the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Differentiating fire regimes and their biophysical drivers in Central Portugal
Abstract. The spatial and temporal patterns of wildfires and their effects over a given area can be described using the concept of fire regime. Here, we characterize fire regimes Central Portugal and investigate the degree to which the differences between regimes are influenced by a set of biophysical drivers. Using civil parishes as units of analysis, we employ three complementary parameters to describe the fire regime over a reference period of 44 years (1975–2018): cumulative percentage of parish area burned, Gini concentration index of burned area over time, and area-weighted total number of wildfires. Cluster analysis is used to aggregate parishes into groups with similar fire regime based on these parameters. A classification tree model is then used to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups. Drivers included slope, summer temperature and spring rainfall, land use/land cover (LULC) type and patch fragmentation, and net primary productivity. Results allowed to distinguish four types of fire regime and show that these can be significantly differentiated using the biophysical drivers, of which LULC, slope and spring rainfall are the most important. Among LULC classes, shrubland and herbaceous vegetation play the foremost role, followed by agriculture. Our results highlight the importance of vegetation type, availability, and rate of regeneration, as well as that of topography, in influencing fire regimes in the study area, while suggesting that these regimes should be subject to specific wildfire prevention and mitigation policies.
- Preprint
(1261 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2022-342', Anonymous Referee #1, 13 Jul 2022
The authors characterize fire regimes over Central Portugal and investigate the degree to which the differences between fire regimes are influenced by a set of biophysical drivers, namely slope, summer temperature and spring rainfall, land use/land cover (LULC) type and patch fragmentation, and net primary productivity. The authors rely on a cluster analysis followed by a classification tree model to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups for 44 yrs period.
Results allowed to distinguish four types of fire regimes and show that these can be significantly differentiated using the biophysical drivers.
The subject discussed in the present article is of high importance, and the paper is well structured and easy to follow. The introduction is very clear and includes interesting and recent references. The objectives are very well presented. Although this study draws on prior studies with similar objectives, the authors try to presented the novelty, namely the use of a large dataset regarding the study of Oliveira and Zêzere (2020) and the study from Bergonse et al. (2022) which assumed a similar fire regime all over the study area. Although the usage of a longer dataset and different fire regimes over Central Portugal might improve the knowledge robustness on that area and topic, the authors need to better address the novelty of this paper comparatively to previous works.
The limitations of the study are not identified nor discussed. This is an important point, as one of the caveats relates to the fact that the datasets used don’t have the same length nor analyse more recent years. Therefore, there are points that need further attention. I believe that this document should be considered for publication after major changes, and if the authors agree to test for different datasets including recent periods.
Below I point some comments and suggestions, which hopefully can help the authors to enhance the manuscript.
Comments:
- Methods: Figure 1: Please add LULC information on this figure (as an additional panel).
- Methods: Line 77-79, Please briefly explain how the High and Very High wildfire hazard classes were determined.
- Methods: Line 108-109, “Prior research developed for the study area indicated an association between fire regime parameters and particular biophysical conditions (Bergonse et al., 2022).” Please remove as it was already mentioned.
- Methods: How do you justify using RFAJ and TPJS was calculated from monthly rainfall data obtained from the Worldclim database (reference 1970-2000)? Was the data used for the same period? Why not using a drought indicator like the standardized precipitation and evaportranspiration index (SPEI)?
- Methods: Lines 127-129: How do you aggregate the information from the different LULC maps from the different years? You only mention how you aggregate the classes not the different years of information. How do you cope with the Land use change? Please clarify.
- Lines 144-146: Please change these lines above as they answer to my question
- Methods: Lines 161-164: MAJOR CAVEAT: databases used for NPP and climate variables.
Why don’t you use NPP from MODIS which reaches present-day? Why do you rely on precipitation and temperature data which account for a period between 1970-2000 knowing that the last years have been record years in this area (Turco et al., 2019; sousa et al., 2019) and that drought conditions have been increasing (Vicente-Serrano et al., 2014)?
Ruffault et al. (2020), identified fire weather regimes objectively by dynamic k-means clustering based on the values of the weather and climate variables associated with each wildfire record, namely, temperature, relative humidity, wind speed, DMC and DC. Their results show that fire risk is higher when short-term meteorological extremes (warm and dry air, strong winds) combine with long-term summer drought, i.e. under the Hot drought, Heatwave and Wind-driven fire weather regimes. Therefore, wind is one of the drivers which is highly correlated and should not be discarded as also pointed by Vieira et al. (2020), nor the combination of factors. Moreover, the authors highlight that the frequency of heat-induced fire-weather is projected to increase by 14% by the end of the century (2071–2100) under the RCP4.5 scenario, and by 30% under the RCP8.5, suggesting that the frequency and extent of large wildfires will increase throughout the Mediterranean Basin. Thus, using more recent data which can account for the latest years is important.
Some additional important references on the topic focusing on the Mediterranean or the Iberian Peninsula:
Turco M, et al. On the key role of droughts in the dynamics of summer fires in Mediterranean Europe. Sci. Rep. 2017;7:81.
Turco M, et al. Climate drivers of the 2017 devastating fires in Portugal. Sci. Rep. 2019;9:13886
Ruffault J, Moron V, Trigo RM, Curt T. Objective identification of multiple large fire climatologies: An application to a Mediterranean ecosystem. Environ. Res. Lett. 2016;11:075006
Ruffault J, Curt T, Moron V, Trigo RM, Mouillot F, Koutsias N, Pimont F, Martin-StPaul N, Barbero R, Dupuy JL, Russo A, Belhadj-Khedher C. Increased likelihood of heat-induced large wildfires in the Mediterranean Basin. Sci Rep. 2020 Aug 14;10(1):13790. doi: 10.1038/s41598-020-70069-z. PMID: 32796945; PMCID: PMC7427790.
Identifying large fire weather typologies in the Iberian Peninsula, M Rodrigues, RM Trigo, C Vega-García, A Cardil - Agricultural and Forest Meteorology, 2020
Vieira I., Russo A., Trigo R.M. (2020) Identifying Local-Scale Weather Forcing Conditions Favorable to Generating Iberia’s Largest Fires . Forests 11(5), 547
Sousa P., Barriopedro D., Ramos A.M., García-Herrera R., Espirito-Santo F., Trigo R.M. (2019) Saharan air intrusions as a relevant mechanism for Iberian heatwaves: The record breaking events of August 2018 and June 2019. Weather and Climate Exetremes, 26, 100224, DOI: http://doi.org/10.1016/j.wace.2019.100224
Vicente-Serrano S. M., Lopez-Moreno Juan-I., Beguería S., Lorenzo-Lacruz J., Sanchez-Lorenzo A., García-Ruiz J. M., Azorin-Molina C., Morán-Tejeda E., Revuelto J., Trigo R., Coelho F., Espejo F. (2014) Evidence of increasing drought severity caused by temperature rise in southern Europe. Environmental Research Letters, doi:10.1088/1748-9326/9/4/044001
How do the usage of more recent databases influence the results as temperature is rising and weather temperature extremes are mounting in this area? And the influence of wind? These need to be tested and compared.
- Methods: when using the CT model you are using a spatial and temporal varying information to assess which is the most important variables in each of the 3-4 clusters? Or the information is aggregated spatially and then related? These options would rely on not so recent meteorological characterization and might not reflect the actual influence of temperature. How do you account for that?
- Discussion: Citing authors previous works on the same area and with similar approach is not a strong comparison. I would suggest the authors to look for similar results from other authors or different areas to support this point (e.g., lines 338, 345).
- Discussion: Lines 409-410: The authors say that “It is therefore possible that the potential effects of summer temperature in burned area are constrained by fuel availability”. As the authors certainly know from the basics of the fire triangle or combustion triangle, which is a simple model for understanding the necessary ingredients for most fires, three elements are needed for a fire to ignite: heat, fuel, and an oxidizing agent (usually oxygen). A fire naturally occurs when the elements are present and combined in the right mixture. A fire can be prevented or extinguished by removing any one of the elements in the fire triangle. Therefore, we can have all the necessary weather and vegetation conditions but if we don’t have ignitions, although the fire weather risk is high, the fire might not even start. Here the authors need to check for the presence of the conditions and not just suggest a possibility.
- The authors lack to show the limitations of the data used and also other aspects which were not addressed in their study.
- The authors don’t highlight how the conclusions on their current and previous work (Bergonseet al., 2022) are different.
Rafaello Bergonse, Sandra Oliveira, José Luís Zêzere, Francisco Moreira, Paulo Flores Ribeiro, Miguel Leal, José Manuel Lima e Santos, Biophysical controls over fire regime properties in Central Portugal, Science of The Total Environment, Volume 810, 2022, 152314, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2021.152314.
- The usage of NPP and temperature seem to be mostly disregarded in the discussion. Justify the usage and discuss their importance.
- When applying for the classification tree model, the importance of each of the parameters is determined based on a linear or non-linear relation? How does that affect the identification of the most important factors?
Citation: https://doi.org/10.5194/egusphere-2022-342-RC1 -
AC2: 'Reply on RC1', Rafaello Bergonse, 25 Jul 2022
We thank the reviewer for the overall positive feedback and the pertinent suggestions provided. We respond to each point raised in detail below.
1) Methods: Figure 1: Please add LULC information on this figure (as an additional panel).
R: A LULC map will be included in a different panel, as suggested.
2) Methods: Line 77-79, Please briefly explain how the High and Very High wildfire hazard classes were determined.
R: When building their wildfire hazard map for the whole of mainland Portugal, Oliveira et al. (2020) defined class breaks based on the configuration of the success-rate curve. This curve was obtained by plotting the fraction of the territory by decreasing hazard level vs. the fraction of total actual burned area. This information will be inserted into the manuscript to clarify this point.
3) Methods: Line 108-109, “Prior research developed for the study area indicated an association between fire regime parameters and particular biophysical conditions (Bergonse et al., 2022).” Please remove as it was already mentioned.
R: The referenced phrase will be removed from the manuscript.
4) Methods: How do you justify using RFAJ and TPJS was calculated from monthly rainfall data obtained from the Worldclim database (reference 1970-2000)? Was the data used for the same period? Why not using a drought indicator like the standardized precipitation and evapotranspiration index (SPEI)?
R: We used the Worldclim database because it is, to our knowledge, the only database available covering most of our study period with a suitable resolution; for further justification regarding the different time scopes of the variables within our dataset, please see point 2 of our response to comment 6, below.
We employed summer temperature and spring precipitation because both variables are related to what we intended to represent (fuel flammability and potential for vegetation growth and posterior fuel availability). Moreover, they are simple and straightforward to interpret and their seasonal values are used to restrict specific activities that may cause fires, being therefore linked to prevention measures applied in the country. For example, the use of fire practices by farmers is usually forbidden after spring, related to temperature and rainfall/humidity thresholds of the season.
The use of spring precipitation as a proxy of the potential growth of vegetation and therefore fuel availability later in the year was suggested by the results previously obtained by Oliveira et al. (2012), described in lines 118-120 of our manuscript. Moreover, in our previous article we have observed spring rainfall to have an important positive influence both over burned area and wildfire frequency (please see Bergonse et al. 2022, Table 4).
The relation between summer temperature and fuel flammability was also justified based on the published literature mentioned in line 124 of the manuscript. The pertinency of this variable was also confirmed by the results of our previous article, in which summer temperature was also shown to have a positive influence over burned area and wildfire frequency (please see Bergonse et al. 2022, Table 4).
We agree that drought indicators such as SPEI could be valuable to uncover other patterns regarding wildfire drivers. Reference to such indicators will be included in the Discussion section, together with appropriate references from the literature.
