Text S1
Contents
1. The Dynamic Land Ecosystem Model: key processes and sub models...................................... 2
2. Model validation ......................................................................................................................... 9
2.1. Model validation for CO2 fluxes .......................................................................................... 9
2.2. Model validation for CH4 fluxes ........................................................................................ 11
2.3. Model validation for N2O fluxes........................................................................................ 16
3. Model uncertainty analysis method .......................................................................................... 17
Reference cited.............................................................................................................................. 19
1
1. The Dynamic Land Ecosystem Model: key processes and sub models
The Dynamic Land Ecosystem Model (DLEM) is a highly integrated, process-based
terrestrial ecosystem model that aims at simulating the structural and functional dynamics of land
ecosystems affected by multiple factors including climate, atmospheric compositions (CO2, O3),
precipitation chemistry (nitrogen composition), natural disturbance (fire, insect/disease,
hurricane, etc), land-use/cover change, and land management practices (harvest, rotation,
fertilization, irrigation, etc). DLEM consists of five vegetation, three soil, and seven debris
boxes, and couples major biogeochemical cycles, hydrological cycle, and vegetation dynamics to
make daily, spatially-explicit estimates of water, carbon (CO2, CH4) and nitrogen fluxes (N2O)
and pool sizes (C and N) in terrestrial ecosystems.
DLEM includes five core components: 1) biophysics, 2) plant physiology, 3) soil
biogeochemistry, 4) dynamic vegetation, and 5) land use and management (Figure 1). Briefly,
the biophysics component simulates the instantaneous fluxes of energy, water, and momentum
within the land ecosystem and their exchanges with the surrounding environment. Plant
physiology component simulates major physiological processes, such as plant phenology, C and
N assimilation, respiration, allocation, and turnover. Soil biogeochemistry component simulates
the dynamics of nutrient compositions and the fundamental microbe’s activities. The
biogeochemical
processes,
including
the
mineralization/immobilization,
nitrification/denitrification, decomposition, methane production/oxidation are considered in this
component. The dynamic vegetation component simulates the structural dynamics of vegetation
caused by natural and human disturbances. Two processes are considered: the biogeography
redistribution when climate change, and the recovery and succession of vegetation after
disturbances. Like most dynamic global vegetation models, DLEM builds on the concept of plant
functional types (PFT) to describe vegetation distributions. The land use and management
component simulates cropland conversion, reforestation (cropland abandonment), forest
managements (harvest, thinning, fertilization and prescribed fire). DLEM emphasizes the
modeling and simulation of managed ecosystems including agricultural ecosystems, plantation
forests and pastures. The spatially-explicit management data sets, such as irrigation, fertilization,
rotation, and harvest can be used as input information for controlling the ecosystems. It also
simulates urbanization processes, and can be used to estimate the impacts of urban impervious
surface and urban lawn management on ecosystem processes.
The basic simulation unit of DLEM is a single grid with corresponding coverage area.
In this unit, vegetated land surfaces are comprised of one of the natural vegetation functional
types (forests, grassland and shrubs) or urban area or cropping system. The classification of
natural forests is based on the leaf structure (needle and broadleaf), leaf phenology (evergreen
2
and deciduous), climate zones (tropical, temperate, and boreal). The grasslands are divided into
C3 grass, C4 grass, and meadow. The shrubs have evergreen and deciduous types. Besides, other
plant functional types including desert, wetland and tundra are also considered.
For simulating vegetation dynamics, DLEM model uses a strategy similar to the dynamic
global vegetation model (DGVM) LPJ. The time step for vegetation dynamic processes is 1 year.
Two kinds of vegetation dynamic processes can be simulated by DLEM: the biogeography
redistribution when climate change, and the plant competition and succession during vegetation
recovery after disturbances. Like most DGVMs, DLEM builds on the concept of plant functional
types (PFT) to describe vegetation distributions. Many different PFTs that adapt to local climate
can coexist in the same grid, competing light, water, and nutrient resources. For the historical
simulations, which focus on historical variations of carbon, nitrogen, and water cycles affected
by changing climate, land-use, and atmospheric compositions, we used prescribed plant
functional type for natural vegetation cover or human-managed system, like cropland and urban.
Figure S1. Structure and Key Processes of Dynamic Land Ecosystem Model (DLEM)
1.1. Biophysical Processes and Radiation Transmissions
3
Radiation is the major energy source driving hydrological and biogeochemical cycles of
terrestrial ecosystems. Daily incident shortwave radiation (SRAD) is used to estimate the
photosynthesis, transpiration, evaporation, and snow melt. SRAD can be put into the model as
user’s input, or generated by the model. The algorisms for generating this variable and relative
humidity are from MT-CLIM 4.3 developed by Thornton et al. (1998). Daily temperature range,
precipitation, DEM, and information of geographical location are used to generate SRAD.
