1. Landscape variation of canopy nitrogen and carbon assimilation in a temperate mixed forest
Zaixing Zhou, Scott V. Ollinger, Lucie Lepine
Earth Systems Research Center, Institute for the Study of Earth, Oceans, and Space
University of New Hampshire, Durham, NH 03824, USA
Introduction a b
Figure 5. Spatial patterns in AVIRIS remotely sensed canopy %N (a) and PnET modeled GPP (b).
Remotely sensed canopy %N ranged from 0.5 to 2.9 with a mean of 1.75 across the study region. The coefficient of variation is 14.3%. While patterns of GPP predicted by PnET were similar to those for canopy %N, GPP was less variable over the study area than canopy %N, with values ranging from 797 to 1622 g C m-2 yr-1, a mean of 1324 g C m-2 yr-1, and a coefficient of variation of 6%. No significant relationships were observed between canopy %N or GPP and slope, aspect, and climate variables including temperature, precipitation, and PAR. After we had detrended GPP with canopy %N, the residual of GPP was significantly positively correlated to growing season PAR (r2=0.69, p<0.001; data not shown).
The availability of nitrogen (N) represents a key constraint on carbon cycling in terrestrial ecosystems and serves as a useful indicator of ecosystem metabolism.
The importance of N as a regulator of C assimilation is well established through the widely observed relationship between leaf-level photosynthetic capacity (Amax) and foliar N concentrations(Wright et al., 2004). At the canopy and stand level, canopy N content has been related to canopy-level photosynthetic capacity (Ollinger et al., 2008), plant respiration (Reich et al., 2008), net primary production (NPP) (Smith et al., 2002), and canopy light use efficiency (Kergoat et al., 2008). Canopy %N status is also linked to the availability of N in soils through mechanisms involving litter decay, net mineralization, plant N uptake, and N loss. Over the past several centuries, most northeastern U.S. forests have experienced human-induced disturbances such as forest harvests, fire, or agriculture. These disturbances have been reported to have long-term impacts on forest carbon and nitrogen cycling (Hooker & Compton, 2003).
In this study, we conducted an analysis to examine the landscape variation of canopy nitrogen and carbon assimilation in a temperate mixed forest at the Harvard Forest (HF), in Massachusetts, U.S.A. We integrated plot–level field observations, eddy covariance (EC) data, remote sensing, and ecosystem modeling. Specifically, we aimed to examine whether canopy %N at the landscape scale reflects impacts of local disturbance history and is related to canopy carbon assimilation.
Figure 6. Relationship between remotely sensed canopy %N and PnET modeled GPP.
At the canopy scale, predicted GPP was positively linearly correlated to canopy %N (r2 = 0.97, p<0.001). Deciduous species with higher canopy N and light use efficiency occur at the higher value end of GPP and in contrast evergreens occur at the lower end. It is noted that within the range of canopy %N (i.e., , where both deciduous and evergreen species exist) predicted GPP per evergreen stand area could still be similar to that per deciduous stand area because evergreen stands had larger amount of foliage mass present.
Methods
The area studied was a 10 km × 16 km landscape surrounding the Harvard Forest (HF) in central Massachusetts, USA, centered near N, W (Fig. 1).
Ecosystem model PnET-II was used to estimate the GPP. The area was delineated into grid cells in a 30-m resolution. For each cell, PnET-II was run for either of deciduous stands or evergreen stands, or both for the mixed.
A canopy N concentration map over HF was developed from airborne imaging spectrometer (AVIRIS). A regression model was established using measured canopy %N data and then applied to the entire image data set to derive spatially explicit estimates of canopy %N. The predicted canopy nitrogen was in line with the observed (Fig 2-a). The relative proportions of deciduous and evergreen forests (Fig 2-b) and the foliar N concentration (Figs 2-c,d) of each component for each pixel were estimated by the linear relationships between remotely sensed canopy N and the corresponding observed values. b c a
Figure 1. Map of the Harvard Forest (HF), Massachusetts, U.S.A., showing the terrain and extent of the study area of 10 km by 16 km.
Figure 7. Relationship between canopy %N, GPP and forest type (a), soil drainage (b), and land use history (c). Different lower case letters on each box represent statistically significant differences in the mean among groups (p<0.05).
Spatial patterns of canopy %N and predicted GPP broadly reflected the distribution of functional types and reflected the relationships observed at the plot level across New England, where the fraction of deciduous species in a stand was positively correlated with canopy %N. Both canopy %N and GPP were lowest in stands with the driest soils (e.g. drainage class 1, with 1.6 mean canopy %N and 1330 gCm-2yr-1 GPP) and the wettest soils (e.g. drainage class 6, with canopy %N and GPP of 1.46 and gCm-2yr-1). This pattern was related to the distribution of forest type composition. Lower values of canopy %N and GPP were associated with the higher fraction of evergreen stands. Canopy %N and GPP were significantly different by land use history across the Prospect Hill tract (p<0.05). Woodlots had the lowest canopy %N and GPP (1.60 and gCm-2yr-1, respectively). Stands with a history of unimproved pasture had the highest canopy %N and GPP (1.84 and gCm-2yr-1, respectively). Relatively fertile soils with a history of cultivated agriculture and improved pasture had intermediate canopy %N and GPP, with higher values in cultivated lands than in improved pasture.
In this study, two tower estimates of GPP and NPP were employed to validate our model prediction. A group of geospatial maps, including DEM, soils, forest stands, and historical land-use, were also obtained to examine their spatial relationships to the variability in canopy %N and GPP estimated. Spatially distributed GPP data from the BigFoot Project and the MODIS GPP product (MOD17) were used to compare with our PnET GPP estimates at the landscape scale. a b c d
Figure 8. Seasonal pattern of estimated GPP from PnET, BigFoot, and MODIS in 2003.