5) Methods: Lines 127-129: How do you aggregate the information from the different LULC maps from the different years? You only mention how you aggregate the classes not the different years of information. How do you cope with the Land use change? Please clarify.
Lines 144-146: Please change these lines above as they answer to my question
R: The mentioned lines will be repositioned as suggested.
6) Methods: Lines 161-164: MAJOR CAVEAT: databases used for NPP and climate variables.
Why don’t you use NPP from MODIS which reaches present-day? Why do you rely on precipitation and temperature data which account for a period between 1970-2000 knowing that the last years have been record years in this area (Turco et al., 2019; sousa et al., 2019) and that drought conditions have been increasing (Vicente-Serrano et al., 2014)?
Ruffault et al. (2020), identified fire weather regimes objectively by dynamic k-means clustering based on the values of the weather and climate variables associated with each wildfire record, namely, temperature, relative humidity, wind speed, DMC and DC. Their results show that fire risk is higher when short-term meteorological extremes (warm and dry air, strong winds) combine with long-term summer drought, i.e. under the Hot drought, Heatwave and Wind-driven fire weather regimes. Therefore, wind is one of the drivers which is highly correlated and should not be discarded as also pointed by Vieira et al. (2020), nor the combination of factors. Moreover, the authors highlight that the frequency of heat-induced fire-weather is projected to increase by 14% by the end of the century (2071–2100) under the RCP4.5 scenario, and by 30% under the RCP8.5, suggesting that the frequency and extent of large wildfires will increase throughout the Mediterranean Basin. Thus, using more recent data which can account for the latest years is important.
R: We will respond to each of the raised issues separately.
- We did use NPP from MODIS. However, we used the previous version (version 06), which was the one available when the dataset for the article was assembled. This can be verified by following the link included in the manuscript (line 149): https://lpdaac.usgs.gov/products/mod17a3hgfv006/).
The newer version of the MODIS dataset (061) is more up-to-date, as it extends to the present day. However, regarding our study period (1970-2018), that is, the period used to characterize the fire regimes) the newer MODIS dataset includes only four more years (2015-2018), which are unlikely to alter the general, long-term tendencies in which this article is focused.
- We relied on climate (precipitation and temperature) data for the period between 1970 and 2000 because data for the remainder of the study period (2001-2018) was, and remains, to our knowledge, unavailable. The differences between the period used to characterize the fire regimes and the periods of available data for the climate variables and Net Primary Productivity have been acknowledged in the Data Collection and Pre-Processing section (lines 160-164). As we also mention in those lines, our approach to fire regime is a long-term approach, that is, our purpose is to define general tendencies over a relatively long time period. Although record years are extremely important to understand in different contexts, they are not so important in relation to our approach, as they would detract from the general tendencies of the fire regimes we wish to characterize.
We will include a subsection in the Results and Discussion section highlighting the differences in the temporal scope of the various variables used as limitation of this work. We will also acknowledge the potential importance of prevailing wind conditions as a fire regime driver, which was not considered in this article.
7) Some additional important references on the topic focusing on the Mediterranean or the Iberian Peninsula:
Turco M, et al. On the key role of droughts in the dynamics of summer fires in Mediterranean Europe. Sci. Rep. 2017;7:81.
Turco M, et al. Climate drivers of the 2017 devastating fires in Portugal. Sci. Rep. 2019;9:13886
Ruffault J, Moron V, Trigo RM, Curt T. Objective identification of multiple large fire climatologies: An application to a Mediterranean ecosystem. Environ. Res. Lett. 2016;11:075006
Ruffault J, Curt T, Moron V, Trigo RM, Mouillot F, Koutsias N, Pimont F, Martin-StPaul N, Barbero R, Dupuy JL, Russo A, Belhadj-Khedher C. Increased likelihood of heat-induced large wildfires in the Mediterranean Basin. Sci Rep. 2020 Aug 14;10(1):13790. doi: 10.1038/s41598-020-70069-z. PMID: 32796945; PMCID: PMC7427790.
Identifying large fire weather typologies in the Iberian Peninsula, M Rodrigues, RM Trigo, C Vega-García, A Cardil - Agricultural and Forest Meteorology, 2020
Vieira I., Russo A., Trigo R.M. (2020) Identifying Local-Scale Weather Forcing Conditions Favorable to Generating Iberia’s Largest Fires . Forests 11(5), 547
Sousa P., Barriopedro D., Ramos A.M., García-Herrera R., Espirito-Santo F., Trigo R.M. (2019) Saharan air intrusions as a relevant mechanism for Iberian heatwaves: The record breaking events of August 2018 and June 2019. Weather and Climate Exetremes, 26, 100224, DOI: http://doi.org/10.1016/j.wace.2019.100224
Vicente-Serrano S. M., Lopez-Moreno Juan-I., Beguería S., Lorenzo-Lacruz J., Sanchez-Lorenzo A., García-Ruiz J. M., Azorin-Molina C., Morán-Tejeda E., Revuelto J., Trigo R., Coelho F., Espejo F. (2014) Evidence of increasing drought severity caused by temperature rise in southern Europe. Environmental Research Letters, doi:10.1088/1748-9326/9/4/044001
How do the usage of more recent databases influence the results as temperature is rising and weather temperature extremes are mounting in this area? And the influence of wind? These need to be tested and compared.
R: We thank the reviewer for the relevant and interesting studies suggested. We will definitely consider them when we revise the manuscript.
As we mentioned in the answer to the previous comment, we did not use a more recent climate database because we are not aware of the availability of one. We will gladly update our results if a more recent database is available.
Although prevailing wind conditions are definitely a potential fire regime driver, wind was not considered in this study. Our intention was to consider the influence of a set of potential drivers, but not to be exhaustive in this regard. The potential role of drivers left out of this study, such as wind conditions, will be considered in a subsection of the Results and Discussion section.
8) Methods: when using the CT model you are using a spatial and temporal varying information to assess which is the most important variables in each of the 3-4 clusters? Or the information is aggregated spatially and then related? These options would rely on not so recent meteorological characterization and might not reflect the actual influence of temperature. How do you account for that?
R: We are not sure we understand the question, but we will outline the essential points of the CT model. For each of the studied parishes (our units of analysis), our dataset includes as attributes the associated cluster (i.e. type of fire regime) and the values of the potential fire regime drivers. Although, as mentioned in previous comments and responses, the temporal scopes of the fire regime descriptors (used to generate the clusters) and the different potential drivers vary somewhat due to data availability limitations (the contrasts in temporal scope were shown in Table 1 and acknowledged in lines 160-164), it is assumed that all values are equally descriptive of general conditions throughout an equivalent long-term period. The information is therefore aggregated spatially (at the parish scale), and then related using the CT model. This CT model was built to assess the capacity of the different biophysical drivers to differentiate between fire regimes, that is, to correctly identify the fire regime each parish is associated to.
It is important to highlight that all biophysical drivers are intended to describe prevailing conditions, and do not take into consideration extremes. For example, we employed mean summer temperature to consider the effect of typical summer conditions, therefore disregarding the effect of exceptional, and thus relatively infrequent, years.
9) Discussion: Citing authors previous works on the same area and with similar approach is not a strong comparison. I would suggest the authors to look for similar results from other authors or different areas to support this point (e.g., lines 338, 345).
R: We will do as suggested.
10) Discussion: Lines 409-410: The authors say that “It is therefore possible that the potential effects of summer temperature in burned area are constrained by fuel availability”. As the authors certainly know from the basics of the fire triangle or combustion triangle, which is a simple model for understanding the necessary ingredients for most fires, three elements are needed for a fire to ignite: heat, fuel, and an oxidizing agent (usually oxygen). A fire naturally occurs when the elements are present and combined in the right mixture. A fire can be prevented or extinguished by removing any one of the elements in the fire triangle. Therefore, we can have all the necessary weather and vegetation conditions but if we don’t have ignitions, although the fire weather risk is high, the fire might not even start. Here the authors need to check for the presence of the conditions and not just suggest a possibility.
R: In the excerpt mentioned by the reviewer, we are indicating potential explanations for the fact that summer temperature, a well-known wildfire driver, was shown not to have a relevant role in our study area. Fuel availability is one possible explanation, as well as a lack of ignitions.
We will reformulate the phrase to read: “It is therefore possible that the potential effects of summer temperature in burned area are constrained by other factors, such as fuel availability or the inexistence of ignitions.” Regarding the former, prior studies mention that the effect of temperature is mediated by the productivity of an area (Pausas e Fernández-Muñoz, 2012; Pausas & Ribeiro, 2013). When fuel is limited, the effect of the temperature is less expressive.
11) The authors lack to show the limitations of the data used and also other aspects which were not addressed in their study.
R: The limitation of the data used are already acknowledged in lines 160-164. We will address this issue together with the potential importance of other variables not considered in this study (such as wind conditions) in a future subsection of the Results and Discussion section.
12) The authors don’t highlight how the conclusions on their current and previous work (Bergonse et al., 2022) are different.
Rafaello Bergonse, Sandra Oliveira, José Luís Zêzere, Francisco Moreira, Paulo Flores Ribeiro, Miguel Leal, José Manuel Lima e Santos, Biophysical controls over fire regime properties in Central Portugal, Science of The Total Environment, Volume 810, 2022, 152314, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2021.152314.
R: Both works share mostly the same dataset and approach, namely parishes as units of analysis, the same three fire regime descriptors, and a common set of potential biophysical drivers. However, they have different objectives, and therefore complementary conclusions. We therefore believe highlighting differences in the conclusions of both works would serve no useful purpose.
In the previous work, relations between each biophysical driver and each of the three fire regime descriptors were analysed using ordinal regression. A single fire regime was assumed for the whole study area, and the results suggested the existence of different fire regimes. This description of the previous work is given in lines 57-64 of the Introduction.
Contrarily, in this more recent work we use cluster analysis to distinguish different fire regimes, we characterize them, and then we use a classification tree model to assess the capacity of the biophysical factors to distinguish between the regimes. Due to the complementarity between both works, we draw on the relations found in the previous article between each fire regime descriptor and different biophysical factors to inform our discussion of the results of CT model (e.g. lines 345, 363, 367).
13) The usage of NPP and temperature seem to be mostly disregarded in the discussion. Justify the usage and discuss their importance.
R: The usage of NPP and temperature is justified, respectively, in lines 147, and lines 123-124 of the Data Collection and Pre-Processing section. Both variables are also considered in the Discussion section. NPP values are discussed in relation to the defined fire regimes (lines 361, 383) and in relation to the implication of these fire regimes to wildfire management (line 433). The importance of temperature is considered in lines 401-412 of the Discussion.
14) When applying for the classification tree model, the importance of each of the parameters is determined based on a linear or non-linear relation? How does that affect the identification of the most important factors?
R: Classification trees are produced by successive binary partitioning, or splitting, of the training data into a growing number of subsets (nodes). Each split is based on a binary condition, defined using the predictor variable (the splitter) that maximizes the homogeneity, or inversely, minimizes the impurity, of the two resulting nodes. In our case, this homogeneity was measured using the GINI criterion, which is based on squared probabilities of membership for each category of the dependent variable (i.e. each of the four fire regimes). GINI reaches its minimum (zero) when all cases in a node fall into a single fire regime.
Each split results in an improvement, which is calculated by comparing the homogeneity of the two resulting nodes with that of the original node. This improvement is attributed to the splitting variable. The importance of each variable for the overall classification procedure is based on the sum of the improvements in all nodes in which the variable appears as a splitter, weighted by the fraction of the training data in each node split (Steinberg, 2009).
