Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Edible re buffers: Mitigation of wildre with
multifunctional landscapes
Xiao Fu
a
, Abigail Lidar
b
, Michael Kantar
c
and Barath Raghavan
a,
*
a
Department of Computer Science, University of Southern California, Los Angeles, 90089 CA, USA
b
Department of Data Science, University of California, Berkeley, Berkeley, 94704 CA, USA
c
Department of Tropical Plant and Soil Sciences, University of Hawai’i at Ma
noa, Honolulu, 96822 HI, USA
*To whom correspondence should be addressed: Email: barathra@usc.edu
Edited By: Joann Whalen
Abstract
Wildres ravage lands in seasonally dry regions, imposing high costs on infrastructure maintenance and human habitation at the
wildland–urban interface. Current re mitigation approaches present upfront costs with uncertain long-term payoffs. We show that a
new landscape intervention on human-managed wildlands—buffers of a low-ammability crop species such as banana irrigated
using recycled water—can mitigate wildres and produce food protably. This new intervention can complement existing re
mitigation approaches. Recreating a recent, major re in simulation, we nd that a medium-sized (633 m) banana buffer decreases
reline intensity by 96%, similar to the combination of prescribed burns and mechanical thinning, and delays the re by 316 min,
enabling safer and more effective reghting. We nd that under climate change, despite worsened res, banana buffers will still
have a protective effect. We also nd that banana buffers with average yield could produce a prot of $56k USD/hectare through fruit
sales, in addition to re mitigation.
Keywords: agroecology, wildland–urban interface, re mitigation, land use
Signicance Statement
Fires are an increasingly large problem for areas at the wildland–urban interface. Creating living re buffers has the potential to pro-
duce a wide range of ecosystem benets in addition to re prevention. Here, we explore a case study of the use of bananas as a living
re buffer in California under current and future climate scenarios, not only do bananas provide re protection now, they provide
increasing capacity in the future and are economical to install and provide an opportunity for additional ecosystem benets.
Competing Interest: The authors declare no competing interest.
Received: June 15, 2023. Accepted: September 19, 2023
© The Author(s) 2023. Published by Oxford University Press on behalf of National Academy of Sciences. This is an Open Access article
distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs licence (https://creativecommons.org/
licenses/by-nc-nd/4.0/), which permits non-commercial reproduction and distribution of the work, in any medium, provided the original
work is not altered or transformed in any way, and that the work is properly cited. For commercial re-use, please contact
journals.permissions@oup.com
Introduction
Sustaining human habitation in seasonally dry regions of the
world is an intensifying challenge with urban expansion and cli-
mate change (1). Anthropogenic temperature increases and con-
sequent global aridity (2) have increased wildre risk; wildres
nearly doubled in frequency in the Western United States from
1984 to 2015 (3). Further, climate change is threatening global
food production not only for staple grain crops but also for import-
ant fruits and vegetables (4). Climate change is decreasing food
production in existing agricultural regions while increasing re
risk (5), providing an opportunity to simultaneously address
both challenges.
Simultaneously, the wildland–urban interface (WUI) (6)—“the
area where houses are in or near wildland vegetation”—has in-
creased over 40% in the United States, along with re risk (7).
WUI land often has high or extreme re risk (8), with such risk
amplied by climate change (9). Despite this risk, the WUI is
increasingly desirable for building structures. Wildre in the
United States imposes annual costs of 70–300 billion USD (10).
California in particular faces major wildre risk, threatening
more than four million homes (11).
There is growing recognition that no one technique will suf-
ciently mitigate the risk of re at WUIs. Instead, many techniques
must be employed, including: (i) enhanced building and land-use
codes, (ii) fuel reduction in wildland and WUI areas, (iii) increased
reghting resources, and (iv) creation and maintenance of re-
breaks. Despite this wealth of options, existing strategies have un-
certain long-term payoffs with large up-front costs (12, 13).
Fire mitigation and reduction is a well-studied area of research
and practice. The size, time, and location of interventions are key
variables to consider to reduce re risk most effectively (14). The
review by Chung shows that available decision variables include
treatment method, timing, and location (15). However, the com-
plexity of re mitigation makes it difcult to identify an optimal
solution in general as the characteristics of a specic locale,
including the soil, topography, climate, zoning, land costs,
PNAS Nexus, 2023, 2, 1–12
https://doi.org/10.1093/pnasnexus/pgad315
Advance access publication 24 October 2023
Research Report
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
vegetation types, heterogeneity of built environment, and com-
peting land uses necessitate the evaluation of numerous
tradeoffs.
Choice of vegetation in WUI areas is key, but standard “re re-
sistant” plants do not provide economically valuable outputs (16).
We study the potential of crop plants on human-managed wild-
lands within WUI, to mitigate re damage while producing an ed-
ible crop.
WUI (6, 7) consists of intermix-WUI and interface-WUI, with
both denitions focusing on the adjacency of urban and wildland
land types. Intermix denes an area where vegetation is typically
“mixed” with urban land (e.g. houses within a forest). Interface
WUI regions can be spatially partitioned into vegetated and popu-
lated areas; such a partition may be suitable for re buffer place-
ment. Intermix-WUI includes human habitation within wildland
and thus does not have a dened boundary for a buffer.
Intermix-WUI also has lower population and requires further buf-
fer placement ne-tuning, so we only consider Interface WUI in
this work.
Multifunctional landscapes for re mitigation
We consider the effectiveness of alternative re buffers designed
for multifunctionality, with careful consideration of the species
and their effectiveness for re mitigation, ease of propagation,
ease of maintenance, and yield of high-value outputs to recoup in-
vestments. We specically consider the use of increasingly avail-
able recycled water in these regions (17, 18).
Planted re buffers must have high water content at all times
(19). Irrigated orchards and vineyards have insufcient moisture
to adequately slow or stop re spread (20, 21). Other types of irri-
gated agriculture are not suited to uneven terrain typical of such
high-re-risk areas. Nonirrigated re buffers are unlikely under
climate change-driven drought conditions to have sufcient
water content (22); the prominent exception is some succulent
re buffers which may be effective but do not produce high-value
outputs. Crops such as banana have very high water content, from
93 to 99% in good moisture conditions to 76 to 88% under drought
(23, 24).
We consider several criteria for potential multifunctional re
buffer crops: (i) minimal management needs, (ii) suitability in pre-
sent and future climates, and (iii) low ammability (25). Such ed-
ible re buffers are thus new agroecosystems in wildland areas
that abut the WUI in which the fuel and management is funda-
mentally different from current circumstances.
Methods
Overview
In Fig. 1, we depict the workow for our experiments and also for a
potential deployment of edible re buffers on the land. Our experi-
ments use canonical tools for re behavior modeling. We use
FARSITE (26) for re simulation, LANDFIRE (27) for landscape
data, and BehavePlus (28) for fuel type modication. We use
standard GIS tools such as ArcGIS (29) for visualization and pri-
mary buffer overlay analysis. Later, we used GDAL (30) and raster-
io (31) for rasters and shapele process with massive number of
fuel maps generation. We use seaborn (32) and geopandas (33)
for data processing and visualization. We use GeoMAC (34),
NOAA (35), and SILVIS Lab WUI data (11, 36, 37). The spatial ana-
lysis and data can be accessed through: https://github.com/
fxdawnn/EdibleBuffer.
