Quantifying and Reducing Uncertainty in Eddy Covariance CO2 Exchange Estimates With Large Data Gaps.
Roving eddy covariance (rEC) applications involve cycling a single, portable rEC system through numerous ecosystems of interest at frequencies of weeks to months. This approach provides capability to measure energy, water and carbon exchanges in many more ecosystems with minimum resources. This approach has been adopted at many flux stations with multiple towers or sites. However, data gaps introduce large uncertainty when sums of annual exchanges are estimated from these measurements. Quantifying reliable carbon uptake/loss estimates, and reducing uncertainty is challenging. Analyses were performed with 5 years of net ecosystem CO2 exchange (NEE) measurements made over an age-sequence of managed eastern white pine (Pinus strobus L.) forests in southern Ontario, Canada. Measurements and synthetic data produced from a permanent closed-path EC system at the 70-year-old site were used to assess uncertainty on annual NEE sums caused by open-path rEC operation at three younger sites (35, 20 and 7 years old). Data was removed both randomly (to simulate short gaps due to typical EC operation), and systematically (to simulate 2-week to 2-month long gaps associated with rEC operation). Numerous gap scenarios were created, with annual data coverage ranging from 35 to 70%. Both Ameriflux Howland (HOW) and Fluxnet Canada Research Network (FCRN) gap-filling models were applied to the gapped-data to produce annual NEE estimates. Both gap-filling methods estimated a baseline uncertainty of ± 25 g C m-2 yr-1, that was associated with short gaps from regular EC operation only. However, when applied to datasets with long rEC gaps, a large discrepancy in gap-filling model performance was evident. The HOW method produced much less overall uncertainty (± 30 to 45 g C m-2 yr-1) in annual sums than the FCRN method (± 45 to > 200 g C m-2 yr-1). Uncertainty in annual sums increased with increases to overall gap frequency and length, as a result of inadequate parameterization of gap-filling models. Uncertainty also increased for scenarios where large gaps appeared in the early growing season. Two treatments were applied in an attempt to improve parameterization of gap-filling models of rEC data: i) pooling of all years of rEC data from a single site, and ii) pooling regression-adjusted measurements from all sites for all years. Both pooled treatments improved performance of the FCRN model considerably, while offering little to no improvement to the HOW method. In particular, pooling data from all sites and years greatly improved FCRN parameterization, reducing uncertainty in annual NEE sums to ± 35 to 45 g C m-2 yr-1, which was comparable to un-pooled HOW model performance, and only marginally larger than baseline uncertainty. These results suggest that roving system methodology can be applicable across other forest age- sequences to produce reliable annual sums of NEE, and that data from other sites and subsequent years can be used to parameterize gap-filling models, thus substantially reducing associated uncertainties in annual C uptake or loss estimates.
An Analysis of the Spatial and Temporal Variation of Carbon and Water Fluxes in Three Temperate Pine Forests Across North America
This study investigates spatial and temporal variations of carbon (C) and water fluxes at three mature temperate pine sites within North America using continuous closed-path eddy covariance measurements from 2003-2005. The three sites had very contrasting climate regimes ranging from very wet (Campbell River site, B.C., Canada) to intermediate (Turkey Point site, Southern Ontario, Canada) to very dry (Metolius site, Oregon, USA). Precipitation (PPT) normals for these three regions are 1451, 1010 and 550 mm respectively. Both Turkey Point and Metolius sites experience summer drought stresses. Net Ecosystem Production (NEP) was greatest at Campbell River and Metolius sites resulting in strong annual C sinks for all three years, with NEP values ranging from 242-383 and 331-363 g C m-2 y-1, respectively. On the other hand Turkey Point was a C sink in 2003 and 2004 (189 g C m-2 y-1 and 131 C m-2 y-1), but was almost C neutral (36 g C m-2 y-1) in 2005 due to an early spring drought and warming event. Metolius and Campbell River experienced small carbon uptake throughout winter months, while at Turkey Point there was no photosynthesis activity due to cold temperatures over winter months (December-March). Campbell River had the highest Water Use Efficiency (WUE) of 5 g C per 1 kg of water (R2 = 0.82), while Turkey Point and Metolius had WUE of 3 and 2.4 g C per 1 kg of water (R2 = 0.80 and R2 = 0.77). The differences in C uptake and WUE among these three sites were mainly due to contrasting environmental conditions/controls (precipitation, soil moisture, and temperature) and site characteristics (leaf area index, rooting depth, soil composition). This study shows that in order to improve large scale or regional estimates of C and water cycles, a stronger understanding of how different sites respond to their environmental controls and site characteristics is imperative.
