| 10 Nov 2025
Composition, frequency and magnitude of future rain-on-snow floods in Germany
Abstract. In Germany, severe trans-basin winter floods are often generated by rain-on-snow (ROS) phenomena. Under suitable conditions, when rain falls on the snow cover, the snow can melt and produce extreme amounts of runoff. In a warming climate, the frequency of ROS events is expected to change locally depending on elevation and regionally based on the general climate conditions. Consequently, the characteristics of ROS-driven winter floods are also anticipated to change. To investigate these changes, streamflow for multiple gauge stations in Germany was simulated using a deep learning model based on an ensemble of downscaled climate projections. Germany, as a representative mid-latitude country with a considerable portion of historical floods generated by ROS, offers extensive spatial and temporal coverage of hydrological observations spanning long temporal scales, and hence warranting efficient training of the deep learning model. We used explainable artificial intelligence to examine flood-generating processes, focusing primarily on ROS, for every simulated flood peak. Changes in frequency, feature importance, and magnitude of ROS flood events were assessed for individual streamflow gauges and for trans-basin floods across four major river basins in Germany. We found that with regard to the ensemble median, the frequency of ROS floods will decrease at the scale of individual gauges, as well as at the trans-basin scale for the Rhine, Elbe, and Weser River basins but increase in the Danube River basin. For all regions, the snowmelt component during ROS floods becomes less relevant, whereas the contribution of rainfall to these events increases. Interestingly, the severity of both the mean and the most extreme ROS trans-basin floods is projected to increase compared to the historical period in all major river basins in Germany, even though several individual gauges may experience a decrease in magnitude. Despite the overall agreement in the trends of the input features across climate models, the resulting trends in ROS floods are considerably disparate. This discrepancy is primarily attributed to the variations in snow dynamics in different climate models.
This study investigates how climate conditions may impact the rain-on-snow (ROS) floods in Germany using an LSTM model trained on downscaled climate projections and explainable artificial intelligence techniques. The results suggest that while ROS flood frequency will decrease in most regions, the severity of extreme ROS floods is expected to increase across all major river basins, and the role of snowmelt in these floods will become less significant. Overall, this study is comprehensive and well-designed. However, I still have several questions and suggestions for improving the current work.
1) The "magnitude" of floods is mentioned in the title. However, the absolute magnitudes of floods are represented by the relative ranks.
2) It is suggested to clearly define the “trans-basin” floods. How to distinguish whether the flood is "trans-basin" or "within-basin"?
3) Lines 5 and 82: The LSTM model may not be the state-of-art deep learning approach nowadays. It would be better to compare its performance to that of the emerging alternatives, e.g., physics-informed deep learning, and the ensemble-based approach is also beneficial given the limitations of any single deep learning model.
4) Lines 18, 471-472, and 523-524: The conclusion that the discrepancy is mainly due to different snow dynamics in climate models should be supported by more detailed arguments and evidence.
5) It would be better to make the Introduction Section more concise and to clearly state the research gaps and study objectives.
6) Figure 1 and the other maps: It is suggested to add a north arrow and a scale bar.
7) Line 98: How to define the "subsurface runoff" and the "day length"?
8) Line 108: The evaluation metric, NSE, has some inherent limitations. Also, given the sampling uncertainty and measurement errors in both temporal and spatial data, it is suggested to present the values of metrics through a statistical distribution instead of a fixed number for a single evaluation period and to apply the metric to the variable of interest like specific flood peaks. The authors can refer to the article below for more information about the limitations of some commonly used evaluation metrics in hydrology.
Reference: “Beyond a fixed number: Investigating uncertainty in popular evaluation metrics of ensemble flood modeling using bootstrapping analysis” (https://doi.org/10.1111/jfr3.12982)
9) Table 1: Could you add a reference for the criterion, "SWE>15mm"?
10) Line 230: How about the potential impacts of human activities on the other gauges? Any thoughts about how to incorporate those infrastructures into the modeling process, especially for the urban areas?
11) Figures 3, 7, 8, and 9: The horizontal axis, "Stations", was sorted by different criteria, such as the mean NSE and the median number of ROS floods. Is it possible to keep it consistent? So it is relatively easy to identify each gauge station.
12) Figure 8: Please add the denotation for the shaded areas to the figure caption.
13) Line 370: It would be better to specify the exact p-value rather than "P<0.05".
14) Line 479: Is it "GMCs" or "GCMs"?