Fire modeling and selection
To understand the potential for edible re buffers to mitigate risk
at the WUI, we simulated a historical re. California has faced nu-
merous major wildland and WUI res in recent years, yet many
historical res yielded inadequate or inappropriate data for ana-
lysis: some progressed too quickly in intermix WUI (e.g. Camp
Fire, 2018), others were at a WUI but were suppressed by extensive
reghting (e.g. Getty Fire, 2019), while still, others had minimal
effect on populated areas. To balance these factors, we selected
the 2017 Tubbs Fire. The Tubbs Fire ts the context of the inter-
vention we explore: location in a semiarid or Mediterranean re-
gion, origination in wildland, progression due to prevailing
winds through the WUI, and rapid advancement that overwhelms
reghting resources leading to signicant loss of life and struc-
tures. Additionally, simulations allowed us to illustrate dramatic
containment differentials, since the Tubbs Fire caused the most
damage at the WUI of any nonintermix WUI re in California his-
tory. Given the re’s location in Sonoma County and its proximity
to Napa County, we were also able to compare banana buffers
with pre-existing alternatives: vineyards and orchards.
Using FARSITE (26), we replicated the Tubbs Fire in simulation.
FARSITE is designed for wildland re simulation, which is well
suited to our context as edible re buffers involve a change of fuels
on human-managed wildlands. No re simulation tools are well
adapted to res within the WUI and urban areas. In our context,
however, the primary analysis is squarely on wildlands, for which
FARSITE was designed, as edible re buffers involve a fuel change
of such human-managed wildlands in which a re may progress
toward urban areas such as in the case of the Tubbs Fire.
We acquired a Tubbs Fire area base map from LANDFIRE,
which provides country-level fuel maps for use in FARSITE and
FlamMap (27). Additional landscape data were obtained from
the LANDFIRE Data Access Tool (LFDAT) (38). To accurately re-
present weather conditions, we supplemented our existing
NOAA weather data with wind speed data from news reports, as
extreme weather was captured with higher delity by local man-
agement agencies than by NOAA stations (which suffered from
power outages) (39). We performed preliminary calibrations on
our parameters based on standard guidelines for re behavior
modeling (26). FARSITE incorporates existing models for surface
re (40), crown re (41), point-source re acceleration, spotting,
and fuel moisture (26, 42). Additionally, FARSITE uses weather
and wind inputs, so we were able to incorporate future weather
scenarios into our simulation.
Given our use of FARSITE, which is a general-purpose wildland
re simulator, using standard fuel maps, we expect that the sim-
ulations we performed using the Tubbs Fire will generalize to
those regions similar to California’s Mediterranean climate,
such as the Mediterranean Basin itself and parts of Mexico,
Chile, Australia, and South Africa. In those areas with frequent
res and/or res in grassland, chaparral, or low-density savanna
we would expect banana re buffers to be more effective than in
our baseline simulation of the Tubbs Fire; in those areas with
dense conifer forest that abuts the WUI, we would expect banana
re buffers to be less effective as crown res often cause long-
distance ember cast. In dense conifer forests with crown res
where, due to the height of the trees and the fuel load, a banana
buffer of much lower height would likely be unable to block the
long-range ember cast. Our evaluation does take ember cast into
account in WUI interface (not intermix) settings that are com-
monplace in California and that have signicant population; these
regions do not have dense conifer forests. The size of the banana
2 | PNAS Nexus, 2023, Vol. 2, No. 10
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
buffer is also taken into account in our evaluation; we evaluate
the extent to which larger buffers can help mitigate ember cast
and thus decrease re spread and provide critical space and
time for reghting efforts to succeed. Further modeling, diverse
spatial deployment, and eld testing would be required to estab-
lish the generality of our simulations.
Determining the optimal urban fuel type
To dene the WUI in our study area, we followed the description
set forth by SILVIS Lab’s WUI dataset (11, 36, 37), as well as a
standard risk assessment framework that characterizes WUIs as
HVRAs (highly valued resources and assets) (43). We classied ur-
ban areas in Santa Rosa based on housing density of above 20
housing units per square kilometer and negated all vegetation
that contributed to re risk (6). Urban boundaries were based on
the US Census Bureau block data, following our denition of
WUI (7). CalFire 2013–2017 housing damage data (44) was used
to assess infrastructure damage and validate our denition of
the protected area. Our classication of urban areas was applied
towards determining an urban fuel type that most closely mod-
eled the satellite data obtained from USGS. We compared the ur-
ban fuel type from the LANDFIRE base map, NB1, against several
other standard fuel models (45) and a custom model, and calcu-
lated the F1 score of each to assess the accuracy of our result
against the ground truth. TL2 was determined to best represent
re propagation in urban areas.
Banana re buffer modeling
In order to identify the suitable area for banana cultivation in
California, global banana occurrence points were downloaded
from Global Biodiversity Information Facility (46). These occur-
rence points were used as input for MaxEnt (47) suitability model-
ing, which was implemented from geo-referenced coordinates
implemented in the software R, under current climate conditions
using 19 bioclimatic variables (48) and under future conditions
representing RCP 4.5 and RCP 8.5 (49). Future models were con-
structed using eight GCMs (BCC-CSM2-MR, CNRM-CM6-1,
CNRM-ESM2-1, CanESM5, IPSL-CM6A-LR, MIROC-ES2L, MIROC6,
MRI-ESM2-0) at a 2.5 arc minute resolution in 2090 (48).
Suitability maps of current and future models were overlaid to ex-
plore which counties have the potential for cropping interven-
tions. Suitability models were considered accurate if they
complied with the following conditions: (i) the ve-fold average
area under the test ROC curve (ATAUC) is greater than 0.7,
Fig. 1. Workow for modeling and analysis of Edible Fire Buffers using existing software tools and datasets, including postsimulation considerations
such as site selection and benets. Green species input components, blue species output components we validated, and red species potential results
we discuss but did not validate.
Fu et al. | 3
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
(ii) the standard deviation of ATAUC (STAUC) is less than 0.15, and
(iii) at least 10% of grids for each model has standard deviation less
than 0.15 (ASD15).
Preliminary buffer testing
To test the potential of food crop rebreaks for minimizing re
damage, we constructed buffers with fuel type NB3, which is clas-
sied as nonburnable agricultural land (45). Keeping all other
baseline parameters constant in FARSITE, we simulated re be-
havior with buffers of the following widths placed on human-
managed wildlands that abut the WUI: 200 m (very small), 390
m (small), 633 m (medium), 1070 m (large), and 1280 m (very
large); for clarity of the gures, we omit large buffers. Width val-
ues were determined by creating buffers in order of increasing
size, with exact value selection constrained by the granularity of
the raster in ArcMap; this yielded values that are distributed
across the range of agricultural land sizes that are standard in
the region.
Fuel types. All fuel modications in our simulations were based
on the LANDFIRE Fire Behavior Fuel Model 40 (FBFM40), which
we selected as our baseline fuel model layer in FARSITE (27). We
tested NB1, TL1, TL2, and a custom urban fuel type in replicating
the Tubbs Fire (45). We used NB3, TL1, and Anderson 2 in buffer
testing (50); to improve accuracy when replicating the character-
istics of a banana buffer, we generated a custom banana fuel mod-
el using BehavePlus (28).
Validating the custom banana and vineyard/orchard fuel models.
Of the nonspatial re behavior modeling systems available, we se-
lected BehavePlus (28) to evaluate the behavior of our custom
fuels. We began with TL1 (45) as a baseline fuel model (as it is
characterized by low spread rate and low ame length), and
then we modied fuel load, fuel bed depth, and fuel moisture pa-
rameters to model banana crops; we provide details in Table S1.
Specically, bananas have a range of total moisture from 76 to
92% depending on drought conditions (24), which, converted to a
dry weight basis is a minimum 316%, higher than the upper cutoff
allowed by BehavePlus (which was designed for wildland species
that rarely have such high moisture content); thus, we use the
maximum value for live fuel moisture. Similarly, for dead fuels,
the banana buffer is highly managed so little to no dead fuels
are to remain and the groundcover will be of a succulent plant
species. However, should management fail to note dead banana
plants, we can account for the dead fuels based upon studies of
banana plant drying, which nds once again that bananas will
yield dead fuels of higher moisture than BehavePlus’s maximum
settings (51), as banana is a nonwoody plant species with very
high water content. We compared simulated re behavior be-
tween our banana fuel type and agricultural NB3, and found
that the two yielded similar results. We tested buffers of varying
widths under NB3 and our custom banana fuel model. Bananas
behave similarly to NB3, and for both fuel types, re spread slows
as buffer size increases.
