230 K.R. Tate et al. Agriculture, Ecosystems and Environment 82 2000 229–246
system C balance may be too coarse, and studies of land-use changes at finer spatial scales are needed to reduce uncertainties in national-scale C balance estimates. © 2000 Elsevier Science B.V. All rights reserved.
Keywords: Net primary production; Soil respiration; Scaling; Remote sensing; Models
1. Introduction
A ‘missing sink’ of carbon C of about 1.8 Pg per year Schimel, 1995 has arisen from attempts to close
the global C budget by balancing atmospheric car- bon dioxide CO
2
concentrations against fossil fuel emissions, land-use change, and ocean C uptake Tans
et al., 1990. Although strong evidence now exists for the terrestrial biosphere being the most likely candi-
date Rayner et al., 1999, much uncertainty still sur- rounds the temporal variation and spatial distribution
of this sink. Spatial and temporal patterns of C in- put to, and release by, soils are especially uncertain
Torn et al., 1997. Possible explanations for this sink Lloyd, 1999 are forest regrowth, CO
2
fertilisation, and N deposition effects on terrestrial ecosystems.
Interest in national C budgets has increased recently for several reasons. First, new national greenhouse
gas reductions’ targets for Annexe One countries have been set under the Kyoto Protocol. If these targets are
ratified, credible national strategies for reporting and reducing CO
2
and other greenhouse gas emissions will need to be developed, using internationally recog-
nised methodologies subject to periodic auditing. Some countries have signalled their intention to use
terrestrial sinks to reduce their CO
2
emissions IGBP Terrestrial Carbon Working Group, 1998. Presently,
New Zealand relies heavily on C uptake and storage in exotic plantation forests mainly Pinus radiata
to offset annual CO
2
emissions of 7.5 Mt CO
2
-C from energy and industrial sources MfE, 1997. This
strategy will increase the need for more integrated, multi-disciplinary approaches involving C monitor-
ing at various scales, experimentation and modelling Cannadel and Mooney, 1999. Second, projections
of the likely effects of these emissions on climate are much less certain at regional and national scales,
where assessments of societal impacts of changing climate are more pertinent than at the global scale.
Third, strategies for achieving multiple environmental aims are likely to be more effective at the national
scale, including reducing greenhouse gas emissions and biodiversity loss, and protecting forests and pro-
ductive soils. Finally, current efforts by signatories to the Kyoto Protocol to make C emissions-reduction
targets binding may be a prelude to more compre- hensive full C-accounting IGBP Terrestrial Carbon
Working Group, 1998. This could be best achieved at the national scale, where databases, monitoring
networks and historical knowledge are likely to be well coordinated Tian et al., 1999.
Accordingly, we are using field and laboratory measurements, databases, satellite remote sensing and
models to assess New Zealand’s national C balance at various spatial scales. We tested the hypothesis
that New Zealand’s terrestrial ecosystems are in C balance, and that the substantial uptake of CO
2
by planted forests and aggrading scrublands is roughly
balanced by potential C losses from indigenous forests and soils, by comparing NPP and soil heterotrophic
respiration. The approach was first to establish a na- tional baseline of soil and vegetation C, against which
to estimate changes Tate et al., 1997. Second, as annual NPP is balanced by soil CO
2
emissions in ecosystems at steady state, each of these C fluxes was
estimated nationally for the major vegetation types: indigenous and exotic forests; scrub; and grasslands
improved, unimproved, and tussock. Collectively, these ecosystems cover about 90 of New Zealand’s
land area Tate et al., 1997. Third, national-scale C balance estimates were compared with plot-based es-
timates of the C balance for indigenous forests, exotic forests, and scrub. Fourth, the potential impact of soil
erosion on national C balance estimates was exam- ined at different scales. Finally, information gaps, and
major sources of uncertainty are discussed.
2. Methods
Calculation of a national C budget NZCB was based on the following mass-balance equation:
NZCB= X
CBAL
LC
− CO
2
-C
fossil fuels
− C
erosion
1
K.R. Tate et al. Agriculture, Ecosystems and Environment 82 2000 229–246 231
where CBAL
LC
is the annual difference between total net primary production above- and below-ground
and soil heterotrophic respiration for the major land-cover types see Section 2.1.1. Carbon dioxide
emissions based on fossil fuel and energy consump- tion are represented by CO
2
fossil fuels
, and C losses from soil erosion by C
erosion
. The methods used to estimate each of these terms are described below.
2.1. National-scale estimates of net primary productivity
2.1.1. Model and data sources Unless otherwise stated, net primary productivity
NPP includes both above- and below-ground C al- location. A subset of the Carnegie–Ames–Stanford
Approach CASA model was implemented Potter et al., 1993 to estimate New Zealand’s annual NPP.