We acknowledge that the criteria for quantifying the overall importance of each predictor variable is unclear in the manuscript. We will therefore integrate the above information into the Data Collection and Pre-Processing section, together with the additional reference below.
Steinberg, D. (2009). CART: Classification and Regression Trees. In X. Wu & V. Kumar (Eds.), TheTop Ten Algorythms in Data Mining (pp. 179–201). CRC Press.
Citation: https://doi.org/10.5194/egusphere-2022-342-AC2
-
RC2: 'Comment on egusphere-2022-342', Anonymous Referee #2, 18 Jul 2022
The topic of the present manuscript is of high importance for the fire community, being the rationale is well presented. However, the novelty of this work is poor. The manuscript’s findings are incremental and do not represent an advance in this field. Namely, the present manuscript is clearly part of a larger study, published in previous papers, with the same rationale, Bergonse et al. (2022) and Oliveira and Zêzere (2020).
Citation: https://doi.org/10.5194/egusphere-2022-342-RC2 -
AC1: 'Reply on RC2', Rafaello Bergonse, 20 Jul 2022
We thank the reviewer for the feedback and comments provided.
Regarding the novelty of the work, the present manuscript is indeed one of the results of a larger study, which includes Bergonse et al. (2022). Both works share a common rationale in that they share the same spatial analysis units and study area, the same three fire regime descriptors, and the same biophysical drivers. However, they have different objectives and analysis techniques. In Bergonse et al. (2022), relations between the biophysical drivers and each of the three fire regime descriptors were separately analysed using ordinal regression equations. Although the study area was assumed to have a single fire regime, the spatial patterns shown by the three fire regime descriptors suggested the existence of distinct regimes.
The current work builds upon the previous results, in that we employ cluster analysis to explicitly identify and then characterize the different fire regimes within the study area, something which has not been done before. We subsequently apply a classification tree model to assess the capacity of the different biophysical drivers to discriminate between the four fire regimes defined. After interpreting the results, we then discuss the implications of the identified fire regimes and their drivers to wildfire management. This also is a completely novel outcome.
Being results of a common, ongoing research project, this manuscript and the previous article share a rationale and can certainly be considered complementary. Each of these works, however, presents distinct and novel results, which is why we feel this manuscript to the suitable for publication in this journal. We believe that the incremental findings presented are valuable and helpful to try understanding, progressively, the complexity of wildfires in Portugal.
We would also like to note that Oliveira and Zêzere (2020) was not a part of the research project mentioned above, having different study area, temporal scope, and analysis technique (random forest). It also employs as dependent variable only one of the three fire regime descriptors mentioned above. In fact, this paper has investigated the relation between the spatial distribution of burned area and different biophysical and social drivers for the parishes of the whole mainland Portugal, in a rather different scope than the one now presented.
Citation: https://doi.org/10.5194/egusphere-2022-342-AC1
-
AC1: 'Reply on RC2', Rafaello Bergonse, 20 Jul 2022
-
RC3: 'Comment on egusphere-2022-342', Anonymous Referee #3, 04 Aug 2022
This manuscript aims to characterize fire regimes in Central Portugal and investigate the degree to which the differences between regimes are influenced by a set of biophysical drivers. The authors used civil parishes as units of analysis and cumulative percentage of parish area burned, Gini concentration index of burned area over time, and area-weighted total number of wildfires over a reference period of 44 years (1975-2018). The authors used cluster analysis to aggregate parishes into groups with similar fire regime and a classification tree model to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups. The methods used seem to be suitable and the manuscript is nicely written. However, I have some criticisms, including some changes concerning the datasets used and the novelty of the present work, that should be addressed before considering the paper for publication in Natural Hazards and Earth System Sciences journal.
Major suggestions/comments:
Novelty of the work:
The results are fairly described and the discussion focused in a reduced number of previous publications, including two previous works of the same authors that exhibits strong similarities with the present work. The novelty (and need) of the present results of is not clearly addressed. The baseline of the present work in terms of data and methods is very similar to the previous two works. The data used is the same and the statistical approaches are slightly different, but very related with the previous ones. The main results are the same: the role played by LULC, slope and spring rainfall in fire behavior.
The present paper adds the classification in 4 FR for central Portugal. However, the FR classification is closely dependent of the data used. This leads to my following comment.
Datasets:
FR regime classification in strongly dependent of the historical data over the region. Therefore, the used of climate data than does not describe the last two decades, when we are facing a change in fire paradigm over Europe, with the occurrence of the so-called megafires, highlights the fragilities of the FR classification.
Besides the ‘old’ meteorological datasets, the higher fire intensity or severity of the observed fire behavior trends was not included in the FR classification. The authors used burned area, however the burned area inside a civil parish may not be a good indication or fire intensity; other parameters (available through remote sensing datasets) should be included.
The inclusion of the suggested datasets, considering the aim of the present work, will strongly improve the quality of the results, highlighting its novelty.
Slope:
“Topography was expressed by slope (80th percentile, in degrees), which can be expected to promote flame propagation”. Why using the 80th percentile and not 90th or 75th. Did you make a sensitivity analysis for this choice? Did the authors include elevation information? Why?
Rainfall:
“RFAJ was calculated from monthly rainfall data obtained from the Worldclim database (1970-2000)”. The authors present an assessment for 44 years (1975-2018) and one of the crucial datasets used is only characterizing half of the period. The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for precipitation data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
…”in the form of raster maps of approximately 30 seconds (about 1 km resolution), which were resampled to a 25 m pixel”. How was done the resampling? Which co-variates were used to do resampling? And, why to do the downscale if the data is further aggregated at parish level?
Temperature:
The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for temperature data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
Net Primary Productivity (NPP):
Please consider to use the most recent version of the NPP product (2000-Present). Alternatively, consider to use the Climate Data Record of NDVI (annual mean or sum) available since 1981 to present (https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C01558).
Lines152-160
This paragraph seems to be out of order. Please consider to reorganize the paragraph (before NPP paragraph).
Lines161-165
The different periods considered for the different datasets could have a strong impact on the results, as the spatial patterns for precipitation and temperature in the last 30 years of the XX century may have strong differences in compared with NPP in the first 20 years of the XXI century.
With the aim to have a fire regime description that really reflects the recent vegetation, climate and fire behavior trends, I strongly suggest to include: a) Temp and Precip from ERA5 from 1979-Present; b) NDVI from 1981-Present. Therefore, the period of analysis would be 1981-Present (41 years).
Lines 360-361 “This is confirmed by FR3’s low Net Productivity Ratio, …, which is indicative of a relatively reduced forest cover.” Please provide a reference or provide the analysis that allow this statement (or remove the sentence)
4.2.3 FR2: Please provide a better characterization of this FR. As it is, seems that this fire regime is not a separate fire regime and may indicate that the classification in 4 fire regimes is not the adequate.
Lines 401-408: The less clear relation with summer temperature is probably related with less adequate database used, that does not reflect the temperature changes in the last two decades. Please check the impact of use of the suggested dataset for meteorological parameters.
Lines 409-412: Is the statement supported by the NPP results of this work? Please explain.
Minor
Lines 90-95: changed format.
Citation: https://doi.org/10.5194/egusphere-2022-342-RC3 -
AC3: 'Reply on RC3', Rafaello Bergonse, 15 Sep 2022
Referee comment:
https://editor.copernicus.org/index.php?_mdl=msover_md&_jrl=778&_lcm=oc158lcm159n&_ms=103240&salt=240954201780718781
This manuscript aims to characterize fire regimes in Central Portugal and investigate the degree to which the differences between regimes are influenced by a set of biophysical drivers. The authors used civil parishes as units of analysis and cumulative percentage of parish area burned, Gini concentration index of burned area over time, and area-weighted total number of wildfires over a reference period of 44 years (1975-2018). The authors used cluster analysis to aggregate parishes into groups with similar fire regime and a classification tree model to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups. The methods used seem to be suitable and the manuscript is nicely written. However, I have some criticisms, including some changes concerning the datasets used and the novelty of the present work, that should be addressed before considering the paper for publication in Natural Hazards and Earth System Sciences journal.
Major suggestions/comments
We thank the reviewer for the overall positive feedback and the pertinent suggestions provided. We have numbered each point and respond to it in detail below.
1) Novelty of the work:
The results are fairly described and the discussion focused in a reduced number of previous publications, including two previous works of the same authors that exhibits strong similarities with the present work. The novelty (and need) of the present results of is not clearly addressed. The baseline of the present work in terms of data and methods is very similar to the previous two works. The data used is the same and the statistical approaches are slightly different, but very related with the previous ones. The main results are the same: the role played by LULC, slope and spring rainfall in fire behavior.
The present paper adds the classification in 4 FR for central Portugal. However, the FR classification is closely dependent of the data used. This leads to my following comment.
R: Regarding the novelty of the work, the present manuscript is one of the results of a larger study, which includes Bergonse et al. (2022). Both works share a common rationale in that they share the same spatial analysis units and study area, the same three fire regime descriptors, and the same biophysical drivers. However, they have different objectives and analysis techniques. In Bergonse et al. (2022), relations between the biophysical drivers and each of the three fire regime descriptors were separately analysed using ordinal regression equations. Although the study area was assumed to have a single fire regime, the spatial patterns shown by the three fire regime descriptors suggested the existence of distinct regimes.
The present work builds upon the previous results, in that we employ cluster analysis to explicitly identify and then characterize the different fire regimes within the study area, something which has not been done before. We subsequently apply a classification tree model to assess the capacity of the different biophysical drivers to discriminate between the four fire regimes defined. After interpreting the results, we then discuss the implications of the identified fire regimes and their drivers to wildfire management. This also is a completely novel outcome.
Being results of a common, ongoing research project, this manuscript and the previous article can be considered complementary. Each of these works, however, presents distinct and novel results, which is why we feel this manuscript to the suitable for publication in this journal. We believe that the findings presented are valuable and helpful towards understanding the complexity of wildfires in Portugal as each small scientific step adds to the knowledge we very much need to deal with such issues.
In the manuscript, the relations between this and the previous work are made explicit in the final part of the Introduction (lines 57 and following).
We would also like to note that the other work mentioned by the reviewer (Oliveira and Zêzere, 2020) was not a part of the research project mentioned above, having different study area, temporal scope, and analysis technique (random forest). It also employs as dependent variable only one of the three fire regime descriptors mentioned above. In fact, this paper investigated the relation between the spatial distribution of burned area and different biophysical and social drivers for the parishes of the whole mainland Portugal, in a rather different scope than the one now presented.
2) Datasets:
FR regime classification in strongly dependent of the historical data over the region. Therefore, the use of climate data than does not describe the last two decades, when we are facing a change in fire paradigm over Europe, with the occurrence of the so-called megafires, highlights the fragilities of the FR classification.
Besides the ‘old’ meteorological datasets, the higher fire intensity or severity of the observed fire behavior trends was not included in the FR classification. The authors used burned area, however the burned area inside a civil parish may not be a good indication or fire intensity; other parameters (available through remote sensing datasets) should be included.
The inclusion of the suggested datasets, considering the aim of the present work, will strongly improve the quality of the results, highlighting its novelty.
R: The issue of the temporal limitations of the climate data is considered in detail below, in comment 4. However, we would like to underline that, in accordance with the adopted fire regime definition (lines 33-34 of the manuscript), the fire regimes were described based on the consequences of fires in terms of burned area through time, and not on the meteorological context in which these fires take place. The characterization of fire regime is based on burned area data for the 44-year period between 1975 and 2018 and is therefore not subject to any fragility derived from climate data limitations. Climate data were subsequently used as possible biophysical factors influencing fire regime, and it is in relation to this aspect of the work that the temporal limitations of the climate data indeed constitute a fragility. This limitation is acknowledged in the Data Collection and Pre-Processing section (lines 161-164) and will be further highlighted in a subsection to be included in the Results and Discussion section, focusing on the uncertainties and limitations of this work.