SRAD is partitioned into reflectance, absorptions by plants, and absorptions by soil
surface. The canopy is divided into sunlit and shaded fractions in which photosynthesis and ET
were estimated independently.
1.2. The hydrological cycle
In the DLEM, water pools in terrestrial ecosystems were classified into six boxes: the
canopy intercepted snow and intercepted water; the ground surface snow; the litter intercepted
water; the upper layer soil water (0 ~ 50 cm), and the lower layer soil water (50 ~ 150 mm). The
water content in each boxes are updated daily based on water input (precipitation and dew) and
the water losses (evaporation, transpiration, sublimation, surface runoff and drainage runoff)
(unit mm/day) driven by the solar radiation and the plant’s physiological processes.
The major processes include the partition of precipitation, canopy interception of rain and
snow, canopy snow sublimation, snowmelt and sublimation from ground snowpack, canopy
evapotranspiration, litter interception of rain or snow, soil surface evaporation, surface runoff
and infiltration, and soil moisture movement.
1.3. The Carbon cycle
The carbon cycle is the most important process in the DLEM; it serves framework for
biogeochemical processes of other nutrients and hydrological processes. The DLEM mainly
simulates carbon fluxes through various pools.
The vegetation carbon pool has six components for trees and shrubs (storage organ, leaf,
heartwood, sapwood, fine root, and coarse root), and five components for herbaceous vegetation
(storage organ, leaf, stem, fine root, and coarse root). Vegetation carbon pool gains carbon
through photosynthesis (Gross Primary Production, GPP), loses carbon through autotrophic
respiration (including maintenance respiration and growth respiration), litter fall, mortality, and
disturbances such as land conversion and fire. The effects of ozone pollution and nitrogen
deposition on photosynthesis are also simulated in the DLEM. The litter carbon pool includes
seven pools for trees and shrubs: coarse woody debris, aboveground very active litter,
aboveground middle active litter, aboveground resistant litter, belowground very active litter,
belowground middle active litter, and belowground resistant litter. The sources of litter pools
include litterfalls from leaves and roots, debris from mortalities, harvest, and land use change.
4
Litter carbon pools can be converted into soil organic matter (SOM) pools and emits CO2 to the
atmosphere through decomposition.
Soil Organic Matter has three pools with different decomposition base rate: very active,
middle active, resistant SOM, and dissolved organic carbon. The balance of SOM depends on the
transformation of litter to SOM, the fractions of conversion from GPP to the dissolved organic
carbon (DOC), the returned organic matter from production decay (e.g. manure), the growth of
microbes, the methane production from DOC, and the decomposition rate. Meanwhile, the
DLEM simulates the life cycles of microbe, and production. It should be noted that the methane
module in the DLEM mainly simulates the production, consumption, and transport of CH4 [Tian
et al., 2010]. Due to relatively small contribution from other substrates [Conrad, 1996; Mer and
Roger, 2001], DLEM only considers the CH4 production from DOC, which is indirectly
controlled by environmental factors including soil pH, temperature and soil moisture content.
The DOC was produced through three pathways, GPP allocation, and decomposition byproducts
from soil organic matter and litterfall. CH4 oxidation, including the oxidation during CH4
transport to the atmosphere, CH4 oxidation in the soil/water, and atmospheric CH4 oxidation on
the soil surface, is determined by CH4 concentrations in the air or soil/water, as well as soil
moisture, pH, and temperature. Most CH4-related biogeochemical reactions in the DLEM were
described as the Michaelis-Menten equation with two coefficients: maximum reaction rate and
half-saturated coefficient. Three pathways for CH4 transport from soil to the atmosphereebullition, diffusion, and plant-mediated transport-are considered in the DLEM [Tian et al.,
2010].
1.4. The Nitrogen cycle
DLEM simulates C-N interaction with “wide-open” N cycle (Rastetter et al., 1997),
where nitrogen inside ecosystem unlimitedly exchanges with exterior environment through
external N supply and N export on atmosphere-land interface (such as N deposition, N fixation,
nitrous trace gas emission etc.) and hydrosphere-land interface (e.g. N leaching loss). In DLEM,
nitrogen cycles are intimately coupled with the carbon cycle by constant C/N ratios of different
biomass pools and SOM. Available nitrogen status can affect the carbon cycle directly through
photosynthesis, respiration, allocation and decomposition processes. Meanwhile, carbon cycle
couples with the nitrogen cycle directly through providing litterfalls with various quantities and
qualities and affecting nitrogen uptake by plants. DLEM simulates the nitrogen in several forms:
organic nitrogen stored in biomass, labile nitrogen stored in plants, organic nitrogen stored in soil
organic matter and litter/woody debris, dissolved organic nitrogen (DON), and inorganic
nitrogen ions such as NH4+ or NO3-.
