River Flows Are a Reliable Index of Forest Fire Risk in the Temperate Tasmanian Wilderness World Heritage Area, Australia

Round 1

Reviewer 1 Report

This paper discussed the use of 60-year record of daily flows from two rivers (Franklin and Davey) in the remote Tasmanian Wilderness World Heritage Area to characterize seasonal patterns in fire risk in temperate Eucalyptus and rainforests. The topic is interesting and worthy of study. The theoretical analysis is well conducted and the overall idea of the article is clear. However, the paper still has some weaknesses that should be considered.

• As mentioned in the introduction section, there exists other ways for forest fire risk assessment. The motivations or advantages of using the river flow data should be given.
• In section 2.2, line 117-118. It will be more intuitive to give the figures of these two logistic regression models.
• In section 2.2, line 120-124. As an implementation step, flow charts are often clearer than words. Please make a figure for a more clear explanation.
• Figure 2 is hard to understand. Please show the linear regression graph first regarding the statement of “There is a strong relationship between river flows and soil moisture ... Linear regressions of … showed excellent predictive performance.”
• The second subgraph of figure 4 should be named “b)”.
• The last sentence of the article, line 259-261. If the approach of using river level proposed in this paper can be transferable to other high rainfall environments in the world, it will be very encouraging. Hence, the author is encouraged to collect some relevant data to further validation of this approach.
• The author should give a more concise conclusion and add future research prospects at the end of the article.
• Please double-check the paper to avoid any typos, e.g., Line 25 “1964 fire fires”.

Author Response

Reviewer No. 1

This paper discussed the use of 60-year record of daily flows from two rivers (Franklin and Davey) in the remote Tasmanian Wilderness World Heritage Area to characterize seasonal patterns in fire risk in temperate Eucalyptus and rainforests. The topic is interesting and worthy of study. The theoretical analysis is well conducted, and the overall idea of the article is clear. However, the paper still has some weaknesses that should be considered.

As mentioned in the introduction section, there exists other ways for forest fire risk assessment. The motivations or advantages of using the river flow data should be given.

We have added the following statement:

‘Given the sparse meteorological data our approach provides a novel approach to identify historical trends in fire risk thereby filling an important knowledge gap for this region.’ 

In section 2.2, line 117-118. It will be more intuitive to give the figures of these two logistic regression models.

We have added a reference to Figure 4 in this section.

In section 2.2, line 120-124. As an implementation step, flow charts are often clearer than words. Please make a figure for a clearer explanation.

We have added a new figure with a flow chart detailing the determination of the three flow thresholds used in the analysis, as figure 2 (note the additional Figure has led to reordering the existing figures).

Figure 2 is hard to understand. Please show the linear regression graph first regarding the statement of “There is a strong relationship between river flows and soil moisture ... Linear regressions of … showed excellent predictive performance.”

We have added additional linear regression plots to Figure 3 showing actual river flows versus modelled soil moisture for each river as separate panels.

The second subgraph of figure 4 should be named “b)”.

Thank you, we have fixed this error, and this is now figure 5.

The last sentence of the article, line 259-261. If the approach of using river level proposed in this paper can be transferable to other high rainfall environments in the world, it will be very encouraging. Hence, the author is encouraged to collect some relevant data to further validation of this approach.

Thank you for the suggestion but this is not practical within the scope of this paper, hopefully our paper will encourage others to explore this novel approach. 

The author should give a more concise conclusion and add future research prospects at the end of the article.

We have added additional text to the conclusion encouraging further research using this approach.

Please double-check the paper to avoid any typos, e.g., Line 25 “1964 fire fires”.

Corrected, and we have made a number of other minor changes to improve clarity of the manuscript.

Reviewer 2 Report

This manuscript introduces a new concept in that can be used in prediction of wildfire spread especially for locations where meteorological coverage is sparse. The manuscript is well structured and easy to follow. Before I vote for publication, I have some comments/questions for the authors.

