the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Sep 2026
Climate model ensembles reveal increasing future extreme precipitation in Northern Italy
Abstract. Assessing future changes in extreme precipitation remains a critical challenge for climate adaptation and flood risk mitigation. While growing observational evidence indicates an intensification of short-duration extreme precipitation, robust quantification of these changes and how it will change in the future remains uncertain. Using one of the longest continuous daily precipitation records available, dating back to 1 January 1813 in Bologna, we evaluate the ability of 22 bias-corrected CMIP6 climate models to reproduce historical extreme precipitation characteristics based on four standardized extreme precipitation indices. Building on this evaluation, we develop a comprehensive assessment framework that combines bias correction, independent validation and multi-model ensemble (MME) schemes. We present a comparison of Bayesian Model Averaging (BMA) with Simple Model Averaging (SMA) and individual model simulations, and apply both MME schemes to project 21st-century changes in extreme precipitation under different emission scenarios. The results show that MME schemes provide a more stable and robust representation of extreme precipitation than most individual models. Among the projected changes, the intensity indices show robust increasing signals across most scenarios and periods, with more pronounced enhancement under high-emission conditions in the far future. By contrast, the frequency-related indices show more complex responses: the annual count of days where precipitation is greater than or equal to 20mm and 10mm (R20mm and R10mm) exhibits a weak increasing tendency and no sustained increasing signal, respectively, with R10mm projected to decline under SSP5-8.5. This study provides a robust framework for assessing future changes in station-scale extreme precipitation and for better characterizing the associated projection uncertainties.
- Preprint
(2045 KB) - Metadata XML
-
Supplement
(1116 KB) - BibTeX
- EndNote
Status: open (until 03 Nov 2026)
-
RC1: 'Referee Comment on egusphere-2026-5518', Anonymous Referee #1, 15 Oct 2026
reply
-
AC1: 'Reply on RC1', Yue Lai, 02 Nov 2026
reply
We thank the Reviewer for providing a constructive and rigorous review, which is very helpful for improving our manuscript. We have replied to each of the comments point by point, and our detailed responses are provided in the attached file.
-
AC1: 'Reply on RC1', Yue Lai, 02 Nov 2026
reply
-
RC2: 'Comment on egusphere-2026-5518', Anonymous Referee #2, 16 Oct 2026
reply
This manuscript presents a clear and well-structured assessment of future changes in extreme precipitation based on 22 CMIP6 climate models, Quantile Delta Mapping, and two multi-model ensemble approaches. The use of the exceptionally long Bologna precipitation record is a notable strength, as is the independent validation framework adopted to evaluate the transferability of the bias-correction procedure and ensemble weights. The comparison between Bayesian Model Averaging and Simple Model Averaging is informative, and the distinction between changes in precipitation intensity and frequency is clearly presented. The figures and tables effectively communicate the results, while the discussion appropriately acknowledges the uncertainty associated with local climate projections.
Overall, the study provides a valuable contribution to the assessment of future precipitation extremes and offers a useful methodological framework for combining bias correction, independent validation, and multi-model ensembles. The conclusions are generally supported by the analyses, and I recommend publication after two minor editorial revisions.
First, the title currently refers broadly to “Northern Italy,” whereas the analysis is based on a single observational series from Bologna. The geographical scope of the title should therefore be narrowed to avoid implying regional representativeness. Moreover, because all the indices are derived from daily precipitation, this temporal scale could usefully be made explicit. A possible title would be:
“Climate model ensembles reveal increasing future daily precipitation extremes at Bologna, Italy.”
Second, the manuscript refers in the Abstract and Introduction to the intensification of “short-duration extreme precipitation.” In the cited literature, this expression frequently refers to hourly or sub-daily precipitation, whereas the present study exclusively analyses daily data. Daily aggregation cannot resolve the processes or extremes normally described as short-duration or sub-daily precipitation. The terminology should therefore be revised throughout so that it is consistent with the temporal resolution of the analysis. Expressions such as “daily precipitation extremes” or, where a broader statement is intended, “heavy precipitation extremes” would be more appropriate.
Subject to these minor revisions, I consider the manuscript suitable for publication.
Citation: https://doi.org/10.5194/egusphere-2026-5518-RC2 -
AC2: 'Reply on RC2', Yue Lai, 02 Nov 2026
reply
We are grateful to the Reviewer for providing a helpful and detailed review. We have replied to each of the comments point by point, and our detailed responses are provided in the attached file.
