George R. Kasica
Delta Method High Temperature Prediction One of the methods that can be used to forecast the upcoming day's high temperature is a method known as the Delta Method. The name of the method comes from the fact that the main item that you are looking at to make your forecast is the change (known by the Greek letter Delta) in the forecast temperatures between several times and altitudes in the atmosphere and using this difference (or delta) to predict the upcoming day's high temperature. I will illustrate the exact process in detail below using the city of San Antonio, Texas during the period from February 5th to February 6th, 2007. Bur first, there are a couple of key restrictions that you must keep in mind in order to apply the delta method correctly and successfully:
Now that we've qualified the conditions needed to apply the delta method, lets look at what conditions were at San Antonio on February 5th and what they were predicted to be on February 6th and then apply the delta method to obtain the high temperature for that day. Finally we'll compare the prediction to the actual high temperature for the day and see how well the method worked. On February 5th we can see by looking at the METAR reports (Click here for a detailed Listing) the cloud cover for the day of the 5th showed basically a mostly sunny sky with just a few clouds in the sky as indicated by the blue highlighted items CLR for clear and FEW or BKN for Few or Broken cloud cover. Likewise looking at the wind directions highlighted in green we can see that for the beginning of the day the winds started out from a north to northwesterly direction before switching to a south to southwesterly direction after about the 1653Z report. Therefore what we will be looking at in the forecast will be for mostly clear skies and winds from something similar to a south to southwesterly direction. Looking at the three forecast model outputs for San Antonio from February 5th at 12Z below we can see in the highlighted blue and green areas in fact that the cloud cover and wind directions are expected to be similar to today's values of clear and south to southwesterly respectively as shown by the CL or FW indications for the cloud cover and the wind directions from 17 to 23 depending on which of the three models you are looking at. Clearly the days are expected to be similar in terms of their cloud cover, and hence incoming solar radiation and their wind directions, so in this case it appears that the delta method should apply quite well. KSAT GFS MOS GUIDANCE 2/05/2007 1200 UTC DT /FEB 5/FEB 6 /FEB 7 /FEB 8 HR 18 21 00 03 06 09 12 15 18 21 00 03 06 09 12 15 18 21 00 06 12 CLD SC SC FW CL CL CL CL CL CL CL CL CL CL FW BK OV SC SC CL BK OV WDR 18 19 16 17 18 22 21 19 19 19 18 17 17 20 20 18 19 18 16 17 20 KSAT ETA MOS GUIDANCE 2/05/2007 1200 UTC DT /FEB 5/FEB 6 /FEB 7 /FEB 8 HR 18 21 00 03 06 09 12 15 18 21 00 03 06 09 12 15 18 21 00 06 12 CLD BK FW CL CL SC CL OV CL BK FW CL CL CL CL CL CL CL CL CL CL CL WDR 20 20 18 16 18 19 21 20 21 20 18 19 19 19 20 20 20 20 17 17 27 SAT SC NGM MOS GUIDANCE 2/05/07 1200 UTC DAY /FEB 5 /FEB 6 /FEB 7 / HOUR 18 21 00 03 06 09 12 15 18 21 00 03 06 09 12 15 18 21 00 CLDS BK CL CL CL CL CL CL CL CL CL CL CL CL CL CL SC SC SC SC WDIR 20 18 15 16 17 23 00 00 20 19 17 17 17 18 17 19 19 19 17The next item we need to apply the delta method to predict the upcoming day's high temperature is the high temperature reading for today. Looking at the METAR reports from above we can see that the high temperature occurred sometime in the 6 hours preceding the 2353Z report and that it was 21.7C or 71F. The last item that we need to utilize the delta method is a piece of data called grid interpolations, which are a computerized prediction for various locations of items such as temperature. These data items can be obtained from the Texas A&M meteorology department web site. Below are the relevant portions of the grid point interpolations for the ETA, NGM and GFS models: ETA --- Station: KSAT Lat: 29.53 Lon: -98.47 Elev: 243 Closest grid pt: 25.1 km. Initialization