A "Make-a-Difference" Experiment to Assess the Value of ARM Data in Carbon Cycle Models
William W. Hargrove, Craig C. Brandt, Henriette I. Jager, and Raymond A. McCord
Data from the ARM program
include many of the measurements needed by carbon modelers to predict
carbon dynamics in terrestrial ecosystems. How much difference, if
any, would using ARM measurements rather than any of several synthetic
climate generators typically used by carbon modelers make in estimates of
respiration and productivity predicted by carbon models? We designed a
"make-a-difference" simulation experiment to compare differences in carbon
flux predictions based on synthetic input data with predictions based
on historical measurements taken from the ARM archive. The experimental
design used alternative data sources that a clever carbon modeler might
employ in the absence of ARM measurements. Since "true" carbon fluxes
are unknown, our experiment examined only the magnitude of differences
created when the same carbon model, Biome-BGC, was driven
with synthetic climate rather than actual ARM measurements. The position
of ARM-derived measurements within population distribution of synthetic
records provided a non-parametric test of significance.
Two widely used weather generators, MTCLIM and
GEM (USDA-ARS), were used to produce artificial
weather input streams in the absence of ARM measurements.
MTCLIM (Mountain Climate Simulator) was originally
developed by Running et al. (1987), and version 4.3 has been substantially
modified by Dr. Peter Thornton, currently at NCAR. MTCLIM
"adjusts" temperature, vapor pressure, and solar radiation values supplied
at a "base" location to values appropriate for a "site" at a different
location, based on latitude, elevation, and mean yearly precipitiation.
MTCLIM is commonly used as a first step in running the
Biome-BGC model, and its output is designed to match the
input format needed by Biome-BGC.
GEM
(Generation fo weather Elements for Multiple applications) was
developed from the basic internal structure of USCLIMATE
(Hanson et al. 1994) and WGEN models (Richardson 1981,
Richardson and Wright 1984), but includes significant improvements.
We used version 6 of GEM, designed for use with the Shaw
water quality model. The GEM6-Shaw generator used in this
study, obtained from Dr. William Frymire at USDA-ARS in Boise, ID, is
endorsed as the weather generation tool recommended by the USDA NRCS
and ARS.
We reviewed the data requirements for Biome-BGC,
a popular carbon model, and identified meteorology and radiation
measurements from the Southern Great Plains (SGP) ARM Central Facility
that were needed to run the model. We simulated a wheat crop growing
at the ARM Central Facility for a 5-year period from 1996 through 2000.
The parameters for c3 grasses used in White et al. (2000) were employed,
except that the frequency of wildfire was set to zero, and the initial
soil nitrogen was set to 0.01 and 0.05 kgN/m2 for litter and mineral
soil nitrogen, respectively. No attempt was made to simulate the wheat
harvest with the Biome-BGC model. The "make-a-difference"
comparison was performed for both the input to the Biome-BGC
model and the output from the model.
Before ARM measurements could be used as input, we needed to identify outliers, fill gaps in the data record using imputation methods, and summarize the measurements at an appropriate temporal scale. Daily means were prepared from ARM SMOS and SIRS data. Maximum and minimum temperatures were selected as maxima and minima from hourly summaries to limit extreme values. Mean daylight temperature and mean daylight vapor pressure deficit (VPD) were calculated using sunrise and sunset calculations based on the location of the Central Facility.
Different strategies were used to produce populations of
synthetic climate data from the two climate generators (Figure 1). GEM is a stochastic
generator. Using coefficients derived from the 30-year Climate Normals
(1960-1990) at the Oklahoma City airport, along with a unique set of
random number seeds, we produced 30 stochastic replicates of climate
for the Central Facility for the 5-year simulation period.
Since MTCLIM is a deterministic model, we used a
spatial technique to obtain a population of 5-year climate estimates.
Measurements from 1996 - 2000 from the 30 Oklahoma MESONET stations
closest to the Central Facility were used to drive MTCLIM.
MTCLIM was used to "adjust" the climates from each of the
MESONET station locations to the Central Facility location.
Finally, we used a combination of GEM followed by
MTCLIM. In the GEM-only pathway, it was assumed
that the climate at the Oklahoma City airport was not substantially
different from that at the Central Facility. In the combined pathway,
MtClim was used to "adjust" each of the 30 GEM-suppplied
climates to the exact location of the Central Facility.
