|Dear authors, please find below my main recommendations concerning your revised manuscript.|
One of the co-authors (Peter Höller) has recently published an in-depth analysis of Austrian avalanche fatalities, with a particular emphasis on tourist avalanches (Höller, 2017). Höllers study covers essentially the same two categories backcountry and off-piste, and a very similar time period. Currently, Höller’s study is not cited. From my perspective, the following results and statements made by Höller should be discussed, as they partly contradict the results and/or may influence the interpretations presented:
– frequency and influence of multi-fatality accidents - 32 accidents with more than 3 victims in the backcountry since 1981/82
– «Although in Austria a slight increase of fatalities in the off-piste area can be seen, this tendency is statistically not significant (p = 0.055).»(p. 6)
– «A trend towards more avalanche fatalities due to off- piste skiing cannot be identified at the moment. »(p. 7)
As already pointed out in one of the initial reviews, the influence of multi-fatality accidents on the absolute annual number of fatalities and on the trend function should be discussed. This may be particularly relevant, as the dataset is split into two categories, with low counts for the off-piste category. Although (Höller, 2017) noted no trend in the number of accidents with many fatalities, single events claiming many lives occurred repeatedly (for instance a backcountry accident in 1982 claimed 12 lives). Therefore, I suggest to additionally explore the trend for the period 1980/81 until 2015/16 using just the counts of fatal accidents in Austria (in this case, multi-fatality accidents have no influence on annual fatality numbers). Additionally, the authors could show trend curves and statistics combining the annual backcountry fatalities for the four Alpine countries Austria, France, Italy, Switzerland. From my perspective, these two approaches would considerably strengthen the analysis, allow a more robust interpretation of the results (particularly if trends are confirmed using the different datasets), and would allow a more in-depth-discussion of the advantages and limitations of the statistical model.
Furthermore, Höller (2017) provides an extensive overview of publications concerning trends and statistics of avalanche fatalities (Switzerland, USA, Italy, France), some of which might be suitable references when discussing trends.
Data and Methods
As you analyze fatality counts only, you could remove line 83 (number of persons involved).
Results - Section 3.1 and 3.2
These two sections introduce the figures and tables, but not the results themselves (this was already pointed out in one of the first reviews). Results are presented and discussed in Section 4 (Discussion) only.
I find this a rather unusual approach. I recommend to introduce the figures together with the results they show.
Discussion - Section 4.1
lines 214 to 219: You discuss the increased number of fatalities in the 1980’s and mention increased snowfall (in Austria, I assume) but cite two sources which explored Swiss snowfall trends (Abegg, 1998; Laternser and Schneebeli, 2003). Furthermore, you note no peak of off- piste fatalities in the 1980’s. As already outlined in one of the initial reviews, these two statements seem contradictory.
lines 264-267: As you point out, AIC/BIC statistics indicate that the curves are non-linear in most cases. Maybe you could formulate more clearly whether this result confirms or contradicts the cited study by Techel et al. (2016) (this did not become clear to me reading these lines).
Discussion - Section 4.2
I could not find any information whether, and to what extent, the spatial simulation matches the actual observed number of fatalities (the maps show this, but they are hard to interpret - printed numbers vs. background colors). Please show the correlation between observations and simulations, or some other measure of similarity. Similarly, you could present the statistical correlation between the fatality numbers at municipal level with the proportion of Alpine terrain or overnight tourists. This would facilitate the interpretation for the reader.
From my perspective, you should also discuss the benefits and limits of the chosen Markov Random Fields method used for the spatial analysis.
Lines 209 - 215, Figure 1
I personally would interpret Figure 1 as a considerable increase between 1969/70 and 1985/86 (probably significant), with only minor changes afterwards (slight increase until 2005, but 90% confidence interval is large in comparison). This interpretation would correspond quite closely to Höller’s results and conclusions, and would also agree quite closely with the trends shown for Switzerland (study you already cite (Techel et al., 2016)). I suggest to show whether fatality counts in the years surrounding 1985 and 2005 differ significantly.
Figures - 7, 8, 9
The color of the individual polygons in Figures 7 and 8 (simulated number of fatalities) show different information than in Figure 9 (Proportion Alpine terrain). This is not fully intuitive. Therefore, I suggest using different color schemes.
The figure captions (Fig. 7 and 8) miss the information what the background color shows and what the numbers are. This information, currently in the text section 3.2, should be added (or moved) to the caption.
Please indicate the clusters CL1 and CL2 on the maps in Fig. 7 and Fig. 8 (you refer to them in the text, but non- Austrians will likely not know which location is which).