1
D O N P R K r k G reatBasin Mojave Sonoran C hihuahuan CA NV AZ NM TX Mexico CO UT WY ID OR SRER BBNP NTS VESR G reatBasin Mojave Sonoran C hihuahuan G reatBasin Mojave Sonoran C hihuahuan CA NV AZ NM TX Mexico CO UT WY ID OR SRER BBNP NTS VESR Figure 1. Distributions of the four major deserts in the Southwest, and locations of the five field sites contributing data to this synthesis project. Sites are: (i) the Valentine Eastern Sierra Reserve (VESR), PI = Michael Loik, (ii) the Nevada Test Site (NTS), PI = Stan Smith, (iii) the Santa Rita Experimental Range (SRER) and the San Pedro River Basin (SPRB), PI = Understanding the effects of altered precipitation on arid and semiarid plants and ecosystems: A Bayesian synthesis Background The intensity, frequency, and variability in the timing of precipitation events are predicted to increase in the southwestern United States over the next few decades. Arid and semi-arid lands are particularly sensitive to altered precipitation regimes and other hypothesized effects of climate change. No quantitative syntheses have been carried-out with data related to the effects of precipitation change on arid or semiarid ecosystems and our current understanding of how the potential responses of plants, soils, and microbial communities will affect carbon and water fluxes is particularly lacking. The objectives of this project are to synthesize existing data related to carbon and water fluxes from leaves to the ecosystem level across the four major deserts of the Southwest. Several research groups are exploring the effects of altered precipitation regimes on ecosystems of the Southwest. Collaborators Huxman, Loik, Smith, and Tissue have and continue to conduct field studies, including precipitation manipulations, that emphasize the effects of variation in pulse, seasonal, and annual precipitation on C (carbon) and H20 (water) dynamics. These studies span sites located in the four major deserts of the Southwest (Fig. 1), and they have produced enormous quantities of data representing different spatial, temporal, and biological scales. Richard W. Lucas and Kiona Ogle Botany Dept, University of Wyoming [email protected] [email protected] Advantages of a Bayesian Approach Bayesian hierarchical modeling. The Bayesian method explicitly link the diverse data sources with the mechanistic models (Fig. 2). The Bayesian model decomposes the data-model synthesis problem into a probabilistic hierarchical framework. Mechanistic models. Data can be analyzed within the context of mechanistic models that represent processes operating at different scales. The models contain ecologically- meaningful parameters that provide important insights into how precipitation variability controls C and H2O dynamics in deserts of the Southwest. Figure 2. A simple hierarchical Bayesian model that couples diverse field data and mechanistic models related to leaf, soil, and ecosystem carbon dynamics. Field data are categorized as stochastic or covariates (i.e., assume measured without error). Stochastic variables arise from distributions whose means are defined by the latent processes. The latent processes represent the “truth” or unobserved quantities, and they are informed by the mechanistic models. Covariates Data Stochastic Latent processes Param eters H yper- param eters NE E Obs N etecosystem exchange LAI Leafarea index Rs Obs Soil respiration B Root& m icrobial biom ass A Obs, Instantaneous photosynthesis C Environm ental covariates SWC Soilw ater content NE E N etecosystem exchange Rs Soilrespiration A Instantaneous photosynthesis A T ot Totaldaily photosynthesis r Instantaneous source- specific respiration NE E NE E obs. error variance Rs Rs obs. error variance A A obs. error variance A site parameters ˆ A species parameters r site parameters ˆ r source parameters a 1 a 2 b 1 b 2 a 1 a 2 b 1 b 2 Ts Soil tem perature Population effects (Param eters describing distributions from w hich site-,species-,and source-specific effects arise) Model of Stomatal Conductance in Larrea tridentata: A Case Study Photosynthesis depends on variable water status of plant (Fig. 3). • Predawn water potential often used as an indicator of plant water status, instantaneous measures would be more accurate. Don’t often have these data. • Hierarchical Bayesian model can be used to predict instantaneous as a latent variable (Figs. 3 & 4). • Modeling instantaneous as a latent variable gives improved result (Fig 4). • The finding that control plants have lower stomatal conductance than watered plants (Fig. 5) offers support for the hypothesis that L. tridentate is successful in arid systems partially because of its ability to tightly regulate its stomatal water loss during periods of water stress. Figure 3. Figure 4. Figure 5. W P estim ates from Winbugs Predicted conductance -0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30 O bserved conductance 0.00 0.05 0.10 0.15 0.20 0.25 0.30 control w atered 1:1 line Predaw n W P Predicted conductance 0.00 0.05 0.10 0.15 0.20 0.25 0.30 control w atered 1:1 line r 2 = 0.59 r 2 = 0.67 Stom atal C onductance ofC ontrol and W atered Plants VPD (kPa) 0 2 4 6 8 10 12 14 conductance (m ol m -2 s-1) -0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 W atered Plants C ontrol Plants Photosynthesis depe nds Photosynthesis depends in part on stomatal in part on stomatal conductance conductance

Data Assimilation Workshop Oct 22-24, 2007 Norman, OK Poster presented by Rich Lucas & Kiona Ogle

  • Upload
    cargan

  • View
    19

  • Download
    0

Embed Size (px)

DESCRIPTION

Model of Stomatal Conductance in Larrea tridentata : A Case Study Photosynthesis depends on variable water status of plant (Fig. 3). Predawn water potential (y) often used as an indicator of plant water status, instantaneous measures would be more accurate. Don’t often have these data. - PowerPoint PPT Presentation