To validate our results, we performed similar experiments us-
ing a conventional vineyard/orchard buffer, which we hypothe-
sized to be less re-resistant than bananas. As well-studied
cropping systems with proven economical benets, vineyards
and orchards provide a comparable model for banana buffers as
a protable land use for a potential re buffer. Based on previous
work assessing land covers, the best match to the vineyard and or-
chard landscapes is Anderson 2 (50), which was used to describe
tree crops—including vineyards and orchards—in Sardinia, Italy
(52), which has a similar climate to California. Both banana buf-
fers and vineyards/orchards were modeled with medium (633 m)
buffer widths as this width demonstrates moderate re mitigation
effects. In addition to this validation in simulation, we note that
vineyards in recent California res have performed poorly as re
breaks, often burning substantially; this is unsurprising as vine-
yards seldom have supplemental irrigation applied late in the sea-
son and are thus dry and woody. Banana orchards are not
common today in California, but there are reports an Australian
banana orchard with grass groundcover burned during a major re-
cent wildre, damaging the banana leaves but leaving pseudos-
tems intact to resume growth soon after the re was
extinguished (53); this points to the importance of our incorpor-
ation of nonammable groundcover.
When reporting reline intensity and arrival time for banana
buffers, we report values directly from FARSITE as produced in
our simulation runs. We compute the mean of the reported re-
line intensity for the pixels of the buffer region in each run of
the simulation. Similarly, we compute the mean arrival time of
the re, in minutes from the ignition, across the pixels of the buf-
fer region in question in each simulation run.
Validating ember spotting behavior. To ensure that the re simu-
lation accurately reects the extreme ember spotting behavior
seen in recent WUI res in California, we ran several validation ex-
periments with and without re buffers and with and without em-
ber spotting enabled in the simulator. With ember spotting
enabled, lowering the wind speed below the actual historical
wind speeds also showed lower re spread, as expected.
Climate scenarios. We used RCP 4.5 and RCP 8.5 to simulate fu-
ture scenarios (49). We acquired temperature projections for the
study area from eight GCMs (BCC-CSM2-MR, CNRM-CM6-1,
CNRM-ESM2-1, CanESM5, IPSL-CM6A-LR, MIROC-ES2L, MIROC6,
MRI-ESM2-0 (48)) and projected the temperature for Tubbs Fire
scenarios. We calculated the mean temperature offset based on
the maximum difference between each predicted temperature
and the 2017 Tubbs Fire area data from October 2017, following
the approach taken in the literature (54–57). We considered ocean
warming as a contributing factor to decreases in relative humidity
on land (58). After projecting the temperature, we modied the lo-
cal relative humidity value for the simulation of re spread based
on the static dew point, yielding, in future scenarios, lower humid-
ity that drives faster re spread. Fire progression under the no buf-
fer scenarios for projected climatic conditions shows that the
urban burn will increase. We compared re spread under RCP
4.5 and RCP 8.5 (49). The RCP models are described in detail in
the IPCC reports and the future model projections are from the
worldclim database which is standard in species distribution
modeling. We used all of the models available. The RCP pathways
have different assumptions about human activity and the GCM
models have different assumptions about the way radiative for-
cing will inuence climate. Exploring multiple scenarios allows
for a better understanding of what potentially could occur in the
future. We focus on RCP 4.5 and 8.5 for the analysis. RCP 4.5 rep-
resents the intermediate scenario that happens to be the most
probable carbon dioxide emission reduction. RCP 8.5 described
the worst scenario where the emissions policies continue as de-
scribed in the stated policies.
4 | PNAS Nexus, 2023, Vol. 2, No. 10
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
Economic return on edible re buffers
Fire buffers of any type are not widely deployed today and are not
currently nancially sustainable. We computed the cost of edible
re buffers, independent of their re mitigation savings, using an
enterprise budget modied from the University of Florida banana
template (59); the modications are shown with references for
changes that were made in order to develop a range of potential
buffer value scenarios in Table S2. We consider a range of yield
scenarios, banana fruit value scenarios, and land cost scenarios.
All cost values are given in 2021 USD.
Practical considerations for conversion of land in California
WUI areas
Converting existing WUI land to edible re buffers is likely to be an
intricate and site-specic process that cannot be fully addressed
here. In brief, our consideration was whether such conversion is
practically feasible, but we leave to future work the exact proce-
dures by which such conversion can or should be done. In this,
we take a perspective gained from our professional experiences
in agricultural and horticultural research and practice in
Mediterranean and tropical regions and successes in banana cul-
tivation in both Northern and Southern California. WUI lands
suitable for edible re buffers are largely covered by low- to
medium-density annual grasses and low-growing shrubs. Some
number of these regions are beginning to type convert as decades
of re suppression and climate change lead to a greater frequency
and intensity of res (60–62). In addition, these lands are often in
an already-disturbed state given the very urban lands/housing
that we aim to protect. The establishment of edible re buffers
on these lands is likely largely limited by infrastructure availabil-
ity, particularly water for irrigation.
Results
Banana and vineyard/orchard fuel types
We considered two possible edible re buffers in our primary ex-
periments: banana and vineyards/orchards. Vineyards and or-
chards are already known to be viable across much of California.
To explore the efcacy of planting banana buffers, we explored
the suitability of using ecological niche models (ENM) in California
(Fig. 2). We rst examined current suitability (2017) then explored
future climate models under two different RCPs (4.5 and 8.5) by
2090. For each future climate model (with different assumptions
about radiative forcing) an ENM was generated, with the models
being averaged for the nal suitability for 2090 under each RCP.
Within each model, each grid cell has projected variables which
are then used to derive a suitability score for a given organism,
which are different under cultivated (65) and natural conditions
(66). Using current information, bananas are already widely suit-
able in California, especially along coastal hills that represent the
majority of WUI and the highly populated areas, and the ability to
grow will improve as warming proceeds toward 2090 in Fig. 2.
Climate change scenarios
We consider the effectiveness of re buffers under two future cli-
mate scenarios in 2090 in addition to the 2007 Tubbs Fire baseline.
The rst is the no mitigation business-as-usual scenario,
Representative Concentration Pathway (RCP) 8.5, and the second
is the moderate mitigation scenario, RCP 4.5 (49). We use the aver-
age of eight future climate models under each scenario (48) to ex-
plore the suitability of banana in California. In modeling future
res, we explore the increase in temperature and the changes in
humidity that would occur. Using the mean of eight models pro-
vided a way to explore the variation in future predictions to get
a better understanding of the efcacy of potential mitigation
strategies. We model the temperature and humidity in these fu-
ture scenarios in the location of the Tubbs Fire.
Fireline intensity
Fire buffers can have a variety of spatial congurations. We
consider rectangular re buffer congurations placed on the
human-managed land that abuts the WUI (6). Real-world buffer
placements are complicated due to land ownership, site availabil-
ity, and other factors. In our simulated evaluation, we explore the
effectiveness of the banana buffers by placing them between the
re ignition site and the populated region to be protected, as our
aim is to evaluate the ability of the buffer to mitigate res that
spread from wild to urban land.
We dene the width of the buffer to be the minimum distance
from the wildland side of the buffer to the side of the buffer
that abuts the WUI; we congure buffers across a range of sizes.
We limit the width of buffers to the sizes of agricultural parcels
in the region, though even larger buffers have the potential to
produce improved re mitigation. Our canonical buffer for
these experiments is a medium 633 m buffer (tested alongside a
very small 200 m buffer, small 390 m buffer, and 1280 m very large
buffer), and we consider a variety of fuel types, including a fully
nonburnable buffer (NB3) (45), baseline regional vegetation (e.g.
a mix of grasses, shrubs, and trees), vineyards/orchards, and
bananas.
