General Scoping Recommendations/Discussion

Initial thoughts on review paper shared by David Godwin - highlighting remote sensing methods/tech supporting wildland fire assessments; published April 30th, 2020.

Although I couldn't attend the meeting, I did watch the entirety of the presentation.  I wanted to thank Joe and Eli for such a great presentation!  I also wanted to provide some more questions that I have after watching the presentation and the follow-up question/answer session:

1) Would it be helpful to start classifying the fire detections by the ecosystem type in which they encompass?  If you're doing the entire SE, you could consider using LANDFIRE or NLCD.  Ultimately you would want to decide how narrowly that you want to subdivide vegetation into categories.  If you're just interested in doing this for Florida, you could consider FNAI.

The reason that I make this suggestion, is because you'll probably find that some vegetation types are accurately defined by remote products (e.g. BA), while others are not.  In my experience, the upland vegetation types are generally easier to map fire than the wetlands.  Inherent problems with changes in water levels affect remote sensing products.

2) I think that it might make sense to subdivide the area of interests by equally sized grids or property boundaries to determine where the remote sensing products work and where they don't.  I really liked Todd's suggestion about considering managed areas like TTRS as units in which you could then monitor for fire activity.

3) If it's determined that some kind of burn severity products are required as part of the output, you should consider converting differenced Normalized Burned Ratio (dNBR) products to Composite Burn Index.  CBI data was been compiled for the conterminous U.S. and Sean Parks has subsequently created a Google Earth Engine workflow to use Random Forest to calculate CBI (see https://www.mdpi.com/2072-4292/10/6/879/htm).  Additionally I have been developing regression equations based on different scales/vegetation products that could be used.  These equations could be implemented in GEE as well and I have already implemented them to convert MTBS dNBR products to CBI.

4) The "scale" of the project really needs to be taken into account.  What is the minimum mapping unit?  At what scale is accuracy assessed?  All of the remote sensing products that are being assessed inherently have different scales at which they're accurate.  Product scale and accuracy need to be taken into account for the scoping of this project.

5) At some point in the scoping it might be nice to determine what the options are for processing and subsequently serving data.  Admittedly, these would just be suggestions.  Whoever ends up subsequently doing the work would decide how to proceed.  At least there would be some already researched options available.

6) Could past fire perimeters somehow be used to help map future?  Many areas in the Southeast are burned in specific burn units on a specific time interval (e.g. 2 years).  If we use a product like the BA product to intersect where areas burn over a specific amount of time, it might be possible to estimate where these burns typically occur and on what interval.  You could then "forecast" where burns will likely happen.  These forecasts could be potentially tied to specific burn permits so you know essentially "who burns where".

7) What steps could be taken to get private landowners "buy in" for potentially participating in the tracking of burned areas?  Maybe it's framed like "if we monitor burned areas we can help determine where hazards might exist for your property".  Or, it could be framed as "we produce these products that can help you with your management of your lands".  Some landowners will likely have potential issues with being monitored.  But, if you can frame it as "we're giving you something that you can use to better manage your lands" it might help.

8) At the end of the day the data "products" that could be produced should be examined.  Depending on your audience, you'll want to range between simple (e.g. summaries, pdfs, kmzs) and actual geospatial data products.  If these products are planned well for different users, you might get more buy-in from private land-owners.  I really think that TTRS is positioned well to be able to "sell" the idea of how this work can lead to better management outcomes, while maintaining landowner privacy.

Q&A notes as shared by Tall Timbers on today's (08.26.2020) technical review call:

1)How are you spatially assigning active fire detections with burned area products?

We have not developed a methodology to do that at this time, however this is of interest as we continue to explore how active fire detections may contribute to burned area products and might be used as ancillary data.

2) How did you tag active fire detections with FFS OBA?

Through temporal matching and then visual analysis. No automated method with our approach, although an automated approach would need to be developed for larger regional analysis.  Given the issues with OBA’s we would have to evaluate  if this would be a useful exercise.

3) How does Tall Timbers get around the inherent problems with the prescribed fire records?

We acknowledge the limitations of these data and use them as general location of the treatment.  Depending on the source, many of the fire records do not represent the true area that was burned.

4) Is there a leading fire mapping standard in use which would support improved compatibility? – Considering both RS and stakeholder database systems.

