Parameterization and Sensitivity Analysis of the BIOME–BGC Terrestrial Ecosystem Model: Net Primary Production Controls

TLDR: In this paper, the authors present documented input parameters for a process-based ecosystem simulation model, BIOME-BGC, for major natural temperate biomes, including turnover and mortality, allocation, carbon to nitrogen ratios (C:N), the percent of plant material in labile, cellulose, and lignin pools, leaf morphology, leaf conductance rates and limitations, canopy water interception and light extinction.

Abstract: Ecosystem simulation models use descriptive input parameters to establish the physiology, biochemistry, structure, and allocation patterns of vegetation functional types, or biomes. For single-stand simulations it is possible to measure required data, but as spatial resolution increases, so too does data unavailability. Generalized biome parameterizations are then required. Undocumented parameter selection and unknown model sensitivity to parameter variation for larger-resolution simulations are currently the major limitations to global and regional modeling. The authors present documented input parameters for a process-based ecosystem simulation model, BIOME–BGC, for major natural temperate biomes. Parameter groups include the following: turnover and mortality; allocation; carbon to nitrogen ratios (C:N); the percent of plant material in labile, cellulose, and lignin pools; leaf morphology; leaf conductance rates and limitations; canopy water interception and light extinction; and the percent of...

Figures

Table 4. Main effects of parameter variation in ENF and C3G biom es. Effect coiumns stiow thie expected effect on NPP (g m-^) of raising a parameter from the m ean minus standard error to the m ean plus standard error. Values are m ean and 95% confidence interval calculated from the t dis tribution with nine d egrees of freedom.

Table 3. (Continued) • Volum e 4 (2000) • Paper No. 3 • Page 84

Table 5. Two-way interaction effects of parameter variation in ENF and C3G bi om es. Ttie interaction coiumns stiow ttie expected interaction effect on NPP (g mi-2) cau sed by raising the shown parameters from m ean minus standard error to m ean pius standard error. Values are m ean and 95% confidence intervai caicuiated from ttie t distribution witti nine degrees of freedom (* indicates significant interaction effect).

Table 1 shows mean values for each parameter. Full parameter descriptions and a discussion of their role in BIOME-BGC, statistical information, species names, and citations are included in appendix A.

Table 6. Comparison of parameterization schiemes. Spatial prediction refers to a m ean value caicuiated from Figure 3 or Figure 4; single value per biom e refers to ttie m ean value caicuiated from the literature searches d e scribed in the preceding sections. Parameters are leaf carbon to nitrogen ratio (C:N|eaf. kg C kg N-'), specific leaf area (SLA, m ̂ kg O ’), leaf life span (months), and the percent of leaf nitrogen in Rubisco (PLNR, dimensioniess). Biomes are evergreen need le leaf forest (ENF) and deciduous broad leaf forest (DBF). DBF leaf life span from BiOME-BGC phenology m odel (White et ai. 1997).

Figure 1. Conceptual chart of blome-level parameter groups and resulting max imum net ecosystem assimilation rates. Parameter categories (left boxes) are as follows. 1) Leaf quality Is a function of CiNieaf, the percent of leaf nitrogen In Rubisco (PLNR), LFRT (leaf and fine root turnover), and specific leaf area (SLA). Low C:N|eaf. high PLNR, high LFRT, and high SLA create high photosynthetic capacity foliage. 2) Leaf costs represent the carbon and nitrogen costs to construct and maintain foliage. 3) Disturbance Includes fire mortality (FM) and whole plant mortality (WPM). 4) Drought resistance represents plant ability to regulate stomatal condu ctance under vapor pressure def icit (VPD) and leaf water potential (LWP) stresses. Bottom panels show the average maximum net ecosystem assimilation rotes (fxmol CO2 m 2 s ’) for ea ch biom e (Buchmann and Schulze 1999).

