Reconstructing the diurnal cycle of temperature from daily climate simulations using three temporal downscaling techniques in a perfect model approach

Nowadays, climate simulations represent petabytes of data. About 30 institutions provided nearly 1.6 PBytes of data for the Coupled Model Intercomparison Project Phase 5 (CMIP5) and estimations for the Phase 6 (CMIP6) data volume range from 15 to 30 PBytes. Despite this large volume of climate data, climate projections data are often available at lower spatio-temporal resolution. Only few climate simulations are available today at a temporal frequency higher than a daily time step. For example, for CMIP6 about a hundred runs are available at hourly time step compared to nearly 2000 runs at a daily or monthly time step. However climate impact studies are usually conducted at finer scales, and so impact model (e.g., energy demand model ) require data at a higher spatio-temporal resolution as input. In this study, we investigate the capabilities of three temporal downscaling techniques to recover the diurnal cycle of temperature from given daily climate simulations (from CMIP6 data base) using a perfect model approach. A quantile-mapping technique, an analogue technique and a linear regression technique were calibrated using a 30-year historical simulations and applied to a 30-year “future” period (from an SSP5-8.5 scenario) using daily average, maximum and minimum temperature as input over a Euro-Mediterranean domain. Results show that overall, the linear regression performs better than the quantile-mapping and the analogue techniques. The performances depend on the geographical region and the season and can be fully explained by the differences between the climate change signal (historical vs future scenario) of daily average temperature and daily maximum/minimum temperature. Indeed, both analogue and quantile-mapping approaches assume that the change in daily maximum/minimum temperature between the historical and future period should be the same as daily average temperature. However, the diurnal cycle of temperature is not only shifted to warmer temperatures but the shape of the cycle changes under future climate scenarios. The linear regression outperforms the other two approaches over the whole domain and for all the seasons by taking into account the daily average, maximum and minimum temperature to reconstruct the diurnal cycle.