• “River flow” is not clearly defined. What do you measure here? Is it flow rate per unit of time? Per some geographical location? etc. I suggest clearly define this and the averaging method that was employed in the introduction.  
• Fire spread is not clearly defined. The abstract starts with “Fire risk can be defined as the probability that a fire will spread”. The probability of fire spread is 100% in all fire if you look at the micro level. How do you define fire spread? in what level? 
• There are some studies that explores the correlations between river basin variations with wildfires. I suggest including them in your literature review section.
• Vapor pressure deficit is a good predictor of the burned area (at least in the western US). Burned area is a good proxy of fire spread. I suggest the authors discuss this in the introduction.
• I am wondering if you can compare your approach with vapor pressure deficit in Tasmania and discuss similarities/differences.
• Have you considered using your method in non-forested area and see the feasibility of your introduced technique?
• I am wondering how the soil moisture is measured and compared to your model in figure 2. Is it an average over several locations at different distances from the river? I think the moisture level heavily depends on the distance from water. My observation in figure 2 doesn’t show an “excellent proxy” for the whole range of ML.
• What is the shaded area in Figure 2? Is it standard error as standard deviation divided by square root of observation frequencies? If so, why does it change at different ML?
• I am interested to see the results for figure 3 when the predicted and actual ML are not close in Figure 2. I other words, any point that is far from the intersection of red and blue lines in figure 2.
• I am wondering if you can add some discussion about the rate of fire spread and its correlation with river flows.

Author Response

Reviewer No. 2

This manuscript introduces a new concept in that can be used in prediction of wildfire spread especially for locations where meteorological coverage is sparse. The manuscript is well structured and easy to follow. Before I vote for publication, I have some comments/questions for the authors.

Thank you for your feedback, but we note that this reviewer has a misapprehension that our response variable is “fire spread”, when in fact it is area burnt.

 “River flow” is not clearly defined. What do you measure here? Is it flow rate per unit of time? Per some geographical location? etc. I suggest clearly define this and the averaging method that was employed in the introduction.  

Thank you, we have clarified the units and measurement in the abstract and in Materials and Methods sections.

• Fire spread is not clearly defined. The abstract starts with “Fire risk can be defined as the probability that a fire will spread”. The probability of fire spread is 100% in all fire if you look at the micro level. How do you define fire spread? in what level? 

We have added text to the Abstract and Introduction and Methods clarifying that fire risk and fire spread relate to area burnt, which is the actual variable we are using in this analysis.

• There are some studies that explores the correlations between river basin variations with wildfires. I suggest including them in your literature review section.

We have undertaken literature search and cannot locate any studies that are similar to ours.  We did find one study that looked at how fire affects river discharges in Siberia, which we added it to our manuscript. We would be grateful if the reviewer can provide some specific citations which would strengthen our study.

• Vapor pressure deficit is a good predictor of the burned area (at least in the western US). Burned area is a good proxy of fire spread. I suggest the authors discuss this in the introduction.

We feel this suggestion if off point.  The focus of our study is area burned as a proxy of fire risk and the relationship with the river record.  We very strongly doubt that river flows could be used to predict vapour pressure deficient given these variables operate at strongly contrasting time scales. 

• I am wondering if you can compare your approach with vapor pressure deficit in Tasmania and discuss similarities/differences.

Again, we feel this suggestion if off point regarding our article.  We acknowledge that more detailed analysis of meteorological variables and fire risk and rate of spread is warranted in Tasmania.

• Have you considered using your method in non-forested area and see the feasibility of your introduced technique?

We focus on forested vegetation because the treeless vegetation is very flammable and becomes available to burn after very short dry spells and is combustible even when the organic soils are saturated (its less sensitive to ecohydrological drought).  We have added this clarification in the text:

‘We focus on forest vegetation because the treeless vegetation in this region is highlight flammable and is available to burn after brief dry spells when organic soils are saturated [19].’

• I am wondering how the soil moisture is measured and compared to your model in figure 2. Is it an average over several locations at different distances from the river? I think the moisture level heavily depends on the distance from water. My observation in figure 2 doesn’t show an “excellent proxy” for the whole range of ML.

The soil moisture was taken as an average across the entirety of our geographic catchment units.  While it is true on a micro-scale that soil moisture is higher in topographic positions close to rivers, we sought to use a simple, catchment-wide measure for this analysis reflecting the occurrence of large fires across the catchments.  Importantly, river flow scales well to soil moisture across the lower ends of the range of river flows, during dry conditions when fire risk is expected to be high and that are of interest to us.   We have added the following text to address this:

‘Such under-prediction of high-flows is not important for this study because under these fires are unlikely to occur.’

We have also added the caption to Figure 4

‘Note the red and black lines overlap in panels a and c. Fires only occur at river levels where there is a close relationship between river flow levels and soil moisture (Figure 3)’.

• What is the shaded area in Figure 2? Is it standard error as standard deviation divided by square root of observation frequencies? If so, why does it change at different ML?