GPP estimates from BigFoot and MODIS were compared to our PnET estimates. The spatial distribution of GPP from this study and BigFoot were more visually similar than MODIS as they had similar spatial resolutions and were more process-based. PnET annual GPP ( ±91.0 gC m-2yr-1) was closer to that from BigFoot (1250.1±361.4 gC m-2yr-1) than from MODIS (1401.0±79.0 gC m-2yr-1). When compared with seasonal GPP, however, PnET GPP was more in line with MODIS than BigFoot (Fig. 15).
Figure 2. Foliar N and relative proportions of deciduous and evergreen forests retrieved by remote sensing. (a) Predicted versus observed canopy nitrogen concentration based on PLS regression model.(b) Forest composition as a fraction of deciduous and evergreen species in relation to the plot-level canopy N. (c) Measured canopy N concentration for deciduous components of field plots in relation to AVIRIS %N. (d) Measured canopy N concentration for evergreen components of field plots in relation to AVIRIS %N.
Figure 9. Seasonal pattern of estimated GPP from PnET, BigFoot, and MODIS in 2003.
As PnET GPP estimates were validated with tower data at a smaller footprint and had an advantage of incorporating spatial variability of canopy %N, we assumed PnET GPP would better represent the landscape area of 7 × 7 km than BigFoot. Hence, compared with PnET estimates degraded to 1km resolution, MODIS GPP was greater by 12.3% in this study area. The ratio of MODIS GPP to PnET GPP, reflecting the difference between MODIS and PnET estimates, was negatively correlated to estimated canopy %N. The difference between MODIS GPP and PnET GPP increased with the decreasing canopy %N, mostly in the mixed stands. It implies that the link between canopy %N and canopy reflectance in the near infrared (NIR) (800−850 nm) in temperate and boreal forests approved by Ollinger et al. (2008) and Lepine et al. (2016) may provide the means for a straightforward and efficient approach for scaling foliar nitrogen via broadband satellite data to broad scale productivity estimates, such as MODIS GPP and NPP products.
Results and Discussion
Figure 3. Predicted versus observed monthly GPP at towers EMS and HEM.
PnET modeled GPP was extracted for a 250 m footprint around the two flux towers, EMS and HEM, and compared to tower-derived GPP. In general, model predictions corresponded reasonably well with measured monthly values. The r2 of predicted versus observed values at EMS and HEM were 0.89 (p<0.001) and 0.90 (p<0.001), respectively. The overall data pooled for two sites had an r2 of 0.88 (p<0.001). Although predicted monthly GPP accounted for similar amount of variance of observations at both EMS and HEM, GPP at EMS was overestimated in the early growing season when observed GPP was ranged from 100 to 200 gC m-2 mon-1(May), and underestimated in the late growing season when observed GPP was greater than 300 g C m-2 mon-1(July and August). The predicted annual values of GPP at EMS and HEM were ± and ± 58.2 g C m-2 yr-1, compared with ± and ±169.2 g C m-2 yr-1 from the flux towers, respectively.
Conclusion
At Harvard Forest and its surrounding area, landscape-scale canopy %N, spatially mapped using high spectral resolution remote sensing, was demonstrated to relate to forest composition, soil drainage, and land use history. Mapped canopy %N was positively correlated to relative fraction of deciduous and evergreen species in the stands as well as the foliar N concentration of each. Land use history showed legacy on the variation of canopy %N. Woodlots had the lowest canopy %N and stands in old unimproved pasture had the highest values, followed by stands historically cultivated and improved for pasture. Integration of remotely sensed canopy %N to the ecosystem process model, PnET, demonstrated that canopy %N regulated the forest canopy GPP, with both GPP and canopy %N showing similar relationships with forest composition, soil drainage and land use history. The comparison of PnET GPP with BigFoot and MODIS GPP indicated that canopy %N explained much of the variation between MODIS GPP and PnET GPP, suggesting that global MODIS GPP estimates may be improved if broad-scale estimates of foliar N are available.
References
Aber JD, Ollinger SV, Federer CA et al. (1995) Predicting the effects of climate change on water yield and forest production in the northeastern United States. Climate Research, 5, 207–222.
Hooker TD, Compton JE (2003) Forest Ecosystem Carbon and Nitrogen Accumulation During the First Century After Agricultural Abandonment. Ecological Applications, 13, 299–313.
Lepine LC, Ollinger SV, Ouimette AP, Martin ME (2016) Examining spectral reflectance features related to foliar nitrogen in forests: Implications for broad-scale nitrogen mapping. Remote Sensing of Environment, 173, 174–186.
Ollinger SV, Richardson AD, Martin ME et al. (2008) Canopy nitrogen, carbon assimilation, and albedo in temperate and boreal forests: Functional relations and potential climate feedbacks. Proceedings of the National Academy of Sciences, 105, 19336–19341.
Wright IJ, Reich PB, Westoby M et al. (2004) The worldwide leaf economics spectrum. Nature, 428, 821–827.
Figure 4. Observed canopy %N versus aboveground net primary production (ANPP) .
We assessed the relationship between ANPP and canopy %N, both collected over a series of plots in the BigFoot area, to examine the spatial correlation of canopy %N with carbon assimilation. Observed canopy %N was significantly positively correlated to observed ANPP (r2= 0.48, p<0.001). Larger residuals of predicted values occurred at the higher end of canopy %N.
Acknowledgements
This study was supported by the National Science Foundation and the US Forest Service. We thank Harvard Forest to provide climate and geospatial data. We thank Andrew Ouimette for his thoughtful discussion and comments.