Quantitative assessment of carbon sequestration reduction induced by disturbances in temperate Eurasian steppe

The study region lays in the central part of Eurasia, spanning from Volgograd Oblast, Russia in the west to Mongolia and Inner Mongolia, China in the east (figure 1). The longitudinal and latitudinal extents are 39°–125°E and 35°–55°N, respectively. The region is generally dominated by a temperate continental climate with different degrees of aridity. Mean annual precipitation in much of the region is less than 400 mm. Temperature conditions have large spatiotemporal variations. Mean annual temperature ranges from −5 °C to 15 °C with a latitude gradient except for the Tian Shan and Altai mountain ranges (Mohammat et al 2013). For seasonal variation, summer is typically hot averaging around 20 °C–26 °C, peak daily temperature could exceed 40 °C, while in winter, the climate is extremely harsh, with low temperatures reaching −30 °C. With longitudinal differences of climate and continentality variations, zonal grassland types vary from meadow steppe (forest steppe), typical (true) steppe to desert/semi-desert steppe. Based on these longitudinal divisions of the Eurasian steppe biome, the region is separated into two sub-regions: Transvolga–Kazakhstan Steppe (TKS) and Mongol Steppe (MS) (figure 1).

Zoom In Zoom Out Reset image size

Political and social factors largely influence the livestock distribution pattern at both spatial and temporal scales among different countries or sub-regions (Chen et al 2015b). To exclude the factors of political/social heterogeneity that control or alter the regional livestock distribution pattern, a resource-oriented scheme was developed. The basic concept is that the livestock distribution is proportional to the local grassland productivity at large scales. We first incorporated official province/prefecture level statistics to generate the regional administrative map of livestock distribution. Thereafter, we applied a net primary productivity (NPP)-driven scheme to predict the pixel level of livestock distribution pattern.

The detailed description and validation of the livestock distribution map are presented in section S1, in the online supplementary data.

2.3.1. The BEPS model

BEPS is a remote sensing based terrestrial biogeochemical model. The model was initially developed to simulate vegetation production in boreal ecosystems (Chen et al 1999, Liu et al 1997). Then the model was coupled with terrestrial water and energy cycling and a CENTURY-derived soil carbon scheme (Chen et al 2007, Ju et al 2006, Liu et al 2003). It has been well validated and applied in multi-scale studies (Zhang et al 2012, Zhang Fangmin et al 2014). The model was recently revised to improve its simulation ability on grassland ecosystems (Chen et al 2016).

Here we only describe the algorithms to calculate the major indexes of terrestrial carbon flux used in this study.

NEP is the difference between NPP and soil respiration (R_h):

$$\begin{equation} {\rm NEP} = {\rm NPP} - {R}_\rm h \end{equation}$$

where NPP is the difference between GPP and autotrophic respiration rate (R_a):

$$\begin{equation} {\rm NPP} = {\rm GPP} - {R}_a \end{equation}$$

GPP is calculated using the canopy 'two-leaf' scheme and the Farquhar biochemical model (Farquhar et al 1980). R_a is simulated as the sum of growth respiration (R_g) and maintenance respiration (R_m). R_g is calculated as a fixed percentage from GPP. For the R_m simulation, we replaced the original scheme of R_m for grassland in BEPS with a component-specific scheme:

$$\begin{equation} R_{m,k} = T_k \times r_{m,k} \times Q_{10,r} \frac{{(T - T_{{\rm ref}} )}}{{10}} \end{equation}$$

T_k is the biomass of the plant component k, and r_m,k is a coefficient of the maintenance respiration. T and T_ref are air temperature and reference temperature for maintenance respiration, respectively. Q_10,r is a temperature sensitivity factor for respiration calculated as a function of temperature (Arora 2003):

$$\begin{equation} Q_{10,r} = 3.22 - 0.046T \end{equation}$$

R_h is calculated by multiplying the carbon release rate (p_k) from the total carbon amount of soil carbon pool k (C_k).

$$\begin{equation} R_{\rm h} = \sum\limits_{k = 1}^8 {p_k C_k } \end{equation}$$

2.3.2. The grazing model

2.3.3. The fire model

The Glob-FIRM model (Thonicke et al 2001) was used in this study. The model predicts the length of fire season using a conception of extinction moisture. The areas affected by fire are estimated as following:

$$\begin{equation} N(w) = w \times {\rm e}^{[k(w - 1)^2 ]} \end{equation}$$

Where N(w) is a temporal and spatial combination fraction of the fire weather for a specific pixel during the year, w is the fraction of days in a year with particular fire condition, which is calculated by the upper soil layer moisture status (Viegas et al 1992).

