the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Jul 2026
Spatial Modelling of Ice-Jam Susceptibility in Southern Quebec: An Enhanced Ice Jam Predisposition Index
Abstract. Ice jams represent a recurrent natural hazard in Quebec, where geomorphological and structural river characteristics play a key role in determining their spatial distribution. This study improves and evaluates the Ice Jam Predisposition Index (IJPI) using a dataset of 35 rivers across southern Quebec. The enhanced model integrates geomorphological parameters, including channel narrowing, sinuosity, and slope, along with structural (bridge) and hydrological (tributary) features to delineate reaches prone to ice-jam formation. Including rapids as an inverse indicator helps distinguish high-energy segments where jams are unlikely to occur. Model validation shows that over 90 % of observed ice jams occurred in reaches with medium to high predisposition, with an overall recall of 0.90, indicating promising performance as a screening tool. The results highlight the dominant influence of flat slopes, channel confinement, and curvature on jam formation, suggesting that the model can serve as a useful spatial tool for regional ice-jam risk management. These results demonstrate the potential of the proposed framework to support operational flood risk management by allowing proactive identification of vulnerable river sections. Decision-makers and emergency management agencies can use the model to prioritise monitoring efforts and improve preparation for ice-jam-related hazards.
- Preprint
(1571 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 26 Sep 2026)
-
RC1: 'Comment on egusphere-2026-2457', Anonymous Referee #1, 24 Aug 2026
reply
-
AC1: 'Reply on RC1', Elnaz Mirzaei, 26 Aug 2026
reply
We appreciate your careful reading of our manuscript and your constructive comments. Please find our responses to each comment below.
Line 36: The citation to Beltaos and Prowse (2009) refers to the general statement about the economic costs of ice-jam floods and the lack of a comprehensive economic assessment, rather than specifically to the 2020 Fort McMurray flood. We agree that the current wording may cause confusion and will clarify this in the revised manuscript.Line 50: The intention was to highlight the remaining challenges in reliably predicting ice-jam flooding, rather than to suggest that little work has been done on river ice break-up forecasting. We recognize that the original wording did not make this distinction clear, especially given the existing work using neural networks, fuzzy logic, and other approaches. We will revise the sentence to clarify this point.
Line 61: The two terms refer to related but distinct prediction problems, and the wording in this section does not make that distinction clear. We will revise this part to distinguish more clearly between predicting river ice break-up and forecasting ice-jam occurrence.
Line 231: The presence of a tributary does not represent a physical reduction in the width of the main channel. Rather, following De Munck et al. (2017), it is treated as an equivalent narrowing to represent the potential obstruction to ice transport caused by ice entering the main channel at a confluence. In this approach, the reduction is based on the tributary width at its outlet, giving greater influence to larger tributaries. We will clarify this point in the revised manuscript.
Line 232: The bridge and tributary coefficients were adopted from De Munck et al. (2017) and were not independently calibrated in this study. We agree that their magnitude may influence the final IJPI and that a sensitivity analysis would be useful to quantify this effect. This will be acknowledged as a limitation of the current approach and considered in future work.
Line 243: We checked the NI values across all reaches of the 35 rivers and confirmed that NI did not reach zero or take a negative value anywhere in the dataset. Thus, Wmodified remained positive for all reaches considered in this study. The current implementation does not explicitly impose a lower bound on Wmodified. Although zero or negative values did not occur in our dataset, we will add an explicit floor on Wmodified in the revised implementation as a safeguard.
Line 249: NI was classified to use the same ordinal scale as SI and slope in the IJPI calculation and to remain consistent with the original approach of De Munck et al. (2017). We acknowledge that this classification limits the detail available for sensitivity analysis. Using continuous values could be considered in future work.
Line 286: SI was classified for the same reason as NI, to place it on a common ordinal scale for aggregation in the IJPI. We recognize that this classification limits sensitivity analysis using continuous SI values, which could be explored in future work.