On a sidenote regarding the properties of the fire regime and their possible change between the period encompassed by the climatic data (1975-2000) and the later years, the following experiment was made. We created a new variable describing the cumulative percentage of parish area burned (CPAB) between 1975-2000 and ranked all study parishes by their value in this variable. We then ranked all parishes as to their value in the CPAB used in the article (i.e., encompassing the whole study period 1975-2018). We then calculated the Pearson correlation coefficient between the two ranked variables, obtaining an R of 0.895, significant at the 0.01 level. This result shows that the relative positions of the different parishes in terms of cumulative percentage of area burned are quite similar, regardless of the period considered. Although we tried this only for this fire regime descriptor, this result strongly suggests that the fire regimes among the studied parishes show a similar behaviour whether we limit the analysis to the climate-data covered period or to the whole 44-year period used in the article.
Regarding the issue of wildfire intensity/severity, fire regimes can be described with greatly varying degrees of complexity. We purposefully employed a simple, straightforward approach, expressing it with three indicators that can be extracted from freely available annual burned area maps, and therefore easily reproduced in other study areas. We agree that severity is an important aspect of fire behaviour that may not be adequately expressed by burned area alone, as are others such as the characteristics of the largest, relatively infrequent fires (which would include the so-called megafires). We will refer to these aspects of fire regime in a future “Uncertainties and Limitations” subsection of the Results and Discussion section, to inform future studies. We believe, however, that the focus on a simpler approach and the absence of these other datasets does not hinder the usefulness or novelty of the work, considering its objectives.
3) Slope:
“Topography was expressed by slope (80th percentile, in degrees), which can be expected to promote flame propagation”. Why using the 80th percentile and not 90th or 75th. Did you make a sensitivity analysis for this choice? Did the authors include elevation information? Why?
R: The set of 12 biophysical variables employed are derived from our previous results in Bergonse et al. (2022), as stated in lines 109-110 of the manuscript. In the referenced article, we initially adopted both slope and elevation as potential biophysical controls, using percentiles 50, 75, 80, 90 and 95. During a multicollinearity analysis, all percentiles were shown to be strongly intercorrelated. We thus chose to keep those more strongly correlated with the remaining percentiles in the same group, leading to the adoption of the 80th percentiles of slope and altitude. Altitude was eliminated further along the multicollinearity analysis process, as its Variance Inflating Factor showed it can be expressed as a linear combination of other variables in the dataset. This process is described in the Data Analysis section of Bergonse et al. (2022).
4) Rainfall:
“RFAJ was calculated from monthly rainfall data obtained from the Worldclim database (1970-2000)”. The authors present an assessment for 44 years (1975-2018) and one of the crucial datasets used is only characterizing half of the period. The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for precipitation data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
…”in the form of raster maps of approximately 30 seconds (about 1 km resolution), which were resampled to a 25 m pixel”. How was done the resampling? Which co-variates were used to do resampling? And, why to do the downscale if the data is further aggregated at parish level?
Temperature:
The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for temperature data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
R: We will focus on the issue of the climate dataset first, and then respond to the comment on the resampling process.
In the manuscript, we acknowledge the disparities between the periods used to characterize the fire regimes and the biophysical drivers in lines 161-165 and in Table 1. We thank the reviewer for the suggestion to use the ERA5 dataset to overcome this issue, which we have investigated. Unfortunately, the spatial resolution of ERA5 makes it too coarse to be applicable to a study on such a detailed scale as the one we employ. ERA5 is made available with a 0.25-degree resolution, which translates to a pixel of approximately 24.8 km. An overlay between an ERA5-derived map and the limits of our study parishes shows that each pixel comprises multiple complete parishes within it, making the ERA5 dataset too generalized for application in this study. In comparison, the Worldclim dataset used has a resolution of approximately 1000 m, which makes it suitable for our scale of analysis. A solution to this issue would be to limit the study to the period 1975-2000, as suggested. However, this would entail a similar problem with the land-use data, which only begins in 1990 (1995 in the cases of two specific variables) (as shown in Table 1).
The use of these imperfectly overlapping datasets, imposed by the unavailability of suitable data, implies the assumption that all are representative of the long-term, general fire regimes and biophysical factors that have characterized the study area within the last four decades. We agree that record fires, such as those seen recently, are important to understand the dynamics of fire in different contexts. However, they are not so important in relation to our approach, as they would detract from the general tendencies of the fire regimes we wish to characterize, moreover when our approach is based on annual burned area data and not on the characteristics of individual fires.
The resampling of the climate data maps was performed using ArcMAP’s Resample tool, using nearest neighbour assignment. This software was used for all spatial analysis operations, as stated in lines 166-7 of the manuscript. The resampling was done to minimize generalization in association to the Zonal Statistics tool used to calculate the mean values for the pixels in each parish. The following example can clarify the rationale behind the resampling. Let us imagine that our purpose is to calculate the mean temperature during the summer months for each of a set of parishes. To do so, we will use the vector map with the parish limits, and a raster map with the variable of interest, that is, temperature during the summer months. Let us assume this raster map has 1000-m pixels. If one of the parish polygons partially overlays a 1000-m pixel without covering its centre, the value of this pixel will be ignored in the calculations, i.e., the mean value calculated for that parish will in practice ignore a part of the surface of the parish. If, however, the raster map has a 25-m pixel, a much smaller part of each parish will suffer from this problem, as each polygon will encompass a much greater number of smaller cells, which will fill its area much more fully.
5) Net Primary Productivity (NPP)
Please consider to use the most recent version of the NPP product (2000-Present). Alternatively, consider to use the Climate Data Record of NDVI (annual mean or sum) available since 1981 to present (https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C01558).
R: The newer version of the MODIS NPP dataset (061) is indeed more up to date, as it extends to the present day. However, regarding our study period (1970-2018, the period used to characterize the fire regimes) the newer MODIS dataset includes only four more years (2015-2018), which are unlikely to alter the general, long-term tendencies in which this article is focused.
Regarding the NDVI dataset from NOAA, we thank the reviewer for the suggestion, which we investigated. The data is provided with a resolution of 0.05 degrees. This translates to a pixel of approximately 2946 m, which is much coarser than the 500-m NPP dataset we employed, even though it could still fit with our analysis. There are always options to make when it comes to the data integrated; in this case, we believe the NPP data is a reasonable choice considering the long-term approach applied, which does not require daily data. We will, nevertheless, make reference to the NOAA dataset in a future Uncertainties and Limitations subsection, to be inserted into the Results and Discussion section. This will inform future studies and research projects.
6) Lines 152-160
This paragraph seems to be out of order. Please consider to reorganize the paragraph (before NPP paragraph).
R: The paragraph will be repositioned as suggested.
7) Lines 161-165
The different periods considered for the different datasets could have a strong impact on the results, as the spatial patterns for precipitation and temperature in the last 30 years of the XX century may have strong differences in compared with NPP in the first 20 years of the XXI century.
With the aim to have a fire regime description that really reflects the recent vegetation, climate and fire behavior trends, I strongly suggest to include: a) Temp and Precip from ERA5 from 1979-Present; b) NDVI from 1981-Present. Therefore, the period of analysis would be 1981-Present (41 years).
R: We thank the reviewer for the suggestions. The issues regarding the climate dataset and the NPP dataset are considered, respectively, in the responses to comments 4 and 5.
8) Lines 360-361
“This is confirmed by FR3’s low Net Productivity Ratio, …, which is indicative of a relatively reduced forest cover.” Please provide a reference or provide the analysis that allow this statement (or remove the sentence)
R: In our previous work, it was observed that NPP was inversely correlated to the percentage of the area of each parish covered with shrubland, but positively correlated to the percentage of eucalyptus forests, pine forests and forests of invasive species (please see Bergonse et al., 2022, supplementary table S.1). We will reference this work at this point.
9) 4.2.3 FR2
Please provide a better characterization of this FR. As it is, seems that this fire regime is not a separate fire regime and may indicate that the classification in 4 fire regimes is not the adequate
R: Throughout section 3.1, our choices regarding the clustering process are described and justified. Figure 2 shows that a 3-cluster solution would indeed express, in a more synthetic way, the major contrasts in fire regime across the study area (as stated in lines 262-263). As we also note, cluster 2 always clearly occupies an intermediate position, regardless of the choice between 3 and 4 clusters (Figure 3 and lines 250 and 255-7). We then justify the option for a four-cluster/fire regime solution based on the practical implications of the fourth cluster (a subset of what was cluster 2 in the 3-cluster solution) regarding wildfire prevention and suppression (as stated in lines 263-266).
We understand the issue raised by the reviewer, in that, when reading the discussion of each cluster in section 4.2, a reader may at first not see the justification for keeping cluster 2, a cluster that has apparently no distinguishing features. However, we feel that such a doubt could only arise from a partial reading of the article, not only because cluster 2 is present in both 3- and 4-cluster solutions, being in either case relevant regarding fire regime variability in the study area (as shown in Figure 2), but also because it possesses intermediate characteristics in both clustering solutions. Indeed, within our study area, this cluster/fire regime it distinguished by this intermediate character.
10) Lines 401-408
The less clear relation with summer temperature is probably related with less adequate database used, that does not reflect the temperature changes in the last two decades. Please check the impact of use of the suggested dataset for meteorological parameters.
R: We considered the issue regarding the climate dataset in our response to comment 4. Although we consider different possible explanations for the results obtained (lines 401-412), we agree with the reviewer in that there may exist a relation between the limitations in temporal extent of the climate dataset and our results. This will be made explicit in the “Uncertainties and Limitations” section that will be incorporated into the Results and Discussion section.
11) Lines 409-412
Is the statement supported by the NPP results of this work? Please explain.
R: The distribution of NPP values among the four FRs (Figure 6-G) shows that the lowest value is associated to FR3, which has the highest Cumulative Percentage of Area Burned (CPAB) of all FRs. This low NPP is in agreement with the highest percentage of area covered by fire-prone, quickly regenerating shrubland, also shown by FR3 (Fig. 6-A). This would support the notion that fuel is the factor controlling burned area in this FR, instead of summer temperature (which has its lowest value in FR3).
Summer temperature has its highest values in FRs 1 and 4. The highest NPP values also occur in these two FRs, in accordance with the highest percentages of area covered by eucalyptus forest, also shown by these two FRs (Fig.6-E). These values, suggesting the combination of available forest fuels and high summer temperatures, only translate into extensive burned area in the case of FR4, which has the second highest CPAB of all FRs (Fig.3-B). In the case of FR1 (with the lowest CPAB of all FRs), factors other than fuel availability seem to constrain burned area. Among the biophysical drivers considered in this work, examples could be the high percentage of agricultural area (fragmenting fuels and implying higher human presence and therefore quicker detection and response) (Fig.6-D), the lowest slope of all four FRs (slowing flame propagation; notably, FR4 has the highest slope values) (Fig. 6-C), and the relatively high fragmentation of fuel (in comparison to FR4) (Fig. 6-H). Additionally, FR1 occurs mostly along the coastal sector, which has higher rates of urbanization and population densities than the inland parishes. It is likely that these will promote faster detection and better response capabilities on the part of the authorities in case of ignition.
In sum, NPP results seem to support the idea that fuel availability constrains to a degree the effect of summer temperatures. However, the role of NPP must be considered together with the effects of other factors, as fire regimes result from the interaction of multiple drivers.
