The Climate Data Store Virtual Assistant

Kevin Marsh, Michela Giusti, Xiaobo Yang, Anabelle Guillory


The Climate Data Store (CDS) is the cornerstone of the Copernicus Climate Change Service (C3S) at ECMWF. It serves tens of thousands of users, providing a wide range of free and open datasets and toolbox applications, which offer easy access to a wealth of climate information. The CDS Virtual Assistant (VA), or ‘Knowledge Duck’, is an exciting new addition to the support provided to C3S users. It can now be seen in action on CDS web pages:

Role of the Knowledge Duck

Role of the Knowledge Duck

User support is an important part of C3S. Various support channels have been set up in recent years as C3S has developed. These include the Knowledge Base, the User Forum, and the Jira-based service desk. Although users can access any one of these channels at any time, it can sometimes be confusing for them to know which one is best to find the answers they need. Therefore, to be better supported in their journey, our users need some guidance to explore the various information sources available. The VA will help ensure maximum and efficient usage of these support resources.

Prior to the introduction of the VA, the concept was tested by the development of a Support Beacon on the C3S website, which allowed users to access relevant pages from the Knowledge Base, and to raise Jira tickets without leaving the C3S website. This new VA is a significant improvement on the Support Beacon, which only provided simple searches of Knowledge Base content. By clicking on the Knowledge Duck icon in the corner of CDS web pages, a pop-up window will open and users can interact directly with the VA (see the screenshot). It provides answers drawn from the Knowledge Base and the User Forum, as well as the ECMWF parameter database and existing documentation stored on the CDS itself. Several pre-configured buttons also allow users to access answers to frequently asked questions about popular areas of interest, such as user accounts, ERA5, CDS API etc. The VA thus enables users to get considerable support whenever they use the CDS.

The VA is currently able to handle hundreds of simultaneous users, but it could be set up to deal with more. It was launched on the CDS on 1 July 2021, and the text box shows key statistics for the first two months of use. It should be noted that these are preliminary statistics. As the VA is an evolving system, a more representative picture will emerge after the system has been in use for a longer period. A full assessment will be produced in 12 months to verify the success of the system.

The Knowledge Duck.
The Knowledge Duck. Located on the bottom right of CDS web pages, the Knowledge Duck can provide answers to common questions of CDS users.

​The Data Support Team monitors user interactions on a daily basis and improves the responses of the VA by either modifying existing responses or adding new ones. Users are invited to provide direct feedback about their interaction with the VA by rating the response they receive. Almost 50% of the feedback received in the first two months was positive, and 15% indicated that the interaction was somewhat useful. Around 25% of the feedback was negative, and no opinion was provided for the remainder. This feedback is also analysed every day by the Data Support team, and the responses from the VA are adjusted so that conversations which led to negative feedback are addressed as quickly as possible. This leads to a rapid improvement in the quality and usefulness of responses from the VA.

Over the first two months of the service, it has been seen that the questions asked by users have increased in complexity, with more conditional free-text questions being asked. While such free-text questions can be challenging for any automated system to answer fully, by constantly improving the responses from the VA it is more likely that the correct information will reach the user. If a user fails to receive a suitable answer from the VA, they are reminded that they can always submit their query via the ECMWF Support Portal (see the article on the Portal).


In the future, the VA will continue to develop and evolve in several ways. The range of information sources used will be extended in order to provide answers to more complex user questions. In some cases, it would also be useful to have a ‘person’ behind the service, so that they could either interact with the user directly, or ‘inject’ a customised response into a user conversation. The system for detecting complex questions or negative feedback could also be improved so that a more agile service can be provided. There is also the possibility of allowing users to raise a Jira query directly from a VA conversation. It is hoped that eventually VAs may be used on the website of the Copernicus Atmosphere Monitoring Serivce (CAMS) and other ECMWF systems to complement existing user support services. Moreover, this may then allow user interactions to be transferred between them (or to retrieve answers from relevant VAs) to provide users with the information they need.

The CDS VA may be a small duck, but it has a very promising future, helping users to navigate the exciting waters of user support!

CDS Virtual Assistant statistics.
CDS Virtual Assistant statistics. The chart summarises VA statistics from July to August 2021.