Fireline intensity with buffers ranging from small (easily
adopted) to very large (difcult to adopt) (Fig. 3B). We show the in-
tensity of the re without buffers, with medium-sized (633 m)
vineyard buffers, and with banana buffers from very small to
very large, considering the 2017 climate conditions of the Tubbs
Fire and projected conditions in 2090 under RCP 4.5 and RCP 8.5
(49). We also compare against control and various fuel treatments
and their mitigations, such as mechanical thinning and pre-
scribed re, in a similar environment as studied by Stephens
and Moghaddas (67). We nd that a medium-sized (633 m) banana
buffer results in a 96% decrease in reline intensity at the WUI,
comparable to the combined effect of prescribed burns and mech-
anical thinning.
In Fig. 3A, we show the time of arrival of the re to the edge of
the wildland edible re buffer that abuts the WUI. We consider
each size of the buffer and once again consider the 2017 climate
conditions of the Tubbs Fire and projected conditions in 2090
under RCP 4.5 and RCP 8.5. We nd that very small buffers
(≤200 m) are inadequate to substantially slow the spread of the si-
mulated Tubbs Fire, regardless of buffer fuel type; we performed
repeated experiments that indicate that this inadequacy of small
buffers is due to ember cast, as winds carry embers over small buf-
fers. The only other buffer type that showed a similarly poor effect
is vineyard/orchard buffers at the end of summer, which are no
better and sometimes worse than the baseline fuels and thus do
not provide a re mitigation benet. We nd that medium-sized
(633 m) banana buffers would slow the arrival of the Tubbs Fire
by 316 min. This would double the amount of time to make a sub-
stantive intervention (e.g. initial attack (68)) on a re. As the time
and intensity of the re are two of the key determinants of the
ability of re crews to stop a re, this would have a major practical
impact on the protection of communities. While under climate
change the re spreads faster, we nd banana re buffers con-
tinue to have a substantial protective benet.
Fu et al. | 5
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
WUI burn area
Further, we show that medium-sized (633 m) and larger banana
buffers provide a substantial re mitigation effect on WUI/urban
re spread. As we mentioned earlier, while FARSITE (26) is not
ideal for measuring WUI/urban re spread, there exists no better
alternative simulator for such projections and our validation ex-
periment showed that it is possible to compute an accurate re
perimeter when recreating the Tubbs Fire; we report the values
with that caveat.
We nd that the medium-sized banana buffer results in a WUI/
urban burn rate of 56.9% of the unmitigated baseline. This com-
pares favorably to a vineyard buffer of the same size, which pro-
vides no clear protective effect (Fig. 4A). Very large buffers (more
than a km in width) are even more effective but likely impractical
to manage. We also show the ndings of re buffers under future
climate change (Fig. 4B and C). We nd that banana re buffers
continue to be effective in these future climate scenarios, in which
there is a baseline, no-buffer urban area burn rate by 2090 of
Fig. 2. MaxEnt model (63) of banana suitability in current (orange) and in 2090 (lilac), with the overlap being peach. The model from 2090 represent the
mean suitability from eight GCMs (48) (BCC-CSM2-MR, CNRM-CM6-1, CNRM-ESM2-1, CanESM5,IPSL-CM6A-LR, MIROC-ES2L, MIROC6, MRI-ESM2-0)
under the RCP 8.5 scenario. Counties outlined in red represent the counties that have had the 10 most intense res (in terms of damage) in California
history. The inset represents Sonoma County, where the Tubbs re started. Currently, 91% of California’s population lives in counties that contain land
suitable for banana cultivation. The base map of California was sourced from US Census Bureau (64).
Fig. 3. Comparison of re buffers sizes (very small =200 m, small =390 m, medium =633 m, large =1070 m, very large =1280 m) and alternative re
buffers. The buffer is placed roughly perpendicular to the direction of movement of the Tubbs re (2017). The urban area is dened through the SILVIS
WUI dataset (11, 36, 37). The buffer is placed between the ignition point and the dened urban area with no banana buffer or with a range of banana
buffers ranging from very small to very large, considering the 2017 climate conditions of the Tubbs Fire and projected conditions in 2090 under RCP 4.5
and RCP 8.5. A) This is a comparison of the arrival time of the re to the side of the human-managed wildland re buffer region that abuts the WUI, The
y-axis depicts the time since the re’s ignition. Boxplots show the range of values for ve scenarios. B) Here, we show reline intensity between different
types of buffers. The Y-axis, reline intensity indicated reghting difculty with higher values indicating a more challenging re. The horizontal lines in
the gure show the reline intensity existing landscape fuels. The boxplots show the mitigation potential of banana buffers under historical and future
(RCP 4.5, RCP 8.5) climate conditions. In our simulations prescribed re and mechanical approaches are the most effective fuel treatment, similar to
previous work (67). Banana buffers performed better than all other treatments with the exception of the prescribed burns, including previous eld studies
(control line at 826.0(kW/m) (67)). The best existing method, prescribed burn (Fire only), reduces reline intensity to 21.0(kW/m). The mechanical
thinning from previous eld experiments (67) was found to only exacerbate reline intensity. By increasing the size of the buffer, bananas can perform as
well as the best existing fuel treatment, prescribed burn. For each scenario, we conducted 45 FARSITE simulations.
6 | PNAS Nexus, 2023, Vol. 2, No. 10
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
154.8% under RCP 8.5 and 135.0% under RCP 4.5 relative to base-
line (2017). We nd that a medium-sized banana buffer at this
WUI results in a reduced urban burn rate, even under predicted
climate change, of 120.1% under RCP 8.5 and 85.9% under RCP
4.5 relative to the same baseline (2017).
Replicating the Tubbs Fire
As our focus is the impact on the WUI, not simply total wildre
spread, we separately veried the replication of our simulation
on the urban perimeter and the total perimeter relative to ground-
truth USGS satellite data (69). For the burn area in the WUI and ur-
ban land in the Tubbs Fire, we achieved F1 scores of {0.68, 0.74,
0.64} with respect to the three available satellite time points and
{0.75, 0.78, 0.74} for the total area burned, exceeding the scores
of general re models (70), despite the fact that FARSITE (26) is pri-
marily intended for wildland res. Our focus in this replication is
on the area of WUI/urban land burned, though our goal in this
work is not to improve upon WUI and urban re modeling tools.
Economic analysis for edible re buffers
Human-managed wildlands have multiple possible uses; for ed-
ible re buffers to be adopted they likely need to create additional
revenue while helping to mitigate re. To explore this potential
value, we studied the cost of planting and harvesting banana by
creating an enterprise budget, with costs chosen based on the
California context for organic production with recycled water,
and yield projections based on banana grown in similar
Mediterranean climates. We consider multiple yields, crop value,
and land cost scenarios. Based on this budget the potential
prot for banana buffers ranges from a low-yield, low-value,
high-land-cost scenario with negligible prot (but no loss) to a
high-yield, high-value, low-land-cost scenario with a prot of
76,136 USD/hectare (see Table S2), accounting for planting, irriga-
tion, inputs, maintenance, harvesting, and other relevant costs.
This analysis is based on the pricing of normal cultivars grown
under organic conditions and does not account for secondary ben-
ets such as a decrease in reghting or insurance costs. There is a
potential increase in value if markets develop for specialty culti-
vars (71). Even if individual buffers do not achieve the high-yield
scenario, any value rather than a liability creates a potential to
consider these buffers as a near-term nancially and ecologically
sustainable solution. In addition, each individual banana buffer
region is unlikely to be affected by re in any given year though,
over a longer timespan of decades, it is likely that each banana
buffer will be affected by the re. As we found in our simulations,
banana buffers largely remain intact even when affected by re
(72), as the banana pseudostems themselves are largely non-
ammable, and even if damaged by re, after removal the plant
will resprout from the corm. Some plant replacement, along
with irrigation pipe replacement, may be necessary to recover
from an extensive re.