Minor
12) Lines 90-95
Changed format.
R: The paragraph will be correctly re-formatted.
-
AC3: 'Reply on RC3', Rafaello Bergonse, 15 Sep 2022
Status: closed
-
RC1: 'Comment on egusphere-2022-342', Anonymous Referee #1, 13 Jul 2022
The authors characterize fire regimes over Central Portugal and investigate the degree to which the differences between fire regimes are influenced by a set of biophysical drivers, namely slope, summer temperature and spring rainfall, land use/land cover (LULC) type and patch fragmentation, and net primary productivity. The authors rely on a cluster analysis followed by a classification tree model to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups for 44 yrs period.
Results allowed to distinguish four types of fire regimes and show that these can be significantly differentiated using the biophysical drivers.
The subject discussed in the present article is of high importance, and the paper is well structured and easy to follow. The introduction is very clear and includes interesting and recent references. The objectives are very well presented. Although this study draws on prior studies with similar objectives, the authors try to presented the novelty, namely the use of a large dataset regarding the study of Oliveira and Zêzere (2020) and the study from Bergonse et al. (2022) which assumed a similar fire regime all over the study area. Although the usage of a longer dataset and different fire regimes over Central Portugal might improve the knowledge robustness on that area and topic, the authors need to better address the novelty of this paper comparatively to previous works.
The limitations of the study are not identified nor discussed. This is an important point, as one of the caveats relates to the fact that the datasets used don’t have the same length nor analyse more recent years. Therefore, there are points that need further attention. I believe that this document should be considered for publication after major changes, and if the authors agree to test for different datasets including recent periods.
Below I point some comments and suggestions, which hopefully can help the authors to enhance the manuscript.
Comments:
- Methods: Figure 1: Please add LULC information on this figure (as an additional panel).
- Methods: Line 77-79, Please briefly explain how the High and Very High wildfire hazard classes were determined.
- Methods: Line 108-109, “Prior research developed for the study area indicated an association between fire regime parameters and particular biophysical conditions (Bergonse et al., 2022).” Please remove as it was already mentioned.
- Methods: How do you justify using RFAJ and TPJS was calculated from monthly rainfall data obtained from the Worldclim database (reference 1970-2000)? Was the data used for the same period? Why not using a drought indicator like the standardized precipitation and evaportranspiration index (SPEI)?
- Methods: Lines 127-129: How do you aggregate the information from the different LULC maps from the different years? You only mention how you aggregate the classes not the different years of information. How do you cope with the Land use change? Please clarify.
- Lines 144-146: Please change these lines above as they answer to my question
- Methods: Lines 161-164: MAJOR CAVEAT: databases used for NPP and climate variables.
Why don’t you use NPP from MODIS which reaches present-day? Why do you rely on precipitation and temperature data which account for a period between 1970-2000 knowing that the last years have been record years in this area (Turco et al., 2019; sousa et al., 2019) and that drought conditions have been increasing (Vicente-Serrano et al., 2014)?
Ruffault et al. (2020), identified fire weather regimes objectively by dynamic k-means clustering based on the values of the weather and climate variables associated with each wildfire record, namely, temperature, relative humidity, wind speed, DMC and DC. Their results show that fire risk is higher when short-term meteorological extremes (warm and dry air, strong winds) combine with long-term summer drought, i.e. under the Hot drought, Heatwave and Wind-driven fire weather regimes. Therefore, wind is one of the drivers which is highly correlated and should not be discarded as also pointed by Vieira et al. (2020), nor the combination of factors. Moreover, the authors highlight that the frequency of heat-induced fire-weather is projected to increase by 14% by the end of the century (2071–2100) under the RCP4.5 scenario, and by 30% under the RCP8.5, suggesting that the frequency and extent of large wildfires will increase throughout the Mediterranean Basin. Thus, using more recent data which can account for the latest years is important.
Some additional important references on the topic focusing on the Mediterranean or the Iberian Peninsula:
Turco M, et al. On the key role of droughts in the dynamics of summer fires in Mediterranean Europe. Sci. Rep. 2017;7:81.
Turco M, et al. Climate drivers of the 2017 devastating fires in Portugal. Sci. Rep. 2019;9:13886
Ruffault J, Moron V, Trigo RM, Curt T. Objective identification of multiple large fire climatologies: An application to a Mediterranean ecosystem. Environ. Res. Lett. 2016;11:075006
Ruffault J, Curt T, Moron V, Trigo RM, Mouillot F, Koutsias N, Pimont F, Martin-StPaul N, Barbero R, Dupuy JL, Russo A, Belhadj-Khedher C. Increased likelihood of heat-induced large wildfires in the Mediterranean Basin. Sci Rep. 2020 Aug 14;10(1):13790. doi: 10.1038/s41598-020-70069-z. PMID: 32796945; PMCID: PMC7427790.
Identifying large fire weather typologies in the Iberian Peninsula, M Rodrigues, RM Trigo, C Vega-García, A Cardil - Agricultural and Forest Meteorology, 2020
Vieira I., Russo A., Trigo R.M. (2020) Identifying Local-Scale Weather Forcing Conditions Favorable to Generating Iberia’s Largest Fires . Forests 11(5), 547
Sousa P., Barriopedro D., Ramos A.M., García-Herrera R., Espirito-Santo F., Trigo R.M. (2019) Saharan air intrusions as a relevant mechanism for Iberian heatwaves: The record breaking events of August 2018 and June 2019. Weather and Climate Exetremes, 26, 100224, DOI: http://doi.org/10.1016/j.wace.2019.100224
Vicente-Serrano S. M., Lopez-Moreno Juan-I., Beguería S., Lorenzo-Lacruz J., Sanchez-Lorenzo A., García-Ruiz J. M., Azorin-Molina C., Morán-Tejeda E., Revuelto J., Trigo R., Coelho F., Espejo F. (2014) Evidence of increasing drought severity caused by temperature rise in southern Europe. Environmental Research Letters, doi:10.1088/1748-9326/9/4/044001
How do the usage of more recent databases influence the results as temperature is rising and weather temperature extremes are mounting in this area? And the influence of wind? These need to be tested and compared.
- Methods: when using the CT model you are using a spatial and temporal varying information to assess which is the most important variables in each of the 3-4 clusters? Or the information is aggregated spatially and then related? These options would rely on not so recent meteorological characterization and might not reflect the actual influence of temperature. How do you account for that?
- Discussion: Citing authors previous works on the same area and with similar approach is not a strong comparison. I would suggest the authors to look for similar results from other authors or different areas to support this point (e.g., lines 338, 345).
- Discussion: Lines 409-410: The authors say that “It is therefore possible that the potential effects of summer temperature in burned area are constrained by fuel availability”. As the authors certainly know from the basics of the fire triangle or combustion triangle, which is a simple model for understanding the necessary ingredients for most fires, three elements are needed for a fire to ignite: heat, fuel, and an oxidizing agent (usually oxygen). A fire naturally occurs when the elements are present and combined in the right mixture. A fire can be prevented or extinguished by removing any one of the elements in the fire triangle. Therefore, we can have all the necessary weather and vegetation conditions but if we don’t have ignitions, although the fire weather risk is high, the fire might not even start. Here the authors need to check for the presence of the conditions and not just suggest a possibility.
- The authors lack to show the limitations of the data used and also other aspects which were not addressed in their study.
- The authors don’t highlight how the conclusions on their current and previous work (Bergonseet al., 2022) are different.
Rafaello Bergonse, Sandra Oliveira, José Luís Zêzere, Francisco Moreira, Paulo Flores Ribeiro, Miguel Leal, José Manuel Lima e Santos, Biophysical controls over fire regime properties in Central Portugal, Science of The Total Environment, Volume 810, 2022, 152314, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2021.152314.
- The usage of NPP and temperature seem to be mostly disregarded in the discussion. Justify the usage and discuss their importance.
- When applying for the classification tree model, the importance of each of the parameters is determined based on a linear or non-linear relation? How does that affect the identification of the most important factors?
Citation: https://doi.org/10.5194/egusphere-2022-342-RC1 -
AC2: 'Reply on RC1', Rafaello Bergonse, 25 Jul 2022
We thank the reviewer for the overall positive feedback and the pertinent suggestions provided. We respond to each point raised in detail below.
1) Methods: Figure 1: Please add LULC information on this figure (as an additional panel).
R: A LULC map will be included in a different panel, as suggested.
2) Methods: Line 77-79, Please briefly explain how the High and Very High wildfire hazard classes were determined.
R: When building their wildfire hazard map for the whole of mainland Portugal, Oliveira et al. (2020) defined class breaks based on the configuration of the success-rate curve. This curve was obtained by plotting the fraction of the territory by decreasing hazard level vs. the fraction of total actual burned area. This information will be inserted into the manuscript to clarify this point.
3) Methods: Line 108-109, “Prior research developed for the study area indicated an association between fire regime parameters and particular biophysical conditions (Bergonse et al., 2022).” Please remove as it was already mentioned.
R: The referenced phrase will be removed from the manuscript.
4) Methods: How do you justify using RFAJ and TPJS was calculated from monthly rainfall data obtained from the Worldclim database (reference 1970-2000)? Was the data used for the same period? Why not using a drought indicator like the standardized precipitation and evapotranspiration index (SPEI)?
R: We used the Worldclim database because it is, to our knowledge, the only database available covering most of our study period with a suitable resolution; for further justification regarding the different time scopes of the variables within our dataset, please see point 2 of our response to comment 6, below.
We employed summer temperature and spring precipitation because both variables are related to what we intended to represent (fuel flammability and potential for vegetation growth and posterior fuel availability). Moreover, they are simple and straightforward to interpret and their seasonal values are used to restrict specific activities that may cause fires, being therefore linked to prevention measures applied in the country. For example, the use of fire practices by farmers is usually forbidden after spring, related to temperature and rainfall/humidity thresholds of the season.
The use of spring precipitation as a proxy of the potential growth of vegetation and therefore fuel availability later in the year was suggested by the results previously obtained by Oliveira et al. (2012), described in lines 118-120 of our manuscript. Moreover, in our previous article we have observed spring rainfall to have an important positive influence both over burned area and wildfire frequency (please see Bergonse et al. 2022, Table 4).
The relation between summer temperature and fuel flammability was also justified based on the published literature mentioned in line 124 of the manuscript. The pertinency of this variable was also confirmed by the results of our previous article, in which summer temperature was also shown to have a positive influence over burned area and wildfire frequency (please see Bergonse et al. 2022, Table 4).
We agree that drought indicators such as SPEI could be valuable to uncover other patterns regarding wildfire drivers. Reference to such indicators will be included in the Discussion section, together with appropriate references from the literature.
5) Methods: Lines 127-129: How do you aggregate the information from the different LULC maps from the different years? You only mention how you aggregate the classes not the different years of information. How do you cope with the Land use change? Please clarify.
Lines 144-146: Please change these lines above as they answer to my question
R: The mentioned lines will be repositioned as suggested.
6) Methods: Lines 161-164: MAJOR CAVEAT: databases used for NPP and climate variables.
Why don’t you use NPP from MODIS which reaches present-day? Why do you rely on precipitation and temperature data which account for a period between 1970-2000 knowing that the last years have been record years in this area (Turco et al., 2019; sousa et al., 2019) and that drought conditions have been increasing (Vicente-Serrano et al., 2014)?
