The manuscript focus is a subject of major importance.  The approach using a land surface model to derive a drought index at a very large scale with an 8-km resolution is quite ambitious. Connecting meteorological forcing based on 22-year historical and 59-year projected climate simulations (with a covering period of 16 years), under two different warming scenarios (RCP 4.5 and 8.5), makes the contribution interesting. A comparison of drought levels with available insurance claim data in France and a particular focus on two regions with contrasting climates is also interesting. The paper is well structured and well written (except for a few minor typing errors as annotated in the pdf file).

The results are interesting although a discrepancy is reported between the historical and projected simulations. The main shortcoming is that the approach takes into account the soil's hydrodynamic properties, the vegetation, and interactions with the atmosphere, but completely ignores the building structure which is of major importance since the insurance claims are relevant to damage to the constructions and the damage level is known to be directly governed by the soil-structure interactions. This point needs to be mentioned in the work limitations. 

The manuscript analyzed droughts which may lead to clay shrinkage using a statistical method based on ISBA model and predicted future droughts according to some GCMs. This work has some significance in helping people cope with drought, but overall, it lacks innovation, and its conclusions lack strong evidence. I have some specific comments as follows.

1. The introduction is overly verbose, and the main theme is unclear.
2. Has the model ISBA been calibrated？If so, please provide details on the calibration process and the calibration results.
3. Line 150. Why choose these GCM-RCM combinations? Judging from the results, there are significant differences between these models.
4. Section 3.1. The historical results lack a comparison with observations.
5. Line 215. Why only choose third quartile of YDMI as an indicator. Does it have special significance? Generally, 50^th percentile is more commonly used.
6. Figure 3-Figure6. From the figure, the trend of the different model are inconsistent. What has caused the significant differences between the different models? Can we draw meaningful conclusions from results with such significant results?