-
AC2: 'Reply on RC2', Yue Lai, 02 Nov 2026
reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 294 | 119 | 63 | 476 | 75 | 57 | 60 |
- HTML: 294
- PDF: 119
- XML: 63
- Total: 476
- Supplement: 75
- BibTeX: 57
- EndNote: 60
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
I liked the paper. The manuscript is very well written, clear and easy to follow, and presents interesting results. The possibility of “validating” bias-correction and model averaging on independent data, characterized by changes, is indeed an advantage respect to many other studies. I found the comparison between simple and Bayesian model averaging particularly valuable, especially the comparable or better performance of the simpler approach during validation. Of course it would have been even stronger to show that these results are valid for more than one raingauge site, but such long records are rare. I have some detailed comments in the following that mainly require some clarification and additional discussion. Therefore I would expect a minor revision of the paper. The comments are not ordered by relevance but according to the paper text.
Line 144: I wonder what the spatial representativeness of the GCM extreme rainfall modelling is. The values associated to one pixel are not representative of one point (i.e., the raingauge in Bologna). At line 172 the Authors state “Precipitation outputs from grid-based GCM simulations represent grid-average values at the grid scale”. A question I had while reading the first time, at this point, is: wouldn’t a spatial downscaling or the use of higher resolution models be necessary? I guess bias correction against a station can itself act as an implicit statistical downscaling.
Table 2: horizontal lines separating the definitions would be helpful here.
Line 178: bias correction is used to resolve the spatial representativeness issue (plus other biases). GCM simulations should be representative of large spatial-scale precipitation “changes”. Can bias-correction be problematic if leading to assuming that the same changes occur at the point-scale? In other words, do GCM projections represent changes in local events? Transferring, for example, a 20% increase in a grid-scale precipitation quantile to the corresponding station quantile introduces an assumption about how changes transfer between scales. Changes in storm intensity, spatial extent, duration, and occurrence can affect point and area-average extremes differently. This would be a problem, I think, for convective events (e.g., summer thunderstorms) but probably this is not so relevant since the paper is about daily precipitation. I would however suggest the Authors to add a comment on this (maybe also referring to Cannon et al., 2015, whom the Authors already cite). Is the validation on a period already characterized by Climate Change the way the Authors assess the capability of bias-correction to capture the changes as well as the statistics of the signal?
Equation (5): maybe a different letter could be used for the model outputs since p is already used as quantile level in previous equations and as pdf in the following ones.
Equation (9): just a clarification for someone who has not used BMA before. The equation suggests that a maximum likelihood procedure is used here. Where are the prior probabilities, mentioned at line 213, that would make the estimation Bayesian?
Section 3.5.2: before looking at the results I asked myself: isn't it trivial that QDM beats raw estimation? The Authors correctly split the calibration and validation periods. But if the spatial representativeness of GCMs is adjusted by the bias-correction, I guess leading to a displaced and more variable signal, I would expect that it is a necessary step to reproduce a point-scale signal. Shouldn't be more interesting to compare several QDM methods? Since the Authors have an improved quantile mapping, why not testing a classical one against it? After reading further, I am indeed surprised that sometimes the raw precipitation index beats the bias-corrected one (e.g., in Figure 4b).
Besides, wouldn’t be interesting to assess how the bias-corrected simulations represent changes/trends in the validation period?
Line 292: I would suggest to start with the relative changes (and their significances) and then talk about the trends. The trends in individual periods are perhaps less relevant than the differences between the periods. Maybe a comment on what the two types of “change” are, and why both should be analyzed, could be added here for clarity.
Line 337: I would suggest to add a comment, even speculative, on why QDM is not improving RMSE and MAE for R99p. Is it because of individual years with large discrepancies? What is instead the improvement in the calibration period, for which I would expect an always positive improvement rate (but maybe I am wrong).
Section 4.2: it is indeed very interesting that SMA performs better (surely not worst) than BMA in the validation period. I would have not expected that. Although based on one example only, I think it is a relevant outcome of the paper. I would add this also in the abstract of the paper.
Section 4.3.1 vs. 4.3.2: as stated before, I would start with the relative changes and then the trends. It is just a suggestion.
Line 386: please define RI in the text. I’ve found it in the note to Table 5 but it took me a while.
Line 390: The text rightly states that “Multi-model ensemble averaging suppresses the contribution of internal climate variability”. Is this the reason why trends are more frequently significant for the model averaging than for the observations and, maybe (not shown in the paper) for the individual GCM models?
Table 6: maybe recall what NF, MF, and FF are in the caption.