Summary
Combining climate models with Earth observations provides an important technique to project climate change under different future scenarios and in turn inform global climate policy. To assess the robustness of the climate projections from these increasingly complex models, innovative and comprehensive software tools such as ESMValTool are needed. The core ambition of ESO4clima project is to enhance ESMValTool in time for the next phase of the international coupled model intercomparison project CMIP7, which will feed into the upcoming climate assessment report of the Intergovernmental Panel on Climate Change (IPCC AR7). The project's targeted enhancements will increase the informativeness and efficiency of rigorous climate model evaluation and analysis using observational data by enabling fuller exploitation of satellite datasets such as ESA's Climate Change Initiative.
What is ESMValTool?
ESMValTool is an open-source, community-developed evaluation and analysis tool widely used in European projects and cited in over 60 peer-reviewed publications. It contributed to IPCC AR6 and has been selected by the CMIP7 Model Benchmarking Task Team as a featured tool.
Background
Earth system models (ESMs), combined with Earth observations, are essential for understanding present-day climate and projecting future changes under different scenarios—informing global climate policy. Over recent decades, climate models have evolved from simple atmosphere-only systems to complex, high-resolution Earth System Models (ESM), with many participating in the Coupled Model Intercomparison Project (CMIP7) - an international effort that coordinates model experiments to compare and improve Earth system models for understanding past, present, and future climate.
To advance robust climate projections, it is critical to assess how well models reproduce observed climate and to systematically analyse, evaluate, and document shortcomings. This is achieved by comparing ESM outputs with Earth observations. Approaches include 1. sampling model outputs - matching the model; to satellite definitions, which is practical yet costly; 2. estimating uncertainties - while reducing false conclusions does not but not improving informativeness; and a third approach, deriving adjusted definitional and sampling differences, with associated adjustment uncertainty information. In this latter approach, adjustments improve the potential for the comparison to be informative about model biases, while exploitation of the uncertainty information helps interpret model-observation differences appropriately according to their statistical reliability. This third approach will be implemented for the first time in ESMValTool within the ESO4clima project. Machine learning will be used to learn adjustments for ESA CCI and CCI+ datasets, creating observation versions optimised for model comparison. These adjustments, combined with uncertainty exploitation, will improve interpretation of model-observation differences. To support CMIP7, ESMValTool will be enhanced with:
- Improved handling of regional model grids
- Integration of new data sources such as xcube
- Better accessibility and interoperability of CCI data
Aims and objectives
The project aims to enhance ESMValTool for CMIP7 to enable more efficient and informative analysis of large climate model datasets and improve the use of ESA Climate Change Initiative observations in model evaluation.
Objectives:
- Develop an improved framework to address observational characteristics and uncertainties in model evaluation
- Implement technical upgrades to handle large volumes of high-resolution model and observational data
To achieve this, a machine-learning (ML) adjustment method will be developed for sea surface temperature (SST) and water vapour (WV) using ERA5 reanalysis data. This involves aggregating CCI data, applying ML-based adjustments, propagating uncertainties, and assessing observation-model differences in light of uncertainty
Project plan
The project consists of four scientific work packages:
- WP1: Improving the comparability of satellite observations and climate models with machine learning
- WP2: Extending ESMValTool for improved support of regional models
- WP3: Enhanced support for satellite data with new data concepts in ESMValTool
- WP4: Capability and capacity development
WP1 develops transferable methods to account for definitional and geophysical observational-model differences, to improve the comparability of satellite observations and climate model output. Machine learning techniques are applied to adjust the satellite data to allow for a more direct and fair comparison with ESM results. In addition, it is investigated how pixel-wise uncertainty information from satellite datasets can be propagated to the temporal and spatial scales relevant for ESM evaluation. The goal is to combine both aspects into a framework that can be implemented into ESMValTool. The demonstration will be for sea surface temperature and water vapour and used to quantify effects on model-satellite comparisons compared with previous work.
The aim of WP2 is to enhance the ability of ESMValTool to analyse regional climate model data. The focus will be on European CORDEX models. Several of these models have errors in their metadata, and this project will introduce fixes for those problems. New regridding capabilities developed will improve ESMValTool support for many ocean models that employ two-dimensional coordinates and thus bring new capability for combining ocean observations with models.
In WP3, the accessibility and interoperability of CCI data for users is enhanced by extending ESMValCore, the framework powering the analyses in ESMValTool, so it can read CCI data directly from the ESA Open Data Portal through the xcube software package. Possibilities to add support for other data sources such as intake-esgf, intake-esm, and/or STAC will be explored. Improvements are achieved by developing the capability to reformat the CCI data while the data are in memory so they follow the standards that climate model output follows. This will be developed as a stand-alone software package that can be used as a plug-in to ESMValCore for ESMValTool users, but also for the wider community, e.g. plain Xarray () or Iris users.
The objective of WP4 is to provide capability for evaluating and analyzing Earth system model data with ESA CCI data sets using ESMValTool. We will do this through training programs, targeted development of tutorials, user support and engagement in the climate modeling and observational data scientific community.