Shaded areas indicated the model 95% confidence intervals – intervals are wider at the ends of the graph because they represent the uncertainty in the line’s slope.

In the caption to Figure 3 we have added:

‘Shaded areas represent 95% model confidence intervals.’

• I am interested to see the results for figure 3 when the predicted and actual ML are not close in Figure 2. I other words, any point that is far from the intersection of red and blue lines in figure 2.

As addressed above, the river flow vs. soil moisture relationship deviates most at the upper end of flow and soil moisture, e.g. Above 7000 ML/day, when fire occurrence is extremely rare or non-existent (figure 4c,d).  We have added text to the caption of Figure 4 clarifying this. 

• I am wondering if you can add some discussion about the rate of fire spread and its correlation with river flows.

Thank you but this is off point, the analysis is focused on area burnt as an expression of fire risk.

Reviewer 3 Report

Review of “River flows are a reliable index of forest fire risk in the temperate Tasmanian Wilderness World Heritage Area, Australia” by Bowman and Williamson.

This manuscript examined the use of river flow as an indicator of fire risk in the Tasmanian Wilderness World Heritage Site. River flow was correlated to a modeled soil moisture product, and the longer river flow records were then used to evaluate fire risk going back to the 1950s and 1960s.

This is an interesting manuscript and overall, it was executed well. The two selected rivers showed a potential to predict fire risk in the surrounding watersheds. The paper is narrowly focused on fire risk at the Tasmanian site and not on the methodology itself. As a consequence, although the paper introduces a new methodology using river flow as a predictor, it provides limited insight into how hydrology may affect the methodology. The last line of the manuscript encourages other researchers to extend the methodology that is used in the study. I believe the paper would be improved by providing guidance/ suggestions about when the methodology would be appropriate, as well as highlighting uncertainties in the approach. River flow is an integrated metric with built-in lags, especially for low flows. Consequently, this method will likely work best when there is a direct correlation between soil moisture in the rooting zone and river flow. I suspect that the approach would be more challenging to implement in watersheds with large groundwater flows, in large watersheds that integrate large areas, and in Mediterranean-climate regions, where highest fire risk is associated with long periods of extended low flows. Further, changes in post-fire vegetation may alter river flow magnitude and thus, alter the relation between river flow and soil moisture in the watershed. For small fires or low severity fires, this latter issue may not be significant, but potentially adds uncertainty to the methodology following large severity fires. I think adding some hydrologic context regarding the methodology in the Discussion would round out the paper.

I don’t have any other major concerns and I expect that the manuscript will be suitable for publication. Below are a few additional comments and suggestions.

L130: Is the low-flow threshold a probability of a large fire occurring in a given year, or across the record?

L201: The analysis underlying the results in Figure 6 seems circular to me. The median threshold is defined as the river flow where half of large fires are below and half are above. This is calculated across the entire time-series, but mostly from fires that occur during the summer. Figure 6 is then generated by looking at the number of summer months when flow is below the median. Both plots are bell shaped, which is what I would expect based on how the median threshold was defined. I see the use of the median threshold when looking at the time-series changes in fire risk, but not Figure 6.

Figure 1: Two things. First, the labels (particularly Franklin) cover up the river. Second, it is extremely difficult for me to see the red river on top of the light green forest background. I recommend a more color-blind friendly river color.

Figure 3a & b: Was the linear regression done piecewise? The regression line looks broken at multiple points. Please clarify.

Figure 3: Caption should indicate mean monthly flows. It looks like you are plotting daily flows.

Figure 3, 4, 5: Fonts are tiny. The font sizes may be improved if the figures are resized for publication, but I’d still make them larger for the reader.

Author Response

This manuscript examined the use of river flow as an indicator of fire risk in the Tasmanian Wilderness World Heritage Site. River flow was correlated to a modeled soil moisture product, and the longer river flow records were then used to evaluate fire risk going back to the 1950s and 1960s.