The major nitrogen processes in terrestrial ecosystem includes nitrogen input from the
atmosphere (through nitrogen deposition and nitrogen fixation), fertilization, nitrogen
5
immobilization/mineralization, plant N uptake, nitrification/denitrification, adsorption/desorption,
nitrogen leaching, and unknown nitrogen loss through fire or other disturbances.
It should be noted that both denitrification and nitrification processes are simulated as
one-step processes as we do not consider the mid-products in each process. Nitrification, a
process converting ammonium into nitrate, is simulated as a function of soil temperature,
moisture, and the NH4+ concentration [Lin et al., 2000]. Denitrification, through which the nitrate
is converted into nitrous gases, is simulated in the DLEM as a function of soil temperature,
moisture, and the NO3- concentration [Lin et al., 2000]. All the products of nitrification and
denitrification that leave the system are N-containing gases. The empirical equation reported by
Davidson et al [Davidson et al., 2000] is used to separate N2O from other gases (mainly NO and
N2).
1.5. Carbon and nitrogen allocation
DLEM simulates carbon and nitrogen allocation by combining the ideas from the
functional equilibrium models (e.g., Friendlingstein et al., 1999), sink regulation models (e.g.,
Marcelis, 1994), and descriptive allometry models (e.g., Wilson, 1988; Marcelis & Heuvelink,
2007). DLEM takes the following assumptions as foundations of carbon and nitrogen partition
at daily time step:
1). Plants have a potential to store proportional carbon and nitrogen to sustain its growth
in harsh situations. If the stored carbon is less than the minimum requirement, 50% of available
GPP (AGPP) (i.e. total GPP minus growth respiration loss) will go to the storage pool until it
gets to minimum storage.
2). Plants have a potential to optimize the capture of limiting resources (Cannell & Dewar,
1994; Litton et al., 2007). DLEM uses some equations revised from Friendlingstein et al. (1999)
to represent limitation effects of light (leaf area index), water, and nitrogen on allocation.
3). Leaves have the first priorities to use the products of photosynthesis and the
remaining carbon is assigned into other tissues (reproduction, root, and stem, accordingly). In
DLEM, the requirement for daily leaf growth is calculated with the leaf phenology and the
maximum leaf carbon content estimated by the size of sapwood pools (for natural vegetation)
and the theoretical maximum leaf carbon (for crops). There are several scenarios for carbon
partitioning:
a. If the available GPP (after being extracted by storage pool) is greater than the
requirement of leaf growth during the growing season, proportional carbon will flow to
the leaf pool to meet the requirements. Then, some constant part (vegetation type-specific)
will flow to reproduction pool. The left carbon will flow to stem and roots with the ratio
calculated by section (2).
6
b. If the available GPP (after being extracted by storage pool) is less than the
requirement of leaf growth during the growing season, all of this available GPP will flow
to the leaf pool. If current LAI is less than 1 and the soil moisture scalar for
photosynthesis is greater than 0.05, the storage pool will be activated to join the
allocation process to meet the leaf carbon requirement. Some constant proportion of this
total available carbon will flow to root. This situation normally occurs at the beginning of
growing season for deciduous vegetation and crops.
c. If current leaf carbon is greater than the potential leaf carbon (calculated with
phenology and maximum LAI), or it is the leaf-off season (phenology period from the
climax to the minimum) of deciduous vegetation, the carbon partition is based on the
ratios calculated in section (2). The extra carbon beyond potential leaf carbon will flow to
stem and roots according to their relative partition coefficients.
4). Availability of nitrogen can limit the overall carbon partitions to different organs. We
assume that all the nitrogen used for carbon allocation to maintain tissue-specific C/N ratio
comes from the storage pool. For leaf allocation, we take its maximum C/N ratio as initial
baseline to estimate its nitrogen requirement. In nitrogen uptake process, the leaf C/N ration can
be changed according to the contemporary nitrogen status of the ecosystem (see detail in the
nitrogen section). If the total nitrogen requirement cannot be met, the extra GPP will flow to
storage pool till its maximum storage capacity. Then the allocation process will feed back to the
former simulation step to adjust GPP as we assume that plants cannot produce more
photosynthate than what they can use in forming biomass and material storage.
