The Landscape Fire Scars Database: mapping historical burned area and fire severity in Chile
- 1Center for Climate and Resilience Research (CR)2, Santiago, Chile
- 2Laboratorio de Ecología del Paisaje y Conservación, Departamento de Ciencias Forestales, Universidad de La Frontera, Temuco, Chile
- 3Image Processing Laboratory, Global Change Unit, University of Valencia, Valencia, Spain
- 4Industrial Engineering Department, University of Chile, Santiago, Chile
- 5Department of Environmental Sciences and Natural Resources, University of Chile, Santiago, Chile
- 6Instituto de Conservación, Biodiversidad y Territorio, Facultad de Ciencias Forestales y Recursos Naturales, Universidad Austral de Chile, Valdivia, Chile
- 7Center for Fire and Socioecosystem Resilience (FireSES), Universidad Austral de Chile, Valdivia, Chile
- 8Fundación Centro de los Bosques Nativos FORECOS, Valdivia, Chile
- 1Center for Climate and Resilience Research (CR)2, Santiago, Chile
- 2Laboratorio de Ecología del Paisaje y Conservación, Departamento de Ciencias Forestales, Universidad de La Frontera, Temuco, Chile
- 3Image Processing Laboratory, Global Change Unit, University of Valencia, Valencia, Spain
- 4Industrial Engineering Department, University of Chile, Santiago, Chile
- 5Department of Environmental Sciences and Natural Resources, University of Chile, Santiago, Chile
- 6Instituto de Conservación, Biodiversidad y Territorio, Facultad de Ciencias Forestales y Recursos Naturales, Universidad Austral de Chile, Valdivia, Chile
- 7Center for Fire and Socioecosystem Resilience (FireSES), Universidad Austral de Chile, Valdivia, Chile
- 8Fundación Centro de los Bosques Nativos FORECOS, Valdivia, Chile
Abstract. Achieving a local understanding of fire regimes requires high resolution, systematic and dynamic databases. Highquality information can help to transform the evidence into decision-making in the context of rapidly changing landscapes, particularly considering that geographical and temporal patterns of fire regimes and their trends vary locally over time. Global fire scar products at low spatial resolutions are available, but high-resolution wildfire data, especially for developing countries, is still lacking. Taking advantage of the Google Earth Engine (GEE) big-data analysis platform, we developed a flexible workflow to reconstruct individual burned areas and derive fire severity estimates for all reported fires. We tested our approach for historical wildfires in Chile. The result is the Landscape Fire Scars Database, a detailed and dynamic database that reconstructs 8,153 fires scars representing 66.6 % of the country’s officially recorded fires between 1985 and 2018. For each fire event the database contains the following information: (i) Landsat mosaic of pre- and post-fire images; (ii) the fire scar in binary format; (iii) the remotely sensed estimated fire indexes (NBR, RdNBR), plus two vector files indicating (iv) the fire scar perimeter and (v) the fire scar severity reclassification. The Landscape Fire Scars Database for Chile and GEE script (JavaScript) are publicly available. The framework developed for the database can be applied anywhere in the world, the only requirement being its adaptation to local factors such as data availability, fire regimes, land cover or land cover dynamics, vegetation recovery, and cloud cover.
Alejandro Miranda et al.
Status: final response (author comments only)
-
RC1: 'Review_Report_Miranda_et_al_2022', Anonymous Referee #1, 25 Apr 2022
Review Report: Miranda et al. 2022
The objective of the study is to reconstruct improved attributes of the historical fire database of Chile by leveraging Landsat imagery and Google Earth Engine (GEE) platform. The proposed workflow within GEE is well executed. The approach can semi-automatically generate different levels of fire products with homogeneous and unambiguous naming conventions. The introduction is well written. Accuracy assessment and validation of the generated products are also reasonable. Challenges of the approach and uncertainties of the results were acknowledged appropriately in the discussion. Figures and Tables are clearly presented. Availability of data sets and codes publicly is very nice.
The methodology and discussion sections are written well, however, it contains some ambiguities that could be clarified in the text.
- The fire area cut-off threshold of 10 ha seems too high provided the 30 m resolution of Landsat imagery.
