Unlocking the forest inventory data: Relating individual tree performance to unmeasured environmental factors

Jeremy W. Lichstein, Jonathan Dushoff, Kiona Ogle, Anping Chen, Drew W. Purves, John P. Caspersen, Stephen Wilson Pacala

Research output: Contribution to journalArticle

31 Citations (Scopus)

Abstract

Geographically extensive forest inventories, such as the USDA Forest Service's Forest Inventory and Analysis (FIA) program, contain millions of individual tree growth and mortality records that could be used to develop broad-scale models of forest dynamics. A limitation of inventory data, however, is that individual-level measurements of light (L) and other environmental factors are typically absent. Thus, inventory data alone cannot be used to parameterize mechanistic models of forest dynamics in which individual performance depends on light, water, nutrients, etc. To overcome this limitation, we developed methods to estimate species-specific parameters (hG) relating sapling growth (G) to L using data sets in which G, but not L, is observed for each sapling. Our approach involves: (1) using calibration data that we collected in both eastern and western North America to quantify the probability that saplings receive different amounts of light, conditional on covariates x that can be obtained from inventory data (e.g., sapling crown class and neighborhood crowding); and (2) combining these probability distributions with observed G and x to estimate hG using Bayesian computational methods. Here, we present a test case using a data set in which G, L, and x were observed for saplings of nine species. This test data set allowed us to compare estimates of hG obtained from the standard approach (where G and L are observed for each sapling) to our method (where G and x, but not L, are observed). For all species, estimates of hG obtained from analyses with and without observed L were similar. This suggests that our approach should be useful for estimating light-dependent growth functions from inventory data that lack direct measurements of L. Our approach could be extended to estimate parameters relating sapling mortality to L from inventory data, as well as to deal with uncertainty in other resources (e.g., water or nutrients) or environmental factors (e.g., temperature).

Original languageEnglish (US)
Pages (from-to)684-699
Number of pages16
JournalEcological Applications
Volume20
Issue number3
DOIs
StatePublished - Apr 1 2010

Fingerprint

forest inventory
sapling
environmental factor
forest dynamics
mortality
nutrient
water resource
calibration
method
temperature

All Science Journal Classification (ASJC) codes

  • Ecology

Keywords

  • Bayesian hierarchical model
  • Dynamic global vegetation model
  • Forest inventory
  • Latent v riable
  • Markov chain Monte Carlo (MCMC)
  • Neighborhood analysis
  • Numerical integration
  • Shade tolerance
  • Understory light

Cite this

Lichstein, Jeremy W. ; Dushoff, Jonathan ; Ogle, Kiona ; Chen, Anping ; Purves, Drew W. ; Caspersen, John P. ; Pacala, Stephen Wilson. / Unlocking the forest inventory data : Relating individual tree performance to unmeasured environmental factors. In: Ecological Applications. 2010 ; Vol. 20, No. 3. pp. 684-699.
@article{d8897023c2f64ef58d46b39186fadca1,
title = "Unlocking the forest inventory data: Relating individual tree performance to unmeasured environmental factors",
abstract = "Geographically extensive forest inventories, such as the USDA Forest Service's Forest Inventory and Analysis (FIA) program, contain millions of individual tree growth and mortality records that could be used to develop broad-scale models of forest dynamics. A limitation of inventory data, however, is that individual-level measurements of light (L) and other environmental factors are typically absent. Thus, inventory data alone cannot be used to parameterize mechanistic models of forest dynamics in which individual performance depends on light, water, nutrients, etc. To overcome this limitation, we developed methods to estimate species-specific parameters (hG) relating sapling growth (G) to L using data sets in which G, but not L, is observed for each sapling. Our approach involves: (1) using calibration data that we collected in both eastern and western North America to quantify the probability that saplings receive different amounts of light, conditional on covariates x that can be obtained from inventory data (e.g., sapling crown class and neighborhood crowding); and (2) combining these probability distributions with observed G and x to estimate hG using Bayesian computational methods. Here, we present a test case using a data set in which G, L, and x were observed for saplings of nine species. This test data set allowed us to compare estimates of hG obtained from the standard approach (where G and L are observed for each sapling) to our method (where G and x, but not L, are observed). For all species, estimates of hG obtained from analyses with and without observed L were similar. This suggests that our approach should be useful for estimating light-dependent growth functions from inventory data that lack direct measurements of L. Our approach could be extended to estimate parameters relating sapling mortality to L from inventory data, as well as to deal with uncertainty in other resources (e.g., water or nutrients) or environmental factors (e.g., temperature).",
keywords = "Bayesian hierarchical model, Dynamic global vegetation model, Forest inventory, Latent v riable, Markov chain Monte Carlo (MCMC), Neighborhood analysis, Numerical integration, Shade tolerance, Understory light",
author = "Lichstein, {Jeremy W.} and Jonathan Dushoff and Kiona Ogle and Anping Chen and Purves, {Drew W.} and Caspersen, {John P.} and Pacala, {Stephen Wilson}",
year = "2010",
month = "4",
day = "1",
doi = "10.1890/08-2334.1",
language = "English (US)",
volume = "20",
pages = "684--699",
journal = "Ecological Applications",
issn = "1051-0761",
publisher = "Ecological Society of America",
number = "3",

}

Unlocking the forest inventory data : Relating individual tree performance to unmeasured environmental factors. / Lichstein, Jeremy W.; Dushoff, Jonathan; Ogle, Kiona; Chen, Anping; Purves, Drew W.; Caspersen, John P.; Pacala, Stephen Wilson.

