1 of 34

LandR & LandR.CS &�fireSense:�Quick overview for NWT

Eliot McIntire, Ceres Barros, Tati Micheletti, Alex Chubaty, Ian Eddy, Jean Marchal, Steve Cumming,

2 of 34

Current Project Location

Western Boreal Initiative
Needed new approaches �to estimate growth, mortality and their climate sensitive equivalents
From WBI perspective, NWT is our pilot

RIA

LandWeb

3 of 34

Outline

Start of an ongoing conversation (https://gitter.im/PredictiveEcology/)
Vegetation

LandR-Biomass – port of Landis-II Biomass Succession

Data clean up (Biomass_speciesData, Biomass_borealDataPrep)
Parameter estimation (Biomass_borealDataPrep)
Parameters – species-level and species-ecoregion level (Biomass_borealDataPrep)

LandR.CS –R package: functions that overlay onto LandR-Biomass

Parameter estimation (LandR.CS)

Validation (in SK)

Fire

FireSense – A climate & vegetation sensitive extension of SCFM (a fire percolation model originally designed by Steve Cumming) (fireSense_IgnitionFit, fireSense_EscapeFit, fireSense_SpreadFit)
Validation

All modules are located:�github.com/PredictiveEcology/moduleName

4 of 34

LandR-Biomass

LANDIS-II Biomass Succession in R

5 of 34

LandR growth & mortality

Landis-II Equations grow and die at “base” rates under current conditions
Luo et al. centred provides growth and mortality “deviations” due to climate deviation from current condition

6 of 34

LandR – Data Clean Up

It turns out that most datasets are not particularly inter-compatible, or inter-consistent
Many, many different types of problems occur

E.g., If you take EOSD product for land cover, and KNN for species cover, age, biomass, they will contradict one another

Age = 140, EOSD = Decidious, Aboveground Biomass = 10 T/ha, Species with Cover: White Spruce 90, Aspen 5, Abies 20
What do we do with this?

Mismatching dates of data products

KNN2010, EOSD 2017, etc.

NAs (i.e., missing data) do not correspond

These issues affects many dimensions
We have this entire workflow public, transparent

Doesn’t mean it is “easy” to just see

All happens in a SpaDES module – Biomass_borealDataPrep (https://github.com/predictiveecology/Biomass_borealDataPrep/ -- note all modules made by our group are same URL structure – github.com, then PredictiveEcology then module name.)

7 of 34

LandR – Parameterization

All happens in a SpaDES module – Biomass_borealDataPrep (https://github.com/predictiveecology/Biomass_borealDataPrep/ -- note all modules made by our group are same URL structure – github.com, then PredictiveEcology then module name.)

8 of 34

LandR

Parameterization

Species-level parameters – 2 sources:

Canada wide table by Cyr et al.
Estimated through simulation experiment from PSPs

mortalityshape
growthcurve

Species-ecoregion parameters –

maxBiomass – A statistical model�lme4::lmer(B ~ logAge * speciesCode + cover * speciesCode + (logAge + cover | ecoregionGroup)��then: predict what B would be when logAge = longevity for the species, and cover = 100%
maxANPP = maxBiomass * maxANPPFactor
Species Establishment Probability – currently based on a simple Species Distribution Model:�glm(cbind(coverPres, coverNum - coverPres) ~ speciesCode * ecoregionGroup, family = binomial)�

9 of 34

Growth curve

Mortality shape

100/X = maxANPP factor

10 of 34

Species traits �(changing some to be spatial)

	Max age	Sexual mature	Shade tolerance	Fire tolerance	Seed distance eff	Seed distance max	Resprout prob	Resprout age min	Resprout age max	Post Fire regen	Leaf longevity	Wood Decay rate	Mortality shape	Growth curve	Leaf Lignin	Hard soft
Lari_Lar	350	40	1	1	50	200	0	0	0	none	1	0.02	25	0	0.2	Soft
Pice_Gla	400	30	2	2	100	303	0	0	0	none	3	0.02	15	1	0.2	Soft
Pice_Mar	250	30	3	2	80	200	0	0	0	serotiny	3	0.02	15	1	0.2	Soft
Pinu_Ban	150	20	1	2	30	100	0	0	0	serotiny	2	0.01	25	0	0.2	Soft
Popu_Tre	200	20	1	2	1000	2000	0.9	10	150	resprout	1	0.07	25	0	0.1	Hard