1.6. Nutrient export from landscape to coastal region
By incorporating the processes of soil erosion with modified universal soil loss equation
(MUSLE) , water routine process based on global river network data sets (GTN30) [Vörösmarty
et al., 2000], and simplified nitrogen removal process in the river systems [Alexander et al., 2000;
Wollheim et al., 2006; Wollheim et al., 2008], we expanded DLEM model with Nutrient Export
(NE) component to track the leached nutrients from terrestrial ecosystems to freshwater aquatic
systems and eventually to the coastal regions. The DLEM-NE model is capable of assessing the
impacts of natural and anthropogenic driving forces on nutrients (currently we consider forms of
DOC, POC, DON, DIN, and PON) leaching and delivering from terrestrial system to coastal
regions. The export of nutrients from landscape to coastal area includes three major processes:
the productions of nutrients in inland watersheds; the leaching of nutrients from land along with
overland flow and base flow; and the transport of nutrients through river networks to river outlets
in coastal region.
1.7. Disturbances and plantation managements
7
DLEM simulates the impacts of intensive land management and disturbances on the
ecosystem structure and functions. The land management includes: forest harvest, forest thinning,
fertilization, weed control, insect & disease control and prescribed fire; the natural disturbances
include: hurricane & storm, wild fire, and pest & disease. Forest management and disturbance
will not completely change the vegetation functional types, but directly influence the carbon,
nitrogen and water flux, forest biomass, litter composition and decomposition, soil organic
matter, wood and wood products and CO2 mitigation by replacing fossil fuel with wood-based
bio-fuel, and indirectly influence other processes of forest carbon, nitrogen and water cycles.
DLEM estimates the wildfire-caused CO2 and non-CO2 GHG emissions from terrestrial
ecosystems. The carbon emission is based on the equations utilized in Seiler and Crutzen [1980];
Thonicke et al. [2001], and Lu et al. [2006].
The fates of wood products are tracked by the DLEM plantation module. In the same year
as the harvest takes place, several intermediate processing and allocation steps are done, until the
carbon resides in the end products, the millsite dump, or is transferred to the bioenergy module.
When the end products are discarded at the end of their lifespan, they can be recycled, deposited
in a landfill, or can be used for bioenergy in the specific module. Carbon is released to the
atmosphere through decomposition at the millsite dump, at the landfill, or via the bioenergy
module. The fates of the harvested wood include: sawnwood, boards wood, paper wood and
firewood. The initial product fractions (Default values) for these three fates are given as
Schelhaas, et al. [2004].
1.8. Agricultural ecosystem module
The DLEM agricultural module enhances the ability of DLEM to simulate the interactive
effects of agronomic /land management practices and other environmental factors on crop
growth, phenology and biogeochemical cycles in croplands. It aims to simulate crop growth and
yield, as well as carbon, nitrogen, and water cycles in agricultural ecosystems. All the processes
of crop growth (e.g. photosynthesis, respiration, allocation) and soil biogeochemistry (e.g.
decomposition, nitrification, fermentation) are simulated in the same way as in DLEM for all
natural functional types and with a daily time-step, but different crops are specifically
parameterized according to each crop type. Besides natural environmental driving factors, the
module pays special attention to the role of agronomic practices, including irrigation, fertilization
application, tillage, genetic improvement and rotation on crop growth and soil biogeochemical
cycles.
1.9. Urbanization processes
Unlike most of current biogeochemical models that either ignore the urbanization processes
or simply treat the urban ecosystem as cropland or grassland, DLEM includes a concise but
8
effective urban module that can simulate the impacts of urbanization on ecosystem structure and
functions. The model treats urban as a mosaic of three different land-use fractions: urban
impervious surface (UIS), urban lawn (ULW), and urban natural (unmanaged) vegetation (UNG).
For simplification, the relative fractions of these three land-use types will not change after
urbanization in DLEM, while the regional coverage of UIS and ULW will change as the results
of urban expansion. Urbanization takes place when the land-use type of the current year is
changed from non-urban types (i.e. the potential vegetation or cropland) into an urban type.
During the urbanization, the land-conversion takes place in UIS and ULW fractions, the prior
local vegetations will keep undisturbed in the UNG. Since trees that are removed from urban
areas are not normally developed into wood products for long-term carbon storage [Nowak and
Crane, 2002], unlike the impacts of land-use change during cropland-conversion, DLEM
assumes that all wood products during urbanization will decompose in one year.
