern California at 2030Z October 13th.  You can make out patchy dry spots over the Fresno area.

The actual high temperature for Friday, October 13th came in as 71F, which is 9 degrees below the climate norm of 80F.  No one expected this huge bust, not even the instructors!  So we got busy trying to find the cause, and determined that it was invasion of a marine layer of cooler surface air that caused the high temperature prediction to bust, for us and for MOS.  The attached streamlines, sounding, and profiler were all used to support this conclusion.  On the streamlines, note the blue arrows that point up to Fresno from the south after coming in offshore from the north (indicating they would travel through a mountain pass).  On the sounding, you can see the small surface inversion which is typical of a marine layer.  And on the profiler, which is a look back at actual conditions, you can see the pink area at 3000 ft. that deepens from right to left (indicative of increasing height of the inversion with time).  The "invasion of the marine layer" was to be our lesson learned for this week.

However, upon further research by the instructor, it was found that a marine layer was not present and thus not the cause of the high temperature bust.  According to the station models (click on files to show satellite images), surface winds were from the south all day, whereas a marine layer comes in from the north.  Also, even though there was no precipitation measured at Fresno until after the forecast period, there were showers (click on files to show radar images) nearby, and the evaporatively cooled air from these showers was carried by the southerly winds to bring temperatures down and form more clouds than MOS predicted, which also worked to cool temperatures.  The real lesson learned here might very well be to always delve deeper for true cause and effect when doing post mortem meteorology.


Quite frankly, my mind is still reeling on this one.  First I was trying to understand the marine layer, which is a new concept we learned from the forecast practice week on Fresno (when I was still concentrating on Orlando).  Then to find out the marine layer was not the cause of the bust, but it was a simple, mesoscale feature we learned in 361 - advection of evaporatively cooled air by southerly winds, evaporatively cooled air produced locally, and more cloud cover than anticipated by MOS - made me realize that I haven't been applying everything I could to these situations.  A lot of this forecasting for daily high/low temperature, precipitation amount and wind speed is part Meteo 101 and part Meteo 361, and I think I've only been focusing on Meteo 101 where we first learned to use MOS.  I've tried to anticipate cloud cover and rain chances, but realize that I haven't been looking at them from a mesoscale point of view, nor even in a synoptic context.  I have been looking at the low pressure systems and frontal boundaries, but for some reason my thinking is fragmented on these features when it comes to the basics of temperature and winds. 

I also realize I haven't been fully utilizing radar and satellite imagery to enhance my perspective of synoptic and mesoscale situations.  I think a synoptic briefing at the beginning of each day would have been helpful, and realize that we are each responsible for approaching our forecast in this manner in order to see "the big picture" first.  This in itself has been a big lesson learned for me - my routine is opposite of what it needs to be.  I need to focus on synoptic, then mesoscale, include climatology, then focus on the details. 

Forecast Week 2 for Fresno, CA KFAT

Lesson Learned:  Sometimes MOS is the best guidance.  And when using the Delta Method, don't use 850 mb for night or winter when mixing is not probable - use 2m, or 1000 mb, for surface conditions, instead.

This week presented a challenge with winds.  The local NWS Area Forecast Discussion from 2:00 pm PDT Sunday October 15th indicated the following:


The prog below, valid at 12Z on October 17th, shows the large pressure gradient present over southern California, which would indicate higher than average winds at the surface.  The local NWS Area Forecast called for NW winds of 10-15 mph, which is only 9-13 knots.  However, only the ETA MOS predicted a wind higher than 9 knots - 11 knots at 06Z 10/17/06 and 14 knots at 09Z 10/17/06.  But since we'd all been experiencing trouble with MOS being too low on the winds (mostly in Orlando - lesson learned there besides), we were all hesitant to take MOS at face value.  The local NWS had a handle on the wind situation, as the actual came in at 12 knots.  But it was a big bust for some of us on our forecasts. 

This prog from the PSU eWall is from the WRF model valid 48 hours out at 12Z 10/17/06, and shows the pressure gradient over southern California on the two left panels.  A large pressure gradient causes higher than average winds at the surface.

The lesson learned from this scenario was twofold:  trust MOS more and don't fight it so hard as if it were always going to bust, and know that MOS figures in downward mixing of momentum in a statistical sense.  It should be noted that since the dynamic models don't handle this downward transfer during the daytime very well, then MOS is bound to be off a bit.  However, an adjustment to a wind forecast based on these conditions should only go a degree or two higher than MOS, rather than a huge difference, as we did.  Besides, downward mixing manifests as gusts rather than sustained wind speeds. 

