Welcome to the University of Arizona's renewable power forecasting website.
You're seeing this text because your browser is not compatible with our site -- most likely because you're using Internet Explorer 8 or earlier. We recommend using Chrome to view our site. Recent versions of Internet Explorer and Firefox will also work with our site.
Enjoy a safer, faster, and better internet by upgrading your browser using the links above!
Welcome to the University of Arizona Renewable Energy Network's public data page.
Select a data source using the menus above.
This page is made possible with support from Tucson Electric Power, Arizona Public Service, and Public Service Company of New Mexico.
About this site
This site shows selected point forecasts from recent WRF model runs from the UA Department of Atmospheric Sciences. We welcome your suggestions, but please know that this site is not funded and is mostly maintained in our spare time. This site is the beneficiary of websites created for renewable power forecasting for TEP and APS.
Our forecasting team is a collaboration between the UA Department of Atmospherics Sciences, the UA Renewable Energy Network, and the UA Department of Physics. You can read more about our work at forecasting.uaren.org.
Thank you to our funding agencies! Arizona Public Service and Tucson Electric Power provide the majority of the funding for the Arizona WRF model. The Arizona Department of Environmental Quality and the SVERI utilities provide additional funding. Will Holmgren is funded by the Department of Energy's EERE Postdoctoral Fellowship Program.
The forecasts on this page can be used for any purpose so long as they are attributed to the University of Arizona. We hope that you will let us know if you use them. We make no guarantees of availability or accuracy.
Please consider supporting our work if you find this page to be useful.
A tour of the site
The site loads with an overview panel of WRF model predictions for Tucson, Phoenix, Flagstaff, and Yuma. Click on "Toggle 1 day / Multi day view" in the upper left corner to toggle between a rolling 24 hour plot of forecasts and actuals, a 2 day forecast, and an 8 day forecast. Your browser is supposed to remember your choice between visits. Your browser will automatically reload the images every 5 minutes.
Click on any image to open it in a new page. This too will automatically reload.
You can see all information for a single sensor by hovering over the "Environmental data" menu and then selecting a sensor. Clicking on "Home" will take you back to the overview page.
The first image is the rolling 24 hour plot of forecasts and actuals. This plot is updated every ~20 minutes and the plot creation time is indicated in the title.
The second image is the 2 day forecast starting at midnight MST. This plot is only updated when new model runs are available.
Next, we have a wide plot with a 7-8 day forecast. We run one model per day out to 8 days, and it is initialized on the 0Z GFS. Remember, we're showing model range, not uncertainty. There's a big difference.
Finally, we create two verification plots of the previous day's model suite. The first image is a plot of forecast and actuals.
The second image shows the difference between forecast and actuals, plus the error statistics for the mean. Error statistics are shown using data resampled to 15 minutes and 1 hour, though only some stations record data more frequently than 1 hour. Also, some stations record data at irregular intervals and this can sometimes lead to strange looking error statistics.
We also create a series of multiday/multiday-ahead verification plots show the accuracy of the weather models over the last 4 days. The "0 day-ahead" plots show forecasts generated using models valid on the day that they were run, the "1 day-ahead" plots show forecasts generated using models run on the previous day, and so on.
Next, there is a summary graph that shows the mean bias error, mean absolute error, and root mean squared error for the 0-5 day-ahead forecasts valid over the previous 4 days.
Finally, there is a Taylor diagram with unbiased RMSE and correlation. Here's a primer on Taylor diagrams.
A few words on the green model range envelope, the mean, and the median. The model range is the +/- 30 minute rolling mean of the +/- 30 minute rolling min/max. The same procedure is used to calculate the model average. We do not apply any smoothing to median.
We will consider adding csv files of the forecast data if there is sufficient interest.
The "Other resources" menu provides links to the main UA WRF site and local radar images and satellite images. It also contains a link to AZ Soundings, a page that just shows the most recent SPC rendering of the 0Z or 12Z sounding for Tucson, Phoenix, Flagstaff and Yuma.
More about the forecasts
See the ATMO WRF page for the latest WRF model details. These forecasts are straight out of the model, no MOS.
How are the plots created? A lot of Python and a little bit of bash.
We have 10s of TB of archived WRF forecasts. Let us know if you're interested in using them in a study.
Click an icon to get more information about that station/system. If available, forecasts for that station/system will be shown below. Most weather station forecasts are only available on the public webpage. Power forecasts are not available to the general public.
Click the Weather Stations, Solar PV, Solar CSP, and Wind checkboxes to hide or show the icons for that station/generation type.
Click the Scale icons checkbox to make the icon area proportional to the peak power of that generator.
Use the Weather Station Type menu to control the variable shown in the forecasts below (if available).