Full text

%40;,7809=5-549(4(%40;,7809=5-549(4(
#*/52(7&5718(9%40;,7809=5-549(4(#*/52(7&5718(9%40;,7809=5-549(4(
:3,70*(2$,77(+=4(30*#03:2(905475:6
!:)20*(90548
:3,70*(2$,77(+=4(30*#03:2(905475:6

!(7(3,9,70>(9054(4+#,48090;09=4(2=8085-9/, !(7(3,9,70>(9054(4+#,48090;09=4(2=8085-9/, 
$,77,8970(2*58=89,335+,2,9!703(7=!75+:*90545497528$,77,8970(2*58=89,335+,2,9!703(7=!75+:*90545497528
0*/(,2&/09,
!,9,7+354+$/574954
$/,%40;,7809=5-549(4(
#9,;,4&":4404.
%40;,7809=5-549(4(0885:2(
"(3(1708/4(",3(40
5225<9/08(4+(++09054(2<5718(9/99688*/52(7<5718:39,+:498.'6:)8
,9:8145</5<(**,88959/08+5*:3,49),4,@98=5:
",*533,4+,+09(9054",*533,4+,+09(9054
&/09,!$/574954#":4404.(4+",3(40!(7(3,9,70>(9054(4+#,48090;09=4(2=8085-9/,
 ?$,77,8970(2*58=89,35+,2,9!703(7=!75+:*90545497528(79/49,7(*9?+50
!# $ 
$/08790*2,08)75:./995=5:-57-7,,(4+56,4(**,88)=9/,:3,70*(2$,77(+=4(30*#03:2(905475:6(9
#*/52(7&5718(9%40;,7809=5-549(4(9/(8),,4(**,69,+-5704*2:805404:3,70*(2$,77(+=4(30*#03:2(9054
75:6!:)20*(90548)=(4(:9/570>,+(+3040897(9575-#*/52(7&5718(9%40;,7809=5-549(4(57357,04-573(9054
62,(8,*549(*98*/52(7<5718385:39,+:

Earth Interactions • Volume 4 (2000) • Paper No. 3 • Page 1
Copyright © 2000. Paper 4-003; 38,228 Words, 5 Figures, 20 Tables.
http://EarthInteractions.org
Parameterization and Sensitivity
Anaiysis of the BiOME-BGC
Terrestriai Ecosystem Modei: Net
Primary Production Controis
Michael A. White,* Peter E. Thornton, Steven W. Running,
and Ramakrishna R. Nemani
Numerical Terradynamic Simulation Group, Missoula, Montana
Received 20 November 1999; accepted 21 June 2000.
ABSTRACT: Ecosystem simulation models use descriptive input parame
ters to establish the physiology, biochemistry, structure, and allocation patterns
of vegetation functional types, or biomes. For single-stand simulations it is
possible to measure required data, but as spatial resolution increases, so too
does data unavailability. Generalized biome parameterizations are then re
quired. Undocumented parameter selection and unknown model sensitivity to
parameter variation for larger-resolution simulations are currently the major
limitations to global and regional modeling. The authors present documented
input parameters for a process-based ecosystem simulation model, BIOME-
BGC, for major natural temperate biomes. Parameter groups include the fol
lowing: turnover and mortality; allocation; carbon to nitrogen ratios (C:N);
the percent of plant material in labile, cellulose, and lignin pools; leaf mor
phology; leaf conductance rates and limitations; canopy water interception and
light extinction; and the percent of leaf nitrogen in Rubisco (ribulose bis-
phosphate-l,5-carboxylase/oxygenase) (PLNR). Using climatic and site de
* Corresponding author address.- Dr. Michael A. White, NTSG, School of Forestry, Uni
versity of Montana, Missoula, MT 59812.
E-mail address: mike@ntsg.umt.edu