2. Model validation
2.1. Model validation for CO2 fluxes
Consistency between model results and field measurements is essential for establishing
the credibility of biogeochemistry models such as DLEM. To evaluate model capabilities, we
compared our model estimates of net ecosystem production (NEP) to short-term measurements
of net ecosystem exchange (NEE) at six eddy co-variance sites in China (Table S1). The eddy
covariance technique has been recognized as one of the most reliable approaches for estimating
the net exchange of carbon dioxide between land ecosystems and the atmosphere. We ran our
models in site-specific mode, using the driving variables specific to the grid cell in which the
field study was conducted. The model results were in reasonable agreement with the
measurements for all sites, with the modeled annual NEE estimates falling within +/- 20 percent
of the eddy flux measurements (Table S1). In addition, the NEP estimated by DLEM also
captured the daily/seasonal variations in the field NEE estimates at eddy covariance sites in
which daily NEE measurements and climate data were available (Figure S2 and S3).
9
Figure S2. Comparison of daily net ecosystem production simulated by DLEM against observed
data in dry farmland of Yucheng, northern China (116o36’ E, 36o57’ N) (Cited from Tian et al.,
2011)
Figure S3. Comparison of daily net ecosystem production simulated by DLEM in temperate
evergreen needleleaf forest in Qianyanzhou, Southeastern China (26o45’ N, 115o04’ E) (Cited
from Tian et al., 2010)
Table S1. Comparison of model and eddy covariance net ecosystem exchange estimates (A
negative value represents a terrestrial carbon sink whereas a positive value represents a carbon
source) (Cited from Tian et al., 2010)
Ecosystems/Location
Time Period
Field observation
( g C m-2 yr-1)
DLEM (g C
m-2 yr-1)
Refs
Temperate mixed forest/Changbaishan
2003
-188
-152
Wu, 2006
10
(42o 24’ N, 128o 06’ E)
Conifer plantation/Qianyanzhou
(26o45’ N, 115o04’ E)
Subtropical evergreen broadleaf
forest/Dinghushan
(23o 10’ N, 112o 32’ E)
Alpine steppe-meadow/Lasha
(37o 40’ N, 91o 05’ E)
Alpine Tundra/Changbaishan
o
(41 53’ to 42o 04’N, 127o 57’ to 128o 11’E)
Winter wheat and summer corn/Yucheng
(116o36’ E, 36o57’ N)
2003 - 2004
-424
-426
Liu et al.,
2006
2005
-436
-332
Zhang,
2006
2004
-6
-9
Xu, 2006
1998 - 1999
40
47
Dai et al.,
2002
October 15,
2002 - October
14, 2003
-198 g C m-2 yr-1
-193 g C m-2
yr-1
Li et al.,
2006
2.2. Model validation for CH4 fluxes
For the site-level validation, we compared our simulated CH4 fluxes with observations at
marshlands in Sanjiang and Ruoergai stations (Figure S4A, B, C, and D). The daily climate data
were provided by the China Meteorological Administration (CMA) (www.cma.org.cn); we used
the climate data from the CMA site that is close to the validation site. The atmospheric CO2
concentration is from National Oceanic and Atmospheric Administration (NOAA). Nitrogen
deposition and O3 pollution were retrieved from our regional dataset, while other soil and
geological properties were from Chinese Ecological Research Network (CERN) or published
literatures. The validations for two sites were shown in Figure S4. The Figures S4A and S4B
showed that the simulated and observed CH4 fluxes are tightly consistent for the first two years
in the Sanjiang Stations, while several observed “hot moments” in CH4 emission were not
captured by simulations. The overall comparison showed the good agreement between
observation and simulations (slope = 0.9323; P < 0.001) (Figure S4B). The simulated daily CH4
fluxes are quite consistent with the observed CH4 fluxes in the Ruoergai station marshland
(Figure S4C and S4D). The overall comparison showed the consistency between simulations and
observations (slope = 0.9953, P < 0.001), while the inconsistencies were showed between
simulated and observed CH4 fluxes during growing season (Figure S4C).
We further collected more site-level CH4 fluxes across China to verify the simulated CH4
flux at regional scale. Due to the scarcity of observations of CH4 flux across China, we only got
monthly CH4 fluxes for three more sites [Cui et al., 1998; Hirota et al., 2004; Wang et al., 2009]
(Figure S4E). These three sites are representative of the three major areas with dense marshland
in China: North China, Northeast China, and Northwest China. The total site-month is 34. The
comparison shows the consistencies between observed CH4 flux and simulated CH4 flux across
11
China (slope = 0.83; P < 0.001). This consistency indicated that the simulated CH4 fluxes are
reliable at regional scale.