Line 301: Slope was calculated from the DEM elevations at the upstream and downstream endpoints of each reach, as described in Eq. (5). Examination of several cross-sections of the DEM showed that the elevation surface is nearly flat across the river channel, suggesting that it represents the water surface rather than the underlying riverbed. Therefore, we cannot confirm that steady-state flow conditions prevailed at the time of LiDAR acquisition. We will clarify in the revised manuscript that the calculated slope should be considered a proxy for the longitudinal channel gradient rather than a direct measurement of the riverbed slope.
Line 305: The absolute value is used only to calculate the magnitude of the local slope. Flow direction is preserved by ordering the reaches from upstream to downstream. A downstream decrease in slope is treated separately in the slope-break calculation, where only a reduction in slope relative to the upstream reach is retained; an increase in slope is assigned a value of zero.
Line 318: Both NI and SI are ultimately converted to the same ordinal scale {0, 1, 2, 3}, so the different spacing of the original class boundaries does not directly assign different weights to the two indices. However, the choice of thresholds can influence how reaches are distributed among the classes and may therefore indirectly affect the final IJPI.
Line 477: Recall will be defined more explicitly as the proportion of observed ice-jam locations correctly identified by the model, consistent with the level of detail provided for Precision.
Line 513: The typo will be corrected to “predictions remained”.
Line 595: Of the 432 jam-intersecting segments, 61 (14.1%) exhibited both Narrowing ≥ 2 and Sinuosity ≥ 2. Regarding their relative influence, the frequency of occurrence within jam-intersecting segments alone does not establish which factor has greater discriminative strength, as this also requires comparison with their occurrence in non-jam reaches. The current weights (Section 3.4) reflect the observed spatial correspondence between predictor classes and documented ice-jam locations.
We hope that our responses adequately address the points raised.
Sincerely, The AuthorsCitation: https://doi.org/10.5194/egusphere-2026-2457-AC1
-
AC1: 'Reply on RC1', Elnaz Mirzaei, 26 Aug 2026
reply
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 121 | 32 | 13 | 166 | 10 | 10 |
- HTML: 121
- PDF: 32
- XML: 13
- Total: 166
- BibTeX: 10
- EndNote: 10
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This is a well written paper and should be of wider interest to the journal’s readership. I recommend publication after the following comments have been addressed:
Line 36: "yet to be conducted"? On what? Fort McMurray? That flood was in 2020 but your citation is from 2009.
Line 50: “limited progress has been made in predicting floods associated with river ice break-up” – I don’t agree with this. There has been a lot of work forecast breakup using neural networks, fuzzy logic, etc.
Line 61: “ice-jam forecasting” or predicting breakup? In line 50 you focused on breakup predictions.
Line 231: “tributary-related reduction” – why is the main channel width reduced with the presence of a tributary?
Line 232: “bridge and tributary coefficients” – it would be interesting to see how sensitive these coefficients are to the outcome; if the values are too high, perhaps they overwhelm the end result, potentially setting too much of a bias towards bridges and tributaries?
Line 243: “0 < NI ≤ 1” – is a value of NI = 0 even possible?
Line 249: I don’t understand why you need to categorise NI into classes (1, 2 and 3) – makes it harder to carry out a sensitivity analysis; why not just use the NI value?
Line 286: {0, 1, 2, 3} – also for sinuosity, why categorise SI into classes and not use the actual sinuosity value instead; this hinders any sensitivity analysis to be made later.
Line 301: you’re assuming that the slope of the water surface is the same as the slope of the riverbed – can you justify that please? How do you know that steady state conditions prevailed during the time the surface elevations were acquired?
Line 305: |𝑍up,𝑖 −𝑍down,𝑖 | - why an absolute value? Is the effect of a slope increase in the flow direction the same as when there is a sudden flattening out of the slope?
Line 318: The min-max bounds of the SI class ranges are set logarithmically (max = 10 × min) whereas the NI class bounds are set linearly – won’t that incorporate a bias of how much influence each of these indices exhibit on the outcome?
Line 477: Can you please also define “recall” with more detail, like you did with “Precision” here?
Line 513: typo: “predictionsremained” should be “predictions remained”
Line 595: In the table, it would have been nice to see how much overlap there is in characteristics, i.e. how many of those counted segments had two or more factors. Were there segments that had both narrowing and low-radius meanders (high sinuosity), and what percentage? Also, which would be the stronger feature – NI or SI?