the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Mar 2024
The Global Forest Fire Emissions Prediction System version 1.0
Abstract. The Global Forest Fire Emissions Prediction System (GFFEPS) is a model that estimates biomass burning in real time for global air-quality forecasting. The model uses a bottom-up approach, based on remotely-sensed hotspot locations and global databases linking burned area per hotspot to ecosystem-type classification at a 1-km resolution. Unlike other global forest fire emissions models, GFFEPS provides dynamic estimates of fuel consumption and fire behaviour based on the Canadian Forest Fire Danger Rating System. Combining forecasts of daily fire weather and hourly meteorological conditions with a global land classification, GFFEPS produces fuel consumption and emission predictions in 3-hour time steps (in contrast to non-dynamic models that use fixed consumption rates and require collection of burned area to make post-burn estimates of emissions). GFFEPS has been designed for use in near-real-time forecasting applications as well as historical simulations for which data are available. A study was conducted running GFFEPS through a six-year period (2015–2020). Regional annual total smoke emissions, burned area and total fuel consumption per unit area as predicted by GFFEPS were generated to assess model performance over multiple years and regions. The model distinguished grass-dominated regions from forested, while also showed high variability in regions affected by El Niño and deforestation. GFFEPS carbon emissions and burned area were then compared to other global wildfire emissions models, including GFAS, GFED4.1s and FINN1.5/2.5. GFFEPS estimated values lower than GFAS/GFED (80 %/74 %), and estimated values similar to FINN1.5 (97 %). This was largely due to the impact of fuel moisture on consumption rates as captured by the dynamic weather modelling. An effort is underway to validate the model, with further developments and improvements expected.
- Preprint
(3378 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 03 May 2024)
-
RC1: 'Comment on gmd-2024-31', Anonymous Referee #1, 14 Apr 2024
reply
Anderson et al. describe the development of the existing Canadian Forest Fire Emissions Prediction System (CFFEPS) into the Global Forest Fire Emissions Prediction System (GFFEPS). The goal is to provide new input data for air quality modelling, taking advantage of the sophisticated way meteorological conditions are taken into account in CFFEPS. I have a number of minor comments (see below) and a few major comments. Addressing the major comments requires a major overhaul of the study and the way it is described.
The first major comment is that in its current state the paper describes the model and output, including a comparison against other inventories. While most literature (Ramo et al., 2021 [https://www.pnas.org/doi/full/10.1073/pnas.2011160118]; Chen et al., 2024 [https://essd.copernicus.org/articles/15/5227/2023/essd-15-5227-2023.html]) points towards more burned area and higher emissions than what has been estimated in the past decade, this paper finds lower numbers. The authors mention that larger validation efforts will follow but I feel they should be part of this paper. I don’t feel it is justified to publish a methodological paper that needs to be revised shortly after, it would be better to include the validation in this paper and modify the approach if needed. Related to that is that the authors need to show that their approach has advantages over existing datasets. For example, does the more dynamic way of modelling fire characteristics lead to better results than FINN (or any other dataset) when comparing to field data (fuels, burned area) or to large-scale top-down constraints (for example Zheng et al., 2023 [https://www.science.org/doi/10.1126/science.ade0805])?
Second, I find the format confusing and the paper is difficult to read. Chapter 4 describes the methodology, but the previous sections also detail about different methods. For a reader it is difficult to keep track of whether CFFEPS or GFFEPS is being discussed. Also, it takes a long time before the Seiler and Crutzen equation is mentioned, that could be done early on. How about
1) Introduction, including the background of CFFEPS and other inventories
2) Methods and datasets with subsections of burned area, fuel, and other parameters. The part based on CFFEPS can be described briefly with reference to CFFEPS papers
3) Results
4) Discussion
The third one is the land cover data. GLC is used, and developed for the year 2000. In almost 25 years many regions in the world have undergone changes in land cover, these should be incorporated to not misclassify fuels and other fire characteristics
The fourth is somewhat related. If I understand correctly, you use lookup tables to assign fuels to each GLC class. In other words, fuels are not variable beyond their GLC class. I feel the science has progressed way beyond this and it would be good to have a spatial and temporal component for the fuels; they can vary a lot over space and time
Minor
L25-30: You compare a low fire year (2021) with regrowth estimates from a longer time period. In general “net” fire carbon emissions are higher than the 0.1 Pg C one would refer from the difference between the quoted 1.84 and 1.75 Pg C
L32: “Estimates show that between 2003 to 2017, biomass burning accounted for 1.68 to 2.27 Pg C yr-1 “. Well, that is one dataset and if the emissions were so wel known then this work would not be necessary. Please include the range of estimates
L34: please change the bullets to a narrative
L64: FRE is the time integral of FRP, not the other way around
L66: FINN does not use MCD64A1
L71: Burned area approaches are not fundamentally restricted by satellite-overpass times as what you label ‘top-down’ studies (in the literature top-down studies often refer to those that use atmospheric observations of for example CO to constrain emissions
L79: GFED also has spatially and temporally variable fuel loads and combustion completeness (also relevant for next paragraph)
L115: multiply -> multiplied
L117: “Fuel consumed is used to calculate total heat flux from the combustion process and then used to calculate plume injection height”. One would expect that a rate would be more useful here
L177: One would wonder whether using FRP would lead to better results than just hotspots
L193: Please consider introducing this equation earlier as it is also used in GFFEPS
L210: Think this Table can go by referring to Giglio et al
L230: In my opinion (see feel free to ignore) Figure 3 and Table 3 do not add much as it is basically unmodified input data. Figure 4 is useful
L269: That is a firm statement, can you back this up with references?
