Age and climate contribution to observed forest carbon sinks in East Asia

The research region in this study is located from latitude 18° N to 54° N and from longitude 70° E to 150° E (figure S1, available at stacks.iop.org/ERL/00/000000), which includes major countries in the East Asian region (we refer to this area as the East Asian region below). We chose this region because we could extract the Chinese region to compare results. The data from the flux sites used in this study included carbon flux data (NEP), which were measured using the eddy covariance technique, forest age, meteorological data (MAT and MAP), and wet nitrogen deposition in the East Asian region. These observation data were extracted from a published literature dataset compiled by Yu et al (2014). These data from flux sites were used to generate statistical models between forest NEP and potential driving factors.

The monthly temperature (MAT) and precipitation (MAP) used to generate the spatial pattern of forest NEP in this study was obtained from the Climate Research Unit (CRU), version TS3.21 (Harris et al 2014). The spatial resolution of TS3.21 was 0.5°, and we resampled them to 0.05° to match the spatial distribution of the forests. Given that the carbon flux data measured by the eddy covariance technique was obtained during the 2000s, we used the CRU data from 2001 to 2010. With these meteorological data, we generated the map of NEP attributed by climate factors. The comparisons of CRU data with flux sites meteorological measurements were shown in figure S2.

The forest distribution data were generated from MODIS land cover type data (Wu et al 2015) (https://lpdaac.usgs.gov/products/modis_overview/modis_products_table/mcd12c1). We extracted the forest area of evergreen needleleaf forests, evergreen broadleaf forests, deciduous needleleaf forests, deciduous broadleaf forests and mixed forests of the region, from latitude 18° N to 54° N and longitude 70° E to 150° E, and set it as the forest distribution map.

In the East Asia region, MAT and age were significantly correlated with forest NEP at the 0.01 level, and nitrogen deposition was significantly correlated with forest NEP at the 0.05 level, while MAP did not show apparent correlation with NEP (table 1). Only MAT and forest age were in the stepwise regression equation when we conducted path analysis. Both MAT and age could indirectly affect forest NEP through the other variable (table 2).

In the East Asian region, a significant correlation existed between the observed NEP and forest age, with 34.7% variation of the NEP accounted for by the forest age (figure 1(a)). The regression equation generated was:

$$\begin{eqnarray}{\rm{N}}{\rm{E}}{\rm{P}} & = & 0.0048\times {\rm{a}}{\rm{g}}{{\rm{e}}}^{2}-2.477\times {\rm{a}}{\rm{g}}{\rm{e}}\\ & & +452.982({R}^{2}=0.347,p=0.007)\end{eqnarray} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

A statistical regression model was generated between MAT alone and observed NEP (equation (2)), which 37.9% variation of the NEP could be explained by MAT.

$$\begin{eqnarray}{\rm{N}}{\rm{E}}{\rm{P}} & = & 13.918\times {\rm{M}}{\rm{A}}{\rm{T}}+141.916\\ & & ({R}^{2}=0.379,p\;\lt \;0.001)\;\end{eqnarray} \tag{ 2 }$$

When we explored the relative contribution of climate on forest NEP on the basis of the tendency of forest carbon sink with age change, the primary effect of age on NEP was de-trended first. Given that forest NEP was determined multifactorially, in addition to age, the effects of other factors were also included in the NEP observations of each flux site, and we calculated the residuals between the observed and modelled NEP by equation (1), which reflect the component of the NEP that could not be explained by age. The NEP residuals no longer correlated with forest age (figure 1(b)). The plot of the NEP residual (figure 1(b)) indicated higher residuals with increasing temperature, i.e., observation sites with higher temperature tend to have higher residuals than those with lower temperature. That is, temperature had an additional contribution to NEP after de-trending the effect of age. To further highlight the pattern, we divided the flux sites into six age groups (0–50, 50–100, 100–150, 150–200, 200–250, 250–300 years) and calculated the average NEP and average MAT for the flux sites above and below the fitting curve separately (figure 2(a)), which showed that the MAT was an apparent determinant of NEP when the forest age was similar. After de-trending the effects of the forest age to NEP, we performed regression analysis between NEP residuals and MAT and found that the NEP residuals were positively correlated with MAT (figure 2(b)). The regression equation is shown in equation (3) and the model revealed that 23.0% variation of the NEP residuals could be explained by MAT.

