fvsclim: prognosis re-engineered to incorporate climate variables

23
FVSCLIM: Prognosis Re- Engineered to Incorporate Climate Variables Robert Froese, Ph.D., R.P.F. School of Forest Resources and Environmental Science Michigan Technological University, Houghton MI Again

Upload: malini

Post on 05-Jan-2016

16 views

Category:

Documents


0 download

DESCRIPTION

FVSCLIM: Prognosis Re-Engineered to Incorporate Climate Variables. Again. Robert Froese , Ph.D., R.P.F. School of Forest Resources and Environmental Science Michigan Technological University, Houghton MI 49931. This presentation has four parts. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

FVSCLIM: Prognosis Re-Engineeredto Incorporate Climate Variables

Robert Froese, Ph.D., R.P.F.School of Forest Resources and Environmental ScienceMichigan Technological University, Houghton MI 49931

Again

Page 2: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

This presentation has four parts

Introduction

Approach

Relevance

Performance

The issue, the question and the model formulations examined

The methods and the data sets

How do revisions affect fit and prediction accuracy?

Does the approach have merit, and what are the next steps?

Page 3: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

This presentation has four parts

The issue, the question and the model formulations examined

The methods and the data sets

How do revisions affect fit and prediction accuracy?

Does the approach have merit, and what are the next steps?

Introduction

Approach

Relevance

Performance

Page 4: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

This presentation has four parts

The issue, the question and the model formulations examined

The methods and the data sets

How do revisions affect fit and prediction accuracy?

Does the approach have merit, and what are the next steps?

Introduction

Approach

Relevance

Performance

Page 5: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

This presentation has four parts

The issue, the question and the model formulations examined

The methods and the data sets

How do revisions affect fit and prediction accuracy?

Does the approach have merit, and what are the next steps?

Introduction

Approach

Relevance

Performance

Page 6: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Wykoff’s (1990) Basal Area Increment Model is the subject of this research

DDS = DBH2t+10 - DBH2

t

but actually..

DDS = DBH2t - DBH2

t-10

BAI = π/4 (DBH2t - DBH2

t-10)

DI = (DBH2 + DDS)0.5 - DBH

ln(DDS) = f(SIZE +SITE +COMPETITION)

Page 7: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Last year I presented results of a validation study of Wykoff’s model

pei yi yi

Page 8: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

“How and Where does Wykoff’s Basal Area Increment Model Fail?”

“I appreciate the opportunity to review your paper. The title certainly grabs your attention, especially if your name is Wykoff and you spent many years developing the subject model.”

I wrote it up as a manuscript…

Bill replied:

Page 9: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The Prognosis BAI model is a multiple linear regression on the logarithmic scale

Wykoff 1990

ln DDS HAB LOC b1 ln DBH b2 LOC :DBH 2

b3 cos ASP SL b4 sin ASP SL b5 SL b6 SL2 b7 EL b8 EL

2

b9 CR b10 CR2 b11 HAB :

CCF

100 b12

(1 PCT )SBAln(DBH 1)

Page 10: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Wykoff (1997) proposed a number of revisions to the model formulation

ln DDS HAB LOC b1 ln DBH b2 LOC :DBH 2

b3 cos ASP SL b4 sin ASP SL b5 SL b6 SL2 b7 EL b8 EL

2

b9 CR b10 CR2 b11 HAB :

CCF

100 b12

(1 PCT )SBAln(DBH 1)

Wykoff 1990

Wykoff 1997

ln DDS HAB LOC b1 ln DBH b2 LOC :DBHb3 cos ASP SL b4 sin ASP SL b5 SL b6 SL

2 b7 EL b8 EL2

b9 CR b10 CR

ln DBH 1 b11 HAB :SBA b12 (1 P90)PBAln(DBH 1)

b13 SBA

DBH

Page 11: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Froese (2003) proposed replacing climate proxies with climate variables

Wykoff 1997

Froese 2003

ln DDS HAB b1 ln DBH b2 DBH b3 ANP b4 GSP b5 GST

b6 cos ASP SL b7 sin ASP SL b8 SL b9 SL2

b10 CR b11 CR

ln DBH 1 b12 HAB :SBA b13 (1 P90)PBAln(DBH 1)

b14 SBA

DBH

ln DDS HAB LOC b1 ln DBH b2 LOC :DBHb3 cos ASP SL b4 sin ASP SL b5 SL b6 SL

2 b7 EL b8 EL2

b9 CR b10 CR

ln DBH 1 b11 HAB :SBA b12 (1 P90)PBAln(DBH 1)

b13 SBA

DBH

Page 12: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The approach involves two parts

• evaluating model revisions– Fit Wykoff (1990), Wykoff (1997) and Froese

(2003) to the new FIA data– Compare fit and lack-of-fit statistics of different

model formulations

• testing on independent data– generate predictions for independent testing data– compare bias of prediction residuals across

model formulations– Compare results using equivalence tests

Introduction

Approach

Relevance

Performance

Page 13: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Froese (2003) pretended to be a physiologist

• ANP: total annual precipitation

• GSL: growing season length(days with nighttime minimum temperature greater than 0°C)

• GSP: total precipitation during the growing season

• GST: mean daily temperature during the growing season

• GSV: mean daily water vapour pressure deficit during the growing season

Page 14: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Froese (2003) also pretended to be a climatologist

Page 15: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Changing model formulation had small effect on fit statistics

Introduction

Approach

Relevance

Performance

Fit to the FIA data:

Page 16: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The Froese (2003) model provided biologically-rational behaviour

• Biologically reasonable sign and magnitude of model coefficients

• Extrapolation issues remain to be resolved

Douglas-fir on median site

Page 17: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Testing revealed that every formulation over-predicts on the validation data

Tested on the Region 1 data:

Page 18: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The 1990 formulation failed to be validated for the monitoring data

Page 19: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The 1997 model performed better, but was still not validated in this situation

Page 20: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The 2003 model performed similarly to the 1997 model but was also not validated

Page 21: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The substitution of climate variables for proxies is validated using equivalence tests

Page 22: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

The model is not appropriately responsive to small and suppressed trees

Results for Pseudotsuga menziesii

Page 23: FVSCLIM:  Prognosis Re-Engineered to Incorporate Climate Variables

Some results are encouraging, some suggest that more work is needed

• Are we (am I) splitting hairs?– Is an RMSE reduction of 2% useful?

• Does it really matter if RMSE reductions are small?

• Can we come up with better DDS model formulations?

• What’s wrong with predictions for small trees?

• Have I modelled climate effects on growth or climate effects on genes?

Introduction

Approach

Relevance

Performance