The fire effect on grasslands is calculated by the prescribed fire resistance. Above-ground biomass and litter are assumed to be fully consumed. So the carbon burned by fire can be calculated as a product of available surface carbon (C_{surface, ave}) and N(w):

$$\begin{equation} C_{{\rm fire}} = C_{{\rm surface,}\;{\rm ave}} \times N(w) \end{equation}$$

Zoom In Zoom Out Reset image size

Due to quick responses and short recovery times of herbaceous plants, lagged effects of disturbances are implicitly included in the BEPS simulation (Oesterheld and McNaughton 1991, Röder et al 2008). In this study, we defined the impacts of disturbances as direct carbon losses due to grazing and fire. The model first simulated the grazing carbon loss and then the fire loss was then simulated after an update of the carbon pool states. At the current stage, we do not consider the interactive effect between grazing and fire events. Hence, the total disturbance impact was calculated as the sum of grazing and fire carbon consumption.

Prior to the regional scale simulations, model outputs were evaluated using several approaches. The model outputs of plant and soil carbon were validated against multi-site field observations distributed in Kazakhstan, Xinjiang, China and Inner Mongolia, China. The net ecosystem exchange (NEE, equal to minus NEP) outputs were validated using the gap-filled eddy covariance (EC) dataset.

To further test the specific performances of the two disturbance models, two other validations were set. The simulated temporal trend of grazing was validated against a multi-year fenced experiment with different livestock density levels; regional results of fire events were compared to an established remote sensing based dataset, the L3JRC product (Tansey et al 2008). Both the burnt area and the date of fire occurrence were validated.

2.6.1. Regional model inputs

2.6.2. Model validation data

Comparisons between modeled and observed results are shown in figure 3. The modeled results were consistent with field measurements. The R² values are 0.84 and 0.58 for plant and soil C, respectively. The corresponding slopes are 0.74 and 0.78.

Zoom In Zoom Out Reset image size

Model outputs of NEE were generally consistent with continuous observations at the eight grassland flux sites (figure 4), and we were able to simulate temporal tendencies in the model. Modeled annual net ecosystem exchange (NEE_MOD) values were higher than measured values (NEE_EC). The R² values between NEE_MOD and NEE_EC ranged from 0.616 (US_BKG) to 0.848 (RU_HA2), and RMSE values ranged from 0.4 gC m⁻² d⁻¹ (RU_HA1) to 1.307 gC m⁻² d⁻¹ (US_BKG) (figure 4(b)).

Comparisons against the fenced observations show that our model set could reproduce seasonal trends of grassland production under different levels of stocking pressure (figure 5). The overall R² is 0.71 (P < 0.01).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Comparisons of burned areas between the L3JRC dataset and model simulation are shown in (figure 6(a); 86% of the pixels are in agreement with the L3JRC dataset. We also compared the regional mean fire occurrence day from 2000–2007; the model identified the regional temporal pattern when fire starts to appear (R² = 0.52, P < 0.05, figure 6(b)).

Zoom In Zoom Out Reset image size

Spatial distributions of mean NPP and NEP from 1999–2008 are shown in figure 7. As a widespread biome transition belt, the TES connects deserts in the southwest and forests in the north; vegetation productivity varies based on these regions. High productivity grasslands are mainly concentrated in areas that are connected with forested areas, such as: meadow steppe in the eastern part of the Mongol Steppe, which is joined to Great Khingan and Siberia; alpine meadow steppe around the Tian Shan and Altai Mountains; and Russian steppe areas that are close to the North Caucasus Forest. Low productivity areas are distributed around the deserts of Kyzylkum and Karakum in the TKS and the Gobi Desert in MS (figure 7(a1)). The model-calculated spatial pattern of NEP is basically consistent with NPP, but may be modified in part by the heterotrophic respiration distribution (figure 7(b1)). Low productivity areas generally acted as carbon sources or were neutral, and high productivity areas acted as carbon sinks. The major differences of spatial distribution were located in the northern provinces in Mongolia (Dornod and Hentiy) and Kazakhstan (Pavlodar and Qaraghandy). High productivity did not result in higher carbon sequestration in the southwest part of the Russian grasslands due to the high soil respiration rate in the red soil of the North Caucasus (Alexandrovskiy et al 2014).