Figure S4. Verification of model performance on simulating CH4 flux over marshland in China
(A: Time-series comparison of observed and simulated CH4 flux in the Sanjiang marshland; B:
scatter plot of simulated and observed CH4 flux for the Sanjiang marshland; C: Time-series
comparison of observed and simulated CH4 flux in the Ruoergai marshland; D: scatter plot of
simulated and observed CH4 flux for the Ruoergai marshland; E: scatter plot showing
comparison between simulated and observed CH4 flux over China (Circles represent
observations from Wang et al.,[2009]; Squares represent observations from Cui et al., [1998];
Triangles represent observations from Hirota et al [2004])) (Cited from Xu and Tian, 2010)
12
20
CH 4 fluxes (mgC/m 2 /day)
Observed
Modeled
10
0
-10
1/1/1997
4/1/1997
7/1/1997
10/1/1997
1/1/1998
4/1/1998
7/1/1998
Da te (month/da y/year)
Figure S5. Comparison of DLEM-estimated CH4 flux with field observations for dry cropland in
Yucheng (36° N, 116° E) (Cited from Ren et al., 2011)
Figure S6. Comparison of DLEM-estimated CH4 flux with field observations for rice paddy in
Qingyuan (23° N, 112° E) (Cited from Ren et al., 2011)
Two sites, including one dry farmland (rotation of 15 winter wheat and summer maize in
Yucheng) and one rice paddy field (two crops of rice 16 in Qingyuan), were selected for model
validations. A comparison of modeled CH4 fluxes and observed CH4 fluxes in dry cropland (Figure
S5) and rice paddy (Figure S6) showed DLEM’s ability to capture not only seasonal patterns, but
also the absolute values of CH4 fluxes. However, two pulses of CH4 flux were simulated by the
module because of two periods of extremely high precipitation. Further investigation showed that the
first peak of CH4 emissions was caused by two-days of heavy precipitation with a total rainfall of
69.3 mm and the second peak of CH4 emissions was associated with a period of heavy precipitation
13
with 60.4 mm per day. It should be noted that the annual precipitation for Yucheng station was 574
mm in 1997. The DLEM agricultural module also simulated the seasonal pattern of CH4 fluxes from
a rice paddy field in the Qingyuan, Southern China.
14
Figure S7. Comparison of the DLEM-estimated CH4 and N2O fluxes with field observations (A:
CH4 flux in Durham Forest (42˚N, 73˚W) (TRAGNET); B: CH4 flux in Alaska wetland (64.8˚N,
147.7˚W) (TRAGNET); C: CH4 flux in Sallie’s fen (43.21˚N, 71.05˚W) ; D: CH4 flux in
Hubbard Brook Forest (43.95˚N, 71.74˚W) (Groffman et al., 2009; Groffman et al., 2006); E:
N2O flux in Hubbard Brook Forest (43.95˚N, 71.74˚W) (Groffman et al., 2009; Groffman et al.,
2006). The error bars in Fig. S7D and S7E represent the standard deviations of four or five
replicated observations; the regression models for these five site-level validations are: Modeled =
0.9389 * observed, r = 0.562, P < 0.001 for A; Modeled = 0.5882 * observed, r = 0.628, P <
0.001 for B; Modeled = 0.8795 * observed, r = 0.502, P < 0.001 for C when fluxes higher than
1000 mg C m -2 day-1 were removed; Modeled = 0.7937 * observed, r = 0.595, P < 0.001 for D;
Modeled = 0.7042 * observed, r = 0.633, P < 0.001 for E) (Cited from Tian et al., 2010)
Two forest sites (Durham forest and Hubbard Brook forest) and two wetland sites
(Alaska wetland and Sallie’s fen) not used in model parameterization were selected for site-level
model verification (Figure S7). We obtained the observational flux data from various sources
including The United States Trace Gas Network (TRAGNET) online dataset
(http://www.nrel.colostate.edu/projects/tragnet/), field observations in Hubbard Brook forest by
Groffman et al. (2006, 2009) and Sallie’s fen (Patrick Crill, personal communication, 2008).
Four simulations for CH4 and one for N2O showed that model results are significantly correlated
with observational data even though the DLEM model underestimated some fluxes (Figure S7A,
B, C, D, and E). While the general seasonal patterns of CH4 flux at these sites were consistent
with the observations, the DLEM model did not capture a few CH4 flux pulses during the peak
growing season in the Sallie’s fen (Figure S7), and underestimated CH4 flux at Alaskan wetland
site (Figure S7B). For the N2O flux, the DLEM model well captured the seasonal pattern and
annual flux of N2O in Hubbard Brook forest, but missed several spikes in observational data
(Figure S7E). This phenomenon of peak fluxes in CH4 and N2O has been observed in a number
of field studies (Chapuis-Lardy et al., 2007; Song et al., 2009), but the underlying mechanisms
still remain unknown.