- The thresholds used for generating fire masks are not clear. For instance, it is not clear whether a global or local threshold was used. Provided the fire scar mask was generated individually, it is appropriate to use local threshold and conduct sensitivity test. The threshold can vary from one fire scar to another based on the condition at the time of image acquisition. Similarly, the sensitivity of thresholds used for spectral filtering could be elaborated a bit.
- Chile seems to be maintaining an up-to-date historical fire point database (CONAF). A major setback for the workflow is likely transferability of the approach to the areas where there is a no or limited point database of historical fire. Alternatives or improvements that could be made to the workflow can be elaborated. For instance, an automation of the identification of thresholds (manually identifying threshold for each fire scar and severity can be challenging); could FIRMS database be useful at least starting 2000 for seeding the workflow?
- How does the Landscape Fire Scars Database fit into other larger/global fire databases (e.g., Global Fire Atlas)? The data set could be relevant for studies at global scale in addition to its use cases like fire management in Chile.
Specifics
L125: “...must be in the form of point data”. Is there any alternative when the fire database/record does not exist?
L125: “100 ha as seed value”, explanation is needed for the choice.
L125: The reason for drawing a buffer around a point is not clear. Is it for extracting attributes for spectral filtering?
L140: When was the ‘mosaic’ and ‘reduce’ function used? Does the approach decide which one to use based on some condition?
L150: The suitability of the use of RdNBR in Chile needs to be justified if it was used for the first time.
Table 1: Fix the abbreviation “Dnbr”. Value inside the square root is missing.
L170: Explicitly mention the “new criteria”. It appears at L180, but it could be clearer.
L175: “initial search distance” what was the value? Is it the same for all fire scars?
L180: “0.3 ha is retained” is confusing. Is it a fire scar? How is it different from the 10 ha threshold?
L210: “Closeness Index”, does it considers the shape similarity between reference and mapped fire scar?
-
RC2: 'Comment on essd-2021-467', Anonymous Referee #2, 27 Apr 2022
The paper contributes a new database for mapping historical burned areas and fire severity in Chile, called the Landscape Fire Scars Database. The manuscript itself is extremely thorough and comprehensive, and the dataset and codes are easily accessible for readers/users. There is a minor typo in line 229, Google Colab (not Collab). Overall, I found the engagement of the literature to be extensive and the motivation of the research to be well-justified. The conversation in the text about the benefits and limitations of the approach was written clearly and informatively. The figures were well made, and the GEE JavaScript codes ran without any issue. I was able to access the data without any issues as well. In general, I find this paper to be a wonderful contribution that will be particularly useful for other developing nations mapping similar landscape changes.
-
RC3: 'Comment on essd-2021-467', Anonymous Referee #3, 03 May 2022
[General comments]
The objective of this article is to present and make available the “Landscape File Scars Database”, a collection of historical fires in Chile built from officially reported fires and Landsat data. Availability of this data is an asset for researchers of various fields and to the fire-science community in particular. The latter will be glad to employ this source for the training or validation of BA regional or global products. The effort to create and make this database available it is a good example of a multi-institutional endeavour that countries developed to different degrees could focused on.
[Specifical comments]
Below I noted some points which could be addressed to improve the article (some of them are described more in detail later in the line-by-line comment section):
1) It is recommended a deeper review of the CONAF existing fire/BA data be added: a section detailing the BA data CONAF products before the creation of this database, so the contribution of the new database is better understood. This would help in the comparison of the number of fires reconstructed (Table 2) that suggests there is a more accurate source available. Also, it would be beneficial to add detailed information about the source of the point database employed to reconstruct the fires.
2) The methodology employed to map burned areas is not totally clear; even though the code is available for GEE users, a nice feature that ensures reproducibility. Firstly, it is not clear when ‘mosaic’ or ‘median’ reducers are employed to reduce imagecollections to images. If the ‘median’ were employed, this would reduce the noise but would soften the burned signal as well, an important feature for severity mapping. Secondly, the way an analyst define the threshold to map the burned area is not clearly defined - I guessed this is an interactive process based on visual assessment, however it is not clearly detailed. In addition, when describing the mapping process in GEE, it is convenient to address the difficulties encountered before writing them in the discussion section (for example, when there are neighbouring fire events, or when omission or commission errors are found in the mapping exercise). Finally, a statistical analysis of the RdNBR thresholds employed in different years/regions would be interesting, although maybe it is beyond the scope of this paper.3) Authors define a 10 ha limit for reconstructing the fire perimeters. This is a very big area, in excess of 100 Landsat pixels, and it would be convenient to discuss and justify this choice carefully.