Based on these results, the mean and total NPP of TES were 88.41 gC m⁻² yr⁻¹ and 392 TgC yr⁻¹for the study period. The mean NPP in the TKS and the MS were similar: 87.7 gC m⁻² yr⁻¹ and 89.5 gC m⁻² yr⁻¹, respectively. The total NPP in these two regions was 240 TgC yr⁻¹ and 152 TgC yr⁻¹, respectively (figure (a2)).

The decadal mean NEP of TES was 12.86 gC m⁻²yr⁻¹, indicating that the region was a net carbon sink. The total NEP was 39.2 TgC yr⁻¹. The values were positive for both sub-regions. The total NEP for the TKS and the MS was 14.4 TgC yr⁻¹ and 24.8 TgC yr⁻¹, respectively (figure 7).

The spatial distribution of the two types of disturbances was very uneven in the TES. For the fire regime, the major influenced areas were the northern and north-eastern TKS. However in the MS, fire rarely happened (figure 7(a3)). For grazing, there was a distinct spatial imbalance in carbon consumption in different countries and in different provinces or prefectures (figure 7(a4)). In the MS, major consumption was concentrated in China, especially in Inner Mongolia. The most extensively used grasslands were found in the traditional natural pastures with relatively high productivity, such as XilinGol Prairie in mid-Inner Mongolia and Hulunbuir Prairie close to the Great Khingan. In the TKS, high consumption was found in Uzbekistan and Turkmenistan, where two deserts (Kyzylkum and Karakum) are located and the major grassland type is desert steppe.

Zoom In Zoom Out Reset image size

Both types of disturbance increased significantly during the study period (R² = 0.82, P < 0.001 for grazing, R² = 0.51, P < 0.05 for fire, figure 8). For grazing, carbon loss was on a near-linear increasing trend in the TKS; and for the MS, the trend was similar, except for a curved valley in 2002, which was probably caused by a long-term dzud (severe winter) in Mongolia in that year (Tachiiri et al 2008). For the fire regime, the main affected region was the TKS, and NPP loss from fire increased significantly over the decade of the study. In the TKS, the means and total disturbance consumption values were 1.76 gC m⁻² yr⁻¹ and 7.8 TgC yr⁻¹, respectively (figure 7(b3)). Carbon consumption from grazing was much higher than for fire, and accounted for 95% of the total carbon loss.

Zoom In Zoom Out Reset image size

For the first seven years, the region acted as a carbon sink, but in the last three years it turned into a weak source. This trend was mainly caused by a rapid NEP decrease, from 34.56 gC m⁻² to −30.96 gC m⁻² in the TKS during between 2002–2008 (with a slight increase in 2007). The inter-annual results indicated that the disturbances alleviated the regional carbon sink from 1999–2005 and aggravated the carbon source effect from 2006–2008 (figure 9).

The disturbance impact was much larger in the MS than in the TKS due to the higher level of grazing in the MS sub-region. According to our results, the mean percentage of grazing consumption from the NEP was 13.5% during the decade. The values were 7.6% and 16.6% for the TKS and the MS, respectively. For fire events, the percentages were 0.7% for the entire region, 0.86% for the TKS, and 0.2% for the MS. If the carbon losses from disturbances were explicitly considered, regional carbon sequestration decreased from 39.2 Tg C yr⁻¹ to 31.4 Tg C yr⁻¹.

Zoom In Zoom Out Reset image size

The spatial distribution of the impact of grazing to NEP is displayed in figure 10. Grazing could reverse the local carbon sink to a source in some areas, if the grazing consumption was higher than the absolute value of NEP. This phenomenon is especially significant in barren areas, such as the grasslands in Turkmenistan and Uzbekistan and the southwest boarder of grasslands in Inner Mongolia, China.