The CASA model uses a parametric approach to de- rive NPP based on intercepted photosynthetically ac-
tive radiation IPAR and a light-use efficiency factor ǫ that converts IPAR to NPP. NPP was calculated
monthly for 1993 over 1 × 1 km
2
grid cells using: a the fraction of photosynthetically active radiation
FPAR derived from monthly normalised differ- ence vegetation index NDVI composites obtained
from the advanced very high resolution radiometer AVHRR instrument on board the NOAA series of
satellites, b monthly solar radiation and c gener- alised land-cover types based on the vegetative cover
map of New Zealand VCM Newsome, 1987, and ǫ
for each of the land-cover types. NDVI data based on the red and near-infrared
reflectance Tucker et al., 1985 from the AVHRR sensor were obtained at 1 × 1 km
2
resolution. How- ever, although the field of view at nadir is 1 km
2
, the off-nadir field of view ranges from 2.4 km along-track
to 6.9 km across-track. Daily AVHRR imagery was, therefore, geo-referenced to the New Zealand Map
Grid and re-sampled to a resolution of 1 km
2
and com- posited into monthly maximum value images Hol-
ben, 1986. A correction for solar zenith angle was also applied Sellars et al., 1994. Land-cover types
were created by generalising 47 vegetation classes Newsome, 1987 into seven cropland, grassland,
grasslandscrub, scrub, grasslandforest, forestscrub, and forest. An ǫ value was assigned to each of these
classes to predict NPP. Then, NPP estimates for these seven classes were transformed into NPP estimates
for six uniform land-cover types indigenous forest, exotic forest, scrub, improved grasslands, unimproved
grasslands, and tussock grasslands by estimating the proportion of the area for each uniform vegetation type
within a mixed class, and multiplying this proportion by the predicted NPP value for that mixed vegetation
class. In this transformation, cropland was included with improved pasture, the grass component of the
grasslandscrub went into unimproved grassland, and the grass component of the grasslandforest was in-
cluded with improved pasture. Two sets of ǫ-values were used to estimate New Zealand’s NPP; one was
based on Potter et al. 1993, and the other used val- ues derived for New Zealand ecosystems. Monthly
climate variables, including daily average short-wave radiation to calculate IPAR from FPAR, were de-
rived from individual meteorological data spatially interpolated onto a 1 × 1 km
2
resolution grid using to- pographic variables Leathwick and Stephens, 1998.
2.1.2. Plot-scale NPP estimates used for model verification
NPP was also predicted for precise site locations Table 1, for which independent estimates were
available, by calculating a mean NPP from the sur- rounding eight pixels. NPP was determined for sites
1 and 5 by intensive gas exchange measurements Benecke and Evans, 1987; Arneth et al., 1998; for
site 1, NPP agreed well with a simulated estimate for another lowland old-growth beech forest site 2 Tate
et al., 1993 Table 1. In situ pulse labelling
14
C was used to estimate NPP for two improved grassland
sites 8, 9 Saggar et al., 1999. The Rothamsted soil C turnover model ROTHC Jenkinson, 1990 was
used to estimate NPP Jenkinson et al., 1992; Tate et al., 1993 at three forest 2–4 and two grassland
7, 10 sites Table 1, by assuming steady-state con- ditions i.e. detrital C input = soil CO
2
-C production. NPP for scrub site 6 was estimated from canopy
information and a canopy-scale productivity model D. Whitehead, pers. comm..
2.2. National assessment of soil respiration 2.2.1. Data sources and national estimates
A national estimate of soil respiration R was made under non-limiting NL soil moisture conditions
232 K.R. Tate et al. Agriculture, Ecosystems and Environment 82 2000 229–246
K.R. Tate et al. Agriculture, Ecosystems and Environment 82 2000 229–246 233
using an Arrhenius-type relationship e.g., Lloyd and Taylor, 1994:
R = AB
2 where B = exp{ − 308.56T − 227.13}, A is a data
set-dependent parameter and T the temperature K. The following steps were followed:
1. Monthly surfaces for the ratio RR
10
were calcu- lated for each 1 × 1 km
2
grid cell of a national tem- perature surface Leathwick, 1998, where R is the
soil respiration rate at actual air temperature and R
10
the soil respiration at 10
◦
C New Zealand’s mean annual temperature. Monthly RR
10
values were averaged to provide a surface of mean annual
relative respiration rates for each grid cell under NL moisture conditions.
2. To quantify the impact of moisture limitation L, RR
10
was calculated by assuming that the L respiration rate would equal the NL rate when
soil moisture was at field capacity −0.01 MPa, be at zero when soil moisture was at wilting
point −1.5 MPa and at a fraction of the NL rate based on actual and potential evapotranspi-
ration for intermediate soil moisture levels. The mean rate of monthly moisture-limited respira-
tion would then equal the NL rate multiplied by the monthly actual-to-potential evapotranspiration
ratio. A monthly water balance was calculated using mean monthly potential evapotranspiration
Leathwick, 1998 and soil moisture storage ca- pacity between −0.01 and −1.5 MPa McDonald
et al., 1988, averaged by major soil classes IPCC, 1996. The actual monthly evapotranspiration was
calculated along with the soil moisture deficit potential − actual evapotranspiration and used to
calculate the actual moisture-limited respiration rate. Monthly average respiration rates RR
10
were again averaged to give a mean annual respi- ration surface for L and NL conditions.