Fire mapping standards vary between agencies and private landowners with regards to mapping and attribution.  There have been some efforts to combine UDSA (FACTS) and DOI (NFPORS) treatment datasets using WFM RD&A’s Fuel Treatment Integrator so data can be standardized for applications such as IFTDSS, WFDSS, EGP, etc.  For RS datasets and products, each has standardized attribution specific to the sensor or product.

5) Can FL database be a surrogate for other state permitting systems?

The FL database is a remote sensing application that produces fire history metrics based on burn probability from LANDSAT BA products.  Each state has it’s own system for tracking burn authorizations.  A standardized permitting system might be best coordinated through the Southern Group of State Foresters.

6) Could TTRS potentially create some kind of a fire probability layer? This could be done by examining convergence of all fire detections or potentially by assigning probability to the modelling process.

Very good question and the short answer is yes Tall Timbers is working on prescribed fire probability mapping based on RS data as well as climatology, fuels, vegetation, etc.  This is active but unpublished research currently being conducted by our Fire Science folks.

7) Would it be worth considering including in the final scope of work a requirement to assess which burn permits were fulfilled (or a subset of permits)? That would allow statements like “Of the X burn permits in 2020, Y% of permits actually resulted in a fire. Of those, Z% were detected by satellites.”

This would be challenging as there is no reporting mechanism for state burn permits to be reported complete or acres burned verified.  Additionally, spatial accuracy of the permit location is problematic given that burn permits can be up to 2km from actual permitted location and single points representing multiple burn units can be up to ~ 10 km away [Nowell et al., 2018]  The use of reference imagery would be insufficient to overcome these challenges.

8) Incorporating permit data will be critical to attribute the satellite-based fire/burned area detection as prescribed fire.

Agree that permit data would be the best source of attribution but unsure of how to overcome challenges.

9) Is the focus of the SE FireMap and current scoping effort focused on only prescribed fire?

No, we accept that the final product will be RS based and there is not differentiation between wildfire and prescribed fire.  The interim report may have been confusing in this aspect as we used permit records and prescribed fire landowner records as a focus for detection analysis.  We recognize that both wildfire and prescribed fire result in ecological change on the landscape and are both important.

10) TTRS should consider evaluating commercial RS products as part of the scoping process.

Tall Timbers has conducted literature review of several mapping efforts using commercial satellites.  We are aware of several studies that have successfully used commercial products to produce burned area and burn severity maps of small scale plots, we have seen no regional applications.  The reason for this is the high cost of commercial imagery.  For example, WorldView-3 that has been used in several studies by Tall Timbers collaborators in the New Jersey Pine Barrens cost ~$22.50 per SqKm.  Considering that Florida is 170,000 SqKm alone and revisit time of < 2 days, obtaining even 1 year of imagery would be cost prohibitive for the SE.  Additionally, beyond a more in depth literature review and providing commercial pricing for sensors to the TOT, Tall Timbers does not have access to commercial imagery to further evaluate.

11) SEFireMap product definition should include additional target accuracy metrics. In addition to   the current desired detection threshold for fire size (5-10 acres), consider omission/commission error and how to appropriately set bounds

We would defer to the published accuracy assessments of sensors and products reviewed in the scoping phase and NRCS to set appropriate bounds for the SE FireMap.

12) Consider fostering discussion with DOD about accessibility of military satellite data? Andy Beavers shared contacts and offered to reach out...

Tall Timbers would be happy to review any DOD remotely sensed burned area or fire detection products.  Once we have a chance to assess the products we would certainly be interested in a discussion of how military satellite data could be incorporated into the SE FireMap.

#7/23/2020 Tall Timbers Interim Report Notes
# SE FireMap Scoping INT REP 2.pdf
How are you spatially assigning active fire detections with burned area products?  This requires a lot of processing, so efficiency is key.
How did you tag active fire detections with FFS OBA?
FMT code that estimates burn severity breakpoints could be easily re-written for GEE.


# Questions for call
# Are there any gaps or proposed modifications we should consider?
Could TTRS potentially create some kind of a fire probability layer?  This could be done by examining convergence of all fire detections or potentially by assigning probability to the modelling process.
How could TTRS improve the fire modeling process?

# What do we like and what do we think is missing?
I really the like the multi-faceted approach in considering many data sources that TTRS has considered.
How does TTRS get around some of the modeling limitations of BA product?  Using their own modeled products?
How does TTRS get around the inherent problems with the prescribed fire data records?  Can they start using area burned as a way to narrow down the fire location?

# How can TTRS improve the scoping process?
Share their proposed processing or actual processing schemas.  It would be nice to see the details.