Climate models need more frequent releases of input data - here's how to do it

"First, model simulations provide insights and predictions on timescales from one year to decades, through initiatives such as the World Climate Research Programme's Coupled Model Intercomparison Project (CMIP) and Explaining and Predicting Earth System Change activities. Second, snapshots of global climate are tracked through various observed and model-based indicators. Third, climate reanalyses such as ERA5 (see go.nature.com/3hxepxf) and MERRA-2 (see go.nature.com/41x7k24), which assimilate historical observations into models, are widely used for monitoring and research to understand how the Earth system has evolved."

"Models are driven by estimates of natural and human influences on Earth's climate system - known as forcings data - that are based on analyses of a wide range of observations, covering temperatures, greenhouse gases and more. Almost all Earth system models use the same forcing data sets, which are formally updated every five to seven years to support successive phases of the CMIP. Thus, the inputs to drive the models are based on older or approximate data that might not reflect conditions today."

Global warming accelerated in the past decade, with the world's average surface temperature in 2024 surpassing pre-industrial levels by 1.5°C. Human activities have been largely responsible for the warming, yet the drivers of recent record temperatures remain uncertain, including the relative roles of human forcings and natural variability. Climate research relies on three main information sources: model simulations (CMIP and related initiatives), observed and model-based climate indicators, and reanalyses such as ERA5 and MERRA-2 that assimilate historical observations. Most Earth system models use common forcing datasets updated only every five to seven years, making model inputs potentially outdated. More frequent, ideally annual, updates to forcing datasets are urgently needed to ensure models reflect current conditions and to enable rapid, accurate attribution and prediction.

Read at Nature