The quantitative point-to-point comparisons of the modeled and observed data also show
that the DLEM captured the seasonal patterns of CH4 and N2O fluxes in terrestrial ecosystems at
site level. The statistical results could be found in Figure S7. Comparisons between CH4 flux
with soil temperature and precipitation indicate that the soil temperature is the major factor
controlling CH4 and N2O fluxes at site level. The soil temperature is negatively correlated with
CH4 uptake at Durham forest and Hubbard forest sites; while the precipitation events did cause
some spikes in CH4 emission (Figure S7). For the Alaska wetland and Sallie’s fen, temperature
control on CH4 emission was obvious, while the precipitation did not show apparent effects on
CH4 emission. For the N2O emission, temperature effect was shown at seasonal scale, while the
15
precipitation effect appeared at daily scale. This hierarchical control on N2O emission was
consistent with a field study (Brumme et al., 1999).
2.3. Model validation for N2O fluxes
We also conducted several site-level validations for N2O flux across China (Figures S810). The results show that the DLEM could capture not only the seasonal pattern but also the
magnitude of N2O flux.
Figure S8. The comparison of simulated N2O flux against observational data for a grassland
ecosystem at Inner Mongolia (44.05°N, 113.85°E) (R2 = 0.3763)
Figure S9. The comparison of simulated N2O flux against observational data at Qingyuan rice
paddy field (23° N, 112° E) (R2 = 0.2379)
16
Figure S10. The comparison of simulated N2O flux against observational data in natural wetland
ecosystem at the Sanjiang Plain station (47.58°N, 133.52°E) (R2 = 0.2959)
3. Model uncertainty analysis method
As in our previous study [Xu 2010], a Bayesian Monte Carlo method was used to
generate the uncertainties of estimated terrestrial fluxes of GHGs fluxes. A combination of local
sensitivity analysis and local uncertainty analysis was utilized to evaluate regional terrestrial
fluxes of CO2, CH4 and N2O (Figure S11). In the DLEM model, there are more than twenty
parameters directly controlling CO2, eight parameters directly controlling CH4 production,
consumption, and transport, and four parameters directly controlling N2O flux. First, we assumed
that the distribution of each parameter follows normal distribution based on our previous
calibration experience [Tian et al., 2010b]. Second, we conducted local sensitivity analysis to
identify the major parameters responsible for terrestrial CO2, CH4 and N2O fluxes, respectively,
and confirm the distribution of each parameter in controlling these fluxes; the local sensitivity
analysis was conducted by evaluating the changes in simulated gas fluxes in response to a 20%
increase and decrease of each parameter. Third, combined with the priori knowledge of
parameters for CH4 and N2O modules, we used improved Latin Hypercube Sampling (LHS)
approach to randomly select an ensemble of 300 sets of 10 parameters responsible for CO2, CH4
and N2O flux, in the Dynamic Land Ecosystem Model (DLEM). Finally, we set up the
simulations by using the sampled 300 sets of parameters; and the simulated results were analyzed
to derive the uncertainties in gas fluxes induced by parameters.
17
Calibrated
parameters
Priori knowledge of
parameters
Local sensitivity
analysis
Variables sensitivity to parameters and potential
distribution of parameters
Latin Hypercube Sampling (LHS) Method
Local Uncertainty Analysis (MCMC)
Parameter-induced uncertainties in regional
estimations of CO2, CH4 and N2O fluxes
Figure S11. Diagram showing the approach combining local sensitivity and uncertainty analyses
in evaluating parameter-induced uncertainties in regional estimation of terrestrial fluxes of CO2,
CH4 and N2O (MCMC: Markov Chain Monte Carlo; Cited from Xu 2010)
The LHS was selected in this study because it is a reliable approach to conduct randomly
sampling with greatly-reduced calculation (Haefner, 2005). The algorithm is as follows: divide
the range of possibility of each parameter being selected (i.e. 0 ~ 1) into 100 (300 in this study)
equally probable ranges, and then randomly (uniformly) select one value from range of this
parameter (± 20% of its calibrated value). Then the Latin Hypercube mechanism was used to pair
all the sampled parameters for forming an ensemble of parameters for simulation. Because the
entire range of each parameter has been equally probable separated and selected, so the randomly
sampled values could completely represent the distribution of parameter population.
After 300 set of simulation results for terrestrial fluxes of CO2, CH4, and N2O were
obtained, the simulations with one of the outliers in CO2, CH4, and N2O fluxes were excluded for
analysis. The final simulation results were utilized for reporting uncertainties associated with
CO2, CH4 and N2O fluxes.
18
Reference cited
Brumme, R., W. Borken, and S. Finke (1999), Hierarchical control on nitrous oxide emission in
forest ecosystems, Global Biogeochemical Cycles, 13(4), 1137-1148.