Citation preview

Page 1: Data Assimilation Workshop Oct 22-24, 2007 Norman, OK Poster presented by Rich Lucas & Kiona Ogle

Great BasinMojaveSonoranChihuahuan

CA

NV

AZ NM

TX

Mexico

CO

UT

WY

ID

OR

SRER

BBNP

NTS

VESR

Great BasinMojaveSonoranChihuahuan

Great BasinMojaveSonoranChihuahuan

CA

NV

AZ NM

TX

Mexico

CO

UT

WY

ID

OR

SRER

BBNP

NTS

VESR

Figure 1. Distributions of the four major deserts in the Southwest, and locations of the five field sites contributing data to this synthesis project. Sites are: (i) the Valentine Eastern Sierra Reserve (VESR), PI = Michael Loik, (ii) the Nevada Test Site (NTS), PI = Stan Smith, (iii) the Santa Rita Experimental Range (SRER) and the San Pedro River Basin (SPRB), PI = Travis Huxman, and (iv) the Big Bend National Park (BBNP), PI = David Tissue.

Understanding the effects of altered precipitation on arid and semiarid plants and ecosystems: A Bayesian synthesis

Background The intensity, frequency, and variability in the timing of precipitation events are predicted to increase in the southwestern United States over the next few decades. Arid and semi-arid lands are particularly sensitive to altered precipitation regimes and other hypothesized effects of climate change. No quantitative syntheses have been carried-out with data related to the effects of precipitation change on arid or semiarid ecosystems and our current understanding of how the potential responses of plants, soils, and microbial communities will affect carbon and water fluxes is particularly lacking. The objectives of this project are to synthesize existing data related to carbon and water fluxes from leaves to the ecosystem level across the four major deserts of the Southwest. Several research groups are exploring the effects of altered precipitation regimes on ecosystems of the Southwest. Collaborators Huxman, Loik, Smith, and Tissue have and continue to conduct field studies, including precipitation manipulations, that emphasize the effects of variation in pulse, seasonal, and annual precipitation on C (carbon) and H20 (water) dynamics. These studies span sites located in the four major deserts of the Southwest (Fig. 1), and they have produced enormous quantities of data representing different spatial, temporal, and biological scales.

Richard W. Lucas and Kiona OgleBotany Dept, University of Wyoming

[email protected] [email protected]

Advantages of a Bayesian Approach

• Bayesian hierarchical modeling. The Bayesian method explicitly link the diverse data sources with the mechanistic models (Fig. 2). The Bayesian model decomposes the data-model synthesis problem into a probabilistic hierarchical framework.• Mechanistic models. Data can be analyzed within the context of mechanistic models that represent processes operating at different scales. The models contain ecologically-meaningful parameters that provide important insights into how precipitation variability controls C and H2O dynamics in deserts of the Southwest.

Figure 2. A simple hierarchical Bayesian model that couples diverse field data and mechanistic models related to leaf, soil, and ecosystem carbon dynamics. Field data are categorized as stochastic or covariates (i.e., assume measured without error). Stochastic variables arise from distributions whose means are defined by the latent processes. The latent processes represent the “truth” or unobserved quantities, and they are informed by the mechanistic models.

Covariates

Data

Stochastic

Latent processes

Parameters

Hyper- parameters

NEEObs Net ecosystem

exchange

LAI Leaf area

index

RsObs Soil

respiration

B Root & microbial

biomass

AObs,

Instantaneous photosynthesis

C Environmental

covariates

SWC Soil water

content

NEE Net ecosystem exchange

Rs Soil respiration

A

Instantaneous photosynthesis

ATot Total daily photosynthesis

r

Instantaneous source-specific respiration

NEE NEE obs.

error variance

Rs Rs obs.

error variance

A A obs. error

variance

A site parameters

A species parameters

r site parameters

r source parameters

a1 a2 b1 b2 a1 a2 b1 b2

Ts Soil

temperature

Population effects (Parameters describing distributions from which site-, species-, and source-specific effects arise)

Model of Stomatal Conductance in Larrea tridentata: A Case Study

• Photosynthesis depends onvariable water status of plant (Fig. 3).• Predawn water potential often usedas an indicator of plant water status,instantaneous measures would be more accurate. Don’t often have these data.• Hierarchical Bayesian model can be used to predict instantaneous as a latent variable (Figs. 3 & 4).

• Modeling instantaneous as alatent variable gives improvedresult (Fig 4).• The finding that control plants have lower stomatal conductancethan watered plants (Fig. 5) offers support for the hypothesis that L. tridentate is successful in arid systems partially because of its ability to tightly regulate its stomatal water loss during periods of water stress.

Figure 3.

Figure 4.

Figure 5.

WP estimates from Winbugs

Predicted conductance

-0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30

Obs

erve

d co

nduc

tanc

e

0.00

0.05

0.10

0.15

0.20

0.25

0.30

controlwatered1:1 line

Predawn WP

Predicted conductance

0.00 0.05 0.10 0.15 0.20 0.25 0.30

controlwatered1:1 line

r2 = 0.59r2 = 0.67

Stomatal Conductance of Control and Watered Plants

VPD (kPa)

0 2 4 6 8 10 12 14

cond

ucta

nce

(mol

m-2

s-1

)

-0.02

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

Watered PlantsControl Plants

Photosynthesis depends in

Photosynthesis depends in

part on stomatal conductance

part on stomatal conductance