MM5 model on Wind Archive plots
-
bwd
- Developer
- Posts: 1247
- Joined: Tue May 13, 2003 8:57 am
- Location: In a van down by the jetty
- Has thanked: 37 times
- Been thanked: 46 times
- Contact:
MM5 model on Wind Archive plots
Something to check out while we wait for the big SE. I have added an option to show a history of the MM5 model on the wind archive plots:
http://www.bigwavedave.ca/archive.php
It gives a general idea of the accuracy of the model.
Here is some info:
- I have been saving the MM5 results since October 2005;
- comparison plots are available for the following sites: Ballenas Is, Columbia Beach, Comox, Discovery Is, Entrance Is, Gordons Beach, Kelp Reef, Lennard Is, Ogden Point, Sheringham Point and Tsawwassen, or any site that is close to a site in the Model Tables:
http://www.bigwavedave.ca/forecasts.php
- some sites are not exactly the same ie. Cook model winds are plotted with Ogden point winds, Island View winds are plotted with Kelp Reef, Lennard/Long Beach...but they give a general idea of the accuracy;
- the pull down list has an (MM5) beside the sites that have model comparison data. You have to use the check box to the right to show the model plots;
- the average measured and model estimated wind for the month is also shown (vector average of wind components);
- note that the model is in 5 knot increments (5,10,15,20 etc), and is only saved from 6am to 9pm. The average wind is computed for the same period for measured/model.
Here is a brief description on how I produce the model tables. I have a program that runs on my home PC that loads the surface windspeed maps from the MM5 model: http://www.atmos.washington.edu/mm5rt/r ... .12km.html
The program gets the windspeed (from the colour) and the wind direction (from the wind barbs) at the various locations shown in the model tables. The sites have been tweaked over time to try and give the best estimate of the winds. At first I picked the exact sites near shore but found I had to move them around on the map to give a better estimate. For example, for Cook and Gordons the site is offshore a bit.
The GFS MM5 model from the Univ of Wash is run twice a day and is updated gradually between 9am and 12 noon (called the 12 run), and between 9pm and midnight (called the 00 run). The results from the model aren't available all at once since it takes a while to produce each hours predictions. At 6am you should see the predictions from the 00 run, and by noon you should see predictions from the 12 run. Sometimes the model run is delayed.
For the model comparision plots I use the 00 run (the run that is done the night before), since this is what we see first thing in the morning. It is expected that the comparison would be slightly better if I used the 12 run. The comparison plots show the 4km grid 00hr model from today (not day 2 or day 3)...so, for example, it shows the model table results that you would would see at 6am for today.
There are 3 resolutions of model runs: 36km, 12km, and 4km. I use the 4km for todays/tomorrows forecast and the 12km for the 3 day outlook.
As mentioned many times, the model tables are not meant as a replacement for EC, NOAA or common sense. They are just an extra tool that we can use. I find them helpful but I don't rely on them. If EC and the model agree then that gives a little more weight to the forecast. If they disagree then go with EC and the current conditions. EC tends to be brief and the model sometimes fills in some of the timing gaps. But as we have seen it can be way off on W/NW for Cook and SE for the south island.
Question, comments?
dave
http://www.bigwavedave.ca/archive.php
It gives a general idea of the accuracy of the model.
Here is some info:
- I have been saving the MM5 results since October 2005;
- comparison plots are available for the following sites: Ballenas Is, Columbia Beach, Comox, Discovery Is, Entrance Is, Gordons Beach, Kelp Reef, Lennard Is, Ogden Point, Sheringham Point and Tsawwassen, or any site that is close to a site in the Model Tables:
http://www.bigwavedave.ca/forecasts.php
- some sites are not exactly the same ie. Cook model winds are plotted with Ogden point winds, Island View winds are plotted with Kelp Reef, Lennard/Long Beach...but they give a general idea of the accuracy;
- the pull down list has an (MM5) beside the sites that have model comparison data. You have to use the check box to the right to show the model plots;
- the average measured and model estimated wind for the month is also shown (vector average of wind components);
- note that the model is in 5 knot increments (5,10,15,20 etc), and is only saved from 6am to 9pm. The average wind is computed for the same period for measured/model.