Cannell, M.G.R., R.C. Dewar, 1994. Carbon allocation in trees: a review of concepts for
modelling. Advances in Ecological Research, 25, 59–104.
Conrad, R. (1996), Soil microorganisms as controllers of atmospheric trace gases (H2, CO, CH4,
OCS, N2O, and NO), Microbiological Reviews, 60(4), 609-640.
Davidson, E. A., M. Keller, H. Erickson, E., L. V. Verchot, and E. Veldkamp (2000), Testing a
conceptual model of soil emissions of nitrous and nitric oxides, BioScience, 50(8), 667680.
Friedlingstein, P . G . Joel C . B . Field and I . Y . Fung (1999) Toward an allocation scheme
for global terrestrial carbon models, Global Change Biology, 5, 755-770
Lin, B., A. Sakoda, R. Shibasaki, N. Goto, and M. Suzuki (2000), Modeling a global
biogeochemical nitrogen cycle in terrestrial ecosystems, Ecological Modelling, 135, 89110.
Litton, C.M.; J.W. Raich, M.G. Ryan, 2007. Carbon allocation in forest ecosystems, Global
Change Biology. 13: 2089-2109.
Lu, A. (2006), Fire Disturbance and Ecosystem Carbon Cycle-Regional Analysis and Modeling,
Dissertation thesis, 128 pp, Chinese Academy of Sciences, Beijing.
Marcelis, L.F.M. (1994), A simulation model for dry matter partitioning in cucumber, Annals of
Botany, 74:43-52.
Marcelis, L.F.M., E. Heuvelink, 2007. Concepts of modeling carbon allocation among plant
organs, in J. Vos, L.F.M. Marcelis, P.H.B. de Visser, P.C. Struik and J.B. Evers (eds.),
Functional-Structural Plant Modelling in Crop Production, 103-111.
Mer, J. L., and P. Roger (2001), Production, oxidation, emission and consumption of methane by
soils: a review, European Journal of Soil Biology, 37, 25-50.
Rastetter, E. B., G. I. Ågren and G. R. Shaver. 1997. Responses of N-limited ecosystems to
increased CO2: A balanced-nutrition, coupled-element-cycles model. Ecological
Applications 7:444-460.
Ren, W., H. Tian, X. Xu, M. Liu, C. Lu, G. Chen, J. M. Melillo, J. Reily, and J. Liu (2011),
Spatial and temporal patterns of CO2 and CH4 fluxes in China's croplands in response to
multifactor environmental changes, Tellus B, In press.
Tian, H., X. Xu, C. Zhang, W. Ren, G. Chen, M. Liu, D. Lu, and S. Pan (2008), Forecasting and
assessing the large-scale and long-term impacts of global environmental change on
terrestrial ecosystems in the United States and China, Springer, New York.
19
Tian, H., X. Xu, M. Liu, W. Ren, C. Zhang, G. Chen, and C. Lu (2010), Spatial and temporal
patterns of CH4 and N2O fluxes in terrestrial ecosystems of North America during 19792008: application of a global biogeochemistry model, Biogeosciences, 7, 2673-2694.
Tian, H., et al. (2011), China's terrestrial carbon balance: contributions from multiple global
change factors, Global Biogeochemical Cycles, In press.
Wilson, J.B., 1988. A review of evidence on the control of shoot:root ratio, in relation to models.
Annals of Botany, 61:433-449.
Wu, J. (2006), Carbon budget of the broadleaved Korean pine forest in Changbai Mountain. Ph.
D. Dissertation. Chinese Academy of Sciences, Shenyang.
Xu, L. (2006), Observation and simulation of the net ecosystem exchange over alpine meadow in
the Qinghai-Tibet Plateau and its responses to global change. Ph.D. Dissertation. Chinese
Academy of Sciences, Beijing.
Xu, X., H. Tian, C. Zhang, M. Liu, W. Ren, G. Chen, and C. Lu (2010), Attribution of spatial
and temporal variations in terrestrial ecosystem methane flux over North America,
Biogeosciences Discussions, 7, 5383-5428.
Xu, X. (2010), Modeling methane and nitrous oxide exchanges between the atmosphere and
terrestrial ecosystems over North America in the context of multifactor global change,
Dissertation thesis, 199 pp, Auburn University, Auburn.
Xu, X., and H. Tian (2011), Methane exchange between marshland and the atmosphere over
China during 1949-2008, Global Biogeochemical Cycles, Under review.
Zhang, L. M. (2006), Ecophysiological controls on seasonal variations of ecosystem carbon
exchange of typical forest ecosystems along NSTEC. Ph.D. Dissertation. Chinese
Academy of Sciences, Beijing.
20