Many of us were attempting the Delta Method on these winds, and we needed to learn the important difference between using the 850 mb level and the 2m, or 1000 mb, level, to calculate surface winds.  In addition to the two-fold lesson learned above, we learned that if downward momentum mixing is expected then the 850 mb level is okay to use since it will reflect surface conditions for a well-mixed boundary layer.  However, if this mixing is not expected to occur, then the 2m surface level needs to be used to reflect true surface winds.  Also, when using either of these methods, it's a a good idea to double check by calculating the level not used and compare. 


I seem to have the most problem with winds, which has always surprised me from the beginning of Meteo 101.  And Orlando really threw me for a loop for awhile.  But with Fresno, I seem to have been completely lost.  This time, I was trying to factor in the low pressure system that was to move back onshore south of Fresno, and I felt certain the winds would pick up, especially when others pointed out the pressure gradient and even the local NWS indicated they'd be way above normal.  However, I failed to consider mountains vs. desert vs. valley - the terrain was the hardest thing for me to picture when trying to determine forecasting for Fresno.  I failed to take into account that MOS was factoring in terrain (years of statistical data for the same place) as well as the mixing from daytime heating.  But with the mixing I'd also forgotten the lesson from Meteo 101 - gusty winds are not what we're forecasting for here, we're looking for sustained.  And I think I unconsciously am confusing the two.


  • As a forecaster, my performance was better than I'd expected on the first city and worse than I'd expected for the second city.  I need to pay more attention to climatology as well as terrain, and not bring my pre-formed assumptions into the picture for each new forecast day (such as that all fronts bring changes in temperature).  In addition, a review of sections in Meteo 101 and Meteo 361 regarding winds is in order.
  • My attitude has been one of toil and trepidation!  I've felt the need to look at almost every site and tool to form a comparison on all four forecast points before I could even post on the boards about one of them, and I admit that this strategy still seems valid in order to form a cohesive view of the situation.  However, it was thought that I might be overanalyzing, so I have to figure out what to keep in my forecasting routine template and what needs to be looked at only in certain situations.  It has occurred to me since that there are certain tools that need to be used as standard for every forecast, and others that only need to be accessed given certain synoptic and mesoscale situations.  Figuring this out will be crucial to my forecasting success.
  • This leads directly into the process I've been using.  I've devised a template in a notebook, which needs to be reordered, but so far has gone like this:
  •                       Climate Norms
  •                       AFD Discussion
  •                       NWS Area Forecast
  •                       NWS Point/Zone Forecasts
  •                       HPC Fronts, QPF, and Current Surface Analysis
  •                      MOS
  •                      eWall MOS (comparison plots, regional max/min temps, and station model plots)
  •                      eWall Progs
  •                      Cloud Cover
  • First I would get the climate norms, then the NWS local forecasts and discussions, MOS, then eWall MOS, then HPC Fronts and eWall Progs.  It was brought to my attention that I should be looking at the synoptic scale first (HPC Fronts and eWall Progs) then zero in to focus on the details (MOS and the local NWS).  So my strategy for daily forecasting will change to incorporate this advice, and hopefully it will make it easier to see the details within the larger scope of the situation.  In addition, perhaps it will make it easier to see something in the overall pattern that will affect the forecast sooner than I have been, which will help in making this routine not such a chore.  Also, I need to incorporate radar and satellite as well as more surface and upper-air analyses into my template.  And one last thing are trends in model guidance - I have not paid attention to trends and these will be extremely important in the upcoming cities.
  • This brings us to the strategies I plan to incorporate into my approach to garner improvement and raise my standing in the contest.  I pretty much covered these in the last bullet, but I will summarize here:
  •                      Change the order of my template to reflect synoptic scale first, then focus on local forecast and details
  •                      Look for trends in model guidance
  •                      Incorporate radar, satellite imagery, surface and upper-air analyses into the template
  •                      Don't try to beat MOS, but let it be a guide in comparison with climatology and progs
  • My new template is forming as we speak, but since beginning Caribou, Maine, I've followed a different format.
  •             Climate Report from previous day for verifications, and Norms/Extremes Climate Data
  •                     HPC Surface Analysis, HPS Fronts, HPC QPF
  •                     Satellite and Radar
  •                     eWall Progs
  •                     Zone and Point Forecasts
  •                     eWall MOS, One-Stop MOS, Tabular
  •                     AFD Discussion and 7-Day
  • As for how to evaluate these new strategies, I think my ease in forming my comparisons, the quality of my posts, and my score ratings in the contest are the best way to gauge whether an improvement occurs or not.  I'm not sure how to evaluate each point other than results.  I was within the first 100 on the first forecast city using my original template, but Orlando was a fairly consistent city for weather.  Fresno has dropped me down into the 600s and I've risen back up to the 500s.  I hope to see an improvement in this standing by looking at the big picture in more depth to form an idea of what is likely to occur on the small scale, keeping climatology in mind this time around.

Debbie Jarvis-Ferguson

Meteo 410

ePortfolio Assignment #2

October 24, 2006