L290: Not clear what the map contributes, can be written down in words
Section 3.3: Maybe some of these parameters are useful to develop variability in fuels, as long as they can be constrained. Equations 2 and 3 seem very ad hoc without references
Section 3.4: There is a new burned area dataset that may be useful here to validate the Tier 1 approach (https://essd.copernicus.org/articles/16/867/2024/essd-16-867-2024.html)
L 400: Giglio et al. 2016 is about active fires, their 2018 paper is more relevant here
L 427: The VIIRS fire community must have a database with stationary sources to help cleaning up the dataset so you only capture landscape fires
Section 4.3. This section is titled combustion completeness but much of the text is about fuel consumption. It becomes increasingly difficult as a reader to follow your workflow, please consider a major overhaul of the structure of the paper.
Section 4.4: this is a novel aspect of this work, would be nice to see a comparison between your approach and the existing datasets (Akagi, Andreae)
L529: How nice would it be to see comparisons beyond Canada and the US
L564: To some degree it is expected that your values match GFAS and GFED as you use more or less the same burned area and fuel consumption numbers. Would be nice to show where you make a difference
L615: Looks we agree on the above
L616: GFED also has variable fuels. It makes me again a bit wondering about the added value of this work but that could be partly because it is often difficult to grasp the methodology. With a better structure, validation, and clearer description of the unique aspects of this work this could be mitigated
Citation: https://doi.org/10.5194/gmd-2024-31-RC1 -
RC2: 'Comment on gmd-2024-31', Anonymous Referee #2, 25 Apr 2024
reply
Anderson et al. set out to create a global version of the Canadian Forest Fire Danger Rating System. Overall, this is an interesting proof of concept; however, I believe the publication of this paper is premature. Please see specific comments below.
Major Comments
Section 3.2: Do the authors have to worry about double counting fires when all the VIIRS sensors are used? Was any preprocessing done on the fire data? Why did the authors not filter out the presumed non-vegetation fire using the type flag? The science quality dataset includes this information.
Lines 310 – 315: It seems the author’s primary justification for using GSI instead of NDVI is the ease of use because remote sensing data requires an extra step to mosaic the data together. How different would the results be using NDVI? Is there a more substantial scientific justification for using GSI instead of NDVI that can be added here?
Section 3.4: The authors may want to add some additional caveats including the fact that the FAO stats may not be accurate in countries where agricultural burning is illegal but widespread. For example, in Ukraine and Russia. Please see the following paper for further information: https://iopscience.iop.org/article/10.1088/1748-9326/abfc04
Furthermore, the Global Cropland Burned Area dataset (Hall et al., 2024) was recently released. It represents the cropland burned area within GFED5 (https://essd.copernicus.org/articles/16/867/2024/). I suggest using either this product or another product specifically designed to map agricultural burned area and compare some of the burned area statistics. The above-mentioned paper focused on Ukraine uses VIIRS active fires, so that is more in line with the author’s methodology.
Section 5.1: The GFFEPS model underestimates the burned area in BONA (two areas with large burned area scars) and TENA. The authors then go on to say that the R2 shows that the BA methodology is appropriate. Surely, the authors require an appropriate accuracy assessment to make this claim.
Line 569 and 579: The authors should compare their BA against GFED5 BA before making this claim since GFFEPS does not account for smaller fires and uses FAO stats for agricultural burning. Also, Africa is dominated by small fires in general, not just agricultural fires. Small fires include the smaller burned patches around larger burn scars, not just an actual small fire (which seems to be how the author interprets them based on lines 665 onwards).
Section 6: Why is there no accuracy assessment/ validation on the burned area product? Since you are using MODIS and VIIRS, the authors can use the BARD dataset (https://edatos.consorciomadrono.es/dataset.xhtml?persistentId=doi:10.21950/BBQQU7). I don't recommend publishing a new product paper without an adequate burned area validation assessment since that is the primary input into the emissions calculations.
Minor comments:
Title: Why is the product called a “Forest Fire” product when the authors are mapping burning in all land cover types?
Line 16: change to “showing”
Line 64: FRE is the time integral of FRP
Lines 74 - 79: The GFED5 Burned Area product has incorporated the GloCAB data (Hall et al., 2024) which provides a cropland burning-specific dataset. https://essd.copernicus.org/articles/16/867/2024/
Line 115: change to “multiplied”
Section 3.1: The justification for using the GLC2000 is weak, especially now that it is almost 25 years old. Have the authors run a sensitivity analysis with other land cover datasets to see how well the GLC2000 dataset has held up in recent years?
Line 269: “sufficiently complete” is quite a strong statement. VIIRS does not include the morning overpass compared to MODIS so the fire location data is already missing a large number of fires. I would remove the last portion of that sentence. It is also worth mentioning that since you are using the active fire product that you only have the afternoon snapshot of fire pixels as opposed to MODIS which has morning and afternoon.
Figure 6 and Figure 7 captions: It would be helpful for readers unfamiliar with these indices to have a brief description of the meaning of the value in the caption.
Line 374: There is a newly released dataset that compiles all the crop-specific emission coefficients from the literature. It is available here: https://doi.org/10.5281/zenodo.7013656
Line 387: Determining the other small fires as “inconsequential” is not an adequate justification for not developing a methodology to improve the representation of small fires, especially given the numerous papers showing how many small fires there are on the landscape. I suggest rephrasing.
Line 427: Why not just remove the persistent sources?
Citation: https://doi.org/10.5194/gmd-2024-31-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 355 | 101 | 23 | 479 | 15 | 19 |
- HTML: 355
- PDF: 101
- XML: 23
- Total: 479
- BibTeX: 15
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1