The ARPA-E PERFORM datasets produced by NREL contain both historical and forecast data on load, wind power, and solar power. NREL offers the following piece of information about the day-ahead forecast data:

Day-ahead forecasts are generated with a 11-hour-ahead lead time, a 48-hour horizon, an hourly resolution, and a daily update rate.

I downloaded the data for wind power in ERCOT (2018) and the two first columns with issue time and forecast time, respectively, are shown below with the first few rows of the dataset (Forecasts quantiles are not included).

Issue time Forecast time
2017-12-30 18:00:00 2018-01-01 00:00:00
2017-12-30 18:00:00 2018-01-01 01:00:00
2017-12-30 18:00:00 2018-01-01 02:00:00
2017-12-30 18:00:00 2018-01-01 03:00:00
$\vdots$ $\vdots$
2017-12-31 18:00:00 2018-01-01 06:00:00

Given the American Meteorological Society's definition of lead time as "the length of time between the issuance of a forecast and the occurrence of the phenomena that were predicted", I assumed that the difference between the forecast time and the issue time had to be 11 hours. However, from row 1 in the table above, we could see that such difference is 30 hours. For the second row, it's 31 hours. In row seven, we have a difference of 12 hours.

When I saw the issue time was the same for various forecasts, I knew my understanding of forecasting and its nomenclature would be tested. I realized that if I have to choose a set of all forecasts for a fixed $n$ hours ahead, given any $n = 1, 2, ..., 48$, I wouldn't be sure of what forecasts (rows) to select.

There is something I am missing and that's what I'd like understand. What is the lead time here? And what does the difference between the forecast time and the issue mean in this case?


1 Answer 1


This might be helpful: https://forecastarbiter.epri.com/definitions/#forecastattrs. Also see this paper https://www.osti.gov/pages/servlets/purl/1762478, which is referenced in the README you got the description from (https://github.com/PERFORM-Forecasts/documentation/blob/main/README.md#forecasts).

My interpretation of the "day-ahead" wind power forecast that you are looking at is:

  • a forecast is issued once per day at 18:00 [daily update rate], e.g., 2017-12-30 18:00
  • the first hour included in that forecast is 05:00 the following day [11-hour lead time], e.g., 2017-12-31 05:00
  • the forecast includes each hour [hourly resolution] out to the one beginning 04:00 (ending at 05:00) on the morning of the 3rd day out [48-hour horizon], e.g., 2018-01-02 04:00.

(I'm assuming they use timestamps representing the beginning of intervals, e.g., 05:00 represents the average of the hour ending 06:00).

So, if you wanted to assemble a set of all forecasts n hours ahead, n would have to be between 11 and 59, and you would only get one forecasted value per day. It would also be subjected to any diurnal patterns in solar, wind, or load, which may or may not be problematic, depending on what you want to do with it.

Here's a rough sketch that might be helpful: sketch of my interpretation of the temporal aspects of PERFORM day-ahead forecasts, with similar format to fig 1 from Doubleday 2020.

Edit: After looking at the description here again and comparing with data in the S3 bucket, it looks like they split data into "day-ahead" and "2day-ahead" datasets. So, I think this description,

Day-ahead forecasts are generated with a 11-hour-ahead lead time, a 48-hour horizon, an hourly resolution, and a daily update rate.

along with my interpretation and my sketch apply to the combined day-ahead and 2day-ahead datasets. Otherwise, there would be overlapping forecast times when the horizon (48 hrs) is longer than one over the update rate (24 hrs). Here's an updated sketch, where I'm guessing at the point where they cut between day- and 2day-ahead:

updated version of previous sketch

  • $\begingroup$ Hi, Will. Thanks for your answer. $\endgroup$
    – Jxson99
    Commented Dec 20, 2022 at 16:25
  • 1
    $\begingroup$ Hope it was helpful! $\endgroup$
    – Will Hobbs
    Commented Dec 20, 2022 at 21:18
  • $\begingroup$ It was helpful indeed. Thanks a lot. $\endgroup$
    – Jxson99
    Commented Dec 21, 2022 at 16:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.