Evaluating Historic Carbon Budget in Temperate Pacific Northwestern Conifer Forest Landscape Using CN-CLASS Model
We used carbon and nitrogen coupled Canadian Land Surface Scheme (CN-CLASS), a process-based model, to simulate the historic carbon stocks and fluxes in a 2500 ha temperate Pacific Northwest conifer forest landscape from 1920 to 2004. Hourly meteorological data was provided from historic climate records. Site maps of soils, topography, vegetation and disturbance type (logging, fire events) were provided by the historic carbon modeling project of the Canadian Carbon Program (CCP). The initial aboveground tree biomass in 1920 and the disturbance matrices were produced by the Carbon Budget Model of the Canadian Forest Sector (CBM-CFS3). Over the study period from 1920-2004, CN-CLASS simulated 188 Mg C ha-1 loss of ecosystem carbon as compared to 200 Mg C ha-1 loss suggested by CBM-CFS3. From 1928 to 1944, burning, historic harvest and slash burning resulted in large losses of carbon to the atmosphere. In this period, the study area was a large carbon source with simulated net carbon loss of 282 Mg C ha-1. From 1945 to 1989, there were very few disturbance events and the study area started to recover with simulated net carbon uptake of 87 Mg C ha-1. During this period, CN-CLASS modeled annual Net Biome Productivity (NBP) ranged from 27 g C m-2 y-1 in 1945 to 590 g C m-2 y-1 in 1964. CN-CLASS simulated summer (July-September) NEP deviations during the undisturbed period (1945-1989), showed a negative relationship with the daily maximum air temperature and precipitation. As harvest of second-growth stands began in 1990s, disturbance again had significant impact on the landscape's carbon budget, and this effect was partially offset by ongoing C uptake in recovering young stands. CN-CLASS underestimated NEP as compared to observed eddy covariance flux measurements because of high temperature sensitivity of its soil respiration algorithm. Simulated mean annual NEP from 1998 to 2004 was 190 g C m-2 y-1 as compared to observed eddy covariance value of 352 g C m-2 y-1. This study will help explore the impact of climate variability and disturbance on landscape-level carbon dynamics.
Evapotranspiration and Water Use Efficiency of Canadian Forests and Wetlands
Plant growth and potential photosynthetic carbon uptake are strongly controlled by water availability and climatic conditions. The study of evapotranspiration (E) among a variety of terrestrial ecosystems is therefore fundamental to understand its role within the water cycle and linkage to carbon fluxes. In this study, we analyze tower-based eddy-covariance data of nine mature forests, two peatland and a grassland site across an east- west continental-scale transect in Canada. The seasonal pattern in E was closely linked to growing season length and rainfall distribution. Although annual precipitation (P) was highly variable (400-1450 mm) among sites, annual E was limited to 400-500 mm for annual P > 700 mm. Site-specific interannual variability in E could be explained by either changes in total P or changes in the amount of downwelling solar radiation due to a varying length of sunny or cloudy periods. A highly positive linear correlation was found between monthly mean values of E and net radiation at a coastal Douglas-fir site (BC-DF49), whereas a hysteretic relationship at all other investigated sites indicated that E lagged behind the typical seasonal progression of meteorological parameters. Furthermore, daytime average dry-foliage Priestley-Taylor alpha was relatively constant (0.55) at BC-DF49 throughout each of the years. However, significant seasonal variations with maxima occurring in the growing season were found among all other sites. Annual means of alpha mostly ranging between 0.5-0.7 implied stomatal limitation to transpiration. Growing season means of alpha were only slightly higher than annual means except for a deciduous (aspen) boreal site (1.23 at SK OA), indicating no or only marginal water supply limitation. At all sites, a high linear correlation between monthly mean values of gross primary production and E resulted in water use efficiency (WUE) being relatively constant. While at most sites, WUE was in a range of 2.6 to 3.6 g C kg-1 H2O, highest WUE of the 12 sites with values around 6.0 g C kg-1 H2O was found at BC-DF49. Significant differences were observed between an eastern ombrotrophic bog (ON-EPL, WUE 1.8 g C kg-1 H2O) and a western treed fen (AB-WPL, WUE 3.0 g C kg-1 H2O), which could be a result of site-specific water cycles and nutrient availability. Results of this study provide important insight into the coupling of water and carbon cycles at the ecosystem level and are very valuable for the validation of regional process-based models used to investigate the influence of climate change on plant functional types.