Biome-BGC was "spun-up" to equilibrium initial conditions
using the 30-year Climate Normals from the Oklahoma City airport.
The model took 1080 years to reach an equilibrium (residual trend =
0.000014), and these steady-state initial conditions were used to start
each of the simulation pathways in the "make-a-difference" experiment.
Stochastic or spatial variability were used to produce a population
of artificial data for the SGP from each of the synthetic weather
generators.
Biome-BGC, like most simulation models, cannot
tolerate missing data or input gaps. For all input variables except
precipitation, we fit linear regressions with all spatial neighbors.
To interpolate values for missing dates, we drove the regression with
data from the neighboring site with the highest r2 that had
data for that date. EF-9 had the strongest predictive relationship
for temperatures at the Central Facility (all r2 > 0.98).
The EF-15 site had the strongest relationship for shortwave radiation
and VPD with the Central Facility.
Site-to-site correlations for precipitation were low (r2
< 0.3). We used a weighted average of precipitation at surrounding sites
as the best predictor for missing precipitation values. EF-9 and EF-15
were used for the Central Facility, while the 30 closest MESONET sites
were used for the MTCLIM pathway.
Figures 2 and 3 compare the 5th-to-95th percentile ranges of cumulative synthetic precipitation for each of the three climate generator pathways with precipitation measurements from ARM on a per-year basis and over the duration of the simulation, respectively. The legend key is the same for all remaining figures, with colored cross-hatching to indicate the 5th-to-95th percentile ranges for each of the three synthetic pathways. Precipitation was highly variable across all 5 simulation years. The ARM measurement of precipitation was within the range of the synthetic estimates, but was often close to the extremes, and changed from being overestimated in 1996 to being underestimated during the four years that followed. The timing of precipitation events was found to be critical to simulated photosynthesis, particularly in late summer.
Figure 4 compares median yearly values for VPD, while Figure 5 shows cumulative VPD values over the simulation period. VPD was generally overestimated by the climate generators, but was underestimated for a critical period in the late summer. Maximum, minimum, and mean daylight temperatures were slightly overestimated, contributing to this overestimate of VPD. The ARM measurement falls at the lower limit of the range in the cumulative plot (Figure 5).
Daylight average shortwave radiation (Figures 6 and 7) were consistently and significantly underestimated by all of the synthetic climate generators. Both the yearly median and the cumulative measurements fell outside the 5th-to-95th percentile ranges for most simulated days. Underestimates in shortwave radiation may have contributed to underestimates of photosynthetic production as compared with ARM measurements.
Hot, dry conditions in late summer caused increased maintenance respiration (Figure 8) and inhibited simulated estimates of Net Primary Productivity (NPP) (Figure 9). The increase in maintenance respiration was underestimated by all climate generators except MtClim. NPP simulated by ARM measurements was overestimated during these brief but important hot, dry periods in late summer.
Climate forcings from all of the synthetic generator pathways led to underestimates of leaf-level carbon assimilation (Figure 10). Cumulative carbon in the leaf litter pool was also underestimated (Figure 11). Soil carbon, however, was overestimated, possibly because the transfer of carbon from litter to soil during decomposition was overestimated due to temperature overestimates.
Net Ecosystem Productivity (NEP) estimates were highly variable,
especially in the late summer (Figure
12). The variable nature of NEP from
year-to-year is shown in Figure 13. In some years, the median
NEP is always positive, but in other years it dips far below the zero
line. Although such variations are shown in the populations from the
synthetic climate estimators, the trajectory of NEP in any single year
is strongly dependent on the exact sequence of hot, dry days occurring in
late summer. Differences in the specific late-summer climate trajectory
accumulate and are often significant throughout
the five-year simulation (Figure 14). Model estimates of NEP
generated from MTCLIM matched those from ARM measurements
most closely, but still significantly underestimated NEP most of the
time (Figure 14).
In some cases, differences in the input drivers were propagated
through the Biome-BGC model, and were visible in the output
estimates. Effects of temperature overestimation on decomposition
rates, and the subsequent imbalance of carbon pools in litter versus
soil have already been discussed. Cumulative yearly estimates of
transpiration (Figure 15) and NPP (Figure 16) can be seen to mirror
patterns of cumulative yearly precipitation
(Figure 2).
