15
Case Study: Distribution and Abundance of Tree Species Along Climate Gradients (Sometimes the PDF is more interesting than the actual model…) Canham, C. D. and R. Q. Thomas. 2010. Frequency, not relative abundance, of temperate tree species varies along climate gradients in eastern North America. Ecology 91:3433-3440

Case Study: Distribution and Abundance of Tree Species Along Climate Gradients

  • Upload
    zarek

  • View
    32

  • Download
    0

Embed Size (px)

DESCRIPTION

Case Study: Distribution and Abundance of Tree Species Along Climate Gradients. (Sometimes the PDF is more interesting than the actual model…). - PowerPoint PPT Presentation

Citation preview

Page 1: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

Case Study:

Distribution and Abundance of Tree Species Along Climate Gradients

(Sometimes the PDF is more interesting than the actual

model…)Canham, C. D. and R. Q. Thomas. 2010. Frequency, not relative abundance, of temperate tree species varies along climate gradients in eastern North America. Ecology 91:3433-3440

Page 2: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

The Data

Relative abundance of the 24 most common tree species in the northeastern US- In ~ 20,000 US Forest Service Forest Inventory and Analysis

(FIA) plots from 19 northeastern states (Maine to Wisconsin, south to Kentucky and Virginia)

Climate (mean annual temperature and average annual precipitation) for each plot, averaged over the period since the previous census- Using 800-m resolution gridded PRISM climate data, with bi-

linear interpolation to true plot locations

Page 3: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

The Basic Question

How does the abundance of tree species vary along climate gradients?

http://www.fs.fed.us/ne/delaware/atlas/s318.html

Page 4: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

Basic Approach

Develop regression models that predict abundance of a given tree species in a plot as a function of climate at the location of the plot…

Initial decisions:1. What to use as a measure of

abundance?I chose relative abundance over absolute abundance

2. What sorts of functions could describe variation in relative abundance along climate gradients

Compare a Gaussian function with a null model that was flat

0 2 4 6 8 10

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

Mean Annual Temperature (oC)

Rel

ativ

e A

bund

ance

Page 5: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

But is a flat line an appropriate null hypothesis?

What about range limits?

And shouldn’t even the Gaussian model have truncated tails?

0 2 4 6 8 100.

020.

040.

060.

080.

100.

120.

140.

16

Mean Annual Temperature (oC)

Rel

ativ

e A

bund

ance

Page 6: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

So, what do the data look like?

5 10 15

0.0

0.2

0.4

0.6

0.8

1.0

ACRU

Mean Annual Temperature (oC)

Rel

ativ

e A

bund

ance

5 10 15

0.0

0.2

0.4

0.6

0.8

1.0

QUPR

Mean Annual Temperature (oC)

Rel

ativ

e A

bund

ance

Acer rubrum (red maple) Quercus prinus ( chestnut oak)

It doesn’t look like there’s much hope for nice Gaussian niche curves…

Page 7: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

A first regression model (in R)

# Square Gaussian - predicts mean of 0 when below lo or above hi

square.gauss.model <- function(a,m,b,lo,hi,X){ ifelse(X < lo,0,ifelse(X > hi,0,a*exp(-0.5*(((X-m)/b)^2)))) }

0 2 4 6 8 10

0.00

0.05

0.10

0.15

Mean Annual Temperature (oC)

Rel

ativ

e A

bund

ance

5 Parameters:• “a” determines height of curve at mode• “m” temperature at peak of curve• “b” breadth of the curve• “lo” lower temperature limit• “hi” upper temperature limit

Note that a separate “null” model is not really necessary, because if “b” is large, the curve is flat

Page 8: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

What might be an appropriate PDF?

ACRU

Relative abundance

Freq

uenc

y

0.0 0.2 0.4 0.6 0.8 1.0

020

0040

0060

0080

0010

000The data are clearly not normally

distributed…

In fact, there are a whole lot of zeros….

Page 9: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

Divide the analysis up into two parts

Predict probability of “presence” (i.e. non-zero abundance) –

Separately predict relative abundance when presentACRU

Relative abundance

Freq

uenc

y

0.0 0.2 0.4 0.6 0.8 1.0

020

0040

0060

0080

0010

000

These two components of abundance have very different

ecological meanings….

Page 10: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

A PDF for relative abundance when present:The Gamma Function

0 20 40 60 80 100

0.00

0.02

0.04

0.06

0.08

0.10

0.12

Gamma PDF

x

P(x

)

mean = 10%mean = 20%mean = 50%mean = 80%

mean = 10%mean = 20%mean = 50%mean = 80%

mean = 10%mean = 20%mean = 50%mean = 80%

Gamma PDF with scale parameter = 22

One likely choice for plots where a species is present…

Page 11: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

First ResultsRelative Abundance (when present)

Bottom line: for all 24 species, relative abundance is highly variable, but shows very little trend in the mean across the entire range of a species’ climate niche limits

Note: these results are for relative abundance, given that a species is present…

Page 12: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

But what about all of those zeros?Creating a PDF that can model both

“presence” and “relative abundance”

“Zero-Inflated Distributions”

ACRU

Relative abundance

Freq

uenc

y

0.0 0.2 0.4 0.6 0.8 1.0

020

0040

0060

0080

0010

000

# ZERO INFLATED NORMAL PDFzinf_norm_PDF <- function(x,mean,sd,pz){ log(ifelse(x==0,pz + (1-pz)*dnorm(0,mean,sd,log=F), (1-pz)*dnorm(x,mean,sd,log=F))) }

# ZERO INFLATED GAMMA PDFzinf_gamma_PDF_climate <- function(x,mean,scale,px){ shape <- mean/scale loglh <- log(ifelse(x==0,pz + (1-pz)*dgamma(0,shape=shape,scale=scale,log=F), (1-pz)*dgamma(x,shape=shape,scale=scale,log=F))) return(loglh) }

Page 13: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

But why stop there? Is presence constant across climate gradients, or

does it vary?

# ZERO INFLATED GAMMA PDF with pz a gaussian function of the# independent (climate) variable

zinf_gamma_PDF_climate <- function(x,mean,scale,pa,pm,pb,px){ shape <- mean/scale

pz <- 1 - pa*(exp(-0.5*(((px-pm)/pb)^2))) # px is temp or precip in the plot

loglh <- log(ifelse(x==0,pz + (1-pz)*dgamma(0,shape=shape,scale=scale,log=F), (1-pz)*dgamma(x,shape=shape,scale=scale,log=F))) return(loglh)}

Page 14: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients

One final complication…

How can I factor range limits into the PDF?- If probability of presence is modeled as a Gaussian function of

climate, but with truncated tails (climatic limits), what likelihood should be assigned to plots outside the estimated climatic limits? (my answer – 1 in a million…)

# CLIMATE DEPENDENT ZERO INFLATED GAMMA, WITH LIMITS PDF - # ARBITRARILY SET Prob(x>0|X=0) = 0.000001zinf_limits_gamma_PDF_climate <- function(x,mean,scale,pa,pm,pb,px){ shape <- mean/scale

pz <- 1 - pa*(exp(-0.5*(((px-pm)/pb)^2)))

loglh <- log(ifelse(mean == 0, ifelse(x == 0,0.999999,0.000001), ifelse(x==0,pz + (1-pz) * dgamma(0,shape=shape,scale=scale,log=F), (1-pz)*dgamma(x,shape=shape,scale=scale,log=F))))

return(loglh) }

Page 15: Case  Study: Distribution and Abundance of Tree Species Along Climate Gradients