Introducing Code for Earth 2023 Selected Teams (Part 2 of 2)

The 2023 edition of Code for Earth is in full swing, with ten developer teams working on projects that tackle challenges in weather, climate, and related domains. These projects represent the diverse areas of focus within the ECMWF's scope of activities, ranging from tropical cyclone data dissemination to wildfire prediction, climate intelligence, spatial resolution enhancement, machine learning for weather forecasting, generation of high-resolution reanalysis data, atmospheric composition diagnostics, land surface variable validation, geospatial data compression, and even a conversational search engine for ECMWF data.

By leveraging the expertise, creativity, and problem-solving skills of these talented teams, Code for Earth is poised to make significant contributions to the field of Earth sciences. Through multidisciplinary collaboration and the embrace of open-source principles, our program aims to drive innovation and foster advancements that benefit the participating teams and the broader Earth sciences community.

Below is a short description of the second group of the selected projects:

DeepR: Deep Learning for Generating High-Resolution Reanalysis Data aims to create a machine learning model that generates high-resolution regional reanalysis data by downscaling global reanalysis data from ERA5. Employing state-of-the-art deep learning techniques such as U-Net, conditional GAN, and diffusion models, DeepR focuses on model effectiveness evaluation using a detailed validation framework. This framework includes various deterministic error metrics, in-depth validations, and interpretability tools to ensure an accurate representation of physical processes and reduce errors.

Atmospheric Composition Dataset Explorer: API for Atmospheric Composition Diagnostics aims to create an API and an interactive application that generate atmospheric composition diagnostics plots based on user specifications. Leveraging the CAMS Atmosphere Data Store, this application automates the creation of frequently used time series, hovmoeller, and geospatial plots. The project aims to provide generic APIs for data retrieval, homogenisation, slicing, sub-setting, aggregation, and visualisation of different CAMS datasets. The addition of new plot types will be made easier, and a GUI will enable users to generate reports based on selected parameters.

Benchmarking Surface Heat Fluxes: Validating Land Surface Variables focuses on validating soil moisture and temperature with the LANDVER validation package. By expanding the package's capabilities to include the validation of latent and sensible surface heat fluxes against Eddy-Covariance measurements, this project aims to improve understanding of ECMWF's Land-Surface Modelling Component ECLand. In addition, this standardised land-surface benchmarking tool provides valuable insights into how soil moisture stress translates into surface heat fluxes and enhances Numerical Weather Prediction.

Compression of Geospatial Data with Varying Information Density aims to improve the compression technique used by xbitinfo to preserve natural variability in geospatial features. By accounting for the varying information density across different locations and datasets, compression efficiency can be increased. This project recognises that datasets like streets, oceans, and the atmosphere have different information densities, and the bitinformation framework should adapt accordingly to efficiently compress these datasets.

ChatECMWF: A Conversational Search Engine for ECMWF Data aims to develop a search engine that provides responses to queries related to ECMWF datasets, charts, and documentation. By leveraging natural language processing techniques and large language models, such as ChatGPT, users can formulate queries in natural language, simplifying access to ECMWF's vast amount of weather data. The search engine maps user requests to API queries, seamlessly providing the required information and reducing the barrier to entry for utilising ECMWF's extensive resources.