How well can we predict subseasonal precipitation?

These maps show the correlation between the weekly precipitation forecast and observed weekly precipitation values, depending upon the month of year when the forecast is made and the weekly forecast leads (1 to 4 weeks into the future).

The correlation values range between -1 and +1. A high positive correlation (toward +1) indicates that the forecast and observed precipitation tend to co-vary in phase, or in agreement with each other; a high negative correlation (toward -1) indicates that the forecast and observed precipitation tend to vary out of phase, or out of agreement with each other.

This skill measure does not give any indication of the bias of the forecast. In other words, even if there is a high positive correlation, the difference between the forecast and observed precipitation could still be quite high.

Use the drop-down menus at the top of the page to select the forecast week. Mouse over the map to select the forecast start time from a control that appears just above the map. Select a combination of the forecast start time and the week in the future to produce a map for a target week.

Keep in mind that forecasts beyond one or two weeks are still an area of active research, tend to have low skill, and should be treated with caution.



In order to document SubX data impact and enable continuing support, users of SubX data are expected to acknowledge SubX data and the participating modeling groups. The SubX model output should be referred to as "the SubX data []" and referenced using the SubX DOI: 10.7916/D8PG249H. In publications, users should include a table (referred to below as Table XX) listing the models and institutions that provided model output used in the SubX system, as well as the digital object identifier of publications documenting the models. In addition, an acknowledgment similar to the following should be included in any publication: “We acknowledge the agencies that support the SubX system, and we thank the climate modeling groups (Environment Canada, NASA, NOAA/NCEP, NRL and University of Miami) for producing and making available their model output. NOAA/MAPP, ONR, NASA, NOAA/NWS jointly provided coordinating support and led development of the SubX system.” where “Table XX” in the paper should list the models and modeling groups that provided the SubX data.

Dataset Documentation

Model Data

Data: Daily forecast and hindcast precipitation data produced on a weekly basis from the NOAA Earth System Research Laboratory (ESRL) FIMr1p1 climate model, as part of the SubX experimental forecast project. The hindcasts and forecasts from this model include 4 ensemble members each week, and forecast leads extend 32 days into the future. Data are available globally on a 1.0° lat/lon resolution grid. The hindcast period ranges from 1999 to 2017.

Data Source: U. S. National Oceanic and Atmospheric Administration (NOAA), Earth System Research Laboratory (ESRL). Along with output from the other SubX models, the data are publicly distributed via the IRI Data Library here: ESRL FIMr1p1. Additional information about the SubX project and the associated forecast models can be found at the SubX Project Website.

Observational Data

Data: Daily gridded precipitation data from the University of California Santa Barbara (UCSB) Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) version 2.0 dataset, available globally from 50°S to 50°N, and spatially regridded to match the 1.0° lat/lon resolution grid of the ESRL FIMr1p1 forecast dataset.

Data Source: University of California Santa Barbara (UCSB) Climate Hazards Group (CHG). The original source of the data is the UCSB CHG web page: CHIRPS.

Analyses: The analysis shown here is the Spearman rank correlation between the ESRL-FIMr1p1 forecast ensemble mean, weekly mean precipitation anomaly for each of four weekly forecast leads, and the matching weekly mean CHIRPS precipitaiton anomalies, by forecast initialization month. The correlation is calculated by the month of the weekly forecast initialization times in order to increase the number of samples included in the correlation. The climatology used to calculate the anomalies is generated from the 1999-2015 mean of ensemble mean precipitation from the model hindcast, and the same base period from the CHIRPS observational dataset. Correlation analysis courtesy of Á. Muñoz, IRI.


Access the dataset used to create this map.

How to use this interactive map

Return to the menu page: Click the blue link called “Forecasts” at the top left corner of the page.


Contact with any technical questions or problems with this Map Room.