My latest North American temperature forecasts

My temperature forecasts for the next 15 days

Jump to the forecasts or to the verification

Brief outline of the method

The starting point for the forecasts are NCEP ensemble forecasts (available at a resolution of 1x1 (latitude-longitude) degrees); I use the control run and 2 ensemble members - 3 forecasts in all.
The forecasts cover the period 6-384 hours after the analysis time - providing 15 days of daily forecasts.
For a given location (at which daily maximum and minimum temperatures must also be available), the average maximum and minimum forecast temperatures at each forecast time are determined from the model output.
Analysis of previous forecasts for each of the 15 forecast lengths (1 day, 2 days, etc) against the subsequent verifying observations for a period of up to 30 days before the start time of the current forecasts, produces a bias correction which is applied to the model-derived forecast to generate my station forecast (the 'issued' forecast). This allows the bias to vary with time of year, etc.
The issued forecasts relate to the period 0000-2400C for maximum and minimum temperatures.

Once the observations become available, they are compared against the issued forecasts in order to analyse the error in the issued forecast, both as a function of forecast length (0-day, 1-day, 2-day, etc.) and calendar date (by combining all forecasts made for that date).
More details of these validation results will be posted once sufficient forecasts have been created.

Guide to the forecasts

The forecasts contain:

Recent observations of minimum, maximum and mean (i.e. the average of minimum and maximum) temperature for up to 10 days. Observations from sites are extracted automatically from SYNOP reports and for a given day cover the period 0000-2400UTC.
Forecasts for 15 days ahead of
- minimum (0000-2400UTC) temperature,
- maximum (0000-2400UTC) temperature and
- mean temperature.
An indication of the forecast confidence based on the standard deviation of the value of the error (equal to the issued forecast minus the observation) of the past 30 forecasts - for each of the forecast lengths (0 days, 1 day, 2 days, etc.). In general the confidence will get less as the forecast length increases.
- 'Hi' denotes a standard deviation in the error of under 1.5C (a 'high' accuracy forecast)
- '-' denotes a standard deviation in the error of 1.5-3.0C (mid-range accuracy), while
- 'Lo' denotes a standard deviation in the error of over 3.0C.
Note that no account has been yet taken of the forecast uncertainty implied by the spread of the individual ensemble forecast values, as it is considered that interpolation from the gridded data to the observation location probably creates a larger uncertainty.

Time zones

An approximate guide to time zones and the times of extremes temperatures is as follows:

Location	Winter local time zone	UTC time of minimum temperature (assumed 6am local time)	UTC time of maximum temperature (assumed 3pm local time)	Summer local time zone	UTC time of minimum temperature (assumed 6am local time)	UTC time of maximum temperature (assumed 3pm local time)
Maine	EST	1100UTC	2000UTC	EDT	1000UTC	1900UTC
Chicago	CST	1200UTC	2100UTC	CDT	1100UTC	2000UTC
Phoenix	MST	1300UTC	2200UTC	MDT	1200UTC	2100UTC
Los Angeles	PST	1400UTC	2300UTC	PDT	1300UTC	2200UTC
Alaska (mainland)	AKST	1500UTC	2400UTC	AKDT	1400UTC	2300UTC
Hawaii	HST	1600UTC	0100UTC	HST	1600UTC	0100UTC

The latest observations

Observations for yesterday

The latest forecasts

Hawaii
Honolulu, Hawaii

Alaska
Kodiak AK

Canada

Edmonton	Winnipeg	St Johns
Halifax	Toronto	Quebec
Whitehorse	Victoria	Fredericton
Resolute	Charlottetown	Regina

United States - listed by state

AL Montgomery	AR Fort Smith	AZ Phoenix
CA Los Angeles	CO Alamosa	CT Hartford
DC Washington Dulles	FL Miami	GA Atlanta
IA Des Moines	ID Boise	IL Chicago
IN Indianapolis	KS Topeka	KY Lexington
LA Lake Charles	MA Boston	ME Caribou
MI Detroit	MN International Falls	MO Kansas City
MS Meridian	MT Helena	NC Raleigh
ND Bismarck	NE Grand Island	NH Concord
NJ Newark	NM Albuquerque	NV Reno
NY New York JFK	OH Columbus	OK Oklahoma
OR Portland	PA Philadelphia	RI Providence
SC Columbia	SD RapidCity	TN Memphis
TX Houston	UT Salt Lake City	VA Richmond
VT Burlington	WA Olympia	WA Spokane
WI Milwaukee	WV Charleston	WY Cheyenne

Verification

To see how good (or bad) recent forecasts actually were, see these comparisons of observations for yesterday against previous forecasts valid for yesterday:

24-hour forecasts,
48-hour forecasts,
72-hour forecasts.

The images above show the standard deviation (SD) and the absolute value (AB) of the forecast error (where the error is the difference between the forecast and the subsequent observation), averaged across all sites for the 30-day periods indicated.

In general the forecast quality reduces with increasing forecast length with maximum temperature forecasts prone to a larger error than those for minimum temperatures.
Forecasts for these locations are currently (1 September 2008) larger than those for the European and British Isles locations as the latter forecasts have been in operation for a longfer period. The algorithm used means that it takes about two months for the forecasts to settle down - in the absence of any other influences.

Comments?

Your comments are welcome, and can be posted to me.

Disclaimer

The forecasts listed above use model data that are kindly provided by NCEP and a computer algorithm of my own devising. Neither my employers nor myself are professional weather forecasters and, while these forecasts may give an indication of conditions to come, we cannot accept any responsibility for the accuracy of this forecast.

The weather over Europe is part of an ever-changing system. The data on these pages should be used as a guide only and there are times when any forecast beyond about 2-3 days is likely to prove unreliable. In addition, it it unwise to make inferences about other weather elements, e.g. rainfall, from temperature forecasts alone.

Always check the latest forecast from Environment Canada or the US National Weather Service before undertaking any weather-dependant activity.

You are visitor number [an error occurred while processing this directive] to this page since 24 July 2008.