Ruffault et al. (2020), identified fire weather regimes objectively by dynamic k-means clustering based on the values of the weather and climate variables associated with each wildfire record, namely, temperature, relative humidity, wind speed, DMC and DC. Their results show that fire risk is higher when short-term meteorological extremes (warm and dry air, strong winds) combine with long-term summer drought, i.e. under the Hot drought, Heatwave and Wind-driven fire weather regimes. Therefore, wind is one of the drivers which is highly correlated and should not be discarded as also pointed by Vieira et al. (2020), nor the combination of factors. Moreover, the authors highlight that the frequency of heat-induced fire-weather is projected to increase by 14% by the end of the century (2071–2100) under the RCP4.5 scenario, and by 30% under the RCP8.5, suggesting that the frequency and extent of large wildfires will increase throughout the Mediterranean Basin. Thus, using more recent data which can account for the latest years is important.
R: We will respond to each of the raised issues separately.
- We did use NPP from MODIS. However, we used the previous version (version 06), which was the one available when the dataset for the article was assembled. This can be verified by following the link included in the manuscript (line 149): https://lpdaac.usgs.gov/products/mod17a3hgfv006/).
The newer version of the MODIS dataset (061) is more up-to-date, as it extends to the present day. However, regarding our study period (1970-2018), that is, the period used to characterize the fire regimes) the newer MODIS dataset includes only four more years (2015-2018), which are unlikely to alter the general, long-term tendencies in which this article is focused.
- We relied on climate (precipitation and temperature) data for the period between 1970 and 2000 because data for the remainder of the study period (2001-2018) was, and remains, to our knowledge, unavailable. The differences between the period used to characterize the fire regimes and the periods of available data for the climate variables and Net Primary Productivity have been acknowledged in the Data Collection and Pre-Processing section (lines 160-164). As we also mention in those lines, our approach to fire regime is a long-term approach, that is, our purpose is to define general tendencies over a relatively long time period. Although record years are extremely important to understand in different contexts, they are not so important in relation to our approach, as they would detract from the general tendencies of the fire regimes we wish to characterize.
We will include a subsection in the Results and Discussion section highlighting the differences in the temporal scope of the various variables used as limitation of this work. We will also acknowledge the potential importance of prevailing wind conditions as a fire regime driver, which was not considered in this article.
7) Some additional important references on the topic focusing on the Mediterranean or the Iberian Peninsula:
Turco M, et al. On the key role of droughts in the dynamics of summer fires in Mediterranean Europe. Sci. Rep. 2017;7:81.
Turco M, et al. Climate drivers of the 2017 devastating fires in Portugal. Sci. Rep. 2019;9:13886
Ruffault J, Moron V, Trigo RM, Curt T. Objective identification of multiple large fire climatologies: An application to a Mediterranean ecosystem. Environ. Res. Lett. 2016;11:075006
Ruffault J, Curt T, Moron V, Trigo RM, Mouillot F, Koutsias N, Pimont F, Martin-StPaul N, Barbero R, Dupuy JL, Russo A, Belhadj-Khedher C. Increased likelihood of heat-induced large wildfires in the Mediterranean Basin. Sci Rep. 2020 Aug 14;10(1):13790. doi: 10.1038/s41598-020-70069-z. PMID: 32796945; PMCID: PMC7427790.
Identifying large fire weather typologies in the Iberian Peninsula, M Rodrigues, RM Trigo, C Vega-García, A Cardil - Agricultural and Forest Meteorology, 2020
Vieira I., Russo A., Trigo R.M. (2020) Identifying Local-Scale Weather Forcing Conditions Favorable to Generating Iberia’s Largest Fires . Forests 11(5), 547
Sousa P., Barriopedro D., Ramos A.M., García-Herrera R., Espirito-Santo F., Trigo R.M. (2019) Saharan air intrusions as a relevant mechanism for Iberian heatwaves: The record breaking events of August 2018 and June 2019. Weather and Climate Exetremes, 26, 100224, DOI: http://doi.org/10.1016/j.wace.2019.100224
Vicente-Serrano S. M., Lopez-Moreno Juan-I., Beguería S., Lorenzo-Lacruz J., Sanchez-Lorenzo A., García-Ruiz J. M., Azorin-Molina C., Morán-Tejeda E., Revuelto J., Trigo R., Coelho F., Espejo F. (2014) Evidence of increasing drought severity caused by temperature rise in southern Europe. Environmental Research Letters, doi:10.1088/1748-9326/9/4/044001
How do the usage of more recent databases influence the results as temperature is rising and weather temperature extremes are mounting in this area? And the influence of wind? These need to be tested and compared.
R: We thank the reviewer for the relevant and interesting studies suggested. We will definitely consider them when we revise the manuscript.
As we mentioned in the answer to the previous comment, we did not use a more recent climate database because we are not aware of the availability of one. We will gladly update our results if a more recent database is available.
Although prevailing wind conditions are definitely a potential fire regime driver, wind was not considered in this study. Our intention was to consider the influence of a set of potential drivers, but not to be exhaustive in this regard. The potential role of drivers left out of this study, such as wind conditions, will be considered in a subsection of the Results and Discussion section.
8) Methods: when using the CT model you are using a spatial and temporal varying information to assess which is the most important variables in each of the 3-4 clusters? Or the information is aggregated spatially and then related? These options would rely on not so recent meteorological characterization and might not reflect the actual influence of temperature. How do you account for that?
R: We are not sure we understand the question, but we will outline the essential points of the CT model. For each of the studied parishes (our units of analysis), our dataset includes as attributes the associated cluster (i.e. type of fire regime) and the values of the potential fire regime drivers. Although, as mentioned in previous comments and responses, the temporal scopes of the fire regime descriptors (used to generate the clusters) and the different potential drivers vary somewhat due to data availability limitations (the contrasts in temporal scope were shown in Table 1 and acknowledged in lines 160-164), it is assumed that all values are equally descriptive of general conditions throughout an equivalent long-term period. The information is therefore aggregated spatially (at the parish scale), and then related using the CT model. This CT model was built to assess the capacity of the different biophysical drivers to differentiate between fire regimes, that is, to correctly identify the fire regime each parish is associated to.
It is important to highlight that all biophysical drivers are intended to describe prevailing conditions, and do not take into consideration extremes. For example, we employed mean summer temperature to consider the effect of typical summer conditions, therefore disregarding the effect of exceptional, and thus relatively infrequent, years.
9) Discussion: Citing authors previous works on the same area and with similar approach is not a strong comparison. I would suggest the authors to look for similar results from other authors or different areas to support this point (e.g., lines 338, 345).
R: We will do as suggested.
10) Discussion: Lines 409-410: The authors say that “It is therefore possible that the potential effects of summer temperature in burned area are constrained by fuel availability”. As the authors certainly know from the basics of the fire triangle or combustion triangle, which is a simple model for understanding the necessary ingredients for most fires, three elements are needed for a fire to ignite: heat, fuel, and an oxidizing agent (usually oxygen). A fire naturally occurs when the elements are present and combined in the right mixture. A fire can be prevented or extinguished by removing any one of the elements in the fire triangle. Therefore, we can have all the necessary weather and vegetation conditions but if we don’t have ignitions, although the fire weather risk is high, the fire might not even start. Here the authors need to check for the presence of the conditions and not just suggest a possibility.
R: In the excerpt mentioned by the reviewer, we are indicating potential explanations for the fact that summer temperature, a well-known wildfire driver, was shown not to have a relevant role in our study area. Fuel availability is one possible explanation, as well as a lack of ignitions.
We will reformulate the phrase to read: “It is therefore possible that the potential effects of summer temperature in burned area are constrained by other factors, such as fuel availability or the inexistence of ignitions.” Regarding the former, prior studies mention that the effect of temperature is mediated by the productivity of an area (Pausas e Fernández-Muñoz, 2012; Pausas & Ribeiro, 2013). When fuel is limited, the effect of the temperature is less expressive.
11) The authors lack to show the limitations of the data used and also other aspects which were not addressed in their study.
R: The limitation of the data used are already acknowledged in lines 160-164. We will address this issue together with the potential importance of other variables not considered in this study (such as wind conditions) in a future subsection of the Results and Discussion section.
12) The authors don’t highlight how the conclusions on their current and previous work (Bergonse et al., 2022) are different.
Rafaello Bergonse, Sandra Oliveira, José Luís Zêzere, Francisco Moreira, Paulo Flores Ribeiro, Miguel Leal, José Manuel Lima e Santos, Biophysical controls over fire regime properties in Central Portugal, Science of The Total Environment, Volume 810, 2022, 152314, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2021.152314.
R: Both works share mostly the same dataset and approach, namely parishes as units of analysis, the same three fire regime descriptors, and a common set of potential biophysical drivers. However, they have different objectives, and therefore complementary conclusions. We therefore believe highlighting differences in the conclusions of both works would serve no useful purpose.
In the previous work, relations between each biophysical driver and each of the three fire regime descriptors were analysed using ordinal regression. A single fire regime was assumed for the whole study area, and the results suggested the existence of different fire regimes. This description of the previous work is given in lines 57-64 of the Introduction.
Contrarily, in this more recent work we use cluster analysis to distinguish different fire regimes, we characterize them, and then we use a classification tree model to assess the capacity of the biophysical factors to distinguish between the regimes. Due to the complementarity between both works, we draw on the relations found in the previous article between each fire regime descriptor and different biophysical factors to inform our discussion of the results of CT model (e.g. lines 345, 363, 367).
13) The usage of NPP and temperature seem to be mostly disregarded in the discussion. Justify the usage and discuss their importance.
R: The usage of NPP and temperature is justified, respectively, in lines 147, and lines 123-124 of the Data Collection and Pre-Processing section. Both variables are also considered in the Discussion section. NPP values are discussed in relation to the defined fire regimes (lines 361, 383) and in relation to the implication of these fire regimes to wildfire management (line 433). The importance of temperature is considered in lines 401-412 of the Discussion.
14) When applying for the classification tree model, the importance of each of the parameters is determined based on a linear or non-linear relation? How does that affect the identification of the most important factors?
R: Classification trees are produced by successive binary partitioning, or splitting, of the training data into a growing number of subsets (nodes). Each split is based on a binary condition, defined using the predictor variable (the splitter) that maximizes the homogeneity, or inversely, minimizes the impurity, of the two resulting nodes. In our case, this homogeneity was measured using the GINI criterion, which is based on squared probabilities of membership for each category of the dependent variable (i.e. each of the four fire regimes). GINI reaches its minimum (zero) when all cases in a node fall into a single fire regime.
Each split results in an improvement, which is calculated by comparing the homogeneity of the two resulting nodes with that of the original node. This improvement is attributed to the splitting variable. The importance of each variable for the overall classification procedure is based on the sum of the improvements in all nodes in which the variable appears as a splitter, weighted by the fraction of the training data in each node split (Steinberg, 2009).
We acknowledge that the criteria for quantifying the overall importance of each predictor variable is unclear in the manuscript. We will therefore integrate the above information into the Data Collection and Pre-Processing section, together with the additional reference below.
Steinberg, D. (2009). CART: Classification and Regression Trees. In X. Wu & V. Kumar (Eds.), TheTop Ten Algorythms in Data Mining (pp. 179–201). CRC Press.
Citation: https://doi.org/10.5194/egusphere-2022-342-AC2
-
RC2: 'Comment on egusphere-2022-342', Anonymous Referee #2, 18 Jul 2022
The topic of the present manuscript is of high importance for the fire community, being the rationale is well presented. However, the novelty of this work is poor. The manuscript’s findings are incremental and do not represent an advance in this field. Namely, the present manuscript is clearly part of a larger study, published in previous papers, with the same rationale, Bergonse et al. (2022) and Oliveira and Zêzere (2020).
Citation: https://doi.org/10.5194/egusphere-2022-342-RC2 -
AC1: 'Reply on RC2', Rafaello Bergonse, 20 Jul 2022
We thank the reviewer for the feedback and comments provided.
