Using climate and site characteristics to predict plant communities after fire treatments
1. Using geospatial environmental
characteristics to determine plant
community resilience to fire and
fire surrogate treatments
Nathan Cline, Bruce Roundy, William Christensen, and
Chris Balzotti
2. The Big Question: Will
cheatgrass dominate if we
treat sagebrush or
woodlands?
Cheatgrass Cover
Before
Treatment
After Treatment
High High
Low ?
Perennial Grass
Cover
Cheatgrass cover
• Identify the site and climate characteristics that
influence cover
3. Project objective and
proposed products
• Create models of cheatgrass and
perennial grass cover using site and
climate characteristics.
• Develop tools for land managers to
use in predicting the probability of
cheatgrass at other sites.
• A field guide
• A geospatial map
5. Data
Vegetation cover
• Cheatgrass
• Perennial grass
• Sagebrush
• Perennial Forbs
Site Characteristics
• Bioclim and
ClimateWNA
• Aspect, slope,
elevation, geospatial
coordinates, solar
radiation
• Treatment and
woodland
encroachment phase
6. Climate Variables
Annual Mean Temperature
Mean Diurnal Range (Mean of monthly (max temp - min temp))
Isothermality
Temperature Seasonality (standard deviation *100)
Max Temperature of Warmest Month
Min Temperature of Coldest Month
Temperature Annual Range
Mean Temperature of Wettest Quarter
Mean Temperature of Driest Quarter
Mean Temperature of Warmest Quarter
Mean Temperature of Coldest Quarter
Annual Precipitation
Precipitation of Wettest Month
Precipitation of Driest Month
Precipitation Seasonality (Coefficient of Variation)
Precipitation of Wettest Quarter
Precipitation of Driest Quarter
Precipitation of Warmest Quarter
Precipitation of Coldest Quarter
Continentality (°C)
Mean annual precipitation (mm)
Mean summer (May to Sep) precipitation (mm)
Annual heat moisture index
Summer heat moisture index
Degree-days below 0°C (chilling degree days)
Degree-days above 5°C (growing degree days)
The number of frost-free days
The julian date on which the frost-free period begins
The julian date on which the frost-free period ends
Precipitation as snow (mm)
Extreme minimum temperature over 30 years (°C)
Hargreave's reference evaporation
Hargreave's climatic moisture index
Hogg's climate moisture index
Hogg's summer (Jun to Aug) climate moisture index
Winter (Dec to Feb) mean temperature (°C)
Summer (Jun to Aug) mean temperature (°C)
Winter (Dec to Feb) precipitation (mm)
Summer (Jun to Aug) precipitation (mm)
7. Analysis
• Spatial Regression Analysis – space is
only important within sites
• Canonical correlation and step-wise
regression analyses
• Canonical correspondence analysis
(CCA)
• Random forest analysis
Analysis was done on the subplot scale (6 to 24 subplots
per site) – 450 subplots
8. Canonical Correlation and
Step-wise regression
R2 VALUES
Run Cheat Sage P Grass P Forbs avgR2 numterms
0 0.555 0.684 0.67 0.526 0.60875
1 0.555 0.684 0.67 0.526 0.60875
6 0.552 0.682 0.667 0.521 0.6055
11 0.535 0.669 0.644 0.508 0.589 24
16 0.47 0.65 0.564 0.39 0.5185 19
21 0.363 0.609 0.551 0.377 0.475 14
25 0.336 0.54 0.4879 0.2651 0.40725 10
26 0.328 0.54 0.4727 0.262 0.400675 9
27 0.318 0.54 0.469 0.202 0.38225 8
28 0.2758 0.5198 0.466 0.169 0.35765 7
29 0.272 0.475 0.4661 0.136 0.337275 6
30 0.269 0.474 0.462 0.095 0.325 5 Elevation MeanAnnTem MTWM mTCM PWQ
31 0.1985 0.4288 0.4387 0.0945 0.290125 4 Elevation MeanAnnTem MTWM PWQ
• Need at last 24 characteristics to achieve > 50%
• P. grass and sagebrush need fewer site characterizes
than cheatgrass and forbs
9. Perennial grass vs. cheatgrass
cover
Perennial grass Cheatgrass
Axis 1 Axis 2 Axis 3
P = 0.01
Variance in cover data
% of variance explained 54.8 11.9 2.5
Cumulative % explained 54.8 66.7 69.2
Pearson Correlation 0.890 0.724 0.562
Canonical correspondence analysis (CCA)
• Cover data included: Perennial grass, cheatgrass,
perennial forbs, & sagebrush
• Site characteristics: aspect, slope, elevation, and 30
climate variables
15. Conclusions
• Our analyses explained between 50-70%
of variation among the four cover
classes.
• Cheatgrass requires up to 24 variables
to explain > 50% of variability.
• Isothermality, temperature and
precipitation during warm and dry
periods, elevation, and solar radiation
may all be important predictors.
• Geospatial maps are coming…
Notas del editor
When treating vegetation in sagebrush step communities, the question is always “ will I get cheatgrass
Talk about the range of variation in these sites
Canonical correspondence analysis (CCA)
The ordination was significant P=0.01
locations: Utah shred data at 470 subplots (42 sites).
Both diagrams are illustrations of the same ordination.
Perennial grass, cheatgrass, perennial forb, and sagebrush cover data where included to create the ordination.
Triangle size: The larger the triangle the higher relative perennial grass and cheatgrass cover.
Most influential characteristics (correlation values > 0.5): Elevation, solar radiation, precipitation of the wetter quarter of the year, precipitation of the driest month, mean temperature of the driest quarter of the year.
Isothermal: (mean monthly diurnal temperature range/annual temperature range)*100 = an indicator of temperature fluctuation
Cheatgrass cover increases with increasing solar radiation and Isothermal percentage. Cheatgrass also appears to increase with decreasing elevation and precipitation during the wettest quarter of the year. Cheatgrass decreases with increasing precipitation during the driest month.
The point on solar radiation fits Condon et al. 2011’s finding that increasing cheatgrass cover is associated with increasing solar radiation.
Isothermal percentage - If we consider temperature a resource, this finding fits Davis 2001’s fluctuating resource hypothesis in that cheatgrass increases with increasing temperature fluctuation.
Elevation and cheatgrass…. I’ll let you discuss chamber’s findings on the subject and how you feel about that.
Perennial grass cover increases with increasing mean temperature during the driest quarter of the year and precipitation during driest month (extension of the growing season?). It decreases with increasing solar radiation and isothermal percentage.
Although RF, a form of machine learning, has been around for some time, its use in ecological studies is relatively new (Cutler et al., 2007; Prasad et al., 2006). RF uses bootstrap samples and a randomized subset of the predictor variables to create a series of classification trees (a forest) that predict species presence. These trees (typically over 500) are then combined for the final model prediction. R with the ModelMap package was used for this analysis