The Weather in the Details


Why have utilities lost millions of dollars on weather-normalization plans? Blame deprecated NOAA calculations.

Fortnightly Magazine - November 2006

The views and findings expressed herein are solely those of the authors and should not necessarily be attributed to PSE.

A hypothetical Northwest utility with a revenue requirement of $50/MWh to $90/MWh and weather sensitivity on the order of 500 MWh to 800 MWh per degree-day would expect revenues to rise by roughly $45,000 for each additional heating degree-day experienced per annum. Reliance on National Oceanic and Atmospheric Administration (NOAA) standard measurements results in approximately 77 additional heating degree-days of weather adjustment as compared with using hourly average heating degree-day measurements.

In other words, NOAA’s measure of heating degree-days between a normal 30-year period and a given test year is consequently too high by 77 degrees when compared with the more accurate hourly estimates for the 30-year period and for the test year. In this case, the hypothetical utility would see a revenue shortfall of between $2 million and $5 million.

Put another way, NOAA’s measure of heating degree-days would yield extra weather normalization of roughly 25 percent, with rates being set too low by similar magnitudes. This shortfall is likely to be larger for gas utilities, where the weather sensitivities are greater, and larger in temperate regions where this difference between NOAA and average hourly heating degree-day is magnified.

Empirical Results

In our study, we analyzed hourly temperature from 1971 through 2005, measured at Washington’s Sea-Tac Airport.1 There were 306,816 such hourly temperatures. We calculated the skewness for each day in this period (daily skewness), as well as the mean daily temperature and the mid-range temperature. All temperatures were recorded top-of-the-hour by NOAA. We then calculated daily heating degree-days comparing the Da and Dm measures (average based and mid-range based). The temperature distribution was found to be positively skewed, with a greater likelihood of warmer extreme temperatures than colder extreme temperatures. Heating degree-days were lower by 100 to140 degree-days per year (depending on the month) using the mid-range estimate as compared with the mean. Skewness in the temperature distribution, therefore, causes the average to be underestimated by 10 to 15 degrees. As the mid-range is larger than the true average temperature, the base temperature less the mid-range estimate typically is smaller than the difference between the base temperature and the true average temperature. For the period 1971 through 2000, the estimate of heating degree-days based on the mid-range (average of minimum and maximum top-of-the-hour temperatures) is 4,817 on an annual basis. This is 125 degree-days lower than the estimate based on the true average.

We examined actual hourly temperatures to compute the hourly degree-days and found that for the period 1971 through 2000, there were 5,093 heating degree-days on an annual basis (method D). Using the daily average of such hourly temperatures lowers the estimate to 4,942 heating degree-days (method a). Using the mid-range estimate of the mean degree temperature published by NOAA further lowers the estimate to 4,817 heating degree-days (method m). There is a further difference of 20 degree-days to get to the published normal of 4,797 due to NOAA errors, spline adjustments, and rounding. Hence, heating degree-days are underestimated by 0.4 percent due to miscellaneous adjustments, 2.6 percent due to the difference between mid-range and daily average temperature, and a further 3.1 percent due to the difference between daily average temperatures and hourly based estimates. The conclusion is that:

D $ a $ m.2

The NOAA method relying on the mid-range produces an estimate of heating degree-days that is significantly lower than actual hourly degree-days.

With respect to weather normalization and rate making, an important consequence of these results is that a smaller degree-day adjustment should be expected between normal periods and test-year periods when using actual hourly degree-days as compared with the difference between normal and test-year using the mid-range average (NOAA method). The reason for this is that the averaging that occurs in forming a normal (30-year average for each month of the normal year) is likely to narrow the difference between the mid-range and average temperature. For instance, using the period of 1971 through 2000 as the normal period and October 2004 through September 2005 as the test-year period, we found that the adjustment between the NOAA normal and the test year is 344 degree-days, but is only 275 degree-days using daily average temperatures. Using the more accurate hourly based estimate implies a difference of 267 degree-days.

The data requirements for calculating degree-days by more precise techniques are minimal. We conclude that an accurate weather adjustment in the rate-making context requires accurate estimates of heating and cooling degree-days. Until NOAA adopts an alternative calculation methodology, our specific recommendation is that electric and natural-gas utilities adopt an hourly or average based measure of heating degree-days when comparing normal and test-year periods in rate-making proceedings. Rote reliance on NOAA calculations leads to excessively large weather adjustments in typical situations.


1. Sea-Tac Airport is a “first-order” weather station with largely complete and accurate historical temperature information.

2. Our results apply to hourly-degree-days as well.