Assessment of precipitation variability sources in a GCM for the implementation of a tuning methodology using in-situ observations

Tuning is now recognised as a key step in climate modelling, and the rise of machine learning techniques is increasing the number of targets used to tune these models. There are several reasons to focus on continental surface tuning. Firstly, a significant part of the sources of uncertainty in regional climate projections lie in the interactions between the atmosphere and the land surface. Secondly, the quality of climate change impact studies highly depends on a good representation of the climate at the surface. Finally, tuning at the surface can benefit from observational sites that provide multivariate, in situ hourly data of many meteorological, radiative and turbulent flux variables.

The objective here is to constrain the water and energy balances at the atmosphere-continental surface interface in the IPSL GCM, using as reference the in-situ observations of the SIRTA instrumented site (Paris suburb). A configuration of the coupled atmosphere (LMDZ) and continental surface (ORCHIDEE) model is set up on a zoomed grid in order to have a 30 km side mesh on the SIRTA point while keeping a reasonable computational cost. In addition, the winds (and possibly the temperature and humidity) are nudged towards the ERA5 reanalyses in order to compare the weather sequences observed at SIRTA with those of the climate model. This nudging technique allows a significant part of the internal variability of the local meteorology simulated by the GCM to be removed and to compare observations and model on a day-to-day basis. An essential step in setting up the tuning of this configuration is to assess the different sources of uncertainty involved. In this presentation, the characterisation of the uncertainties associated with the choice of configuration and the internal variability will be addressed more specifically, with a focus on clouds and precipitation.

In order to characterise the uncertainty linked to the internal variability, we compare the precipitation variability of a simulation ensemble with perturbed initial conditions with that of a perturbed physical ensemble obtained by machine-assisted exploration of the free parameters of the models. The internal variability of the precipitation simulated at SIRTA is found to be of the same order of magnitude as the parametric sensitivity, especially during convective periods, which questions the possibility of a tuning against SIRTA observations. We use a rainfall product (combining radar and rain gauges) from Météo-France in order to evaluate both the representation of spatial and temporal variability in a wider area around SIRTA and the associated uncertainty for tuning. We also present results concerning the uncertainty due to the configuration based on sensitivity tests to the grid configuration and the nudging setup. Finally, we evaluate the part of the precipitation variability due to the soil response by imposing an evaporation factor on the study area. We show how this configuration can be used in the atmosphere model tuning strategy, as it allows to get rid of the rainfall evaporation feedback.