Day 1 Q&A -
The questions and comments from the chat during Day 1 are recorded below. They are organized according to presenter.
How would you address the uncertainties of using different data sources? -Jaifu Mao
Generally, there are different sources of uncertainty: #1 Different datasets of the same variable (e.g. burned area) cause uncertainty because they are based on different sensors with different spatial resolutions, overpass times, and retrieval methods. This uncertainty can be addressed by either finding out which dataset is the most reliable under certain conditions or by using an ensemble approach in which we just use all datasets or combine them into a best estimate. #2 Datasets of different variables (e.g. burned area and fire radiative power) are sometimes inconsistent (e.g. FRP detects a fire but not burned area). A simple approach would be to introduce correction factors by e.g. estimating burned area from FRP. For a fire database, it would be absolutely necessary to provide quality flags with each dataset/estimate and to quantify uncertainties in single datasets and ensemble estimates.
For the type of global database you envision, what are some in situ observations that you see as 1) most helpful for modeling efforts and 2) able to be collected at the spatial scale we need to be comprehensive enough to inform/include in global models. -Rob Crystal-Ornelas
The most helpful observations would be fuel loads and moisture content before and after fire. These measurments allow us to estimate various other information such as fire danger, fire severity, combustion completeness and hence would provide ground constraints for fire emissions. In order to understand fire behaviour, these measurements would be available for different plants and ecosystems. Obviously, it would be difficult or impossible to provide representative data for the scale of an e.g. 500 m MODIS pixel. But if we focus more on higher resolution data (e.g. Landsat or Sentinel-2, 10-30 m), it will probably be possible to get a range of samples.
I’m wondering how your proposed ML (in the future research direction) considers the feedback between vegetation and fires when predicting fires at a monthly scale? -Sally Wang
I think in order to really understand fire, we need to go to higher temporal resolutions than monthly. The feedback from fires on vegetation and back to fire would be ideally captured by the observations of vegetation. Because the fire impacts should be present in the observations and hence affect the predictions at the next time step. As another alternative, one could make use of machine learning models that can account for time lags and feedback. Currently, we are working with long-/short-term memory networks (LSTMs) which might capture such feedback.
Can you comment on the difference in emissions for the smoldering fires vs flaming fires? -Jackie Shuman
There are large difference in emissions factors, with relatively more CH4 and CO during smoldering, see also: https://acp.copernicus.org/articles/21/8557/2021/acp-21-8557-2021.html
Do you have the intention to look back in the satellite record to assess changes in the lightning vs. human vs overwintering? Do you think overwintering is new or more now than previously? -Nancy French
I would love to go back in time, but it gets difficult before the MODIS era as we need the active fire to retrieve the ignition locations. Overwintering fires are definitely not new, but they may have become more widespread since they are linked to extreme fire years (which have become more frequent).
Do you think it would be possible to predict future dynamics of overwintering fires based on factors like permafrost thaw dynamics and carbon emissions? -Konstantina
Knowing how often fires occur in boreal peatlands would be very important for this (and accounting for permafrost thaw would be great!). I am also hoping we can visit locations of overwintering fires on site at some point to better understand how drainage conditions and soil characteristics influence their occurrence.
Jim - you mentioned that extreme fires are the new normal but we also know that management has been a major suppressor of fire since European colonisation in many places. Prior to that indigenous populations used fire for management and dendrochronological records from the western US show pretty extreme fires prior to settlement. Could all of our speakers comment on how necessary it is to include long-term historical records in our databases and understanding of changing fire regimes and their influence on the carbon cycle? -Anthony Walker
There’s pretty cool work with tree rings showing how fire scars in tree rings are very scarce within the past 100 years due to fire suppression policy in western US. -Barbara Bomfin
I think this is necessary. It seems that the FireMIP models are all constrained towards the satellite period and diverge in the past. Ideally, one could also try to constrain fire models against e.g. charcoal records and the satellite observations to both get better estimates on historical/palaeo changes and the actual information on fire dynamics. This could also help to get estimates on the past effects of fire on the C cycle. - Matthias Forkel
There is a pretty clear relationship between daily temperature and burned area in the CA data that to me shows climate is structuring some of the new extremes. Totally agree though that management is important. For California, the following ref is my goto paper for looking at long-term management effects. A. H. Taylor, V. Trouet, C. N. Skinner, S. Stephens, Socioecological transitions trigger fire regime shifts and modulate fire–climate interactions in the Sierra Nevada, USA, 1600–2015 CE. Proceedings of the National Academy of Sciences 113, 13684-13689 (2016). -James Randerson
Bill and others as well - what is the scale / size of these datasets? -Shreyas Cholia
Table provided by Bill Riley:
Would it be important to also have datasets for areas that didn't burn in order to have information about changes in ecology related to climate change/invasive species/ etc. (or have pre-fire data for future fires that occur)? Or would that be a different database? -Kaelin Cawley
Yes, a very important point. Often we use chronosequences, which include measurements in non-burned nearby sites. A synthesis of these types of experiments in different regions would be helpful.
