This is a well-written review of a valuable data set and includes an interesting assessment of the data regarding spatial and temporal patterns of global fire emissions. The work also compares results to previously developed related datasets and reviews the value of this improved product for a variety of uses for air quality, climate studies, and more. The presentation is generally very good. I have only a few suggested technical edits and a the like, listed below.

 Data and Method (page 4)

The overview section is very important for this work because it serves as an opportunity to briefly describe the approach, so readers will not be obligated to read the full accounting of methods if it is not required for their purposes. A few minor edits will make this section even more helpful:

Line 113: This paragraph uses the word “higher” without reference to what it is “higher” than. I think it is higher than MODIS resolution, etc.. In general, when comparative words are used, there should be a clear connection to what they are referring to. I suggest a general re-wording to make this more clear.

Line 116: The 0.25 X 0.25 deg grid cell size of the dataset is mentioned here for the first time. The phrasing seems to assume that the reader knows that GFED products are of this spatial scale. I think it would be beneficial to be more explicit regarding the spatial scale and temporal time step in the opening paragraph of this section or in the introduction, if that seems more appropriate. As with the previous comment, revising the wording will remedy this issue.

Line 129 (Eq. 1): The equation is given here out of order. Typically, equation variables are explicitly called out after the equation is introduced, rather than before and after, as done here. And, the left side equation variable is not explicitly defined until 2 paragraphs later (line 140). As a reader, it took some work going back and forth from prior to and after the equation to know the variables’ meanings/definitions.

Line 153: If I am correct and following this properly, you can reference back to Eq. 1 in this location to clarify the relationship between Eq 1 and Eq2. e.g:  “…over all vegetation types (Eq. 1),…”

Line 155 (Eq. 2): Again, the Eq 2 left side term is not defined. You could reference it directly in the previous paragraph and call it out explicitly: “The total GFED5 burned area for each 0.25deg grid cell during the MODIS era (2001-2020; BA_GFED5(x,t)) was estimated …” The subsequent sentence referencing section 2.4 is OK, but the reader is still left without an explicit definition of each term on the right side of the equation.

A revision of this short section will make this paper accessible to a wide audience for this important dataset.

Discussion (page 21):

Your discussion section is excellent. Informative, thought-provoking, and concise. I have two comments:

First: Line 646: In this section you use the terms “top down” and “bottom up”. I strongly suggest this terminology be changed (I would like the community to stop using it), as it is confusing and not fully consistent.  For example, in this paper you use “top down” to mean estimation of aerosols based on atmospheric sensing retrievals from AOD (lines 654-655) and CO retrievals using MOPITT (lines 667-668). “Bottom up” is not explicitly defined, but I think the way it is written in the text equates “multiplicative approach of Seiler and Crutzen” (line 647) to “bottom up” as well as “satellite and in-situ measurements” (line 651), which are based on fuels and fuel combustion metrics found either via on-site measurements or fire energy methods (I am assuming). So, your definitions are fairly clear, with “bottom up” being an approach that measures components that drive emissions either directly or indirectly and “top down” being emissions estimated from atmospheric observations.

The issue is that other uses of the terms are divergent from this. Usually the “accounting” approach (Seiler & Crutzen, AKA “bottom up”) is clear and consistent, with a variety of ways to come up with the amount of material that is combusted. However, some use “top down” in defining any use of remote sensing, including FRP-based combustion estimates e.g., Wiggins et al. 2021.  I have even seen GFED being referred to as a “top down” approach because it uses remote sensing for burn area estimation. To avoid this confusion, I suggest dropping the terminology and adopting what is used for general emissions estimation for sources other than biomass burning.

In emissions accounting outside of biomass burning, the community would use “activity-based methods” for “bottom up” accounting-style approaches. Your “top down” approaches would be called “atmospheric approaches”. And any that combine these would be “hybrid approaches”, which would include any data assimilation or modeling methods. A recent NASEM report on GHG emissions accounting includes this generalized terminology  (see Page 3 of this report: https://nap.nationalacademies.org/catalog/26641/greenhouse-gas-emissions-information-for-decision-making-a-framework-going)

If you do choose to keep the terminology. Please work on the wording in this section to be clear on the definitions of top down and bottom up.

Second: Line 754: I am not sure what is meant by “cumulative effects of climate change” is it just changes in cloud cover? What other climate changes might make fire observations less now than before?

Tables and Figures:

A few minor notes:

Table 4 and 5 show different numbers for Total burn area trend for BONA – a typo, I think. (-1.91 in Table 4; -0.91 on Table 5)

I found Table 5 a bit confusing due to the order of the columns, but this is very minor. I would have put the full period GFED5 to the left (first column). The others are “paired” for a meaningful comparison, so maybe vertical lines that show the way to compare would help. 

GENERAL COMMENTS

The manuscript presents the updated burnt area estimates in the

latested version of the well-established GFED inventory, GFED5. In

order to fill the entire time series of 1997-2020 with monthly values

and 0.25deg resolution for the entire globe, burnt area observations

from the MODIS satellite instruments along with active fire

observations from the MODIS, VIRS and ATSR satellite instruments are

being used. A new methodology is employed compared to earlier work:

The omission and commission errors of the MODIS burnt area products

with 500m spatial resolution are corrected based on a comparison to

high-resolution (20-30m) observations of burnt area by the Landsat and

Sentinel-2 satellites at several reference sites and time

periods. Using a very large number of fitting parameters, the GFED5

burnt area time series is thus anchored to these high-resolution

observations of burnt area.