$$\begin{eqnarray}{\rm{N}}{\rm{E}}{\rm{P}}\;{\rm{r}}{\rm{e}}{\rm{s}}{\rm{i}}{\rm{d}}{\rm{u}}{\rm{a}}{\rm{l}}{\rm{s}} & = & 9.43\times {\rm{M}}{\rm{A}}{\rm{T}}-82.95\\ & & \;\times ({R}^{2}=0.23,p=0.013)\end{eqnarray} \tag{ 3 }$$

Zoom In Zoom Out Reset image size

We also generated the statistical model between the NEP residuals and MAP (figure 2(c)), but the results indicated that MAP had no apparent relationship with the NEP residuals at 0.05 significance level (R² = 0.028, p = 0.413).

Another factor that might have affected the forest NEP is nitrogen deposition. If the age effect was neglected, the observed NEP seems significantly related with the wet nitrogen (R² = 0.444, p = 0.003) (figure 3(a)), which is similar to the relationship observed by Yu et al (2014). When the effect of the forest age on NEP was de-trended, the residual of NEP did not significantly correlate with wet nitrogen deposition anymore (R² = 0.133, p = 0.164) (figure 3(b)). However, we would like to emphasise that it could not prove that nitrogen deposition had no significant effect on forest NEP when age was considered; it only showed that nitrogen deposition might not be a suitable predictor for regional scale NEP.

Zoom In Zoom Out Reset image size

With the CRU meteorology data from 2001 to 2010, MODIS land cover map and the statistical relationship between NEP and MAT, the spatial distribution of forest NEP could be estimated for the East Asian region (figure 4). The spatial distribution of NEP estimated by climate factors alone (equation (2)) was exhibited in figure 4(a), the mean forest NEP was 281.428 g C m⁻² yr⁻¹ and the total annual NEP was 1.123 Pg C yr^-1. While when the primary effect of forest age was de-trended and the regression equation between the NEP residuals and MAT (equation (3)) was employed, the estimated value of mean NEP residuals was 11.575 g C m⁻² yr⁻¹ and the total annual NEP residuals was 0.046 Pg C yr⁻¹ (figure 4(b)). The NEP residuals which indicated forest NEP contributed by climate on the basis of the tendency of forest carbon sink with age change of this region exhibited a decreasing gradient from south to north. Positive NEP was mainly located in the south and southeast, while negative NEP was located in the north and northeast.

Zoom In Zoom Out Reset image size

To further explore and better compare the results, we extracted the region of China (figures 5(a), (b)). The mean NEP of China estimated by equation (2) was 312.63 g C m⁻² yr⁻¹, and the total annual NEP was 0.540 Pg C yr⁻¹ (figure 5(a)). While the mean NEP residuals estimated by equation (3) was 32.79 g C m⁻² yr⁻¹, and the total annual NEP residuals was 0.057 Pg C yr⁻¹ (figure 5(b)). It indicated that when the age effect was ignored and only climate factors were considered, the carbon sink contributed by climate factors would be significantly overestimated. To strengthen the evidence, we also employed an assessment scheme generated by Zhu et al (2014), where forest age was not considered as well, to estimate largest potential of NEP (equation (4)):

$$\begin{eqnarray}{\rm{N}}{\rm{E}}{\rm{P}} & = & 34.25\times {\rm{M}}{\rm{A}}{\rm{T}}+0.7\times {\rm{M}}{\rm{A}}{\rm{P}}-0.03\\ & & \times {\rm{M}}{\rm{A}}{\rm{T}}\times {\rm{M}}{\rm{A}}{\rm{P}}-256.09\end{eqnarray} \tag{ 4 }$$

