May 21, 2026
As previously announced (see below), the IRI Data Library will soon shut down. While we had estimated that the shutdown might happen as soon as April, we now know that we will retain our current staff through the end of June. After that, we hope to continue operating for a few more months, possibly with a decline in reliability and responsiveness because of a reduction in staff. The full shutdown will most likely happen by the end of October 2026, possibly earlier.
We now invite you to start testing a new service (working name: forecast.ccsr) that will partially replace the IRIDL. Forecast.ccsr currently contains hindcasts and forecasts from four of the five active members of the NMME, for three variables: precipitation, 2-meter temperature, and sea surface temperature. Other variables and the fifth model (CFSv2) will be added gradually over the next few weeks, to be followed later by SubC, S2S, historical archives of past NMME model versions, and a small number of observational variables.
While the new web site is visually bare-bones at this point, the OPeNDAP functionality for data download should be fully functional. Please test it thoroughly and let us know about any issues as soon as possible. Each variable has its own web page, and at the top of each such page is a button labeled “Copy OPeNDAP URL.” After clicking that button, you can paste the URL into your OPeNDAP client of choice, and then use that client to perform operations such as downloading a geographical or temporal subset.
We have generally followed the metadata conventions used in the IRIDL, with these notable exceptions:
- Seasonal forecasts in the IRIDL used a calendar called “360” and units of “months since …”, neither of which is allowed by the CF Conventions. While the Conventions do define a calendar called “360_day”, we have chosen to represent forecast data on forecast.ccsr using the “standard” (Gregorian) calendar. As for the units of the L (lead) coordinate, they are still conceptually in months, but we have omitted the “units” attribute since the unit “months” is expressly prohibited by the latest version of the Conventions. Auxiliary coordinates “target” and “target_bnds” are available on each variable to resolve any doubts about how to interpret L.
- Precipitation is expressed in units of mm (total for the month) rather than mm/day. We made this change because we have seen many programs, including some of our own, that calculate seasonal totals inaccurately from rates (by not accounting correctly for the varying lengths of months). Calculating a seasonal total from monthly totals is simpler and thus more likely to be done correctly.
- Consistency in variable names and units is enforced. For example, on forecast.ccsr all 2-meter temperature variables are named t2m (rather than variously t2m, tmean, or tref depending on the model), and all are in ˚C (rather than some in ˚C and others in Kelvin).
We invite you to comment on these decisions, as well as any other problems that you may notice. Please send comments as soon as possible; after a comment period of a few weeks, we will finalize the metadata format.
With the new service comes a new help desk email: forecast-ccsr-help@columbia.edu. The IRIDL help desk help@iri.columbia.edu will also continue to be monitored as long as the IRIDL is running.
Work is under way to adapt PyCPT to download data from forecast.ccsr. We will make an announcement here and on the PyCPT and IRIDL mailing lists when that is ready for testing.
January 30, 2026
Alternative Data Library Sources
We have published a list of alternate sources for data currently hosted by the Data Library. The following pages provide information:
November 11, 2025
Sunset of the IRI Data Library
After three decades serving climate data and information to the worldwide research and humanitarian communities, the IRI Data Library (IRIDL) will soon undergo major changes.
For more than a decade, operation of the IRIDL has been funded primarily by international development grants to the IRI. Such grants have declined in recent years, and will soon no longer be sufficient to continue employing the existing IRIDL staff, which is already significantly smaller than it was at its peak. We anticipate that by April 2026 the IRIDL will no longer be staffed sufficiently to continue operating in its current form.
In the time that remains, the IRIDL team has two main goals. Our primary focus is on ensuring that IRI’s remaining funded projects continue to meet their commitments. To this end, we are exploring various options for a smaller, simpler data infrastructure that will be sustainable at the current level of funding. Secondarily, to the extent possible without endangering our funded commitments, we will do what we can to help current users of the IRIDL find other ways to satisfy their data needs.
Sustainable infrastructure
Operating the existing Data Library requires a team of several people with different specialized sets of skills. We are exploring various options to reduce the complexity of the system so that it can be operated by a smaller and less specialized staff. Ideas we are exploring include the following:
- Replace the redundant architecture of the current IRIDL (two separate, self-sufficient installations in different buildings, each with a distributed filesystem) with a single server. This will make the system simpler to administer, at the cost of increased downtime during maintenance and hardware failures.
- Eliminate software written in the Ingrid programming language, which was invented for the IRIDL, is not widely known, and for which expertise has been lost to attrition over the years. We are evaluating open source data serving software such as Hyrax and TDS, as well as alternative formats such as Zarr and Icechunk that clients can read directly from cloud storage. No definitive technical decisions have yet been made in this area.
An instance of this simplified architecture will be created at Columbia Climate School’s Center for Climate Systems Research (CCSR) to host data from the NMME, SubC, and S2S projects, thanks to continuing funding from NOAA and DoD. We hope to make the new data service available for limited beta testing by December 2025 and for public access by January 2026.
A new version of PyCPT (IRI climate forecasting software) will be prepared that can download data from sources other than the IRIDL, likely including the new data service at CCSR.
If you are currently engaged with IRI on other funded projects, the data, applications, and support funded by that project will continue to be available privately to you. Users of other datasets should begin exploring other sources for the data they need.
The IRIDL Maprooms website will no longer be maintained. No replacement for the maprooms is planned.
If you have expertise, time, or funds to contribute towards building or hosting a more sustainable Data Library, we invite you to contact us at help@iri.columbia.edu.
Helping users transition
In order to help users of the IRIDL rebuild their data workflows with different tools, we will publish IRIDL code and data, and answer as many transition-related help requests as we can.
In the coming months, we plan to make the following resources available:
- A spreadsheet listing upstream or alternative download sources for each dataset currently served by the IRIDL, and identifying datasets that are no longer available elsewhere. For datasets that will be “orphaned” by the loss of the IRIDL, we will help transfer data to anyone who wishes to take over hosting them.
- Source code for the IRIDL’s data catalog. The data catalog consists of executable instructions (in the Ingrid programming language) that map different providers’ idiosyncratic data formats, directory and file naming conventions, and metadata schemas into the IRIDL’s standard representation, as well as metadata such as revision history and pointers to documentation.
- Source code for scripts that are used to download data from upstream providers.
Users of Maprooms can already find the data and analysis information through links on their respective maproom page, and then use the resources above, when made available, to reproduce their workflow of interest.
As this information becomes available, the IRI help desk (help@iri.columbia.edu) will increasingly prioritize helping users find alternatives, rather than helping them use resources that will soon be lost.
To receive updates on the status of the sunset plan, please watch this page and/or visit https://iri.columbia.edu/subscribe to subscribe to the Data Library mailing list.
Regretfully,
The IRI Data Library team