11 of 34

LANDIS Biomass Succession = LandR_Biomass

Process-driven spatially explicit model for simulating forest succession

Growth: determined by age, species, ecoregion, and inter-cohort competition
Mortality: ‘development mortality’ and ‘age-related mortality’

Data is simulated at the cohort level (same species/age class) in units of biomass

Tracks Aboveground Biomass, by cohort and species over time

12 of 34

LandR.CS

LANDIS-II Biomass Succession in R (LandR)

with Climate-Sensitive Growth and Mortality (LandR.CS)

13 of 34

Adding Climate 🡪 LandR.CS

Taylor, Boulanger et al:

Use an external process model to get Maximum Biomass, Maximum ANPP.

Here, we use Luo et al. 2017, 2018

Estimate climate effects in PSPs (possibly taking into account competition within plots) on growth and mortality

14 of 34

LandR.CS

Based on years of vegetation work in W. Boreal Forest, culminating with Luo et al 2018, 2019 while he was in my lab

15 of 34

Multivariate mixed effect model (example form)

16 of 34

Rate of change of the changes�with climate

Current rates with climate

17 of 34

Sensitivies to climate depend on context

In dry areas, large improvements in DeltaAGB if there is more moisture

18 of 34

Luo et al 2019

Canopy improving

At expense of understory losses

Canopy

Net is neutral

19 of 34

Individual trees as source of divergences in outcomes

Tree cores – generally taken from “climate” sensitive trees, which are canopy or dominant trees
We hypothesized that improvements with canopy trees growth would be neutral at the plot-level because of competition with understory

20 of 34

Semi-automated refitting

Luo et al was only in AB, SK, MB
We can’t extrapolate the parameters
So we refit the models with PSP data “on the fly” (building up a continually updated database)

21 of 34

PSP treatment

Aggregate PSPs from various sources

Subset to a ‘study area’

Filter data for several criteria�e.g. min. 3 measures, 1950 - present

Estimate biomass (Lambert 2005), calculate both observed and unobserved growth and mortality within each measurement interval

22 of 34

Climate Data

2 metrics: Annual Temperature Anomaly (ATA) = the deviation in MAT from the reference MAT, and Cumulative Moisture Index (CMI) = precipitation - evaporation

Calculated 1950 - present for every year at each plot, as well as future landscape

23 of 34

Model Fitting

Growth fit with a GLMM-PQL:�Growth ~ logAge * (ATA + CMI) + ATA * CMI, random effect = Plot ID

Mortality fit with GAMLSS due to zero inflation

24 of 34

Mortality partial effects

ATA

CMI

25 of 34

LandR.CS

Climate-sensitive multiplier �on growth (aNPP) �and mortality �(both age/development)

Uses statistical models fit to PSP data and climate data

26 of 34

Applying to LandR

CS = baseline growth * Predicted growth w/ current CMI and ATA

Predicted growth w/ reference CMI and ATA

If baseline mortality = 0, CS = 0.

Due to the mortality algorithms, cohorts will still die at longevity even with reduced mortality

27 of 34

FireSense

Vegetation and Climate Sensitive Landscape Fire Simulation Model

28 of 34

Climate & Vegetation Sensitive Fire model:

FireSense

The spatial generalization of a relatively simple percolation model, where each vegetation type and maps of weather create the heterogeneous landscape through which the fires spread

29 of 34

Percolation model

3 stages

Ignition
Escape
Spread

Have to estimate each from historical data

30 of 34

Approach

Fuels

Collect data on fuels
Convert biomass by tree species & Land Cover to Fuel classes – Take

first 5 axes of PCA of tree species +
HighlyFlammable_nonForest +
LowFlammable_nonForest +
Young (<15years)

Climate

Montly drought code for summer months

Iterative

31 of 34

Fuel Classes

32 of 34

Sample of Fuels (NE BC)

33 of 34

Sample Fire Spread Fitting

34 of 34

Sample Fire Spread Fitting