Regarding the novelty of the work, the present manuscript is indeed one of the results of a larger study, which includes Bergonse et al. (2022). Both works share a common rationale in that they share the same spatial analysis units and study area, the same three fire regime descriptors, and the same biophysical drivers. However, they have different objectives and analysis techniques. In Bergonse et al. (2022), relations between the biophysical drivers and each of the three fire regime descriptors were separately analysed using ordinal regression equations. Although the study area was assumed to have a single fire regime, the spatial patterns shown by the three fire regime descriptors suggested the existence of distinct regimes.
The current work builds upon the previous results, in that we employ cluster analysis to explicitly identify and then characterize the different fire regimes within the study area, something which has not been done before. We subsequently apply a classification tree model to assess the capacity of the different biophysical drivers to discriminate between the four fire regimes defined. After interpreting the results, we then discuss the implications of the identified fire regimes and their drivers to wildfire management. This also is a completely novel outcome.
Being results of a common, ongoing research project, this manuscript and the previous article share a rationale and can certainly be considered complementary. Each of these works, however, presents distinct and novel results, which is why we feel this manuscript to the suitable for publication in this journal. We believe that the incremental findings presented are valuable and helpful to try understanding, progressively, the complexity of wildfires in Portugal.
We would also like to note that Oliveira and Zêzere (2020) was not a part of the research project mentioned above, having different study area, temporal scope, and analysis technique (random forest). It also employs as dependent variable only one of the three fire regime descriptors mentioned above. In fact, this paper has investigated the relation between the spatial distribution of burned area and different biophysical and social drivers for the parishes of the whole mainland Portugal, in a rather different scope than the one now presented.
Citation: https://doi.org/10.5194/egusphere-2022-342-AC1
-
AC1: 'Reply on RC2', Rafaello Bergonse, 20 Jul 2022
-
RC3: 'Comment on egusphere-2022-342', Anonymous Referee #3, 04 Aug 2022
This manuscript aims to characterize fire regimes in Central Portugal and investigate the degree to which the differences between regimes are influenced by a set of biophysical drivers. The authors used civil parishes as units of analysis and cumulative percentage of parish area burned, Gini concentration index of burned area over time, and area-weighted total number of wildfires over a reference period of 44 years (1975-2018). The authors used cluster analysis to aggregate parishes into groups with similar fire regime and a classification tree model to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups. The methods used seem to be suitable and the manuscript is nicely written. However, I have some criticisms, including some changes concerning the datasets used and the novelty of the present work, that should be addressed before considering the paper for publication in Natural Hazards and Earth System Sciences journal.
Major suggestions/comments:
Novelty of the work:
The results are fairly described and the discussion focused in a reduced number of previous publications, including two previous works of the same authors that exhibits strong similarities with the present work. The novelty (and need) of the present results of is not clearly addressed. The baseline of the present work in terms of data and methods is very similar to the previous two works. The data used is the same and the statistical approaches are slightly different, but very related with the previous ones. The main results are the same: the role played by LULC, slope and spring rainfall in fire behavior.
The present paper adds the classification in 4 FR for central Portugal. However, the FR classification is closely dependent of the data used. This leads to my following comment.
Datasets:
FR regime classification in strongly dependent of the historical data over the region. Therefore, the used of climate data than does not describe the last two decades, when we are facing a change in fire paradigm over Europe, with the occurrence of the so-called megafires, highlights the fragilities of the FR classification.
Besides the ‘old’ meteorological datasets, the higher fire intensity or severity of the observed fire behavior trends was not included in the FR classification. The authors used burned area, however the burned area inside a civil parish may not be a good indication or fire intensity; other parameters (available through remote sensing datasets) should be included.
The inclusion of the suggested datasets, considering the aim of the present work, will strongly improve the quality of the results, highlighting its novelty.
Slope:
“Topography was expressed by slope (80th percentile, in degrees), which can be expected to promote flame propagation”. Why using the 80th percentile and not 90th or 75th. Did you make a sensitivity analysis for this choice? Did the authors include elevation information? Why?
Rainfall:
“RFAJ was calculated from monthly rainfall data obtained from the Worldclim database (1970-2000)”. The authors present an assessment for 44 years (1975-2018) and one of the crucial datasets used is only characterizing half of the period. The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for precipitation data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
…”in the form of raster maps of approximately 30 seconds (about 1 km resolution), which were resampled to a 25 m pixel”. How was done the resampling? Which co-variates were used to do resampling? And, why to do the downscale if the data is further aggregated at parish level?
Temperature:
The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for temperature data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
Net Primary Productivity (NPP):
Please consider to use the most recent version of the NPP product (2000-Present). Alternatively, consider to use the Climate Data Record of NDVI (annual mean or sum) available since 1981 to present (https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C01558).
Lines152-160
This paragraph seems to be out of order. Please consider to reorganize the paragraph (before NPP paragraph).
Lines161-165
The different periods considered for the different datasets could have a strong impact on the results, as the spatial patterns for precipitation and temperature in the last 30 years of the XX century may have strong differences in compared with NPP in the first 20 years of the XXI century.
With the aim to have a fire regime description that really reflects the recent vegetation, climate and fire behavior trends, I strongly suggest to include: a) Temp and Precip from ERA5 from 1979-Present; b) NDVI from 1981-Present. Therefore, the period of analysis would be 1981-Present (41 years).
Lines 360-361 “This is confirmed by FR3’s low Net Productivity Ratio, …, which is indicative of a relatively reduced forest cover.” Please provide a reference or provide the analysis that allow this statement (or remove the sentence)
4.2.3 FR2: Please provide a better characterization of this FR. As it is, seems that this fire regime is not a separate fire regime and may indicate that the classification in 4 fire regimes is not the adequate.
Lines 401-408: The less clear relation with summer temperature is probably related with less adequate database used, that does not reflect the temperature changes in the last two decades. Please check the impact of use of the suggested dataset for meteorological parameters.
Lines 409-412: Is the statement supported by the NPP results of this work? Please explain.
Minor
Lines 90-95: changed format.
Citation: https://doi.org/10.5194/egusphere-2022-342-RC3 -
AC3: 'Reply on RC3', Rafaello Bergonse, 15 Sep 2022
Referee comment:
https://editor.copernicus.org/index.php?_mdl=msover_md&_jrl=778&_lcm=oc158lcm159n&_ms=103240&salt=240954201780718781
This manuscript aims to characterize fire regimes in Central Portugal and investigate the degree to which the differences between regimes are influenced by a set of biophysical drivers. The authors used civil parishes as units of analysis and cumulative percentage of parish area burned, Gini concentration index of burned area over time, and area-weighted total number of wildfires over a reference period of 44 years (1975-2018). The authors used cluster analysis to aggregate parishes into groups with similar fire regime and a classification tree model to assess the capacity of a set of potential biophysical drivers to discriminate between the different parish groups. The methods used seem to be suitable and the manuscript is nicely written. However, I have some criticisms, including some changes concerning the datasets used and the novelty of the present work, that should be addressed before considering the paper for publication in Natural Hazards and Earth System Sciences journal.
Major suggestions/comments
We thank the reviewer for the overall positive feedback and the pertinent suggestions provided. We have numbered each point and respond to it in detail below.
1) Novelty of the work:
The results are fairly described and the discussion focused in a reduced number of previous publications, including two previous works of the same authors that exhibits strong similarities with the present work. The novelty (and need) of the present results of is not clearly addressed. The baseline of the present work in terms of data and methods is very similar to the previous two works. The data used is the same and the statistical approaches are slightly different, but very related with the previous ones. The main results are the same: the role played by LULC, slope and spring rainfall in fire behavior.
The present paper adds the classification in 4 FR for central Portugal. However, the FR classification is closely dependent of the data used. This leads to my following comment.
R: Regarding the novelty of the work, the present manuscript is one of the results of a larger study, which includes Bergonse et al. (2022). Both works share a common rationale in that they share the same spatial analysis units and study area, the same three fire regime descriptors, and the same biophysical drivers. However, they have different objectives and analysis techniques. In Bergonse et al. (2022), relations between the biophysical drivers and each of the three fire regime descriptors were separately analysed using ordinal regression equations. Although the study area was assumed to have a single fire regime, the spatial patterns shown by the three fire regime descriptors suggested the existence of distinct regimes.
The present work builds upon the previous results, in that we employ cluster analysis to explicitly identify and then characterize the different fire regimes within the study area, something which has not been done before. We subsequently apply a classification tree model to assess the capacity of the different biophysical drivers to discriminate between the four fire regimes defined. After interpreting the results, we then discuss the implications of the identified fire regimes and their drivers to wildfire management. This also is a completely novel outcome.
Being results of a common, ongoing research project, this manuscript and the previous article can be considered complementary. Each of these works, however, presents distinct and novel results, which is why we feel this manuscript to the suitable for publication in this journal. We believe that the findings presented are valuable and helpful towards understanding the complexity of wildfires in Portugal as each small scientific step adds to the knowledge we very much need to deal with such issues.
In the manuscript, the relations between this and the previous work are made explicit in the final part of the Introduction (lines 57 and following).
We would also like to note that the other work mentioned by the reviewer (Oliveira and Zêzere, 2020) was not a part of the research project mentioned above, having different study area, temporal scope, and analysis technique (random forest). It also employs as dependent variable only one of the three fire regime descriptors mentioned above. In fact, this paper investigated the relation between the spatial distribution of burned area and different biophysical and social drivers for the parishes of the whole mainland Portugal, in a rather different scope than the one now presented.
2) Datasets:
FR regime classification in strongly dependent of the historical data over the region. Therefore, the use of climate data than does not describe the last two decades, when we are facing a change in fire paradigm over Europe, with the occurrence of the so-called megafires, highlights the fragilities of the FR classification.
Besides the ‘old’ meteorological datasets, the higher fire intensity or severity of the observed fire behavior trends was not included in the FR classification. The authors used burned area, however the burned area inside a civil parish may not be a good indication or fire intensity; other parameters (available through remote sensing datasets) should be included.
The inclusion of the suggested datasets, considering the aim of the present work, will strongly improve the quality of the results, highlighting its novelty.
R: The issue of the temporal limitations of the climate data is considered in detail below, in comment 4. However, we would like to underline that, in accordance with the adopted fire regime definition (lines 33-34 of the manuscript), the fire regimes were described based on the consequences of fires in terms of burned area through time, and not on the meteorological context in which these fires take place. The characterization of fire regime is based on burned area data for the 44-year period between 1975 and 2018 and is therefore not subject to any fragility derived from climate data limitations. Climate data were subsequently used as possible biophysical factors influencing fire regime, and it is in relation to this aspect of the work that the temporal limitations of the climate data indeed constitute a fragility. This limitation is acknowledged in the Data Collection and Pre-Processing section (lines 161-164) and will be further highlighted in a subsection to be included in the Results and Discussion section, focusing on the uncertainties and limitations of this work.
On a sidenote regarding the properties of the fire regime and their possible change between the period encompassed by the climatic data (1975-2000) and the later years, the following experiment was made. We created a new variable describing the cumulative percentage of parish area burned (CPAB) between 1975-2000 and ranked all study parishes by their value in this variable. We then ranked all parishes as to their value in the CPAB used in the article (i.e., encompassing the whole study period 1975-2018). We then calculated the Pearson correlation coefficient between the two ranked variables, obtaining an R of 0.895, significant at the 0.01 level. This result shows that the relative positions of the different parishes in terms of cumulative percentage of area burned are quite similar, regardless of the period considered. Although we tried this only for this fire regime descriptor, this result strongly suggests that the fire regimes among the studied parishes show a similar behaviour whether we limit the analysis to the climate-data covered period or to the whole 44-year period used in the article.
