Interactive comment on “ Climate , CO 2 , and demographic impacts on global wildfire emissions ”

Abstract. Wildfires are by far the largest contributor to global biomass burning and constitute a large global source of atmospheric traces gases and aerosols. Such emissions have a considerable impact on air quality and constitute a major health hazard. Biomass burning also influences the radiative balance of the atmosphere and is thus not only of societal, but also of significant scientific interest. There is a common perception that climate change will lead to an increase in emissions as hot and dry weather events that promote wildfire will become more common. However, even though a few studies have found that the inclusion of CO2 fertilisation of photosynthesis and changes in human population patterns will tend to somewhat lower predictions of future wildfire emissions, no such study has included full ensemble ranges of both climate predictions and population projections, including the effect of different degrees of urbanisation. 
 
Here, we present a series of 124 simulations with the LPJ–GUESS–SIMFIRE global dynamic vegetation–wildfire model, including a semi-empirical formulation for the prediction of burned area based on fire weather, fuel continuity and human population density. The simulations use Climate Model Intercomparison Project 5 (CMIP5) climate predictions from eight Earth system models. These were combined with two Representative Concentration Pathways (RCPs) and five scenarios of future human population density based on the series of Shared Socioeconomic Pathways (SSPs) to assess the sensitivity of emissions to the effect of climate, CO2 and humans. In addition, two alternative parameterisations of the semi-empirical burned-area model were applied. Contrary to previous work, we find no clear future trend of global wildfire emissions for the moderate emissions and climate change scenario based on the RCP 4.5. Only historical population change introduces a decline by around 15 % since 1900. Future emissions could either increase for low population growth and fast urbanisation, or continue to decline for high population growth and slow urbanisation. Only for high future climate change (RCP8.5), wildfire emissions start to rise again after ca. 2020 but are unlikely to reach the levels of 1900 by the end of the 21st century. We find that climate warming will generally increase the risk of fire, but that this is only one of several equally important factors driving future levels of wildfire emissions, which include population change, CO2 fertilisation causing woody thickening, increased productivity and fuel load and faster litter turnover in a warmer climate.


C7914
between different factors can be rather large and are important to consider, when quantifying the effect of a variable.The effects of different variables are not simply additive as assumed in eq. 4. The authors here not only negelect the effect of synergies but map them into the effect of fertilization (eq. 6).This procedure can bias the derived effect of CO 2 .The way the method is presented here is quite complicated.Following more closely the proposed method by Stein and Alpert (1993) could simplify the presentation and analysis and most importantly remove possible biases.See for instance Calvo and Prentice (2015) for a similar application of the method.

Reply:
We thank the reviewer for pointing out the method by Stein and Alpert.After careful consideration, we believe that an analogous setup of our study using the said methodology could be approximated by the following combination of time windows from existing transient simulations: be performed in addition to the existing 112.Since all require a 1000-year spin-up, the fact that some do not need to be to run until 2100 creates only insubstantial savings in computer time.But apart from a 250% increase in the computational effort, there is one more fundamental problem here that leads us to believe that employing the method by Stein and Alpert needs some substantial additional thought before it can be implemented in the present context: The problem is that the method lends itself most naturally to time slice experiments, such as those presented by Martin Calvo and Prentice (2015).But all simulations listed above are transient simulations that have not been run to steady state and therefore depend on initial conditions.They are therefore not directly comparable (cf.input variables listed above with and without *).And because of the transient setup, our sensitivities are derived from temporal changes between two periods (Equ.4), and not from the difference between time-slice experiments each run to steady state.
In fact, a better representation of the setup in our study than Equ. 4 would be the following: we run a simulation with only climate varying and ask what the change in emission is between two time periods each (1901-1930 vs. 1971-2000 and 1971-2000 vs. 2071-2100).Then we add varying CO 2 and ask the question how much the change in emissions between the two time periods is affect in this particular simulation.This change is attributed to the combination of climate and CO 2 .Finally, we also include varying population density and attribute the temporal change to the combined climate, CO 2 and population effect.Then we decompose the effects by subtracting the combined climate and CO 2 from the full effect, and the climate from the combined climate and CO 2 effect.Essentially we are asking the question: what is the additional change in emissions between two set time periods when making an additional input variable time variant instead of constant, starting from climate only via climate and CO 2 to the full effect.What we do not consider is a simulation where climate and population are variable, but CO 2 is fixed.The reason for this omission is that we simply do not consider this setup of interest and that we are not aware of any model study that has used C7917 it.
We agree that the way Equ..4-7are presented could easily be misinterpreted to represent single-factor sensitivity experiments that are assumed to add up the full change.This is not the case, because for one we are dealing with a combination of factorial experiments and time change, and also because we do not perturb the system one by one but sequentially.Nevertheless, we have chosen to fix CO 2 and population in the middle of the transient period in order to minimise the size of the difference between the fixed and the variable input.
In order to make this clearer, we can change Equ.4-7 to the following: with A further note on the method by Stein and Alpert: In our opinion, their method, just as the sequential- can in fact be expressed as one sub-set of the Taylor series), it is important to note that their exchange term is not the same as the two-dimensional cross term a xy of the Taylor series: f (x + ∆x, y + ∆y) = f (x, y) + a x ∆x + a y ∆y + a xx ∆x 2 + a yy ∆y 2 + a xy ∆x∆y + . . .(7) and for that reason we disagree with the notion that their terms are necessarily easier to interpret than the ones we use.As Stein and Alpert note in their article, a difference field derived from output between two experiments that vary in one input field can be difficult to interpret when the underlying model is non-linear.However, we believe that their method does not solve this issue, but only offers one possible gradual improvement.(Another possibility would for example be to derive additional Taylor expansion coefficients based on finite difference approximations, such as a xx , a yy and a xy .)Given finite resources, we believe that investing them into a larger ESM ensemble has been the better choice than employing the method by Stein and Alpert.Doing this in conjunction with transient simulations could be the subject of a later study, but would probably involve the output from only one or two climate models.