In: Ecological Applications, Vol. 20, No. 3, 01.04.2010, p. 684-699.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Unlocking the forest inventory data

T2 - Relating individual tree performance to unmeasured environmental factors

AU - Lichstein, Jeremy W.

AU - Dushoff, Jonathan

AU - Ogle, Kiona

AU - Chen, Anping

AU - Purves, Drew W.

AU - Caspersen, John P.

AU - Pacala, Stephen Wilson

PY - 2010/4/1

Y1 - 2010/4/1

N2 - Geographically extensive forest inventories, such as the USDA Forest Service's Forest Inventory and Analysis (FIA) program, contain millions of individual tree growth and mortality records that could be used to develop broad-scale models of forest dynamics. A limitation of inventory data, however, is that individual-level measurements of light (L) and other environmental factors are typically absent. Thus, inventory data alone cannot be used to parameterize mechanistic models of forest dynamics in which individual performance depends on light, water, nutrients, etc. To overcome this limitation, we developed methods to estimate species-specific parameters (hG) relating sapling growth (G) to L using data sets in which G, but not L, is observed for each sapling. Our approach involves: (1) using calibration data that we collected in both eastern and western North America to quantify the probability that saplings receive different amounts of light, conditional on covariates x that can be obtained from inventory data (e.g., sapling crown class and neighborhood crowding); and (2) combining these probability distributions with observed G and x to estimate hG using Bayesian computational methods. Here, we present a test case using a data set in which G, L, and x were observed for saplings of nine species. This test data set allowed us to compare estimates of hG obtained from the standard approach (where G and L are observed for each sapling) to our method (where G and x, but not L, are observed). For all species, estimates of hG obtained from analyses with and without observed L were similar. This suggests that our approach should be useful for estimating light-dependent growth functions from inventory data that lack direct measurements of L. Our approach could be extended to estimate parameters relating sapling mortality to L from inventory data, as well as to deal with uncertainty in other resources (e.g., water or nutrients) or environmental factors (e.g., temperature).

AB - Geographically extensive forest inventories, such as the USDA Forest Service's Forest Inventory and Analysis (FIA) program, contain millions of individual tree growth and mortality records that could be used to develop broad-scale models of forest dynamics. A limitation of inventory data, however, is that individual-level measurements of light (L) and other environmental factors are typically absent. Thus, inventory data alone cannot be used to parameterize mechanistic models of forest dynamics in which individual performance depends on light, water, nutrients, etc. To overcome this limitation, we developed methods to estimate species-specific parameters (hG) relating sapling growth (G) to L using data sets in which G, but not L, is observed for each sapling. Our approach involves: (1) using calibration data that we collected in both eastern and western North America to quantify the probability that saplings receive different amounts of light, conditional on covariates x that can be obtained from inventory data (e.g., sapling crown class and neighborhood crowding); and (2) combining these probability distributions with observed G and x to estimate hG using Bayesian computational methods. Here, we present a test case using a data set in which G, L, and x were observed for saplings of nine species. This test data set allowed us to compare estimates of hG obtained from the standard approach (where G and L are observed for each sapling) to our method (where G and x, but not L, are observed). For all species, estimates of hG obtained from analyses with and without observed L were similar. This suggests that our approach should be useful for estimating light-dependent growth functions from inventory data that lack direct measurements of L. Our approach could be extended to estimate parameters relating sapling mortality to L from inventory data, as well as to deal with uncertainty in other resources (e.g., water or nutrients) or environmental factors (e.g., temperature).

KW - Bayesian hierarchical model

KW - Dynamic global vegetation model

KW - Forest inventory

KW - Latent v riable

KW - Markov chain Monte Carlo (MCMC)

KW - Neighborhood analysis

KW - Numerical integration

KW - Shade tolerance

KW - Understory light

UR - http://www.scopus.com/inward/record.url?scp=77950909934&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77950909934&partnerID=8YFLogxK

U2 - 10.1890/08-2334.1

DO - 10.1890/08-2334.1

M3 - Article

C2 - 20437956

AN - SCOPUS:77950909934

VL - 20

SP - 684

EP - 699

JO - Ecological Applications

JF - Ecological Applications

SN - 1051-0761

IS - 3

ER -