Spatial distributions of forest aboveground biomass and landscape dynamics associated with conservation status and ownership in New England, USA
We combined remote sensing derived forest aboveground biomass (AGB) estimation and the Conservation Biology Institute/World Wildlife Fund Protected Area Database using GIS techniques and spatial pattern analysis to illustrate how different conservation status and ownership could affect the landscape dynamics and spatial distributions of AGB in New England states, USA. The AGB means between all pairs of protection status and ownership categories were significantly different (P < 0.05). The highest mean AGB was observed in the protected public lands (156 Mg/ha), 44% higher than the lowest AGB mean (108 Mg/ha) observed in private regulated lands (privately owned but under the regulatory control by a state agency), or 30% higher than that in privately owned lands on average (120 Mg/ha). Seventy-seven percent of the regional forests with AGB > 200 Mg/ha, totaling about 9,300 km2, were located outside protected areas and were concentrated in western MA, southern VT, southwestern NH, and northwestern CT. The fragmentation rate in protected public lands between 1992 and 2001 was the least with greater rates were observed in privately regulated and non-regulated lands. Changing rates for the 4 representative fragmentation indices (patch density (PD), edge density (ED), landscape shape index (LSI), and mean patch size (MPS)) ranged from 1% in MPS to 6% in PD in protected public lands during the 9-year period whereas the mean changing rates ranged from 21% in LSI to 32% in PD in private lands. Thus, ownership and conservation status appears to have a strong impact on the dynamic changes of landscape structures in the region. These results indicate that if maintenance and enhancement of relatively unfragmented, high-AGB forest is a goal, expansion of protected areas appears to be an important management strategy.
Spatial Partitioning of CO2 Fluxes Based on Canopy Structure Within a Heterogeneous Managed Boreal Wetland Ecosystem
Vegetation canopy structural characteristics play an important role in the transfer of mass and energy exchanges through time. The spatial variability of biomass surrounding the eddy covariance flux measurement system (EC) will result in differences in a) the amount of surface area available for flux exchanges, b) aerodynamic roughness of the ecosystem, and c) the source area (biophysical influences) on fluxes, depending on wind direction. The following study classifies CO2 fluxes based on wind direction and land cover/vegetation type using a combination of EC flux measurements, footprint model parameterization, and airborne lidar within a heterogeneous boreal wetland ecosystem. CO2 and H2O fluxes have been examined within the Utikuma Regional Study Area, Alberta, using EC methods since 2005. This site is unique because, in most cases, EC are deployed in flat and homogeneous land cover types with large fetch. The wetland/upland complex examined here is heterogeneous and is characterised by low-lying wetlands to the south and south-west of the EC and upland aspen forests to the north and north-east. Further, airborne lidar provides spatially explicit, high resolution three-dimensional measurements of the vegetation canopy, understory, and ground surface that are both time consuming and expensive to measure using typical forest mensuration/survey methods. The influences of vegetation structure, specifically surface area of leaves (leaf area index), aerodynamic properties of vegetation surrounding the EC, and land cover types on fluxes are examined. Spatial partitioning of fluxes based on land cover type and wind direction is used to examine both wetland and upland exchange processes.
Stochastic Soil Moisture Balance Modelling for Predicting Tree Mortality: A Case Study in Oilsands Reclamation.
Reclamation of oilsand sites north of Fort McMurray, AB, involves restoring the ecosystem to a functional boreal forest, where operators remain liable for environmental impacts until such time that the area of disturbance is reclaimed. Numerous studies have examined the variability of fluxes of energy and water vapour during the growing season that results from temporal variations in radiant input, soil and biophysical properties, and moisture conditions. However, it is uncertain to what degree reclamation strategies have impacted carbon, water, energy, and nutrient exchanges. Specifically, as a result of the constructed nature of the soil cover, uncertainty exists regarding the long-term viability of reclaimed forests due to the interaction of available soil moisture and increasing evaporative demand as the trees mature. A site specific probabilistic assessment of soil moisture and plant stress is derived for a 16 year old jack pine stand planted on reclamation cover in northern Alberta (RJP92). Results are obtained using a process-based quantitative model which accounts for the stochastic nature of rainfall along with measured soil and biophysical properties. The model is parameterized and tested against field data for the 2007 and 2008 growing season respectively. Functional relations between soil moisture tension, bulk surface resistance, and NEE along with published values serve to link the probability distribution of soil moisture to dynamic plant water stress. Based on the soil and biophysical parameterization for RJP92, the long term viability of the jack pine stand is assessed. This is accomplished by simulating the inter-annual variability of the rainfall parameters which are modelled as independent gamma-distributed random variables, where climate is considered intransient. Results indicate that, given the range of climate variability over the past century, the probability of stand failure is minimal assuming there are no changes in evaporative controls as the stand matures. However, results also indicate that RJP92 is water stressed for a significant portion of the growing season, where the available water holding capacity in the till portion of the soil cover represents a major control on stand productivity. Although not investigated explicitly, this analytical framework lends itself to assessing the impact of climate change on the jack pine stand by varying the rainfall parameter distributions as well as the evaporative component of the model.