Zoom In Zoom Out Reset image size

On the basis of this assessment scheme, we calculated the model NEP of forests and its spatial distribution in China (figure 5(c)). Although it also exhibited a decreasing gradient from the southeast to northwest, similar to figure 5(b), the magnitude of the forest averaged NEP in China was 406.94 g C m⁻² yr⁻¹ and the total annual NEP was 0.704 Pg C yr⁻¹, which was of the similar magnitude as our estimation by equation (2) (312.63 g C m⁻² yr⁻¹ and 0.540 Pg C yr⁻¹) where forest age was not considered either, but much higher than our estimation by equation (3) (32.79 g C m⁻² yr⁻¹ and 0.057 Pg C yr⁻¹) where forest age was de-trended first.

A significant high spatial discrepancy (figure 5(d) (figure 5(c) minus figure 5(b))) exists between the largest potential NEP estimated by Zhu's assessment scheme (figure 5(c)) and our results (figure 5(b)), in which the age effect was de-trended first. The largest difference of forest NEP occurred in southeast China where forest NEP contributed by climate factors had been the most overestimated. Given that southeast China has a relative young forest age (Dai et al 2011, Zhang et al 2014), it could be deduced that the main reason for this higher NEP in southeastern China as attributed to forest age (equation (1)), but not attributed to climate factors. That is, in actual NEP estimations, if we ignore the effect of forest age and directly build the statistical model between NEP and climate, we would significantly overestimate the effect of climate factors on NEP.

In this study, we compared the results of forest NEP estimated by climate factor MAT with and without age as a restriction factor. Climate factors may better fit alone with NEP, but their effects were on the basis of age-related trend of forest productivity. After de-trending the effect of age, variance of NEP explained by climate factors was reduced from 0.379 to 0.23, and age turned to be the primary factor affecting NEP (R² = 0.347), which consistent with many other studies (Pregitzer and Euskirchen 2004, Fang et al 2014, Du 2015). The similarity between the relationship pattern of NEP and forest age in the East Asian region in this study and in Yu et al's study (2014) of the entire Asian region indicated the stability of the relationship of NEP and age. In further estimation, if forest age is de-trended, climate contribution to forest productivities among regions will be comparable and it could be advantageous for scientists to predict the effects of climate change on forest ecosystems and their regional disparities.

MAT and MAP reflected the average heat and water condition on the regional and inter-annual scale. In most studies, this relationship was found between MAT and forest NEP (Magnani et al 2007, Piao et al 2009, Loudermilk et al 2013), while relatively less studies used MAP in the NEP statistical models. In this study, after de-trending the effect of forest age, climate factor MAT was the only factor significantly correlated with NEP. The reason that MAP did not fit well with forest NEP in the East Asian region might due to the fact that precipitation was not a major limitation of ecosystem productivity in this region, particularly in areas where forest stands were developed. Another reason may be that only part of the precipitation was related to the ecosystem carbon gain due to water-use efficiency and that the ecosystem water-use efficiency differs strongly depending on factors such as vegetation types, meteorological conditions and disturbances.

The significant correlation between NEP and wet nitrogen deposition weakened after de-trending the effect of forest age on NEP in the East Asian region. It may indicated that the effect of wet nitrogen deposition on NEP could be largely explained by forest age on a scale like that of the East Asian region. The potential reason was most likely related to human activities because regions where frequent afforestation activities occur were often consistent with regions where intense human industrial activities occur. Thus, the regions where forest age was relatively lower might also have relatively higher wet nitrogen deposition. However, nitrogen deposition data is relatively less and its effect on carbon sink is complex and uncertain (Fleischer et al 2015). Therefore, control experiments performed on smaller scales would be more helpful to understand the mechanism underlying these issues.

