Please find my review as an attached pdf.

Review: 

Improving historical trends in the INFERNO fire model using the Human Development Index

Summary: 

This is a detailed and extensive piece of work that attempts to use HDI to improve the JULES-INFERNO fire model.  

I'm considered to be a SME and I'm somewhat confused by the approach. It is not clear why the authors have taken this approach, primarily because so many critical factors in wildfire propagation are not included.  By taking a global data set, 'de-weathering', the assumption seems to be that the remaining correlation is between GFED and HDI, which patently isn’t the case.  

Although INFERNO necessarily requires a simplification of the complex fire processes, my concern is that the approach taken here does not include processes that are a critical part of the system. 

I'm not a statistician, but the work largely neglects the physical process driving fire behaviour.  I wonder if an EOF analysis would be more appropriate.  

The study shows a 'de-weathered burnt area fraction'. It is unclear to me what the physical meaning of this metric is and how can it sensibly be interpreted with context to current fire regimes and understanding of fire weather and fire behaviour. I would be surprised if the meaning is clear to other readers. 

The processes driving fire activity globally are extremely complex and the various contributing factors are not fully acknowledged in the study or incorporated into the approach.  A primary driver of fire is fuel availability and there is significant uncertainty in future fuel regimes (including fuel structure, dryness and landscape continuity) in a changing climate. Shifting hydrological regimes influencing fuel availability are a factor driving global fire activity in recent years. The focus in this paper seems disproportionately towards HDI and imbalanced with the other ingredients in fire regimes. 

This is a very detailed and methodical study, but the statistical approach does not reconcile with the physical process that drive global fire activity, which are highly heterogeneous. The approach assumes that by 'de-weathering' the GFED dataset the remaining fire mapping is dependent on HDI. 

Specific comments.

•    Fig 3 shows the 'log transform 'deweathered' burnt area fraction % (of global fire emissions database)', (which is derived from satellite hot spot data and >12 years old). It is unclear how this should best be interpreted in context of fire regimes, or future emissions scenarios.

•    GFED isn’t explained in the text (it is in the appendix, but it's key to the main study and the Giglio study is from 2013, so a relatively old dataset (The Giglio reference date is 2013, but Figure 7 states 1997-2016).  Giglio's paper shows a map with GFED regions; these are very heterogeneous (just one example is that Mediterranean Europe and Greenland are treated as homogeneous in the current study, while there are great differences in HDI and fire regimes across that spatial area). 

•    The Canadian FWI has been used.  The FWI is intended for use in forested areas of Canada. It was not designed to be used in Eurasia, tropical areas, grasslands or agricultural areas, rainforest vegetation or peat fuel. It captures short and medium term fuel drying. It does not capture seasonal grass growth or long-term drought. As it was intended to capture fuel availability in forest types, FWI does not readily translate to fuel availability across other landscapes, particularly on climate time scales. 

•    The FWI is applied here globally outside it's intended aim. The calculation method (average time period used) is not explained clearly in the text. From Figure 1 it is unclear if this is averaged over all seasons and all years. Averaging at high latitudes will mask any extremes, which are typically the times when impactful fires occur. 

•    HDI is averaged 1997-2016. This time period may not be is appropriate as input to projections (China for example is changing rapidly). HDI values are applied over large areas (countries) with heterogeneous land use and population density, which is inconsistent with typical fire regimes, which can vary considerable over relatively small areas, particularly with topographic and vegetation structure and noting that topography is not included in the framework. 

•    My understanding is that an objective of the modelling is to capture emissions, but there is no distinction made between low intensity and high intensity fires, in either the GFED4 data or FWI, or whether the fire is an agricultural fire or wildfire (which will have varying fuel consumption and therefore produce varying emissions), the seasonal timing of the fire activity, or the vertical extent of the fire, which are important considerations for emissions.  For emissions, it is typically the large intense fires that are critically important as these are the ones which produce upper-troposphere and stratospheric injection of particulates and are often associated with incomplete combustion.    Although INFERNO necessarily requires a simplification of the complex fire processes, my concern is that the approach taken here does not include processes that are a critical part of the system. 

•    Fuel availability is not considered. Fire processes are different in different landscapes and a global approach to describing fire regimes is problematic due to the variation in fuel type and fuel availability..  

•    Fuel type and structure consumed is not discriminated, although this will be a significant factor in calculating emissions. 

•    Fuel availability is a key factor in landscape fires. Fire trends can and do vary hugely with different vegetation structure (eg grasslands often burn annually). The consideration of vegetation type and structure and fuel availability is rather superficial in this study, however in many parts of the world, this is likely to be a key ingredient in changing fire-climate regimes. 

•    Fuel continuity is a key factor influencing fire size. This can be (indirectly) related to HDI due to landscape features such as highways, railway lines, which often act as barriers to fire spread or can be used to support containment lines.  Natural boundaries such as watercourses and topographic features can also constrain fire spread.  Population density will also contribute to fragmentation of landscapes. 

•    The term 'suppression', is used several times but is not clearly defined.  Many landscape fires are not actively suppressed, or are only suppressed or contained in proximity to infrastructure and assets. Suppression and containment will vary considerably geographically and methods such as hand tools, backburning, fuel reduction will vary considerably dependent on resources, assets and fuel loads.  Mitigation efforts such as planned burns and fuel management will also vary with HDI (and resources). In addition, many fires extinguish overnight, dependent on favourable weather conditions, fuel availability and fuel continuity, but these processes are not captured in the current study.  

Short comments

•    The fire-science terminology could be used more precisely. Examples include 'fire models' and 'fire simulators'.  

•    L40 A discussion on how the processes are included in CMIP frameworks would be useful

•    L105 FWI is not the influence of climate 

•    L140 topography is a key factor in fire spread as is local meteorology.  

•    L180 It is not clear to me how the 1860 -2016 dataset is applied

•    L200 What are the origins of this approach for global vegetation structure, particularly its relevance to fire prone environments. 40% seems high for many trees. The vegetation types described aren’t reflective of many fire regimes. 

•    L250 this could be due to fuel availability and continuity, but these factors haven't been included. 

•    How are large and small fire sizes categorised (0.2, 0.6 etc is not a fire size)

•    Maps of the regions are not shown until the appendix and the regions are highly internally heterogeneous with regard to HDI and fire regimes. 

•    What is 'burnt area fraction' a fraction of? Giglio 2012 specifies this as the fraction of each grid cell that burns each year but that is not clear in the current study.

•    Fig 10. It is hard to compare these plots due to the varying y axes. The interannual variability seen in the observations is interesting, but is not well captured in the smoothed JuELS-INFERNO-HDI, which is a concern. 

•    There are many acronyms, some of which are not expanded (eg GFED4). Also several small typos.