The climate simulators, both alone and together, did a good job of
simulating the conditions of an average year, as reported elsewhere
(Johnson et al. 1996), but failed to capture the variation present in
the measured climate. Wheat is limited in the Biome-BGC
model by nitrogen, high temperature, and precipitation. These results are
consistent with those reported for Biome-BGC in a published
sensitivity analysis (White et al. 2000). Simulated photosynthesis shuts
down during hot, dry periods, while maintenance respiration increases,
causing simulated NPP and NEP to go strongly negative. The particular
timing of these hot, dry periods, especially in late summer, has large
effects on the magnitude of the carbon flux estimates produced by
Biome-BGC.
Estimates of leaf area index and vegetation carbon were about the same for all pathways. Estimates of these pools were not so temporally variable, so the "average" behavior of the climate generators produced estimates of these equal to estimates produced from the ARM measurements. However, there were large differences in estimates for NEP, litter carbon, and leaf-level assimilation, and the differences in carbon pools tended to grow and accumulate throughout the 5-year simulation period.
The temporal scale used for comparing results in the "make-a-difference" experiment is important. Comparison of annual means show few differences, in most cases (but see NEP, Figure 12), since the climate generators perform well, on average. Comparison of particular actual years, comparison of seasons (esp. late summer), and multi-year comparisons typically show that the use of ARM measurements make a large difference in carbon flux estimates.
We plan to repeat the "make-a-difference" experiment using another
carbon model, SiBD (Sellers et al. 1996). The SiBD model has greater
emphasis on carbon fluxes rather than pools. It requires more detailed
input data at a higher temporal frequency, including downward longwave
radiation. The emphasis on fluxes may favor a "no differences" result,
but the increased temporal frequency and reliance on radiometric inputs
may make a "difference" in the comparison.
Gap-filling and the imputation of missing measurement values continues to be a central issue when using ARM measurements in simulation models. We are planning a sister experiment to determine whether different gap-filling techniques "make-a-difference" in carbon model outputs. We will perform a "hole-punching" experiment to produce artificial gaps of varying lengths in different parameters through different seasons, and will then use a jackknife approach to compare estimates with those produced using the originally removed measurements.
We believe that the "make-a-difference" design presented here may provide a template which is generally useful for quantifying the value added by ARM data for many different alternative uses. The use of ARM measurements to force climate models may be particularly productive. We intend to continue to foster and facilitate alternative uses for ARM data, both within the carbon modeling community and elsewhere.
Hanson, C.L., K.A. Cumming, D.A. Woolhiser, and C.W. Richardson. 1994. Microcomputer program for daily weather generation. U.S. Dept. Agric., Agric. Res. Svc. Pub. No. ARS-114, 38 pgs.
Johnson, G.L., C.L. Hanson, S.P. Hardegree, and E.B. Ballard. 1996. Stochastic weather simulation: Overview and analysis of two commonly used models. J. Appl. Meteor 35: 1878-1896.
Richardson, C.W. 1981. Stochastic simulation of daily precipitation, temperature, and solar radiation. Water Resources Research 17: 182-190.
Richardson, C.W., and D.A. Wright. 1984. WGEN: A model
for generating daily weather variables. U.S. Dept. Agric.,
Agric. Res. Svc. Pub. No. ARS-8, 83 pgs.
Running, S.W., Nemani, R.R., and R.D. Hungerford. 1987. Extrapolation of synoptic meteorological data in mountainous terrain and its use for simulating forest evapotranspiration and photosynthesis. Can. J. For. Res. 17:472-483.
Sellers, P.J., D.A. Randall, G.J. Collatz, J.A. Berry, C.B. Field,
D.A. Dazlich, C. Zhang, G.D. Collelo, and L. Bounoua. 1996. A revised
land surface parameterization (SiB2) for atmospheric GCMs. Part I:
Model formulation. Journal of Climate 9:676-705.
White, M.A., P.E. Thornton, S.W. Running, and R.R. Nemani. 2000.
Parameterization and sensitivity of the Biome-BGC terrestrial ecosystem
model: Net primary production controls. Earth Interactions 4:1-85.
For additional information contact:
*Oak Ridge National Laboratory, managed by UT-Battelle, LLC, for the U.S. Dept. of Energy under contract DE-AC05-00OR22725.
"The submitted manuscript has been authored by a contractor of the U.S. Government under contract DE-AC05-00OR22725. Accordingly, the U.S. Government retains a nonexclusive, royalty-free license to publish or reproduce the published form of this contribution, or allow others to do so, for U.S. Government purposes."