To test the accuracy of estimating NEP by forest age and climate factors separately, we used our decoupling method and forest flux sites data of the East Asian region (from latitude 18° N to 54° N) and the complete dataset of the Asian region (from latitude 2° N to 65° N) collected from Yu et al (2014), to estimate the NEP contributed by forest age and climate factors separately. We then compared the sum of the age and climate contributed NEP with the flux site observed NEP. Equations (1) and (3) were used to estimate the forest age contributed NEP and climate contributed NEP after de-trending the effect of age in the East Asian region. Equations S1 and S2 were used to estimate the forest age contributed NEP and climate contributed NEP after de-trending the effect of forest age in the whole Asian region (figure S4, table S2). Furthermore, NEP estimations were presented in 10° latitude bins (figure 6). No significant difference existed between the flux site NEP and the sum of the age and climate contributed NEP in either region scale, which indicated that our method of estimating NEP was reliable.

Zoom In Zoom Out Reset image size

The carbon flux data and forest age data strongly relied on flux sites; the accuracy of NEP estimation was highly influenced by the number and spatial distribution of the sites. Most of the old forests were in cold regions and most young forests were in warm regions (figure 1(a)). To balance these defect, cold region with young forests and warm region with old forests should be given more consideration when selecting the location of flux tower sites.

Another uncertainty was caused by plant functional types (PFTs) aggregation because aggregate data could hide or misrepresent the relationship of NEP with age and climate of each PFTs. To address this issue, each PFTs data should be analysed before data aggregation (Clark et al 2011). In this study, we divided the forest ecosystem into PFTs employing two different ways to test the stability of the relationship of NEP with age and climate (figure 7), so it could make up for the limited number of flux sites to some extent. The decline trends of NEP with age were similar among different PFTs, the slopes of models (NEP with age) were almost the same (figures 7(a), (c)). Although the mean NEP of needleleaf forest was greater than broadleaf forest (figure 7(a)), and the mean NEP of evergreen forest was greater than deciduous forest (figure 7(c)), the differences were not necessarily resulted from PFTs but because of the relative higher temperature range (figures 7(b), (d)). No obvious trend of NEP residuals with MAT showed in evergreen forests (figure 7(d)), it might because evergreen forests distributed in a relatively narrow temperature range. Therefore, data aggregation could cause estimation error, but with more forest sites and larger age and temperature range, the general characteristic of NEP with age and MAT could be represented better at forest ecosystem level.

Zoom In Zoom Out Reset image size

Approximately 23% of the NEP residuals could be explained by MAT after de-trending the effect of forest age. The scatter plot indicated that the variation of the NEP residuals in the younger forest age stage was higher (figure 1(b)), which could suggest that temperature changes may had stronger effects on young forests. In China, frequent and intense afforestation activities in recent decades have reduced the average forest age (Dai et al 2011), and combined with the monsoon climate, temperature changes will have a stronger effect in the China region. The coefficient of independent variable MAT in equation (2) is larger than in equation S2 (table S2), indicating that temperature changes have a stronger effect in the East Asian region where China is located compared to the whole Asian region. Thus, facing the global warming situation, the forest of the latitude band where China is located has a greater potential for carbon sequestration.

If forest age was not taken into consideration when estimating the forest NEP, then the effects of climate might be magnified. The forest NEPs estimated mainly from climate factors were about 10 to 12 times higher (depended on estimated methods) than the NEP contributed by climate factors (MAT) after de-trending the effect of the forest age (table S3) in China, and they were dramatically higher than the value presented by Pan et al (2011), which was estimated using inventory data and long-term field observations coupled to statistical or process models.

On the basis of forest inventory data, Pan et al (2011) revealed that the annual total carbon sink, which is contributed by all natural and human factors, in China's forests was 0.18 Pg C yr⁻¹ from 2000 to 2007. When this magnitude was used as a reference value, we found that climate contribution on NEP on the basis of the tendency of forest carbon sink with age change accounted for 31.7% of the total NEP in our study.