We further quantified the grazing impact to NEP for each country and sub-region and found a great spatial imbalance presented in figure 10. Grazing has minor impacts in Kazakhstan, Kyrgyzstan, Russia, Xinjiang, China and Mongolia with a percentage lower than 1% from NEP. However, in arid countries like Uzbekistan and Turkmenistan, over-grazing widely exists. Grazing consumption exceeded NEP in 71.9% and 92.6% of grasslands in these two countries, respectively. These results demonstrate the strong impact of grazing in altering the carbon budget in these areas.

Zoom In Zoom Out Reset image size

The two types of disturbances we studied lead to profound impacts on carbon sequestration in the TES. If the carbon losses from disturbances were explicitly considered, carbon sequestration was significantly reduced in this region. According to our study, the total consumption from the two types of disturbances was 7.8 Tg C yr⁻¹, accounting for 14.2% of the regional NEP, and grazing is the major contributor to carbon loss. Although the importance of grazing has been recognized for a long time, it has not been well quantified into the regional or global carbon sequestration calculations. Therefore, our results could contribute to understanding the large uncertainty in carbon sequestration abilities in arid and semi-arid ecosystems around the world (Ahlström et al 2015).

Chen et al (2015b) proposed the concept of coupled human and natural systems (CHANS), which focuses on the strong interactions between biophysical and socioeconomic systems among TES regions. The CHANS framework suggests that the importance of anthropogenic influences might produce stronger impacts on the grasslands than climate change (Chen et al 2015a). Although it is still difficult to explicitly disentangle the relative contributions of the two factors, our study provides strong quantitative evidence for the anthropogenic part. Furthermore, the carbon consumption caused by both types of disturbance significantly increased during the study period. These trends correspond to the official statistics and previous observations. FAO statistics (FAOSTAT, http://faostat.fao.org/) showed that the livestock numbers in all nations and regions in TES have increased since the late 1990s. Chen et al (2013) reported that more than 40% of grasslands in Central Asia were used for animal husbandry in 2009 compared to 2000 based on remote sensing data. With the continually expanding demand of livestock products and a 'warming and drying' trend, both the scale of animal husbandry and fire danger are predicted to increase (Flannigan et al 2013, Godber and Wall 2014). Hence, the effects of disturbance in TES will probably continue to expand and strongly impact regional carbon cycling in the future.

Moreover, the uneven distribution of disturbances and different management styles need to be considered and mitigated. Our results suggested that similar net grazing consumption occurred in southern countries in the TKS and in Inner Mongolia, China in the MS, but the effect of grazing on carbon sequestration in the southern TKS was much higher than in Inner Mongolia. These regional patterns are related to fundamental differences in grassland quality and productivity. But more importantly, the absolute number of livestock in the two countries (i.e. Uzbekistan and Turkmenistan) is far beyond a suitable stocking rate, leading to the overall over-grazing condition. We have not found detailed statistics from these two countries, however, it is reported that the average stocking rate near settlements in Kyrgyzstan is about 2.74 sheep head hm⁻², which is 3.2 times higher than the recommended scale (0.85 sheep head hm⁻²) (Mirzabaev et al 2016). This result may reflect uncontrolled and excessive stocking rates in Uzbekistan and Turkmenistan, since livestock density in the two countries is obviously higher than in Kyrgyzstan. Because the grasslands are managed by state-controlled cooperatives or farmer associations, this situation is closely related to the national condition. Historically, the national economy in the two countries largely has relied on grassland resources since the collapse of the USSR (Lioubimtseva and Henebry 2009). The huge numbers of livestock and years of over-utilization has directly caused the deserts to expand (Robinson et al 2003). Now grassland resources are diminished but livestock numbers continue to increase after the USSR has collapsed (Chen et al 2013).