4) In the validation process, a database of 194 fire scar perimeters has been employed as a comparison source (2015-2018 years), only 78 of them reconstructed. Omitting 60% of the fire scars gives the impression that many fires are not reconstructed for various reasons (and theoretically not because of a lack of images in those years). In addition, Table 2 shows that 66.6% of fires are reconstructed (hence, omission of 33%). The above should be clearly addressed and discussed, as it is an important limitation of the database.
5) It seems the accuracy of the assessment is not clearly established. When considering the 78 fires perimeters, I was expecting to see validation metrics comparing those perimeters database/ manually derived. I found only one paragraph (section 3.1) on the accuracy of the perimeters, plus the metrics shown are not clear (“global accuracy result is 0.79”). Section 2.3 refers to the methodology followed to carry out an accuracy assessment with the Closeness Index, however no results derived from this methodology for the 78 patches are shown, only some illustrative results in Figure 4. For completeness, I would also add error matrix derived metrics (user/producer accuracy or complementary omission/commission errors) in order to have similar metrics comparable to other research studies. Also convenient, I would add comparative information between your approach and coarse/resolution BA products (MCD64A1, FIRECCI products, GABAM; or the global wildfire dataset (https://doi.org/10.6084/m9.figshare.10284101) This would give an idea of the accuracy of the database to potential users.
6) The lack of availability of Landsat imagery is one of the sources of omission of the database. It would be interesting to follow up doing an imagery availability analysis across Chile through the years. Linked to this, the regional availability depending on the cloud cover mentioned in the text could be better contextualized.
About the data publicly available:
1) It was straight forward to download the database - I download it without any problem.
2) It is reassuring the quality control described in the manuscript warrants file-concordance between fires.
3) It would be helpful to upload the information within this database to the GEE servers so that users may be able to use and assess it directly (for example an asset with the perimeters and severity). I would emphasise in the manuscript the reason why this database is important. For example, to me, it is not clear why post- and pre-imagery is added, and why the NBR/ RdNBR is included (I can only guess most of the people will use the perimeters/Severity associated from the process).
[Line by line comments]
Line 56 -> I would add an additional reference to the fire_cci BA products (MERIS/ AVHRR / MODIS /OLCI based).
https://doi.org/10.1016/j.rse.2015.03.011 /
https://doi.org/10.1016/j.jag.2021.102473
https://doi.org/10.1016/j.rse.2019.111493
https://doi.org/10.5194/essd-10-2015-2018
https://www.mdpi.com/2072-4292/13/21/4295
https://doi.org/10.3390/rs13214295Line 60 (or 69)-> I would include a link to the GABAM database (Landsat, 30m) (although it is later referenced)
https://vapd.gitlab.io/post/gabam/
https://doi.org/10.3390/rs11050489
Line 97 -> It would be a good addition to upload the database to GEE and share the assets.
Line 106 -> Figure 1 does not correspond to the study site but to the methodological workflow
Line 112 -> the link does not point to a web page with burned area statistics but to a general web page to the CONAF
Line 119-> I’m not a GEE expert but I believe GEE native language is not Javascript (it is the most popular client library because of the code editor https://code.earthengine.google.com/ )
Line 120 -> Incorrect reference: it refers to Figure 1 and not Figure 2
Line 133 ->” We use the atmospherically corrected surface reflectance and orthorectified images from Landsat 5 (1984-2013), 7 (1999-) and 8 (2013-)” The collection and GEE tag could be indicatedLine 140 -> ‘Pixels of snow, clouds, and cloud shadows are excluded from each image on the basis of the pixel quality band provided by Landsat.’ I think these methodological details should be covered more in detail.
Line 141 -> “For each image collection, we applied either the mosaic or the median reducer function to get a unique image of the landscape conditions at moments as close as posible before and after a fire event.” This affirmation must be clarified. How do you get the closest burning date with the median reducer? In principle, employing the median reducer would decrease the burned signal strength
Line 151 -> “This index has shown better results in Mediterranean areas” Does this sentence refer to mapping burned areas or to burning severity?