My question is related to the DNN-fire model in the Earth system model. Do you already have some first results that show how this model behaves under future climate conditions? Does it produce realistic results or might it be a problem that it is trained based on present conditions? Matthias Forkel
We have not analyzed future simulations yet with the DNN model. I am pessimistic about how realistically standard ML models can be propagated into the future, primarily because of (1) moving outside the training space and (2) not accounting for vegetation changes (I guess that’s related to #1). One possible benefit of the DNN model is that it was designed to keep the underlying “process model” relationships, which might be expected to be conserved into the future. But, if the system changes from evergreen to deciduous trees, e.g., the ML model would no longer be reasonable.
In your studies, how long does it take for soil to recover after fire events (depending on severity)? What factors affect in general? -Lilian Davila
When comparing the effects from different fire severities, how do you account for differences in the pre-fire stand that causes different fire severities? E.g., microtopographic features may result in different moistures, which could affect burn severity. These same topographic features may influence post-fire SOC too. -Nicholas Dove
It emphasizes the importance of having pre/post-fire measurements & samples taken as temporally close to each other as possible. This takes a massive team, or limits the number of field plots possible to access. Without these fine-scale data, our measurements will also include some variability we're not able to resolve very well. We've used DEMs, pre-fire veg measurements, and weather records to help with this.
Do you know of syntheses or meta-analyses of the effect of pre-fire conditions on SOC responses to fire? I’m guessing, from your talk, not, but was hoping … -Bill Riley
I have a field crew in CA this summer collecting data to answer that question, and a NIFA award starting this winter to explore the mechanisms in more detail. Stay tuned!
How do you think the changes in microbial communities post-fire influence trajectories of plant succession? -James Randerson
I’d say there’s still a strong “chicken-or-egg” element on the table. We see vegetation communities change post-fire. We see microbial communities change post-fire. To what extent does one influence the other? I’m not sure I’ve seen conclusive data clearly demonstrating a *causal* effect of microbes on post-fire plant community composition - e.g., the absence of this specific microbe has led to a specific plant not being able to recolonize or flourish post-fire. That said, we know that microbes have important and specific relationships with plants, whether we’re thinking about symbioses or pathogens. Additionally, since microbes play a critical role in post-fire nutrient cycling, especially with respect to nitrogen, the effects of fire on the overall microbial community is likely also important.
That said, again, microbes are so responsive, and have potentially high colonization/dispersal potential, whether from soils below, where lethal temperatures were not reached, or from above (e.g., Kobziar’s work on aerial dispersal of microbes during fires). So, I might posit that soil microbes will more tend to reflect the post-fire conditions (e.g., pH shifts and changing nutrient profiles in the soil) than be dominated primarily by which taxa survive or are killed (at least in terms of function).
I think it’s super interesting to compare how we think about fire data from the top (remote sensing, large-scale modeling) down vs the bottom (in situ measurements) up. How can we bridge the gap between the scale of model and satellite grids and the scale of variability in ecosystem structure, fire history, severity, and fine scale heterogeneity you see on the ground? I think this will be an important challenge for this whole group! -Benjamin Sulman
I agree, this is a key challenge! With soils, measuring severity @ the soil sample location is much more informative for explaining variability in e.g. soil C than using severity metrics derived from remote sensing imagery (e.g. 30m Landsat pixel size). -Jessica Miesel
To ecosystem/fire modelers, though non-modelers probably have insights: I'm struggling with the balance of anthropogenic influences on fire: - Added ignitions as access to remote areas increases. - Suppression. - Reduced fire spread due to land-use mosaic. - Larger fires in remote areas due to build-up of fuels and global warming. Could models (and/or obs.) potentially offer insights to help policy makers optimize fire management approaches for different parts of the world? -Samuel Levis