| dc.description.abstract | Feedbacks between the climate system and the high-latitude carbon
cycle will substantially influence the intensity
of future climate change. It is therefore crucial that the net ecosystem
exchange of CO2 (NEE) between the high-latitude land surface and the
atmosphere is accurately quantified, where NEE refers to the difference
between ecosystem respiration (R) and photosynthesis (gross ecosystem
exchange, GEE): NEE=-GEE+R in umol/m^2/s. NEE can only be directly
measured over areas of 1 km^2 through eddy covariance, and modeling
approaches such as the Vegetation Photosynthesis Respiration Model (VPRM) are
required to upscale NEE. VPRM
is a remote
sensing based model that calculates R as a linear function of air
temperature (Ta) when air
temperature is above a given threshold (Tlow), and sets respiration to a
constant
value when Ta<Tlow. GEE is estimated according to remote sensing
observations of vegetation indices, shortwave radiation, air temperature, and
soil moisture. Although in situ findings have shown
that snow and Arctic species composition have a
substantial
influence on high-latitude NEE, model estimates of high-latitude NEE have
typically been generated without Arctic-specific vegetation classes, and
without using remote sensing observations to represent
the effects of snow on NEE. The hypothesis driving this
work was therefore that uncertainty in estimates of high-latitude NEE could
be reduced by representing the influences of Arctic
vegetation classes and snow. The central objectives were
to determine feasible approaches for reducing uncertainty in VPRM estimates
of NEE by representing the influences of snow and Arctic vegetation,
create PolarVPRM accordingly, and analyze inter-annual variability in PolarVPRM
estimates of high-latitude North American NEE (2001-2012).
The associations between snow and NEE, and the potential to describe
these influences on NEE using remote sensing observations, were
examined using time lapse camera observations of snow cover area (SCA) and eddy
covariance measurements of NEE from Daring Lake, Northwest Territories,
Canada. Analyses indicated
good agreement between SCA derived from camera, Landsat and Moderate Resolution
Imaging Spectroradiometer (MODIS) observations. SCA was also found to influence
the timing and magnitude of NEE. MODIS SCA was therefore incorporated into VPRM,
and VPRM was calibrated using eddy covariance and meteorological observations
collected in
2005 at Daring Lake. VPRM was run through years
2004-2007 over both Daring Lake and Ivotuk, Alaska, USA, using four model
formulations, three of which represented the effects of SCA on respiration
and/or photosynthesis, and another which did not use MODIS SCA. Comparisons
against eddy covariance observations indicated that uncertainty was reduced in
VPRM estimates of NEE when respiration was calculated as a linear function of
soil temperature when
SCA>50%, and as a linear function of air temperature when SCA<50%,
thereby reflecting the influence of snow on decoupling soil/air temperatures.
Representing the effect of SCA on NEE therefore reduced uncertainty in VPRM
estimates of NEE.
In order to represent spatial variability in high-latitude
estimates of NEE due to vegetation type, Arctic-specific vegetation classes were
created for PolarVPRM by combining
and aggregating two existing vegetation classifications: the Synergetic Land
Cover Product and the Circumpolar Arctic Vegetation Map. Levene's test
indicated that the PolarVPRM vegetation classes divided the pan-Arctic
region into
heterogeneous distributions
in terms of net primary productivity, and passive microwave derived
estimates of snow and growing season influences on NEE. A
non-parametric statistical approach of Alternating Conditional Expectations
found significant, non-linear associations to exist between passive microwave
derived estimates of snow and growing season drivers of NEE. Furthermore,
the shape of these associations varied according to the vegetation class over
which they were examined. Further support was therefore provided to the idea
that uncertainty in model estimates of NEE could be reduced by calculating snow
and growing season NEE separately within each vegetation class.
PolarVPRM estimates of NEE in 2001-2012 were
generated at
a three hourly and 1/6 x 1/4 degree resolution across
polar North
America (55-170 W, 55-83 N). Model
calibration was conducted over three sites: Daring Lake, Ivotuk, and Atqasuk,
Alaska, USA. Model validation was then conducted by comparing PolarVPRM
estimates of year-round daily average NEE
to non-gap-filled eddy covariance observations of daily average NEE acquired
over the three calibration sites, as well as six other Arctic sites.
PolarVPRM performed well over all sites, with an average mean absolute
error (MAE) of 0.20 umol/m^2/s, and had
diminished
error rates when the influence of SCA on
respiration was explicitly represented. Error
analysis indicated that peak growing season GEE was underestimated at Barrow
because GEE at this site showed a stronger response to the amount
of incoming shortwave radiation than at the calibration site, suggesting
that PolarVPRM may underestimate GEE over wetland and barren vegetated
regions. Despite these uncertainties, PolarVPRM was found to generate more
accurate estimates of monthly and three-hourly NEE relative to eddy covariance
observations than two established models, FLUXNET Model-Tree Ensemble (MTE) and CarbonTracker.
Relative to eddy covariance observations and PolarVPRM estimates, MTE
tended to overestimate snow season respiration, and CarbonTracker tended to
overestimate the amount of midday photosynthesis. Analysis of PolarVPRM output
across North America (north of 55 N) found an increase in net annual carbon
efflux over over time (2001-2012). Specifically, increased rates of respiration
are estimated when soil and air temperatures are warmer. Although
increases in growing season vegetation indices and air temperature enable
greater
photosynthetic uptake by Arctic vegetation, forests and shrublands
uptake less CO2 in the middle of the growing season when air temperatures rise
above the physiological optima for photosynthesis. As a result, PolarVPRM
estimated a decline in net photosynthetic uptake over time. Overall, PolarVPRM
output indicates that North American regions north of 55 N are
losing strength as a carbon sink in response to rising air temperatures. | en |