3. Mean annual RR
10
surfaces for NL and L con- ditions were overlaid with the VCM Newsome,
1987 to provide area-weighted mean RR
10
val- ues for each land-cover type. In mixed vegetation
classes, the areas of each component e.g., scrub in the scrub-grassland class were estimated based
on their relative abundance within the mixed cate- gory. The mean NL and L annual RR
10
values for a mixed class were assumed to apply to each
component within the class. Then, mean annual R
R
10
values for the six land-cover types were generalised from the 47 vegetation classes as for
the NPP simulations. Total area and area-weighted mean annual RR
10
values for each land-cover type were then calculated.
4. The mean annual RR
10
values were converted to land-cover specific respiration by estimating
land-cover specific R
10
values. To estimate R
10
, A
values the slope of the regression line were first estimated for 11 ‘sites’ Table 3 using lin-
ear regression with zero constant intercept, where site-specific respiration was the independent vari-
able and B Eq. 2, using actual temperature data was the dependent variable. Soil CO
2
production and temperature were measured using a closed
chamber system connected to a portable infrared gas analyzer IRGA SRC-1 and EGM, PP Sys-
tems, Herts., UK Jensen et al., 1996. Data that were clearly subject to moisture L’s were ex-
cluded. Analysis of variance of soil respiration vs B
for the 11 sites was used to indicate within and between land-cover effects on estimated A values
and calculated R
10
values t CO
2
-C ha
− 1
per year. 5. Finally, national respiration rates t CO
2
-C ha
− 1
per year and total respiration Mt CO
2
-C per year were calculated for each land-cover type by multi-
plying the mean annual area-weighted RR
10
values by the R
10
estimates for each land-cover type. Potential ‘temporal scaling’ errors in the RR
10
val- ues were identified by comparing RR
10
values based on Eq. 11; Lloyd and Taylor, 1994 calculated using
mean monthly, seasonal, or annual temperatures. 2.2.2. Estimates of heterotrophic respiration R
h
To determine R
h
, the few root C allocation esti- mates available for New Zealand ecosystems Table 1
were used to quantify autotrophic respiration from roots R
a
. Root C allocation ranged from 24 of to- tal NPP for a P. radiata forest Arneth et al., 1998 to
80 for alpine tussock grassland Tate et al., 2000, and generally agreed with published values for forests
and grasslands Ruimy and Saugier, 1994. For the six land-cover types, the following values were used to
estimate R
a
by assuming, on a C basis, that R
a
equates with the proportion of NPP allocation below ground
Table 1: indigenous forests 0.40, exotic forests 0.24, scrub 0.40, improved grassland 0.55, unim-
234 K.R. Tate et al. Agriculture, Ecosystems and Environment 82 2000 229–246
proved grassland 0.70, tussock grassland 0.80. This assumption was based on data for fine root pro-
duction and respiration in temperate forests Ryan et al., 1996; Keith et al., 1997, where R
a
was esti- mated to be roughly equivalent to below-ground NPP.
In improved established grasslands, root biomass shows little temporal variation Saggar et al., 1997,
suggesting root biomass is near steady state; under these conditions, a 1 : 1 distribution of total respi-
ration between R
a
and R
h
also appears reasonable, based on the estimates of R
a
total soil respiration for grassland soils summarised in Hanson et al. 2000.
2.3. National soil erosion National estimates of C losses from soil erosion
were based on regional sediment-yield data Glasby, 1991, initially assuming that landsliding 0.85 m
depth was the key erosion process. Average soil C concentration was estimated at 1.89 average soil
C concentration to 0.85 m depth for all pedons in the National Soils Database; McDonald et al., 1988.
Potential scaling errors associated with estimating national soil C loss by erosion were then investigated
by quantifying the contribution of sediment and C from different erosion processes shallow landslides
soil slips, gully, and sheet erosion in the Waipaoa River basin 2200 km
2
and the Lake Tutira catch- ment 32 km
2
located in the east coast region of the North Island Page et al., 1994a,b; Trustrum et al.,
1998, 1999. In the Waipaoa River basin, estimates of sediment production for the period 1920–2000
i.e. 80 years following land clearance by different erosion processes were derived from detailed analysis
of individual catchment sediment budgets Trustrum et al., 1998, 1999. Soil C removed by each process
was estimated using average soil C content for a range of disturbed and undisturbed, but deforested,
soil profiles M. McLeod, pers. comm.. For shallow landsliding and gully erosion assumed to be confined
to the upper 0.85 m of the soil profile, an average soil C content of 1.89 was assumed see above.
For sheet erosion assumed to erode the upper 0.1 m of soil, a soil C value of 5.25 M. McLeod, pers.
comm. was used to estimate C losses. The amount of C sequestered on the floodplain in the lower reaches
of the basin in the same 80-year period was estimated on the basis of the volume of the alluvial sediments
and their average C content 1.3 as determined by coring. Likewise, C stored on the continental shelf
was estimated using an average C content of 1.0.
3. Results and discussion