Here is a brief description on how I produce the model tables. I have a program that runs on my home PC that loads the surface windspeed maps from the MM5 model: http://www.atmos.washington.edu/mm5rt/r ... .12km.html
The program gets the windspeed (from the colour) and the wind direction (from the wind barbs) at the various locations shown in the model tables. The sites have been tweaked over time to try and give the best estimate of the winds. At first I picked the exact sites near shore but found I had to move them around on the map to give a better estimate. For example, for Cook and Gordons the site is offshore a bit.
The GFS MM5 model from the Univ of Wash is run twice a day and is updated gradually between 9am and 12 noon (called the 12 run), and between 9pm and midnight (called the 00 run). The results from the model aren't available all at once since it takes a while to produce each hours predictions. At 6am you should see the predictions from the 00 run, and by noon you should see predictions from the 12 run. Sometimes the model run is delayed.
For the model comparision plots I use the 00 run (the run that is done the night before), since this is what we see first thing in the morning. It is expected that the comparison would be slightly better if I used the 12 run. The comparison plots show the 4km grid 00hr model from today (not day 2 or day 3)...so, for example, it shows the model table results that you would would see at 6am for today.
There are 3 resolutions of model runs: 36km, 12km, and 4km. I use the 4km for todays/tomorrows forecast and the 12km for the 3 day outlook.
As mentioned many times, the model tables are not meant as a replacement for EC, NOAA or common sense. They are just an extra tool that we can use. I find them helpful but I don't rely on them. If EC and the model agree then that gives a little more weight to the forecast. If they disagree then go with EC and the current conditions. EC tends to be brief and the model sometimes fills in some of the timing gaps. But as we have seen it can be way off on W/NW for Cook and SE for the south island.
Question, comments?
dave
Last edited by bwd on Sun Nov 05, 2006 12:25 pm, edited 1 time in total.
-
more force 4
- Sponsor
- Posts: 1459
- Joined: Wed Oct 15, 2003 8:57 am
- Location: Victoria, BC
- Has thanked: 23 times
- Been thanked: 8 times
- Contact:
Cool! Thats pretty near, great work Dave! It certainly looks surprisingly close most of the time. When it overrepresents wind on windy days, it seems that there are usually real gusts even higher. Is there a way to zoom in on the graphs a little? Looks like some fun to see if it performs better at some seasions than others, and by sailing spot.
How did you get the average wind direction? Did you use circular normal distribution functions? (otherwise, the average of 350 and 10 degrees is ?) I always thought they were kinda cool.
Didn't you mention that IOS or UBC or someone was doing a formal analysis of the model performance along these lines?
How did you get the average wind direction? Did you use circular normal distribution functions? (otherwise, the average of 350 and 10 degrees is ?) I always thought they were kinda cool.
Didn't you mention that IOS or UBC or someone was doing a formal analysis of the model performance along these lines?
-
bwd
- Developer
- Posts: 1247
- Joined: Tue May 13, 2003 8:57 am
- Location: In a van down by the jetty
- Has thanked: 37 times
- Been thanked: 46 times
- Contact:
Thank you for the comments everyone 
Ve=Speed*sin(Direction) and Vn=Speed*cos(Direction)
Then you average the Ve and Vn components for the month:
avgVe=sum(Ve)/n, avgVn=sum(Vn)/n, n=number of records
and from that you compute a vector averaged speed and direction:
Speed = avgVe^2+avgVn^2 and Dir = atan2(avgVe,avgVn)
...sorry about the math.
"Comparision of MM5 and Meteorological Buoy Winds from British Columbia to Northern California", by Scott Tinis, Richard Thomson, Clifford Mass and Barbara Hickey published in Atmosphere-Ocean 44(1) 2006.
I have a copy if anyone wants to read it. No buoys in JdeFuca were looked at, and only 2 in Georgia Strait. Here are some conclusions that I got out of it:
- wind speeds derived from the 12km MM5 model ranged from 81-101% of observed wind speed.
- Diurnal Sea Breeze: coherence between the model and observed winds is moderate to high
- Storms: there was a 35 deg bias in the mean model winds during the fall period, despite that the coherence is high.
- from two buoys in the Strait of Georgia (Sentry Shoal and Halibut Bank), the northern buoy (Sentry shoal) mean wind and direction were similar between the model and observed in both summer and fall. For the southern buoy (Halibut bank), the mean modelled wind compared poorly with oberservation.
