A Kalman Filtering approach to reduce uncertainty on global and regional climate change

We describe a new statistical method to narrow uncertainty on estimates of past and climate change. Our approach can be viewed as an adaptation of Kalman Filtering, or Kriging, for Climate Change. The definition of what we call "signal" and "noise" are different from those used in typical weather forecasting systems, but the formalism is pretty similar, and estimation of the "model error" and "observational error" covariance matrices play a central role.

This approach allows us to simultaneously constrain projections, metrics of sensitivity, and to assess human influence on the past climate (attribution). It provides a consistent picture of on-going changes, through merging model simulations and observations in a Bayesian fashion. Cross-validation suggests that our method produces robust results and is not overconfident.

Beyond GSAT results, I will focus on application of this method to narrow uncertainty on regional or local scale warming -- which is a step forward from the AR6. Even at the local scale, we find that observational constraints narrow uncertainty on future warming, and that local observations provide useful information. The case of France will be used as an illustrative example, then I'll describe local results worldwide and show how they constrain warming patterns. I will briefly browse other applications, including some related to the water cycle, and discuss implications of this work.