Time: 07-02-05 1200 UTC PARAMETER/TIME 000 006 012 018 024 030 036 042 048 ------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ DAY / HOUR 05/12 05/18 06/00 06/06 06/12 06/18 07/00 07/06 07/12 ------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ TEMPS 2 M (F) 43 64 63 50 46 67 67 54 48 850 MB (C) 5 5 6 6 5 6 7 9 13 NGM --- Station: KSAT Lat: 29.53 Lon: -98.47 Elev: 243 Closest grid pt: 25.1 km. Initialization Time: 07-02-05 1200 UTC PARAMETER/TIME 000 006 012 018 024 030 036 042 048 ------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ DAY / HOUR 05/12 05/18 06/00 06/06 06/12 06/18 07/00 07/06 07/12 ------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ TEMPS 1000 MB (C) 12 15 19 18 16 18 21 20 16 850 MB (C) 5 5 6 6 6 7 9 11 11 AVN/GFS ------- Station: KSAT Lat: 29.53 Lon: -98.47 Elev: 243 Closest grid pt: 25.1 km. Initialization Time: 07-02-05 1800 UTC PARAMETER/TIME 000 006 012 018 024 030 036 042 048 054 060 066 072 ------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ DAY / HOUR 05/18 06/00 06/06 06/12 06/18 07/00 07/06 07/12 07/18 08/00 08/06 08/12 08/18 ------------------- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ TEMPS SFC (2 M) (F) 62 57 48 45 64 59 51 50 65 64 59 61 68 850 MB (C) 5 6 6 6 8 9 12 12 12 13 13 14 15In order to apply the delta method what we do is we compute the difference between the predicted temperatures from 0Z on February 6 to 0Z on February 7th which would include the day we are trying to forecast the high temperature for, February 6th. We then add that difference (or delta) to today's high temperature to obtain the prediction for tomorrow. The calculations are shown below for the various levels of the atmosphere, either 2 meters above the ground in the case of the ETA and GFS models or 850mb pressure level for all three (ETA, NGM and GFS) models. ETA 850mb difference 06/00 6C to 07/00 7C = +1C (+2F) 2m difference 06/00 63F to 07/00 67F = +4F High For 2/5/07 71F High Temp 850mb Delta 71 + 2 = 73F High Temp 2m Delta 71 + 4 = 75F NGM 850mb difference 06/00 6C to 07/00 9C = +3C (+6F) High For 2/5/07 71F High Temp 850mb Delta 71 + 6 = 77F GFS 850mb difference 06/00 6C to 07/00 9C = +3C (+6F) 2m difference 06/00 57F to 07/00 59F = +2F High For 2/5/07 71F High Temp 850mb Delta 71 + 6 = 77F High Temp 2m Delta 71 + 2 = 73F To further refine our answers we could take a consensus or average of the various results we obtained above for each type of calculation: 850mb Consensus (73 + 77 + 77) / 3 = 75.6 say
77F Lastly we could also take a consensus of both of the various methods the 850mb and the 2m results to give us one number: Overall (77F + 74F) / 2 = 75.5 say 76F Although this consensus is not strictly part
of the delta method, it may allow us to get something of a better result by
averaging the various methods As a result of our work above we have arrived at the forecast high temperature of around 76F for San Antonio, Texas for February 6th. Let's look at the relevant portion of the climatology report from the San Antonio National Weather Service Office and see how well we did: ...THE SAN ANTONIO CLIMATE SUMMARY FOR FEBRUARY 6 2007... CLIMATE NORMAL PERIOD 1971 TO 2000 CLIMATE RECORD PERIOD 1886 TO 2006 WEATHER ITEM OBSERVED TIME RECORD YEAR NORMAL DEPARTURE LAST VALUE (LST) VALUE VALUE FROM YEAR NORMAL .................................................................. TEMPERATURE (F) YESTERDAY MAXIMUM 71 410 PM 83 1925 65 6 69The actual high temperature was 71F, about 5 degrees higher than our consensus number, but if we look at the various individual answers we obtained above, several of them were much closer in fact. The ETA 850mb result and GFS 2m result were just 2 degrees off at 73F, so clearly the method does work, and rather well, although the value of taking a consensus number is debatable based on just this one example. Comparing our forecast highs to the expected numbers from the various MOS outputs above we see that they had predicted 69, 68 and 71(red highlighted items), far less accurate than our delta method, even with the questionable consensus numbers. |