Regarding the issue of wildfire intensity/severity, fire regimes can be described with greatly varying degrees of complexity. We purposefully employed a simple, straightforward approach, expressing it with three indicators that can be extracted from freely available annual burned area maps, and therefore easily reproduced in other study areas. We agree that severity is an important aspect of fire behaviour that may not be adequately expressed by burned area alone, as are others such as the characteristics of the largest, relatively infrequent fires (which would include the so-called megafires). We will refer to these aspects of fire regime in a future “Uncertainties and Limitations” subsection of the Results and Discussion section, to inform future studies. We believe, however, that the focus on a simpler approach and the absence of these other datasets does not hinder the usefulness or novelty of the work, considering its objectives.
3) Slope:
“Topography was expressed by slope (80th percentile, in degrees), which can be expected to promote flame propagation”. Why using the 80th percentile and not 90th or 75th. Did you make a sensitivity analysis for this choice? Did the authors include elevation information? Why?
R: The set of 12 biophysical variables employed are derived from our previous results in Bergonse et al. (2022), as stated in lines 109-110 of the manuscript. In the referenced article, we initially adopted both slope and elevation as potential biophysical controls, using percentiles 50, 75, 80, 90 and 95. During a multicollinearity analysis, all percentiles were shown to be strongly intercorrelated. We thus chose to keep those more strongly correlated with the remaining percentiles in the same group, leading to the adoption of the 80th percentiles of slope and altitude. Altitude was eliminated further along the multicollinearity analysis process, as its Variance Inflating Factor showed it can be expressed as a linear combination of other variables in the dataset. This process is described in the Data Analysis section of Bergonse et al. (2022).
4) Rainfall:
“RFAJ was calculated from monthly rainfall data obtained from the Worldclim database (1970-2000)”. The authors present an assessment for 44 years (1975-2018) and one of the crucial datasets used is only characterizing half of the period. The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for precipitation data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
…”in the form of raster maps of approximately 30 seconds (about 1 km resolution), which were resampled to a 25 m pixel”. How was done the resampling? Which co-variates were used to do resampling? And, why to do the downscale if the data is further aggregated at parish level?
Temperature:
The data used is not representative of the fire regimes in Portugal, namely considering the fire behavior in the XXI century. Please use an alternative database for temperature data that characterize the entire period evaluated, e.g. ERA5 (1979 to present), or alternative change the period of analysis to 1975-2000.
R: We will focus on the issue of the climate dataset first, and then respond to the comment on the resampling process.
In the manuscript, we acknowledge the disparities between the periods used to characterize the fire regimes and the biophysical drivers in lines 161-165 and in Table 1. We thank the reviewer for the suggestion to use the ERA5 dataset to overcome this issue, which we have investigated. Unfortunately, the spatial resolution of ERA5 makes it too coarse to be applicable to a study on such a detailed scale as the one we employ. ERA5 is made available with a 0.25-degree resolution, which translates to a pixel of approximately 24.8 km. An overlay between an ERA5-derived map and the limits of our study parishes shows that each pixel comprises multiple complete parishes within it, making the ERA5 dataset too generalized for application in this study. In comparison, the Worldclim dataset used has a resolution of approximately 1000 m, which makes it suitable for our scale of analysis. A solution to this issue would be to limit the study to the period 1975-2000, as suggested. However, this would entail a similar problem with the land-use data, which only begins in 1990 (1995 in the cases of two specific variables) (as shown in Table 1).
The use of these imperfectly overlapping datasets, imposed by the unavailability of suitable data, implies the assumption that all are representative of the long-term, general fire regimes and biophysical factors that have characterized the study area within the last four decades. We agree that record fires, such as those seen recently, are important to understand the dynamics of fire in different contexts. However, they are not so important in relation to our approach, as they would detract from the general tendencies of the fire regimes we wish to characterize, moreover when our approach is based on annual burned area data and not on the characteristics of individual fires.
The resampling of the climate data maps was performed using ArcMAP’s Resample tool, using nearest neighbour assignment. This software was used for all spatial analysis operations, as stated in lines 166-7 of the manuscript. The resampling was done to minimize generalization in association to the Zonal Statistics tool used to calculate the mean values for the pixels in each parish. The following example can clarify the rationale behind the resampling. Let us imagine that our purpose is to calculate the mean temperature during the summer months for each of a set of parishes. To do so, we will use the vector map with the parish limits, and a raster map with the variable of interest, that is, temperature during the summer months. Let us assume this raster map has 1000-m pixels. If one of the parish polygons partially overlays a 1000-m pixel without covering its centre, the value of this pixel will be ignored in the calculations, i.e., the mean value calculated for that parish will in practice ignore a part of the surface of the parish. If, however, the raster map has a 25-m pixel, a much smaller part of each parish will suffer from this problem, as each polygon will encompass a much greater number of smaller cells, which will fill its area much more fully.
5) Net Primary Productivity (NPP)
Please consider to use the most recent version of the NPP product (2000-Present). Alternatively, consider to use the Climate Data Record of NDVI (annual mean or sum) available since 1981 to present (https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C01558).
R: The newer version of the MODIS NPP dataset (061) is indeed more up to date, as it extends to the present day. However, regarding our study period (1970-2018, the period used to characterize the fire regimes) the newer MODIS dataset includes only four more years (2015-2018), which are unlikely to alter the general, long-term tendencies in which this article is focused.
Regarding the NDVI dataset from NOAA, we thank the reviewer for the suggestion, which we investigated. The data is provided with a resolution of 0.05 degrees. This translates to a pixel of approximately 2946 m, which is much coarser than the 500-m NPP dataset we employed, even though it could still fit with our analysis. There are always options to make when it comes to the data integrated; in this case, we believe the NPP data is a reasonable choice considering the long-term approach applied, which does not require daily data. We will, nevertheless, make reference to the NOAA dataset in a future Uncertainties and Limitations subsection, to be inserted into the Results and Discussion section. This will inform future studies and research projects.
6) Lines 152-160
This paragraph seems to be out of order. Please consider to reorganize the paragraph (before NPP paragraph).
R: The paragraph will be repositioned as suggested.
7) Lines 161-165
The different periods considered for the different datasets could have a strong impact on the results, as the spatial patterns for precipitation and temperature in the last 30 years of the XX century may have strong differences in compared with NPP in the first 20 years of the XXI century.
With the aim to have a fire regime description that really reflects the recent vegetation, climate and fire behavior trends, I strongly suggest to include: a) Temp and Precip from ERA5 from 1979-Present; b) NDVI from 1981-Present. Therefore, the period of analysis would be 1981-Present (41 years).
R: We thank the reviewer for the suggestions. The issues regarding the climate dataset and the NPP dataset are considered, respectively, in the responses to comments 4 and 5.
8) Lines 360-361
“This is confirmed by FR3’s low Net Productivity Ratio, …, which is indicative of a relatively reduced forest cover.” Please provide a reference or provide the analysis that allow this statement (or remove the sentence)
R: In our previous work, it was observed that NPP was inversely correlated to the percentage of the area of each parish covered with shrubland, but positively correlated to the percentage of eucalyptus forests, pine forests and forests of invasive species (please see Bergonse et al., 2022, supplementary table S.1). We will reference this work at this point.
9) 4.2.3 FR2
Please provide a better characterization of this FR. As it is, seems that this fire regime is not a separate fire regime and may indicate that the classification in 4 fire regimes is not the adequate
R: Throughout section 3.1, our choices regarding the clustering process are described and justified. Figure 2 shows that a 3-cluster solution would indeed express, in a more synthetic way, the major contrasts in fire regime across the study area (as stated in lines 262-263). As we also note, cluster 2 always clearly occupies an intermediate position, regardless of the choice between 3 and 4 clusters (Figure 3 and lines 250 and 255-7). We then justify the option for a four-cluster/fire regime solution based on the practical implications of the fourth cluster (a subset of what was cluster 2 in the 3-cluster solution) regarding wildfire prevention and suppression (as stated in lines 263-266).
We understand the issue raised by the reviewer, in that, when reading the discussion of each cluster in section 4.2, a reader may at first not see the justification for keeping cluster 2, a cluster that has apparently no distinguishing features. However, we feel that such a doubt could only arise from a partial reading of the article, not only because cluster 2 is present in both 3- and 4-cluster solutions, being in either case relevant regarding fire regime variability in the study area (as shown in Figure 2), but also because it possesses intermediate characteristics in both clustering solutions. Indeed, within our study area, this cluster/fire regime it distinguished by this intermediate character.
10) Lines 401-408
The less clear relation with summer temperature is probably related with less adequate database used, that does not reflect the temperature changes in the last two decades. Please check the impact of use of the suggested dataset for meteorological parameters.
R: We considered the issue regarding the climate dataset in our response to comment 4. Although we consider different possible explanations for the results obtained (lines 401-412), we agree with the reviewer in that there may exist a relation between the limitations in temporal extent of the climate dataset and our results. This will be made explicit in the “Uncertainties and Limitations” section that will be incorporated into the Results and Discussion section.
11) Lines 409-412
Is the statement supported by the NPP results of this work? Please explain.
R: The distribution of NPP values among the four FRs (Figure 6-G) shows that the lowest value is associated to FR3, which has the highest Cumulative Percentage of Area Burned (CPAB) of all FRs. This low NPP is in agreement with the highest percentage of area covered by fire-prone, quickly regenerating shrubland, also shown by FR3 (Fig. 6-A). This would support the notion that fuel is the factor controlling burned area in this FR, instead of summer temperature (which has its lowest value in FR3).
Summer temperature has its highest values in FRs 1 and 4. The highest NPP values also occur in these two FRs, in accordance with the highest percentages of area covered by eucalyptus forest, also shown by these two FRs (Fig.6-E). These values, suggesting the combination of available forest fuels and high summer temperatures, only translate into extensive burned area in the case of FR4, which has the second highest CPAB of all FRs (Fig.3-B). In the case of FR1 (with the lowest CPAB of all FRs), factors other than fuel availability seem to constrain burned area. Among the biophysical drivers considered in this work, examples could be the high percentage of agricultural area (fragmenting fuels and implying higher human presence and therefore quicker detection and response) (Fig.6-D), the lowest slope of all four FRs (slowing flame propagation; notably, FR4 has the highest slope values) (Fig. 6-C), and the relatively high fragmentation of fuel (in comparison to FR4) (Fig. 6-H). Additionally, FR1 occurs mostly along the coastal sector, which has higher rates of urbanization and population densities than the inland parishes. It is likely that these will promote faster detection and better response capabilities on the part of the authorities in case of ignition.
In sum, NPP results seem to support the idea that fuel availability constrains to a degree the effect of summer temperatures. However, the role of NPP must be considered together with the effects of other factors, as fire regimes result from the interaction of multiple drivers.
Minor
12) Lines 90-95
Changed format.
R: The paragraph will be correctly re-formatted.
-
AC3: 'Reply on RC3', Rafaello Bergonse, 15 Sep 2022
Data sets
Fire regime parameters and potential biophysical fire regime drivers - NUTS2 Centro, Portugal Bergonse, Rafaello; Oliveira, Sandra; Zêzere, José Luís; Moreira, Francisco; Ribeiro, Paulo Flores; Leal, Miguel; Lima e Santos, José Manuel https://doi.org/10.5281/zenodo.6552804
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
444 | 237 | 44 | 725 | 27 | 29 |
- HTML: 444
- PDF: 237
- XML: 44
- Total: 725
- BibTeX: 27
- EndNote: 29
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1