Line 187: “the event’s severity is calculated from the RdNBR in a continuous raster format and categorized based on the ranges proposed by Miller and Thode (2007).” I think it would help writing down the ranges proposed in the manuscriptTable 1: RdNBR is not fully described (square root of what in the divider?)
Line 169 -> “Step (iv) involved the selection of the RdNBR index value for each wildfire that best captures the burned area based on visual interpretation. “ . Please reword and clarify what this sentence mean.
Line 202: Please complete the Brull reference “Brull, J.: Análisis de la severidad de los incendios de magnitud de la temporada de incendios forestales 2017-2018, 2018.”
Line 206: What was the spatial distribution of the evaluation samples? If the minimum size was 200 ha and 60% were not mapped by the database (78/194), the omission of the database seems high. It would be important to calculate the omission percentage both in number and area percentage.
Line 210: Why not use error matrix based traditional metrics like User/producer accuracy (or the complementary Omission, commission errors)? I understand the usefulness of the polygon-based comparison due to both source and validation being polygons but I believe having and omission/commission rate would be more significant.
Line 256: “Using the data for all 12,250 fires recorded by CONAF between 1985 and 2018 with a burned area greater than 10 ha,”. How is this information collected in CONAF? A description of the methods employed would be helpful.
Table 2: “R is the number of reconstructed fire scars contained in our database, and UR is the number of fire scars in the database that could not be reconstructed due to the unavailability of satellite images.” R and UR are not listed in the table. Are those the Yes/No columns? Please edit.
Line 260: A typical map of the number of Landsat scenes available across Chile would be interesting to understand changes through the years.
Line 270: “The total number of fires >0.01 ha exhibits a positive linear relationship with the total number of fires > 10 ha also recorded by CONAF between 1985 and 2018 (R2 = 0.86).” I cannot establish between which two variables this relationship is performed. First, I think it would be helpful to clarify what is the CONAF dataset. Then, the slope and intercept of the regression would add valuable information about the tendency of over/ under estimation.Line 272: “indicating that the distribution of the reconstructed data is regionally representative (Table 2, Figure 2)” -> please add the scatterplots as the reader may expect them.
Line 290: Fire scar evaluation: Line 298: “Nevertheless, the global accuracy result is 0.79” This is an important result but it is not easily understood: is it the aggregated 78 ‘Dnorm’ value? Please specify.
Line 300: Some of the limitations addressed here are new for the reader, I think they should be noted before in the results/methodology sections. For example, issues like having more than one fire event in neighbouring areas should have been addressed in the methodology section. The same applies for problems related to commission errors.Line 381: I would make clearer 5 days temporal resolution starts in 2017 with the second satellite, and that although some bands are at 10 m spatial resolution, critical bands accurate BA mapping like SWIR are at 20 m.
Line 390 “No evident pattern associated with the latitudinal or vegetation-type change was observed in applying the threshold value to identify scars”. It would be interesting to analyse the validity of the threshold values throughout various regions/years in Chile.Figure 2: Instead of using negative longitudes and latitudes, it is preferable to use South / West. For clarity, avoid adding the background shadows in the detailed maps.
Figure 3: I appreciate this is a plot to illustrate the computed variables, however I would include also information about the place/date of the fires illustrated.
Figure 4: I would define ‘Dnorm’ in the footnote for clarity
Alejandro Miranda et al.
Data sets
Fire Scars: remotely sensed historical burned area and fire severity in Chile between 1984-2018 Miranda, Alejandro; Mentler, Rayen; Moleto-Lobos, Italo; Alfaro, Gabriela; Aliaga, Leonardo; Balbontín, Dana; Barraza, Maximiliano; Baumbach, Susanne; Calderón, Patricio; Cardenas, Fernando; Castillo, Ivan; Gonzalo, Contreras; de la Barra, Felipe; Galleguillos, Mauricio; Gonzalez, Mauro; Hormazabal, Carlos; Lara, Antonio; Mancilla, Ian; Muñoz, Francisca; Oyarce, Cristian; Pantoja, Francisca; Ramirez, Rocío; Urrutia, Vicente. https://www.pangaea.de/tok/6dcc6e08241c5076ef6bff47bbe73014308d4881
Alejandro Miranda et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
307 | 107 | 15 | 429 | 6 | 6 |
- HTML: 307
- PDF: 107
- XML: 15
- Total: 429
- BibTeX: 6
- EndNote: 6
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1