2)
Another assumption that has the potential to change the explored trends is the crop mask.Regions with high changes in crop cover often show strong changes in fire, these regions are masked in the present study although at least on the regional scale these gridcells might strongly contribute to the trends.See for instance Andela et al. (2014) for the importance of cropland cover increase on trends in Africa.Also a number of studies on cropland abandonment exist.
Reply: We have deliberately chosen to keep the crop mask constant because we wanted to avoid accounting for the direct effect of wildland being replaced by cropland.This would have either required the inclusion of a model for agricultural burning, which we do not have, or would constitute a trivial effect when -as here -we only consider wildfires, because they cannot occur on croplands solely by definition.However, we believe that the main effect of cropland expansion is not limited to suppressing C7919 wildfires on croplands, but also due to the infrastructure they tend to be accompanied by, which creates numerous barriers to fire spread.This is discussed extensively and demonstrated on the basis of historical tree ring data by Guyette et al. (2002), and also noted by Andela and van der Werf (2014; 3rd line on page 793): "This rapid initial decline [of burned area as a function of cropland extent] suggests that agricultural activity does not affect burned area only in the land that is actually converted, but also in the remaining savannas.[...,] large uncontrolled fires are less likely to occur owing to changes in fire management and lower fuel continuity due to, for example, more developed road networks (Shaffer 2010)." As far as the effect of land abandonment is concerned (not only cropland abandonment), the study by Moreira et al. (2011) points out that at least in southern Europe, this phenomenon has generally increased flammability of the landscape.This is the reverse of the effect simulated for Africa, but plays a role in Europe where the SSP3 scenario predicts a substantial population decline.These effects are included through the empirically derived shape of the burned-area vs. population density relationship (Equ.1).
3) The effect of CO 2 fertilization has been adressed by some previous studies that should be mentioned and discussed (for instance Calvo and Prentice, 2015, Lasslop andKloster, 2015, Kelley andHarrison, 2014).
Reply: Thank you for pointing out these studies.
4) 4) Technical Comments p.1502, l. 10: also here synergies are neglected, moreover, can only fuel change or could it also be the fuel combustion completeness.
Reply: See the reply to this reviewer's Comment 1 on the merits of the Stein and Alpert method.In this case, we express the change in emissions due to burned area change in a first-order forward computation using the previous time slice's ratio between emissions and burned area and the change in burned area attributed to CO 2 change derived C7920 from a sequential / transitional simulation setup.Synergy terms are not neglected, but do not appear in this configuration.However, we agree that it is important to point out that we summarize all of the remaining emission change due to CO 2 to the fuel effect, including effects that differ from the first-order forward calculation.This calculation therefore only serves as a first approximation for separating these factors.
As for the question of fuel load vs. combustion completeness, the reviewer rightly points at an inconsistency in the formulation: emissions change due to either changes in burned area, or changes in the amount of fuel combusted per area -see for example Figure 6, where this quantity is called "combustible fuel load".By contrast, the text referred to mentions emission changes via changes in "fuel load".5) p.1502, l. 17: in order "to" Reply: Thank you.6) p. 15029, l. 15: there are also some studies indicating an increase at least at the end of the 20th century that are worth mentioning, for instance Mouillot and Field (2005).
Reply: A comprehensive discussion of the issue of historical changes of fire emissions can be found in van der Werf et al. (2013), which mentions the study by Mouillot and Field (2005)Â ň -a historical reconstruction -but also discusses various arguments based on ice core proxies.7) p. 15030 l.23: a large part of savannas are used as pasture and fire is used as a tool to avoid woody encroachment.pasture areas may most likely be maintained as grasslands and therefore woody encroachment could actually lead to an increased use of fire to maintain the pastures.Is your model applicable to such systems over such time scales where strong changes in human land use can be expected?As far as I understand the only way you consider land use is by masking cropland areas?
Reply: SIMFIRE has been trained on actual observations of burned area that reflect the effect of land use.These effects are subsumed under the effect of population C7921 density.The model by Bistinas et al. (2014) also considers cropland and grazing land as statistical contributing factors for predicting burned area.However, we do not believe that existing scenarios of these factors in conjunction with population density are of sufficient quality yet to be included in these simulations.
Since our simulations are based on a model trained on recent data, we implicitly assume that management practices do not adapt themselves to woody encroachment.However, Wigley et al. (2010) report a strong and rather universal increase in woody vegetation cover based on aerial photography from South Africa between the 1930s and the early 2000s.The increase also happens in lands managed for grazing, despite of the well-known efforts of herders to decrease shrub cover.Furthermore, grazing might even lead to decreased shrub cover, contrary to the efforts of herders, because of over-grazing (Bond and Midgley 2012).There is also evidence for woody thickening over large scales in Australia based on a 27-year long satellite record (Donohue et al. 2009).These observations support our modeling results and we therefore believe that the effects of management tend to be insufficient to counteract a general trend towards woody thickening and that the assumption of temporally invariant management is therefore a reasonable approximation.
p. 15032, l. 25: fire frequency in terms of burned area?Reply: Yes.Sometimes "fire frequency" is indeed used as number of fires per time interval.
perturbation / time change method used here, simply asks a concrete what if question based on finite differences: What if input A changes by a given amount, what if input B changes, and what if both change?(And analogous for more than two separately varied input fields.)The interaction term is then simply the change in output when both input variables change minus the sum of the changes when only one input is modified.(Or equivalent for more input variables).Even though the equations presented by Stein and Alpert resemble those of a Taylor expansion (and each term C7918