Pasture management practices are the other important contributor to the current regional patterns. In China, livestock numbers have experienced a boom period since the 2000s to meet the social and economic needs for animal products in China (Rae and Zhang 2009). To alleviate pressure from unmanaged pasture use and to preserve grassland resources, the Chinese government has promoted several regional programs like the Grain to Green Program (GTGP) and the Grazing Withdrawal Program (GWP) since the 1990s. The primary goal of both programs is to conserve pastures under a different level of degradation by transforming grazing and cropping activity to be sustainable. According to our previous study in Inner Mongolia, construction of infrastructure, including animal sheds and corrals, increased by 20%, and the size of fenced grassland increased by nearly three times from 1999–2008. Accordingly, grassland NPP increased by 40.6 g C m⁻² yr⁻¹ between 2001 and 2009 (with climatic factors excluded) (Mu et al 2013c). On the other side of TES, in the TKS, the influence of high grazing intensity on carbon cycling has not received enough attention from the scientific community or from government officials. Although some management suggestions to conserve carbon stocks, such as rotational and migratory grazing, have been supported by empirical evidence (ICARDA 2001, 2002), field studies are still urgently needed in arid areas. Furthermore, the huge amount of investment needed to carry out rehabilitation might be a greater obstacle to reversing degradation on a large scale (Blench and Sommer 1999, Mirzabaev et al 2016). Over the long term, economic development is the primary task for developing countries in this region, but finding a balance between living improvements of human beings and ecosystem sustainability is an essential topic to consider.

We validated model outputs in several ways. The results showed general consistency of our model with continuous carbon flux observations, field data, and remote sensing product. However, model uncertainties are sometimes unavoidable when simulating a complex system with multiple factors. First, uncertainty exists in the model input of leaf area index (LAI). Unstable surface reflectance, foliage clumping, and soil effects in regions with sparse vegetation are among the major sources of LAI deviation (Liu et al 2012). The latter two factors generally cause under-estimation of LAI, which could have caused the under-estimations of NPP and NEP in this study. Second, the uncertainty could arise from mathematical descriptions and corresponding parameterization of biogeochemical processes in the model. We compared our model results with previous studies in this region (table 1). Our results are more comparable with models that share the same photosynthesis scheme (i.e. Farquhar model), but our results are often lower than light use efficiency (LUE) models.

In the current soil carbon model, the mechanical description of microbial activities is relatively weak in soil carbon circulation and the algorithms to calculate environmental factors on soil decomposition are not well supported (Luo et al 2015). Moreover, the disturbance intensity, especially the stocking rate, could significantly affect the magnitude of changes in carbon pools and fluxes. Light grazing might contribute to accumulation of carbon, whereas moderate and heavy grazing would induce carbon loss in both plant and soil pools and lead to soil erosion and degradation (Yan et al 2013, Zhou et al 2016). Responses of grasslands to different disturbance intensities should be considered in future models and simulations.

With the input of data based on governmental official statistics, regional livestock consumption could be constrained at a certain range. The model validation results also support the effectiveness of the grazing model. However, the current model assumes a constant daily intake rate of livestock, which ignores the effects of environmental change such as climatic warming and CO₂ enhancement on the livestock intake rate (Collier and Zimbelman 2007, Nardone et al 2010). In addition, more frequent extreme weather events cause more severe stresses that are closely related to higher livestock mortality (Vitali et al 2009).

For the fire regime, the major disagreement between the model results and the field observations is from the mountainous areas in north Kazakhstan and Kyrgyzstan. We suspect that deviations in the model come from the discrepancy between the model scheme and real climate patterns in these areas. The model used soil water content as a scalar to determine fire occurrence. That is, if surface soil water content is lower than 'extinction moisture', then the area has the potential to burn. In these mountainous areas, the climate pattern is different from the rest of the region. Rainfall mainly happens in the winter and spring, and the temperature is mild throughout the year, rarely exceeding 30 °C, even during the mid-summer (Rachkovskaya and Bragina 2012). Thus it is possible that the soil water content is low but energy conditions for fire occurrence are not met.

The period of our study avoids significant land use and cover change (LUCC) from political events, e.g. the collapse of the former USSR. However, some studies have highlighted the potential impacts from LUCC in recent years. For example, Chen et al (2013) reported that the grasslands were gradually reclaimed after the overall recovery during the 1990s in Central Asia. Mu et al (2013c) suggested that the increasing grassland areas lead to carbon accumulation in Inner Mongolia. Based on our results, we speculate that persistent over-grazing will largely accelerate irreversible land degradation, especially in the southern countries in the TKS. In addition, the emergence of mining and oil production, as well as the current trend of urbanization are also potential factors affecting LUCC in the TES (Peng et al 2009, Pomfret 2016, Suzuki 2013).