In addition to the value of the food produced, there is a substan-
tial improvement in land value by changing the groundcover.
Esthetic preferences show that green spaces in and around urban
areas, particularly those with trees, are preferred (73). Landscape
multifunctionality is becoming increasingly important in all as-
pects of the built environment (74). Municipalities invest in devel-
oping public areas that provide multiple services; these public
areas include parks, roads, schools, and business parks. Public
rebreaks that also provide healthy, nutritious food could provide
prot, savings in terms of re damage, and additional jobs, creat-
ing win-win scenarios.
Discussion
Fuel reduction
Fuel reduction is a key practice in re risk management. The U.S.
Forest Service performs spatial risk analysis by examining simu-
lated res given a range of hypothetical fuel treatment scenarios,
including change of the canopy cover and fuel map. However, the
actual costs and complexity of such treatment make it difcult for
land managers to reduce both current and future re risk (15).
Similarly, for forest fuel-reduction treatments such as prescribed
re and its mechanical surrogates, the effect is transient so long-
term mitigation is hard to achieve (75).
Fireghting resources
The predicted increase in re occurrence and severity will de-
crease the success rate of the initial attack making re control
more difcult and necessitating an increase in reghting
resources (76). The 3–5 h of the initial attack is considered to be
crucial for mitigating re spread and suppression costs (68).
Suppression efforts at this stage may include air tankers and con-
trol line resources such as pumps, hoses, bulldozers, and shovels
Fig. 4. Comparison of re buffers’ effectiveness in mitigating urban damage in present and future climate scenarios for the Tubbs Fire (2017). The buffer
is placed roughly perpendicular to the moving direction of the Tubbs re. The urban area is dened through the SILVIS WUI dataset (11, 36, 37). The buffer
is placed between the ignition point and the dened urban area. Boxplots show the signicant set results of the results for different buffer size scenarios
(small =390 m, medium =633 m, very large =1280 m) simulations of the FARSITE re simulation for each buffer. A) Comparison of the urban burn area
of baseline (no buffer) scenario and medium-sized vineyard and banana buffers. We conducted 72 FARSITE simulations. B) Comparison of urban burn
area in banana buffer of different widths in projected RCP 4.5 2090 conditions. We conducted 96 FARSITE simulations. C) Comparison of urban burn area
in banana buffer of different widths in projected RCP 8.5 2090 conditions. We conducted 96 FARSITE simulations. In both B) and C), very large buffers,
which are over 1 km in size, entirely stop re spread.
Fu et al. | 7
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
(77). The initial attack fails when suppression resources arrive too
late, re intensity is too high, or when reghters fail to contain
the spread to a predetermined target area (78). If the initial attack
is insufcient and an advancing re crosses a dened evacuation
trigger, then an evacuation order is released to nearby communi-
ties (79). Reducing the rate of spread of res during the initial at-
tack could reduce the need for evacuations, mitigate evacuation
risk, and allow reghting personnel to execute substantive sup-
pression efforts.
Firebreaks and re buffers
Creating rebreaks through prescribed burns or mechanical
methods has also proven to be effective. Recent work has pro-
posed multiple strategies, including: (i) replacement of the species
mix to less ammable natives (80), (ii) ecological vegetation man-
agement in high-risk areas such as by utility companies on lands
near power lines (81), and (iii) conversion to low-growing succu-
lents, irrigated agriculture (20), or other green rebreaks (82). In
addition to preconstructed rebreaks, reghting techniques in-
corporate the creation of bulldozer lines during initial attack
(83). However, each of these strategies imposes up-front and on-
going costs.
Building and land use
The last two decades have seen legislative efforts to mandate and
to reimburse individual homeowners and municipalities to re-
spond to re risk (84). The increase in res has led to changes in
building materials (12) and new regulations for WUI areas.
These regulations have helped change patterns of vegetation
management through an increased clearing of dead plant mater-
ial and by limiting the use of highly ammable plants (85). We dis-
cuss current California land use below and note the current
human management of wildlands that abut the WUI.
Practical considerations for edible re buffers
While we have focused on banana as a key crop for edible re buf-
fers, next we discuss some requirements for and limitations to its
use. More study is needed on the economic and ecological impacts
of edible re buffers.
Microclimatic suitability
Not all Mediterranean areas are appropriate for banana cultiva-
tion, though present WUI-adjacent areas in California are nearly
all suitable for banana cultivation as they frequently occur in
thermal belts with little to no frost, so with careful cultivar selec-
tion, bananas are likely widely appropriate. Additionally, due to
the aridity of Mediterranean climates, it appears that buffers
need to be larger than in regions such as China where small vege-
tative rebreaks are employed effectively (82).
Water needs
Banana is a high-water-need crop. We nd through an informal
analysis in Southern California that recycled water is widely avail-
able to meet this need, and its availability is increasing quickly
with recycled water infrastructure expansion (17). We performed
this analysis by manually collecting and examining (proprietary)
recycled water maps from the individual water districts across
the ten counties of Southern California with a total population
of over 23 million people. Recycled water availability is not as
widespread in Northern California due to the greater availability
of freshwater, but such infrastructure is similarly being expanded
across the region. In addition, many high-re-risk WUI areas have
been developed more recently, and these areas have ubiquitous
recycled water, currently used for ornamental landscaping. In
addition, banana buffers could leverage even lower-cost, undisin-
fected secondary recycled water (86). Recycled water also carries
with it plant nutrients that can decrease or eliminate the need
for supplemental fertilizer. As California is a region with increas-
ing hydrological volatility and long dry seasons, there is a need to
put available water resources to better use. Maximizing the reuse
of municipal water, in the form of recycled water, to grow food
while also protecting those same WUI areas from re is a virtuous
cycle.
Development of re-resistant cultivars
Plant moisture content is a trait that appears to have genetic vari-
ance within species and can be selected for (87). Thus breeding
programs can identify species and populations within species
that could be used for edible re buffers. This could be a fruitful
use of public funds: to select new, multifunctional varieties.
Groundcover and establishment
The high-re-risk areas we consider are often hilly and covered in
re-prone vegetation. While banana plants themselves are re re-
sistant, the land underneath the banana canopy must be man-
aged as well. Annual dryland-adapted grasses quickly cover
such WUI lands, sometimes within a single season after a re.
Such grasses would signicantly decrease the effectiveness of
any buffer or break. Thus, edible re buffers should be established
along with low-growing, nonammable groundcover species;
common examples include Senecio mandraliscae and low-growing
cultivars of Aloe ciliaris. Such groundcovers can be maintained
along with the crop species with a cellulose-based biodegradable
weed fabric to prevent the emergence of grasses during establish-
ment. To avoid a monoculture of banana buffers that succumb to
the spread of a pest or pathogen (e.g. Panama Wilt), thereby
undermining the re buffer, deployments of banana buffers
should employ spatiotemporal diversity of cultivars. Such diver-
sity is also likely to be benecial to adapt to local microclimatic
needs and market demand. Testing multiple groundcovers is feas-
ible because, unlike most horticultural crops, upkeep on banana
farms is minimal (88).
Land considerations
Soil suitability for banana cultivation is a possible concern, but in
the regions we consider, soils are often well-drained clay loam
(e.g. the regions of concern in California, including the area of
the Tubbs Fire), with good suitability for banana cultivation with
water and fertility supplementation (e.g. through recycled water).
If the land has recently suffered a burn there is a risk of soil hydro-
phobia that exacerbates runoff; such soils would benet from
limited-depth mechanical tillage before banana cultivation be-
gins. Pathogenic nematodes, fungi, and viruses for banana are
currently not present in California and thus not a factor in this
context, but may need consideration in other regions. The major
limitations to banana cultivation are not biological; banana grows
well on hillsides and in erosion-prone soils and banana root sys-
tems are well adapted to shallow soils (88), yielding potential
erosion-control benets. Rather, zoning (e.g. residential, agricul-
tural, parkland) and political authority (e.g. cities, counties,
HOAs, utilities) are likely to present challenges. The primary pur-
pose of edible re buffers for re mitigation is in line with existing
land uses and zoning in human-managed wildland areas that
abut the WUI. The availability of recycled water eases the
8 | PNAS Nexus, 2023, Vol. 2, No. 10
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
deployment of novel land use change proposals such as edible re
buffers.
Extreme re behavior
Recent years have seen a rise in “extreme” re behavior, including
more frequent occurrence (9) of lightning-producing pyrocumulo-
nimbuses and pyrovortices (89). It seems unlikely that any buffers
or breaks—even if completely nonammable—can stop the
spread of such res. However, it may still be the case that edible
re buffers, in combination with other mitigation efforts such as
low-ammability urban landscaping and building materials, can
meaningfully respond to worsening extremes, especially in high-
population nonintermix WUI regions. Further, katabatic “Santa
Ana”-type winds have led to increased severity in many of the
most damaging effects.
Practical, sustainable options that improve the use of water,
provide food, improve esthetics, and protect people and homes
are needed for municipalities to take concrete action to change
the status quo; edible re buffers are one such response.
Further, the increasing re burden under future climates (2) ne-
cessitates creative solutions, and edible re buffers provide the
potential for substantial relief under both RCP 8.5 and RCP 4.5,
showing how creative combinations of climate science, ecology,
and agriculture can tackle wicked problems. Looking at the histor-
ic use of banana in California, its future suitability, and the value
such edible re buffers would yield, banana becomes a prime test
case for how to use edible re buffers to sustainably mitigate wild-
re risk.
Alternative buffer scenarios
We explored banana in depth as a possible crop for edible re buf-
fers, but it is not the only possible option, nor is a monoculture ne-
cessary for buffer effectiveness. Other low-ammability,
high-value crops could appropriately mitigate res and serve as
re buffers. The key properties of such re buffers are that they
are not only low-ammability and yield a prot but also that
they are suitable in other dimensions for the zoning, terrain,
and context of many WUI lands. Specically, standard vegetable
crops (e.g. of leafy greens or vegetables), while low-ammability,
are not suitable given the need for signicant labor and machin-
ery, and the need for relatively at terrain. Many WUIs, especially
in California, have variable terrain and are in an environment (e.g.
residential neighborhoods) where large machinery and the work-
force of a working farm are not suitable. Banana, like most peren-
nial crops, requires less labor and machinery and is more
adaptable to varied terrain. In addition, automated systems could
be developed to automate aspects of the banana harvest; such an
environment would provide a prime location to test such new
technology. Thus perennial, low-ammability species are ideal.
Ginger is a low-ammability culinary crop that would be suit-
able for growing in the same regions as banana, though mechan-
ical harvesting may be required to keep harvest costs low. Indeed,
there is a long history of agroforestry in banana orchards, and
many different cover crops have been successful (90). Including
an intercrop such as ginger could increase the value of the total
crop, take advantage of microclimates (as ginger grows well in
partial shade and is low growing), and increase species diversity
in the buffer region.
The most widespread high-value crop in Mediterranean cli-
mates is wine grapes. We evaluated the effectiveness of vineyards
as re buffers and found that they may not be suitable due to their
ammability. While currently not a widely grown crop, carob is a
potential high-value, high-yield, and low-ammability crop ap-
propriate for high-re-risk Mediterranean climates. There is
some evidence that carob is less ammable than other similar
dryland species (25). Since carob can grow without supplemental
irrigation in dry-summer regions, it may be appropriate in settings
where irrigation is either unavailable or too costly.
Finally, some regions may consider physical nonammable re
buffers, such as ones made from concrete or metal, as opposed to
crop-based buffers. However such buffers would impose a high in-
stallation cost, would require annual maintenance (to keep soils
from building up on top thereby enabling grass growth), would
provide no revenue or additional benets, and would exacerbate
problems such as water and soil runoff. Like bulldozer lines,
they are an expensive and intrusive intervention that may be suit-
able in certain circumstances but unlikely to yield win-win
outcomes.
Current California land use
It is also important to consider pre-existing land use where edible
re buffers could be grown. Unlike truly wild areas, the lands we
consider for such buffers in California are human-managed re-
gions that are not truly wild due to current re management prac-
tices, urban development, and the proliferation of introduced
annual grass species that have outcompeted native perennial
grasses (91, 92). In addition, future climate conditions with in-
creased re frequency and severity are projected to induce losses
of ecosystem services; these can be mitigated by reducing green-
house gas emissions (93) and possibly, in human-managed wild-
land areas, through the creation of edible re buffers. Moreover,
in many existing WUI areas of California, landscaping using the
genus Strelitzia, which is visually very similar to banana, is wide-
spread. Finally, dening clear management techniques that are
acceptable in this peri-urban agriculture is essential, and any reg-
ulations need to be rigorously enforced to ensure no negative con-
sequences for human health.
Multifunctional landscapes
High-value and nonammable crops can improve ecological ben-
ets while mitigating wildre hazards. Complex agroecosystems
are becoming more prevalent in land use planning to ensure eco-
logical benets while considering bio-diversity (94). Banana buf-
fers are a step towards multifunctional landscapes. Moreover,
banana buffers may integrate with existing ecosystems more
seamlessly than prior re management approaches (20).
Complex agroecosystems that provide both ecological and eco-
nomic benets fulll more societal goals and encourage further
landscape design exploration.
Future work
In this article, we have shown through simulation that banana
buffers can be effective in mitigating the spread of WUI res.
Simulation feasibility potential fuel treatment reduces the over-
head of real-world control burn trials. Given the increasing danger
and prevalence of WUI res, it is necessary to conduct
well-curated controlled-burn experiments to ensure ecological,
economic, and sustainable wildre mitigation. The practical feasi-
bility of such burn trials requires further study as there are limita-
tions due to the unavailability of burn permits in the counties in
California where there would be the highest suitability, due to
high re risk (95).
Fu et al. | 9
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
Funding
The authors declare no funding.
Author contributions
X.F.: performed research, analyzed data, wrote the paper; A.L.:
performed research, analyzed data, wrote the paper; M.K.: per-
formed research, wrote the paper; B.R.: designed research, per-
formed research, analyzed data, wrote the paper.
Data availability
The data that support the ndings of this study are available in the
databases listed in the Methods. Data and code are available at the
following DOI:10.5281/zenodo.6373729.
References
1 Mora C, et al. 2018. Broad threat to humanity from cumulative cli-
mate hazards intensified by greenhouse gas emissions. Nat Clim
Change. 8(12):1062–1071.
2 Mann ML, et al. 2016. Incorporating anthropogenic influences
into fire probability models: effects of human activity and cli-
mate change on fire activity in California. PLoS One. 11(4):
e0153589.
3 Abatzoglou JT, Williams AP. 2016. Impact of anthropogenic cli-
mate change on wildfire across western US forests. Proc Natl
Acad Sci USA. 113(42):11770–11775.
4 Ray DK, et al. 2019. Climate change has likely already affected
global food production. PLoS One. 14(5):e0217148.
5 Fitzpatrick MC, Dunn RR. 2019. Contemporary climatic analogs
for 540 north American urban areas in the late 21st century.
Nat Commun. 10(1):1–7.
6 Stewart SI, Radeloff VC, Hammer RB, Hawbaker TJ. 2007.
Defining the wildland–urban interface. J For. 105(4):201–207.
7 Radeloff VC, et al. 2018. Rapid growth of the US wildland-urban
interface raises wildfire risk. Proc Natl Acad Sci USA. 115(13):
3314–3319.
8 Kramer HA, Mockrin MH, Alexandre PM, Radeloff VC. 2019. High
wildfire damage in interface communities in California. Int J
Wildland Fire. 28(9):641–650.
9 Goss M, et al. 2020. Climate change is increasing the likelihood of
extreme autumn wildfire conditions across California. Environ
Res Lett. 15(9):094016.
10 Thomas D, Butry D, Gilbert S, Webb D, Fung J. 2017. The costs and
losses of wildfires. NIST Special Publication, 1215.
11 Martinuzzi S. 2015. The 2010 wildland-urban interface of the
conterminous United States. Research Map NRS-8. Newtown
Square, PA: US Department of Agriculture, Forest Service, Northern
Research Station. 124 p. [includes pull-out map]. Vol. 8. p. 1–124.
12 Stockmann K, Burchfield J, Calkin D, Venn T. 2010. Guiding pre-
ventative wildland fire mitigation policy and decisions with an
economic modeling system. For Policy Econ. 12(2):147–154.
13 North M, Collins BM, Stephens S. 2012. Using fire to increase the
scale, benefits, and future maintenance of fuels treatments. J For.
110(7):392–401.
14 Schoennagel T, Nelson CR, Theobald DM, Carnwath GC,
Chapman TB. 2009. Implementation of national fire plan treat-
ments near the wildland–urban interface in the western United
States. Proc Natl Acad Sci USA. 106(26):10706–10711.
15 Chung W. 2015. Optimizing fuel treatments to reduce wildland
fire risk. Curr For Rep. 1(1):44–51.
16 CalFire. 2019. Fire-resistant landscaping.
17 Bischel HN, Simon GL, Frisby TM, Luthy RG. 2012. Management
experiences and trends for water reuse implementation in nor-
thern California. Environ Sci Technol. 46(1):180–188.
18 Tortajada C. 2020. Contributions of recycled wastewater to clean
water and sanitation sustainable development goals. NPJ Clean
Water. 3(1):1–6.
19 De Lillis M, Bianco PM, Loreto F. 2009. The influence of leaf water
content and isoprenoids on flammability of some mediterranean
woody species. Int J Wildland Fire. 18(2):203–212.
20 Salis M, et al. 2016. Evaluating alternative fuel treatment strat-
egies to reduce wildfire losses in a Mediterranean area. For Ecol
Manage. 368:207–221.
21 Gomez JA, Amato M, Celano G, Koubouris GC. 2008. Organic olive
orchards on sloping land: more than a specialty niche production
system?. J Environ Manage. 89(2):99–109.
22 Trenberth KE, et al. 2014. Global warming and changes in
drought. Nat Clim Change. 4(1):17–22.
23 Mahouachi J, López-Climent MF, Gómez-Cadenas A. 2014.
Hormonal and hydroxycinnamic acids profiles in banana leaves
in response to various periods of water stress. Sci World J.
2014.1–9.
24 Ismail MR, Yusoff MK, Mahmood M. 2004. Growth, water rela-
tions, stomatal conductance and proline concentration in water
stressed banana (Musa spp.) plants. Asian J Plant Sci. 3(6):709–713.
25 Srec
ec S, Kremer D, Karlovic K, Purgar DD, Erhatic R. 2017.
Possible role of carob tree (Ceratonia siliqua L.) in fire protection
of agro-forest systems of croatian south adriatic islands regard-
ing the similarities with other mediterranean countries. In: 17th
International Multidisciplinary Geoconference SGEM 2017.
26 Finney MA. 1998. FARSITE, Fire Area Simulator–model development
and evaluation. US Department of Agriculture, Forest Service,
Rocky Mountain Research Station.
27 Rollins MG. 2009. Landfire: a nationally consistent vegetation,
wildland fire, and fuel assessment. Int J Wildland Fire. 18(3):
235–249.
28 Andrews PL, Bevins CD, Seli RC. 2005. Behaveplus fire modeling
system, version 4.0: User’s guide. Gen. Tech. Rep. RMRS-GTR-106
Revised. Ogden, UT: Department of Agriculture, Forest Service,
Rocky Mountain Research Station. 132p. 106.
29 I ESRI. 2020. Arcgis: release 10.8. Technical Report, Esri Inc,
Redmond.
30 GDAL/OGR contributors. 2020. GDAL/OGR geospatial data abstrac-
tion software library. Open Source Geospatial Foundation.
31 Gillies S. 2013. Rasterio: geospatial raster i/o for Python pro-
grammers. Technical Report, Mapbox.
32 Waskom M, et al. 2017. mwaskom/seaborn: v0.8.1 (september
2017). Technical Report, Zenodo, September.
33 Jordahl K, et al. 2019. geopandas/geopandas: v0.6.1. Technical
Report, Zenodo, October.
34 Walters SP, Schneider NJ, Guthrie JD. 2011. Geospatial multi-
agency coordination (geomac) wildland fire perimeters, 2008.
Technical Report, US Geological Survey.
35 NOAA. 2020. Past weather by zip code - data table.
36 Radeloff VCH, et al. 2017. The 1990–2010 wildland-urban inter-
face of the conterminous United States-geospatial data.
Technical Report, SILVIS Lab, Dept of Forest & Wildlife Ecology,
University of Wisconsin-Madison.
37 Radeloff VC, et al. 2005. The wildland–urban interface in the
United States. Ecol Appl. 15(3):799–805.
38 Stratton RD. 1998. Guidebook on LANDFIRE fuels data acquisition, cri-
tique, modification, maintenance, and model calibration. US
10 | PNAS Nexus, 2023, Vol. 2, No. 10
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
Department of Agriculture, Forest Service, Rocky Mountain
Research Station.
39 Stratton RD. 2006. Guidance on spatial wildland fire analysis: models,
tools, and techniques. United States Department of Agriculture,
Forest Service, Rocky Mountain ….
40 Rothermel RC. 1972. A mathematical model for predicting fire spread
in wildland fuels. Vol. 115. Intermountain Forest & Range
Experiment Station, Forest Service, US ….
41 Rothermel RC. 1991. Predicting behavior and size of crown fires in the
Northern Rocky Mountains. Vol. 438. US Department of Agriculture,
Forest Service, Intermountain Forest and Range ….
42 Finney MA. 2006. An overview of flammap fire modeling capabil-
ities. In: Andrews PL, Butler BW, comps. Fuels Management-How to
Measure Success: Conference Proceedings; 2006 March 28–30;
Portland, OR. Proceedings RMRS-P-41, Vol. 41. Fort Collins (CO):
US Department of Agriculture, Forest Service, Rocky Mountain
Research Station. p. 213–220.
43 Scott JH, Thompson MP, Calkin DE. 2013. A wildfire risk assess-
ment framework for land and resource management.
Technical Report, U.S. Department of Agriculture, Forest
Service, Rocky Mountain Research Station.
44 Henning AH, Cox J, Shew D. 2016. CAL FIRE’s damage inspection
program: its evolution and implementation. In: NFPA Conference
and Expo; June; Las Vegas, NV. p. 13–16.
45 Scott JH. 2005. Standard fire behavior fuel models: a comprehensive set
for use with Rothermel’s surface fire spread model. US Department of
Agriculture, Forest Service, Rocky Mountain Research Station.
46 GBIF.org. 2019. Occurrence download.
47 Elith J, et al. 2011. A statistical explanation of maxent for ecolo-
gists. Divers Distrib. 17(1):43–57.
48 Fick SE, Hijmans RJ. 2017. Worldclim 2: new 1-km spatial reso-
lution climate surfaces for global land areas. Int J Climatol.
37(12):4302–4315.
49 Field CB, Barros VR. 2014. Climate change 2014–impacts, adaptation
and vulnerability: regional aspects. Cambridge University Press.
50 Anderson HE. 1981. Aids to determining fuel models for estimating fire
behavior. Vol. 122. US Department of Agriculture, Forest Service,
Intermountain Forest and Range ….
51 Phoungchandang S, Woods JL. 2000. Solar drying of bananas:
mathematical model, laboratory simulation, and field data com-
pared. J Food Sci. 65(6):990–996.
52 Salis M, et al. 2015. Analyzing seasonal patterns of wildfire expos-
ure factors in Sardinia, Italy. Environ Monit Assess. 187(1):4175.
53 Sexton-McGrath K. 2018. Bushfire tears through banana planta-
tion of sustainable produce innovators natural evolution. ABC
Far North.
54 Pironon S, et al. 2019. Potential adaptive strategies for 29 sub-
saharan crops under future climate change. Nat Clim Change.
9(10):758–763.
55 Fumia N, et al. 2022. Wild relatives of potato may bolster its adap-
tation to new niches under future climate scenarios. Food Energy
Secur. 11(2):e360.
56 Khoury CK, et al. 2015. Distributions, ex situ conservation prior-
ities, and genetic resource potential of crop wild relatives of
sweetpotato [Ipomoea batatas (L.) Lam., I. Series Batatas]. Front
Plant Sci. 6:251.
57 Xie W, et al. 2018. Decreases in global beer supply due to extreme
drought and heat. Nat Plants. 4(11):964–973.
58 Byrne MP, O’Gorman PA. 2018. Trends in continental tempera-
ture and humidity directly linked to ocean warming. Proc Natl
Acad Sci USA. 115(19):4863–4868.
59 ufl.edu. 2014. Bananas: estimated production costs in Miami-
Dade County.
60 Pausas JG, Keeley JE. 2021. Wildfires and global change. Front Ecol
Environ. 19(7):387–395.
61 Fried JS, Torn MS, Mills E. 2004. The impact of climate change on
wildfire severity: a regional forecast for northern California. Clim
Change. 64(1):169–191.
62 Barbero R, Abatzoglou JT, Larkin NK, Kolden CA, Stocks B. 2015.
Climate change presents increased potential for very large fires
in the contiguous United States. Int J Wildland Fire. 24(7):892–899.
63 Phillips SJ, Anderson RP, Schapire RE. 2006. Maximum entropy
modeling of species geographic distributions. Ecol Modell.
190(3-4):231–259.
64 Smith J. 2014. US census bureau. [accessed 2022 Apr 20]. https://
web.archive.org/web/20190301191426/https://www.census.gov/
geo/maps-data/data/tiger-cart-boundary.html.
65 Evans JM, Fletcher Jr RJ, Alavalapati J. 2010. Using species distri-
bution models to identify suitable areas for biofuel feedstock
production. Gcb Bioenergy. 2(2):63–78.
66 Radomski T, et al. 2022. Finding what you don’t know: testing SDM
methods for poorly known species. Divers Distrib.28(9):1769–1780.
67 Stephens SL, Moghaddas JJ. 2005. Experimental fuel treatment
impacts on forest structure, potential fire behavior, and pre-
dicted tree mortality in a California mixed conifer forest. For
Ecol Manage. 215(1–3):21–36.
68 Hirsch KG, Corey PN, Martell DL. 1998. Using expert judgment to
model initial attack fire crew effectiveness. For Sci. 44(4):539–549.
69 Hawbaker TJ, et al. 2020. The landsat burned area algorithm and
products for the conterminous United States. Remote Sens
Environ. 244:111801.
70 Radke D, Hessler A, Ellsworth D. 2019. Firecast: leveraging deep
learning to predict wildfire spread. In: IJCAI. p. 4575–4581.
71 Tsvetkov I, et al. 2018. Plant organic farming research–current
status and opportunities for future development. Biotechnol
Biotechnol Equip. 32(2):241–260.
72 Mostafa M, Uddin N. 2016. Experimental analysis of compressed
earth block (CEB) with banana fibers resisting flexural and com-
pression forces. Case Stud Constr Mater. 5:53–63.
73 Wang R, Zhao J, Liu Z. 2016. Consensus in visual preferences: the
effects of aesthetic quality and landscape types. Urban For Urban
Green. 20:210–217.
74 Costanza R, et al. 2014. Changes in the global value of ecosystem
services. Glob Environ Change. 26:152–158.
75 Stephens SL, et al. 2012. The effects of forest fuel-reduction treat-
ments in the United States. BioScience. 62(6):549–560.
76 Fried JS, et al. 2008. Predicting the effect of climate change on wild-
fire behavior and initial attack success. Clim Change. 87(1):251–264.
77 Martell DL. 2015. A review of recent forest and wildland fire man-
agement decision support systems research. Curr For Rep. 1(2):
128–137.
78 Arienti MC, Cumming SG, Boutin S. 2006. Empirical models of
forest fire initial attack success probabilities: the effects of fuels,
anthropogenic linear features, fire weather, and management.
Can J For Res. 36(12):3155–3166.
79 Cova TJ, Dennison PE, Kim TH, Moritz MA. 2005. Setting wildfire
evacuation trigger points using fire spread modeling and GIS.
Trans GIS. 9(4):603–617.
80 Curran TJ, Perry GLW, Wyse SV, Alam MdA. 2018. Managing fire
and biodiversity in the wildland-urban interface: a role for green
firebreaks. Fire. 1(1):3.
81 Clarke DJ, White JG. 2008. Towards ecological management of
Australian powerline corridor vegetation. Landsc Urban Plan.
86(3–4):257–266.
82 Cui X, et al. 2019. Green firebreaks as a management tool for wild-
fires: lessons from China. J Environ Manage. 233:329–336.
Fu et al. | 11
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023
83 Arrubla JAG, Ntaimo L, Stripling C. 2014. Wildfire initial response
planning using probabilistically constrained stochastic integer
programming. Int J Wildland Fire. 23(6):825–838.
84 Lauer CJ, Montgomery CA, Dietterich TG. 2020. Evaluating wild-
land fire liability standards—does regulation incentivise good
management? Int J Wildland Fire. 29(7):572–580.
85 Brenkert-Smith H, Champ PA, Flores N. 2012. Trying not to get
burned: understanding homeowners’ wildfire risk–mitigation
behaviors. Environ Manage. 50(6):1139–1151.
86 Sheikh B, Nelson KL, Haddad B, Thebo A. 2018. Grey water: agri-
cultural use of reclaimed water in California. J Contemp Water Res
Educ. 165(1):28–41.
87 Abrams MD. 1994. Genotypic and phenotypic variation as stress
adaptations in temperate tree species: a review of several case
studies. Tree Physiol. 14(7-8-9):833–842.
88 Chia CL. 1981. Bananas commodity fact sheet BA-3 (A) fruit.
Technical Report, Hawaii Cooperative Extension Service,
CTAHR, University of Hawaii.
89 Potter BE. 2012. Atmospheric interactions with wildland fire be-
haviour—II. Plume and vortex dynamics. Int J Wildland Fire.
21(7):802–817.
90 Tixier P, et al. 2011. Model evaluation of cover crops, application
to eleven species for banana cropping systems. Eur J Agron. 34(2):
53–61.
91 Theobald DM, Romme WH. 2007. Expansion of the US wildland–
urban interface. Landsc Urban Plan. 83(4):340–354.
92 Jackson LE. 1985. Ecological origins of California’s mediterranean
grasses. J Biogeogr. 12(4):349–361.
93 Lee C, Schlemme C, Murray J, Unsworth R. 2015. The cost of cli-
mate change: ecosystem services and wildland fires. Ecol Econ.
116:261–269.
94 Runck B, et al. 2023. State spaces for agriculture: a meta-systematic
design automation framework. PNAS Nexus. 2(4):pgad084.
95 York R, Roughton A, Tompkins R, Kocher S. 2020. Burn permits
need to facilitate—not prevent—“good fire” in California. Calif
Agric. 74(2):62–66.
12 | PNAS Nexus, 2023, Vol. 2, No. 10
Downloaded from https://academic.oup.com/pnasnexus/article/2/10/pgad315/7325141 by guest on 25 October 2023