Earth Interactions • Volume 4 (2000) • Paper No. 3 • Page 2
scription data from the Vegetation/Ecosystem Modeling and Analysis Project,
the sensitivity of predicted annual net primary production (NPP) to variations
in parameter level of ± 20% of the mean value was tested. For parameters
exhibiting a strong control on NPP, a factorial analysis was conducted to test
for interaction effects. All biomes were affected by variation in leaf and fine
root C:N. Woody biomes were additionally strongly controlled by PLNR, max
imum stomatal conductance, and specific leaf area while nonwoody biomes
were sensitive to fire mortality and litter quality. None of the critical param
eters demonstrated strong interaction effects. An alternative parameterization
scheme is presented to better represent the spatial variability in several of
these critical parameters. Patterns of general ecological function drawn from
the sensitivity analysis are discussed.
KEYWORDS: Biogeocbemical processes; Plant ecology; Land/atmosphere
interactions
1. Introduction and background
Terrestrial net primary production (NPP, g m“^), equal to gross primary production
minus autotrophic respiration, represents the carbon available for plant allocation
to leaves, stems, roots, defensive compounds, and reproduction and is the basic
measure of biological productivity. Tree growth, forage available for grazing, food
production, fossil fuel production, and atmospheric CO2 levels are all strongly
controlled by NPP. Accurate quantification of NPP at local to global scales is
therefore central topic for carbon cycle researchers, foresters, land and resource
managers, and politicians. For recent or current NPP estimates, satellite remote
sensing can be used (e.g.. Potter et al. 1993) but for research investigating pre-
1970s time periods or future climate scenarios, simulation models are required.
Models have been used to simulate regional water and carbon cycles under
current and historical climates (Nemani et al. 1993; Running 1994), soil carbon
dynamics (Motovalli et al. 1994), effects of nitrogen saturation (Aber et al. 1997),
and the location of global carbon sources and sinks (Houghton et al. 1998; Ran-
derson et al. 1997). Models can also be used to develop basic theoretical under
standings of ecosystem function that cannot be tested with field methods (Chur
kina and Running 1998; Schimel et al. 1996). Perhaps most importantly, models
are used to address the political and management need for estimates of ecosystem
responses to chmate changes (Intergovernmental Panel on Climate Change 1995).
In particular, as fossil fuel consumption exponentially increases atmospheric CO2
(Keeling 1994) there is a growing need to provide credible estimates of ecosystem
storage or release of carbon (Hunt et al. 1996; Schimel et al. 2000). NPP is a
conunon component of these modeling approaches.
Large-scale biogeochemical (BGC) modeling, the topic of this research, is a
specific type of modeling that seeks to mechanistically represent ecosystem cycles
of carbon, water, and nutrients at regional to global scales through an integrated
consideration of biology and geochemistry. The simulated land surface is divided
into grid cells described by vegetation type (land cover), slope, aspect, elevation,
albedo, and soil depth and texture (e.g., from Zobler 1986) from which soil water

Earth Interactions • Volume 4 (2000) • Paper No. 3 • Page 3
holding capacity and water release properties may be calculated (Clapp and Hom-
berger 1978). Nitrogen deposition, CO2 concentration, and climate data (usually
monthly or daily) describe the atmosphere. Mathematical equations representing
an abstraction of reality are then used to simulate ecosystem cycles of carbon
(assimilation and respiration), nitrogen (mineralization, immobilization, leaching,
volatilization, and denitrification), and water (evaporation, transpiration, and run
off).
The theoretical basis for NPP predictions and other model processes is usu
ally based on realistic laboratory or field research, yet this same model realism
often translates to a seemingly endless proliferation of difficult to obtain driving
inputs, or parameters. In some cases, parameters are measured for a particular
study, but when left unconstrained by measurement, parameters can be used as
tuning knobs capable of producing a wide range of outputs. We feel that for these
reasons, parameter selection and documentation, not model theory, are the main
factors currently limiting the accuracy and believability of global and regional
model simulations. As Aber (Aber 1997) stated: “ALL of the parameters used in
the model should be listed, and ALL values for those parameters given, along
with the references to the sources of those parameters.” Aber also argued for
complete descriptions of model structure and sensitivity. To address these and
related concems, our goals in this research are to
• provide an account of the source (or lack thereof) for parameters in BI
OME-BGC, a commonly used terrestrial ecosystem process model, for
major temperate biomes;
• assess the sensitivity of NPP to independent variation in every parameter;
• conduct a factorial sensitivity analysis of the most critical parameters;
• investigate pattems of ecosystem function revealed by the sensitivity anal
ysis; and
• present a blueprint for an alternative parameterization scheme for critical
parameters.
2. Materials and methods
2.1 BIOME-BGC
Using prescribed site conditions, meteorology, and parameter values, BIOME-
BGC simulates daily fluxes and states of carbon, water, and nitrogen for coarsely
defined biomes at areas ranging from 1 m^ to the entire globe. Plant physiological
processes respond to diumal environmental variation (Geiger and Servaites 1994),
but BIOME-BGC uses a daily time step in order to take advantage of widely
available daily temperature and precipitation data from which daylight averages
of short wave radiation, vapor pressure deficit, and temperature are estimated
(Thornton et al. 1997; Thornton and Running 1999). Nonlinear diumal photosyn
thetic responses to radiation levels will not be captured by the use of daylight
average radiation, but models initially designed to operate at daily timescales may
still be used to accurately represent short-term variation in carbon fluxes (Kimball
et al. 1997b).