The data is highly relevant for the scientific community and the

manuscript is overall well written and suitable for publication in

ESSD. The authors have presented the context of burnt area estimation

very thoroughly. However, the wider context of other research on

vegetation fires and their emissions should be given a bit more

comprehensively (and with more primary references); I recommend that

the more senior co-authors edit the manuscript, in particular the

Introduction and Discussion sections, in this respect.

Generally, many mathematical relationships are described in the text

instead of formulas. In my opinion, the manuscript would be clearer,

if more formulas were used.

SPECIFIC COMMENTS

The authors fit very many parameter to generate the final burnt area

product. That fact that this seems necessary has repercussions, e.g.

on the interpretation of the MODIS burnt area product. In order to

fully understand the implications on how much information is coming

from which input, I consider it necessary to be more explicit about

the number of fit parameters. My understanding is the following and I

think a discussion (possibly corrected where I misunderstood)

should be added to the manuscript:

For normal land cover types (12) in each GFED region (14) and each

tree cover bin (10), there are two parameters each. This yields 12 * 14 *

10 * 2 = 3360 parameters. Fig. 3 shows that about half of the tree

cover bins have no data, so the number of parameters for deriving

burnt are from the MODIS product for the normal land cover types is

closer to, say, 1700.

For periods with MODIS coverage by only one satellte, the set of above

listed parameters is fitted separately for Aqua and Terra. This yields

another 2 * 1700 = 3400 parameters.

For six different crop types, global conversion factors are derived

(6). For each of peatland burning and deforestaion fires (2), a single

scalar is derived and globally applied (2). These 6 + 2 = 8 parameters

are used to derive the corresponding burnt area from MODIS active fire

observations.

Judging by the vastly different numbers of parameters apparently

required to derive realistic burnt area from MODIS burnt area (1700 +

3400 = 5100) and MODIS active fire (8), admittedly for different fire

types, it appears that MODIS active fire observations may contain more

information on burnt area than MODIS burnt area observations. I don't

really believe this, but I think the authors need to discuss it in

order to justify their use of MODIS burnt area observations.

For the pre-MODIS era, different scaling parameters of VIRS/ATSR

active fires to burnt area are used for each GFED region (14), each

dominant vegetation class (16) and each seasonal period

(3). Additionally, for each parameter, the goodness of fit is used to

decided whether to use a climatology, doubling the number of

parameters. This yields another 14 * 16 * 3 * 2 = 1344 parameters.

Additional parameters are described in lines 481-498. I struggle to

understand this paragraph exactly and recommend to rephrase it.

Overall, it seems to be necessary to use more than 5100 + 8 + 1344 =

6452 fit parameters to derive a realistic burnt area time series from

the satellite products of burnt area and active fires. Corrections

seem particularly necessary and difficult for burnt area

observations. This poses serious questions on the information content

of the satellite observations, the danger of overfitting and future

methodologies for burnt area estimation, for example whether machine

learning and inclusion of further data sources might be the

appropriate approach. The authors should discuss implications of this

for their product and future developments, including in Section 6.

The discussion, in particular on emission estimates (Section 4.1),

completely ignores well-established active fire-based inventories like

FINN, GFAS and QFED. This context is required here.

Concerning the statement on evidence for underestimation of fire

emission (line 649 ff), I think this is overly simplified, since there

are also cases (regions and chemical smoke constituents) that appear

to be overestimated. For particulate matter, the question of

underestimation depends heavily on whether "emissions" are considered

at the top of the flame or many kilometers downwind of the fire since

strong and quick formation of secondary organic aerosols happens

inbetween. Therefore, experience, for example in the development of

the global aerosol model of the Copernicus Atmosphere Monitoring

Service, has shown that much of the demostrated discrepancy between

bottom-up and top-down emission estimates can be attributed to

unresolved processed in the model. Generally, the discussion of the

discrepancies between bottom-up and top-down estimates must include a

discussion of other sources of error than from burnt area, i.e. from

fuel load, combustion completeness and emission factors.

The statements in Section 4.2 on Implications could, in my opinion,

generally have been made for the earlier version GFED4s as well. It

would be nice to show the added benefit of the new version; It's

obviously more accurate, but what implications does that have?

TECHNICAL CORRECTIONS

According to my understanding of the English language, the word

"section" should be capitalised when followed by a number, like

"figure". This applies throughout the manuscript, e.g. lines 107-110.

l. 567: Delete the comment in bracket, or explain what the parameter

is important for.

l. 653: We have even found a factor of 3.4, i.e. larger than the cited

range, cf. Kaiser et al. BG 2012.

Section 2.4.3 seems to be a repetition of previously given information

and could be deleted.

Fig. 8: If you plotted the x-axis in log scale, it would be possible

to the differences in the shape see diurnal cycles as well as

magnitude of the emissions.

TRANSLATE with x

English

 

TRANSLATE with

COPY THE URL BELOW

Back

EMBED THE SNIPPET BELOW IN YOUR SITE

 

Enable collaborative features and customize widget:

 

TRANSLATE with x

English

TRANSLATE with

COPY THE URL BELOW

Back

EMBED THE SNIPPET BELOW IN YOUR SITE

Enable collaborative features and customize widget: