Fire weather and fire behavior seasonality in the Northern Plains

Overview

Initial analysis in the Fire Ecology paper showed that the distribution of acceptable burn days is shifting across seasons.
In response to the first review, these analyses address consequences of seasonal shifts in prescribed fire activity by (1) describing how weather conditions on acceptable burn days vary by season, and (2) predicting how fire behavior might be expected to vary in different seasons within the bounds of acceptable burn days.
A Principal Components Analysis shows that acceptable burn days in different half-seasons have unique sets of conditions. Broadly speaking (and additional cluster analysis could test/confirm this, if we want), conditions on acceptable burn days fall into three categories: dormant season (early spring, early fall), early growing season (late spring, early summer), and late growing season (late summer).
The main axis of variation in the PCA indicates that among-half season variability is driven by seasonal temperature and atmospheric moisture variables, specifically, average air temperature and dew point. Within season variation among acceptable burn days was associated with fuel moisture, relative humidity, and wind speed.
Fire behavior simulations suggest differences in expected fire spread across seasons. Most substantially, late-summer fires are expected to have very low rates of spread on account of high proportions of live herbaceous fuels. High proportions of the cool-season invasive grass Poa pratensis reduces predicted fire spread in all seasons but is partially mitigated by high levels of live fuel curing (60% live fuel moisture) and weather conditions at the upper bound of the acceptable burn window.

Multivariate analysis of weather seasonality

Principal Components Analysis of fire-related weather variables from 2006-2013 (period of complete cases across all variables) show the typical conditions on acceptable days for several RAWS in the Grand Forks and Bismarck fire weather regions. In permutational tests accounting for non-independence among stations, region and year did not explain variation. Half-season was a significant factor (P < 0.001, R^2 = 0.50) and all half-season clusters were significantly different than each other in pairwise perMANOVA (P < 0.01 after multiple comparison p-value adjustment). Differences among half-season weather conditions were strongly associated with dewpoint and daily average air temperature. Within half-seasons, variability among typical acceptable fire weather days was mostly associated with relative humidity, fine fuel moisture, and wind speed. Species scores are scaled to fit the site score plot, see second table below for raw axis loadings.

PC1 + PC2 explain 73% of variation.
	PC1	PC2	PC3	PC4	PC5
Eigenvalue	2.30	1.34	0.89	0.46	0.01
Proportion Explained	0.46	0.27	0.18	0.09	0.00
Cumulative Proportion	0.46	0.73	0.91	1.00	1.00

Variable loadings along PC1 and PC2. PC1 is primarily a seasonal vector defined by air temperature and dewpoint, while PC2 is a fire weather vector defined by fuel moisture, relative humidity, and wind speed.
	PC1	PC2	PC3
DWP	3.3	0.3	-0.4
FM	-0.5	-3.0	-0.1
RELH	1.8	-2.1	-1.2
SKNTms	-1.4	0.9	-2.9
Tempave	3.1	0.9	-0.1

Results of post-hoc pairwise permutation MANOVA testing the Euclidean distance matrix against half-season groups.
	Early fall	Early spr	Early sum	Late spr
Early spr	0.01	NA	NA	NA
Early sum	0.01	0.01	NA	NA
Late spr	0.01	0.01	0.01	NA
Late sum	0.01	0.01	0.01	0.01

Fire behavior models

We used the Rothermel (1972) fire spread equation to predict horizontal spread rates of headfires in simulated fuelbeds to describe how different fuel and weather conditions interact in each of the half-seasons considered in the study.

Establishing parameters

The Rothermel fire spread equation is informed by three categories of parameters:

Topography
Weather
Fuel

For topography we set slope to 0.

For weather we looked across acceptable burn days to determine the range of fire weather conditions within each half-season. Because our multivariate analysis specifically found no evidence for a year effect in acceptable burn day conditions–i.e. the conditions of a typical burn day didn’t change over the time frame of the study–we looked across all acceptable burn days and determined the upper and lower bounds of acceptable fire weather conditions (as well as the median, or typical, acceptable burn day) within each season.

As there was very little seasonal variability in these parameters, we used the upper, lower, and median values for fuel moisture and wind speed as weather variables in Rothermel fire spread simulations.

Highest, lowest, and median fuel moisture and wind speed on acceptable burn days in each half-season across all years in the study period. Note little variability in upper and lower conditions.
HalfSeason	variable	low	mod	high
Early fall	FM	9	13	19
Early fall	SKNTms	2	4	5
Early spr	FM	8	11	20
Early spr	SKNTms	2	4	6
Early sum	FM	7	11	18
Early sum	SKNTms	2	4	5
Late spr	FM	9	12	20
Late spr	SKNTms	2	4	6
Late sum	FM	8	11	17
Late sum	SKNTms	2	4	5

Highest, lowest, and median fuel moisture and wind speed across all half-seasons and all years. These values define upper, lower, and median fire weather scenarios in Rothermel fire spread simulations.
variable	low	mod	high
FM	7	12	20
SKNTms	2	4	6

For fuel parameters we sought to model fire spread through a typical northern mixed-grass prairie fuelbed under two levels of Poa pratensis invasion and three levels of live fuel curing (60%, 90%, and 120% live fuel moisture). We began with a default GR fuel model from Scott & Burgan (2005) and modified the following parameters, based on a total fuel load of 4.4 tonnes/ha:

Fuel load scenarios by half-season under two P. pratensis invasion scenarios. LH = Live herbaceous fuel, FD = Fine dead fuel. Actual fuel loads calculated from given live:dead fuel ratios from a total of 4.4 tonnes/ha (increased to 4.5 later in the season to account for growth)
Season	Invasion	% LH	% FD	LH (tonnes/ha)	FD (tonnes/ha)
early spring	low	0.05	0.95	0.220	4.180
early spring	high	0.10	0.90	0.440	3.960
late spring	low	0.20	0.80	0.880	3.520
late spring	high	0.32	0.68	1.408	2.992
early summer	low	0.40	0.60	1.760	2.640
early summer	high	0.70	0.30	3.150	1.350
late summer	low	0.40	0.60	1.800	2.700
late summer	high	0.50	0.50	2.250	2.250
early fall	low	0.19	0.81	0.855	3.645
early fall	high	0.42	0.58	1.890	2.610

Fire spread simulation results

Simulated fire spread for northern mixed-grass prairie under two levels of P. pratensis invasion, three levels of live fuel curing, and three fire weather conditions defined as the median and upper and lower bounds of acceptable burn days across each half-season.

Fuels data

The following tables and figures summarize potentially useful data to determine fuel proportions.

Oakville data

Hadley & Buccos (1967) report a lower live component in communities dominated by Poa spp than communities with Poa present but not listed first; they don’t report the specific species although P. pratensis is suspected since the Poa-dominated associations include Bromus inermis and are described on disturbed sites. This seems counter-intuitive to the effect of POPR and makes one wonder if 1967 isn’t too old of a reference considering the current condition of widespread POPR dominance and long-term thatch build-up? More specifically, they didn’t report on POPR-invaded prairie so much as disturbed POPR-dominated sites within the prairie. Furthermore the Poa-dominated communities are very low productivity vs. the native communities (with Andropogon and Stipa) in which Poa are present but not dominant. The more I think about it the more difficult it is to interpret these data for our purposes.

Oakville data from Hadley & Buccos (1967) suggest summer live fuel proportions of 60% and 50% are reasonable for stands not dominated by Poa and those that are, respectively.
location	season	community	live.prop	dead.prop
Oakville	summer	Bro-Poa	0.61	0.39
Oakville	summer	Dis-Hor-Poa	0.57	0.43
Oakville	summer	Poa-And-Sti	0.56	0.44
Oakville	summer	Poa-Mel	0.49	0.51

CGREC data

Micayla has been collecting aboveground biomass from the Central Grasslands REC and hand-sorting into live and dead components. I suggest we draw our parameters from 2nd and 4th quintiles (lower and upper edges of the boxes in bottom graph) of her data to simulate two levels of live fuel proportion within each month.

2nd and 4th quintiles from Tukey’s five-number summary of hand-sorted data from CGREC.
date	component	low	high
May	dead	0.680	0.795
May	live	0.205	0.320
Jun	dead	0.310	0.610
Jun	live	0.390	0.690
Jul	dead	0.475	0.615
Jul	live	0.385	0.525
Aug	dead	0.460	0.590
Aug	live	0.410	0.540
Sep	dead	0.580	0.810
Sep	live	0.190	0.420

Rothermel output

Actual values returned by Rothermel fire spread equation as implemented by ros function in the R Rothermel package.
Half season	Invasion	Fire weather	Live fuel	ROS
early spring	low	low	low	18
early spring	low	low	medium	19
early spring	low	low	high	19
early spring	low	moderate	low	71
early spring	low	moderate	medium	75
early spring	low	moderate	high	78
early spring	low	high	low	175
early spring	low	high	medium	188
early spring	low	high	high	198
early spring	high	low	low	11
early spring	high	low	medium	11
early spring	high	low	high	12
early spring	high	moderate	low	39
early spring	high	moderate	medium	44
early spring	high	moderate	high	47
early spring	high	high	low	94
early spring	high	high	medium	108
early spring	high	high	high	119
late spring	low	low	low	14
late spring	low	low	medium	17
late spring	low	low	high	19
late spring	low	moderate	low	51
late spring	low	moderate	medium	64
late spring	low	moderate	high	74
late spring	low	high	low	119
late spring	low	high	medium	152
late spring	low	high	high	182
late spring	high	low	low	7
late spring	high	low	medium	10
late spring	high	low	high	12
late spring	high	moderate	low	25
late spring	high	moderate	medium	34
late spring	high	moderate	high	44
late spring	high	high	low	55
late spring	high	high	medium	80
late spring	high	high	high	105
early summer	low	low	low	10
early summer	low	low	medium	15
early summer	low	low	high	18
early summer	low	moderate	low	34
early summer	low	moderate	medium	51
early summer	low	moderate	high	68
early summer	low	high	low	74
early summer	low	high	medium	115
early summer	low	high	high	163
early summer	high	low	low	1
early summer	high	low	medium	7
early summer	high	low	high	11
early summer	high	moderate	low	4
early summer	high	moderate	medium	21
early summer	high	moderate	high	37
early summer	high	high	low	9
early summer	high	high	medium	46
early summer	high	high	high	84
late summer	low	low	low	10
late summer	low	low	medium	15
late summer	low	low	high	18
late summer	low	moderate	low	33
late summer	low	moderate	medium	51
late summer	low	moderate	high	68
late summer	low	high	low	74
late summer	low	high	medium	115
late summer	low	high	high	162
late summer	high	low	low	5
late summer	high	low	medium	8
late summer	high	low	high	11
late summer	high	moderate	low	17
late summer	high	moderate	medium	28
late summer	high	moderate	high	40
late summer	high	high	low	37
late summer	high	high	medium	62
late summer	high	high	high	95
early fall	low	low	low	15
early fall	low	low	medium	17
early fall	low	low	high	19
early fall	low	moderate	low	52
early fall	low	moderate	medium	64
early fall	low	moderate	high	74
early fall	low	high	low	122
early fall	low	high	medium	154
early fall	low	high	high	183
early fall	high	low	low	6
early fall	high	low	medium	9
early fall	high	low	high	11
early fall	high	moderate	low	20
early fall	high	moderate	medium	31
early fall	high	moderate	high	42
early fall	high	high	low	44
early fall	high	high	medium	69
early fall	high	high	high	99