Earth Interactions • Volume 4 (2000) • Paper No. 3 • Page 4
BIOME-BGC simulates the development of soil and plant carbon and nitro
gen pools; no input of soil carbon information or leaf area index (LAI, m^ leaf
area per m^ ground area) is required. LAI controls canopy radiation absorption,
water interception, photosynthesis, and litter inputs to detrital pools and is thus
central to BIOME-BGC. Model structure is discussed by Thornton (Thornton
1998) and is available online (www.forestry.umt.edu/ntsg), and will not be pre
sented here. Briefly, though, NPP is based on gross primary production simulated
with the Farquhar photosynthesis model (Farquhar et al. 1980) minus maintenance
respiration [calculated as a function of tissue nitrogen concentration (Ryan 1991)]
and growth respiration (a constant fraction of gross primary production). Theory
and applications of BIOME-BGC and its predecessor, FOREST-BGC, are widely
available (e.g.. Hunt et al. 1996; Kimball et al. 1997b; Kimball et al. 1997c;
Running 1994; Running and Coughlan 1988; Running and Gower 1991; Running
and Hunt 1993; Running and Nemani 1991; White et al. 1999).
In BIOME-BGC, 34 parameters within several main categories are used to
distinguish separate biomes. 1) Turnover and mortality parameters are used to
describe the portion of the plant pools that are either replaced each year or re
moved through fire or plant death. 2) The allocation of photosynthetically accu
mulated carbon to leaf, stem, and root pools is controlled by a series of allometric
parameters. 3) Carbon to nitrogen ratios define nutrient requirements for new
growth, plant respiration rates, photosynthetic capacity, and litter quality. 4) The
percentage of lignin, cellulose, and labile material in fine roots, leaves, and dead
wood controls litter recalcitrance and influences decomposition rates. 5) Three
morphological parameters control the distribution of LAI at the leaf and canopy
level. 6 ) Several ecophysiological parameters are used to control rates of and
limitations to leaf conductance. 7) Single parameters are used to control water
interception, canopy radiation absorption, and the rate of carbon assimilation.
Conceptually, the parameter groups describe biomes by rejecting excessive detail
and unobtainable parameters while maintaining broadly significant vegetation de
scriptions.
2.2 P aram eterization
For each parameter we conducted a literature search for each biome and calculated
mean and standard deviation. There were two choices when assigning values: use
the mean for each biome or conduct multiple comparison tests to group biome
values together into statistically similar groups. Natural variability within biomes
and, in some cases, limited sample sizes led the statistical approach to produce a
homogeneous parameterization wherein biomes were remarkably indistinguish
able. Since the ecological relevance of biome differences is well recognized (T.
M. Smith et al. 1997) we chose the first option and did not test for statistically
significant differences.
Data were usually available for evergreen needle leaf forest (ENF) and de
ciduous broadleaf forest (DBF), but in the grass literature, C4 data were rare and
many authors reported “grasslands” without C3/C4 discrimination. We therefore
parameterized a single grass biome. The C4 grass (C4G) is simulated with simple
mechanisms to concentrate CO2 levels and to increase quantum yield efficiency.

Frequently asked questions

Q1. What have the authors contributed in "Parameterization and sensitivity analysis of the biome-bgc terrestrial ecosystem model: net primary production controls" ?

The authors present documented input parameters for a process-based ecosystem simulation model, BIOMEBGC, for major natural temperate biomes. E-mail address: mike @ ntsg. umt. edu Earth Interactions • Volume 4 ( 2000 ) • Paper No. 3 • Page 2 scription data from the Vegetation/Ecosystem Modeling and Analysis Project, the sensitivity of predicted annual net primary production ( NPP ) to variations in parameter level of ± 20 % of the mean value was tested. An alternative parameterization scheme is presented to better represent the spatial variability in several of these critical parameters. Patterns of general ecological function drawn from the sensitivity analysis are discussed.