This is an interesting manuscript and overall, it was executed well. The two selected rivers showed a potential to predict fire risk in the surrounding watersheds. The paper is narrowly focused on fire risk at the Tasmanian site and not on the methodology itself. As a consequence, although the paper introduces a new methodology using river flow as a predictor, it provides limited insight into how hydrology may affect the methodology. The last line of the manuscript encourages other researchers to extend the methodology that is used in the study. I believe the paper would be improved by providing guidance/ suggestions about when the methodology would be appropriate, as well as highlighting uncertainties in the approach. River flow is an integrated metric with built-in lags, especially for low flows. Consequently, this method will likely work best when there is a direct correlation between soil moisture in the rooting zone and river flow. I suspect that the approach would be more challenging to implement in watersheds with large groundwater flows, in large watersheds that integrate large areas, and in Mediterranean-climate regions, where highest fire risk is associated with long periods of extended low flows. Further, changes in post-fire vegetation may alter river flow magnitude and thus, alter the relation between river flow and soil moisture in the watershed. For small fires or low severity fires, this latter issue may not be significant, but potentially adds uncertainty to the methodology following large severity fires. I think adding some hydrologic context regarding the methodology in the Discussion would round out the paper.

I don’t have any other major concerns and I expect that the manuscript will be suitable for publication. Below are a few additional comments and suggestions.

Response

Thank you for these very helpful suggestions.  We have captured in the following additional text in the Discussion which we have slightly reworked.

‘We know of no prior studies that have used river flows to assess forest fire risk [36], although note that one study has shown a relationship between extensive fires in river catchments in Siberia and reduced late summer/autumn river flows [37]. Serious consideration should be given to monitoring flows of other undammed rivers in the region to provide managers with a simple estimate of forest fire risk and encourage other researchers to explore the relationship between historical river flow records and landscape fire activity. We suspect this approach of using river levels may be transferable to other high rainfall environments globally, which are likely to be increasingly impacted by forest fires due to climate change [36,38,39]. An important constraint with using river flows is it is likely most effective where soil moisture and river flows are tightly coupled (Figure 3). Our approach is likely unsuitable for rivers where this coupling between flows and soil moisture is weaker, such as watersheds with large groundwater flows or where the catchment area is large and there are pronounced climatic gradients, in regions with prolonged dry seasons conducive for landscape fire (such as Mediterranean and monsoonal climates) when river flows are consistently low. Additionally, our approach provides a baseline for subsequent research to discover how the size and severity of fires affects river flows.’

L130: Is the low-flow threshold a probability of a large fire occurring in a given year, or across the record?

Response

That threshold is based on the probability of a fire occurring in a given month. Text has been added to the methods to clarify this.

L201: The analysis underlying the results in Figure 6 seems circular to me. The median threshold is defined as the river flow where half of large fires are below and half are above. This is calculated across the entire time-series, but mostly from fires that occur during the summer. Figure 6 is then generated by looking at the number of summer months when flow is below the median. Both plots are bell shaped, which is what I would expect based on how the median threshold was defined. I see the use of the median threshold when looking at the time-series changes in fire risk, but not Figure 6.

Response

Thank you for the comment; we do not believe the inclusion of figure 6 to be circular, its intent is simply to graphically summarise how often the summer specifically experiences periods with flows below a threshold when there is a median probability of forest fire in the surround ecohydrological domain.  The therefore shows the likely months when forest fires occur in this region.  We have updated the caption of figure 6 to clarify this.

The caption now reads:

Figure 6. Characterization of forest fire season based on river flows. Frequency of the number of summer (November-March) months in the historical river gauge records fire for the (a) Franklin River and (b) Davey River when flows were below the median probability of forest fires occurrence in the respective surrounding ecohydrological domains identified by this analysis.

Figure 1: Two things. First, the labels (particularly Franklin) cover up the river. Second, it is extremely difficult for me to see the red river on top of the light green forest background. I recommend a more color-blind friendly river color.

Response

We thank the review for their suggestion; we have moved the labels to the left, replaced the red river lines with bolder pink lines, and made the green a paler shade.

Figure 3a & b: Was the linear regression done piecewise? The regression line looks broken at multiple points. Please clarify.

Response

Apologies, the increment our graphing software used to render the line was set to 5% intervals so the line looked broken, we have produced the graph again with a 1% intervals which reflects the shape more accurately.

Figure 3: Caption should indicate mean monthly flows. It looks like you are plotting daily flows.

Response

Thank you, a correction has been made to the caption to clarify that the points represent months.

Figure 3, 4, 5: Fonts are tiny. The font sizes may be improved if the figures are resized for publication, but I’d still make them larger for the reader.

We have increased the font size for figures 3 and 5, but have not done so for figure 4 as the complexity of these graphs would make that difficult – however, this figure will be larger in publication than in the provided document.

Round 2

Reviewer 1 Report

The paper has been revised properly and can be accepted as it is.

Author Response

Thanks you.