- differences between 12km and 4km resolution model was negligible (except for a california buoy)
- the model should be examined closely before it is used for such things as Search and Rescue.
Hopefully I've summarized it accurately. Best to read it yourself though,
dave
Good question, unfortunately no. In the future I'd like to have the data stored in a database and do interactive plotting (select site, date/time etc) and then you could plot any segment that you want. But that will have to wait. My home PC makes the plots and uploads them so they are a fixed size. One way to zoom is to install Firefox (you should be using it anyway), and install the ImageZoom browser extension. This will let you zoom into any image, webcams included. I find the best way to see the model is to click the checkbox a few times to turn the MM5 model plot on and off - then you can see the measured data and model a bit better (as it turns on/off/on etc)more force 4 wrote:Is there a way to zoom in on the graphs a little?
I was wondering if anyone would ask that. Yes, averaging directions is not a good idea since as you say the average of 350 and 10 is 180. So what is normally done is to take each speed and direction reading and compute the components East and North using:more force 4 wrote:How did you get the average wind direction? Did you use circular normal distribution functions? (otherwise, the average of 350 and 10 degrees is ?)
Ve=Speed*sin(Direction) and Vn=Speed*cos(Direction)
Then you average the Ve and Vn components for the month:
avgVe=sum(Ve)/n, avgVn=sum(Vn)/n, n=number of records
and from that you compute a vector averaged speed and direction:
Speed = avgVe^2+avgVn^2 and Dir = atan2(avgVe,avgVn)
...sorry about the math.
Yes there is a paper fom IOS/UofW:more force 4 wrote:Didn't you mention that IOS or UBC or someone was doing a formal analysis of the model performance along these lines?
"Comparision of MM5 and Meteorological Buoy Winds from British Columbia to Northern California", by Scott Tinis, Richard Thomson, Clifford Mass and Barbara Hickey published in Atmosphere-Ocean 44(1) 2006.
I have a copy if anyone wants to read it. No buoys in JdeFuca were looked at, and only 2 in Georgia Strait. Here are some conclusions that I got out of it:
- wind speeds derived from the 12km MM5 model ranged from 81-101% of observed wind speed.
- Diurnal Sea Breeze: coherence between the model and observed winds is moderate to high
- Storms: there was a 35 deg bias in the mean model winds during the fall period, despite that the coherence is high.
- from two buoys in the Strait of Georgia (Sentry Shoal and Halibut Bank), the northern buoy (Sentry shoal) mean wind and direction were similar between the model and observed in both summer and fall. For the southern buoy (Halibut bank), the mean modelled wind compared poorly with oberservation.
- differences between 12km and 4km resolution model was negligible (except for a california buoy)
- the model should be examined closely before it is used for such things as Search and Rescue.
Hopefully I've summarized it accurately. Best to read it yourself though,
dave
-
more force 4
- Sponsor
- Posts: 1459
- Joined: Wed Oct 15, 2003 8:57 am
- Location: Victoria, BC
- Has thanked: 23 times
- Been thanked: 8 times
- Contact:
Yep, those were the functions I was fuzzily remembering. I said circular normal distribution - thats taking these one step further for statistically testing, helped Anne with a paper once on grave orientations or something. Oh, and for the orientation of culturally modifed tree scars vs slope aspect. I hated math in high school (failed or made C, usually) but this stuff I find fun for some twisted reason.bigwavedave wrote:Thank you for the comments everyone
. So what is normally done is to take each speed and direction reading and compute the components East and North using:
Ve=Speed*sin(Direction) and Vn=Speed*cos(Direction)
Out of interest, how did the IOS et al study measure a match? Did they allow for any time difference (sometimes it might perfectly predict wind speed and direction changes through the day, but be off by 2 hours, for instance)?
It would be really fun (for geeks!) to get the tables and do some number crunching windsurfing/kiting-related. E.g., count as 'success' any time it predicted winds over 15 knots at a sailing beach and the real wind was over 15 knots within 3 hours or so, and 'failure' if either there was no wind over 15 knots, or there was and MM5 hadn't predicted it. It would be a lot more 'real world' than the stats saying it is 95% accurate or whatever - we don't really care when it accurately predicts 5 knots.
