Research paperQuantification of the response of global terrestrial net primary production to multifactor global change
Introduction
Net primary production (NPP) is a principal component in the global biosphere carbon cycle and it represents the net carbon fixed by the global plant community (Chapin et al., 2002). As a fundamental attribute of the global biosphere, NPP plays an essential role in providing humans with necessary food, timber and fibre (Vitousek et al., 1986, Costanza et al., 1998, Running et al., 2004). Because NPP is a composite reflection of the combined climatic, geochemical, ecological and human effects on the biosphere (Nemani et al., 2002, Nemani et al., 2003), it is sensitive to multiple environmental changes such as climate and atmospheric changes (Chapin et al., 2002). Therefore, to assess the spatial and temporal patterns of NPP and to quantitatively analyse the relationships between NPP and its related environmental factors, these factors have received increasing attention in global change studies during the past several decades (Piao et al., 2005, Hsu et al., 2012, Liang et al., 2015, Pan et al., 2016).
Previous studies have indicated that the variability of NPP is controlled by a broad range of biotic and abiotic factors operating mainly through changes in plant physiological activities and phenology (Geider et al., 2001, Richardson et al., 2010, Stoy et al., 2014, Xia et al., 2015). Climate change and increasing CO2 concentrations were recognized as the key factors in the change in global terrestrial NPP (Melillo et al., 1993). Rising air temperatures, altered precipitation patterns and elevated atmospheric CO2 interact with each other and exert a combined impact on ecosystem structure and function (Canadell et al., 2007). In addition to climate change and the fertilization effects of rising atmospheric CO2, land use change, such as afforestation and deforestation (Zhou et al., 2015), and anthropogenic nitrogen deposition (Stevens et al., 2015) also have an important effect on terrestrial NPP. Therefore, the complex interactions between these factors in the global terrestrial ecosystem pose a considerable challenge for NPP modelling studies. In general, these interactions are not well known, and it is difficult to attribute the relative contribution of multifactor global changes.
Empirical models have been used to quantify the relationship between NPP and related environmental variables (Zaks et al., 2007, Del Grosso et al., 2008, Cleveland et al., 2015), and it assume that effects of environmental variables on NPP change are linear and independent of each other. However, evidences from both field experiment and theoretical analysis have shown nonlinear ecosystem responses to the environmental changes (Berry and Bjorkman, 1980; Peng et al., 2013a), and highlighted the potential limitations from the linear regression analysis. A considerable number of ecosystem process models also have been applied to analyse the spatiotemporal patterns of NPP and its responses to global change in terrestrial ecosystems (Cramer et al., 1999, Pan et al., 2014). However, huge uncertainties remain in the different ecosystem models in estimating global NPP. Siegenthaler and Sarmiento (1993) quantified the annual global NPP to be 51.97 Pg C, which was much lower than the estimation of Sundquist (1993), who estimated global NPP to be 60 Pg ; Cramer et al. (1999) conducted a comparison of 17 process-based models, and the estimated global NPP had a wide range of 44–66 Pg C yr−1, resulting from how the water balance was represented in the models. Similarly, Friedlingstein et al. (2006) found the differences in the same 17 global NPP models were due largely to the belowground processes that cause different responses of NPP to multifactor global change. Thus, the limitations of NPP estimation are largely attributed to complicated processes in the biosphere, and the process models have been unable to consider all of the complicated nonlinear relationships involving ecosystem and environmental variables. Compared with empirical models and process-based models, the artificial neural network (ANN) method has the greatest potential to address the nonlinear problems because of its accurate mapping capability (Liu et al., 2010). The ANN method is known for its strengths in handling many types of prediction and classification complexities. This method has been used successfully to map global terrestrial N2O emissions (Zhuang et al., 2012) and to simulate methane emissions (Dengel et al., 2013, Zhu et al., 2013), the soil organic carbon dynamics (Dai et al., 2014, Yang et al., 2014, Were et al., 2015), and the C flux of a Chinese fir plantation in subtropical China (Wen et al., 2014).
Most of the field measurements of NPP have been conducted and published for global terrestrial ecosystems during the past several decades. The detailed observational data and the ANN method may offer an opportunity to analyse the effects of multiple environmental factors on global terrestrial NPP. In this study, we synthesized 2196 measurements from a global compilation of NPP data on global terrestrial ecosystems. We chose a three-layer back-propagation neural network (BPNN) method (Svozil et al., 1997, Saxén and Pettersson, 2006, Liu et al., 2012) to estimate NPP and selected five key variables: including precipitation (Hsu et al., 2012), air temperature (Clark et al., 2003), leaf area index (Schloss et al., 1999), fraction of photosynthetically active radiation (Bicheron and Leroy, 1999) and atmospheric CO2 concentration (Norby et al., 2005), which are considered to be main factors controlling the NPP dynamics of global terrestrial ecosystem (Chapin et al., 2002). The main objectives of this study were to (1) examine the performance of the ANN model in estimating the NPP of the global terrestrial ecosystem, (2) analyse the spatiotemporal patterns of global NPP during the period 1961–2010, and (3) quantify the relative contributions of major environmental factors controlling the change in global NPP.
Section snippets
Data
We have collected and complied most of the available observational NPP data from the Oak Ridge National Laboratory (ORNL) Distributed Active Archive Center database (http://daac.ornl.gov/NPP/npp home.shtml). These study sites contain NPP measurements from more than 30 countries and cover a range of vegetation types and biomes. Each site contains a collection of NPP observational records (Bailey, 1989, Jager et al., 2000). These NPP observed values were originally recorded as yearly measurements
The performance of ANN model
The linear correlation coefficients between the observed NPP and estimated NPP using the linear model, which uses a single variable as the input, are shown in the first row of table S2. We found that the correlations between NPP and P, T, LAI, fPAR and CO2 were strong. The correlation coefficients between the observed NPP and the simple ANN model-predicted NPP, which used one of the factors as the input in the developed ANN model, are also listed in the second row of table S2. The results
Comparison of ANN-estimated NPP with previous estimates
In this study, the global NPP estimated by the ANN model was 61.46 Pg C yr−1 during the period 1961–2010, which is comparable with the value reported by the IPCC of 61 Pg C yr−1 (IPCC, 2013), and it also falls within the range of a previous terrestrial NPP estimate of between 44 and 66 Pg C per year by 17 global NPP process-based models (Cramer et al., 1999). However, the simulation result of the ANN model is relatively higher than other global estimates of NPP (Table 5). For example, our result is
Conclusions
Based on published site-level NPP measurements of global terrestrial ecosystems and associated environmental factors, an ANN model was used to estimate global terrestrial NPP. The developed ANN model performs relatively well, and it was capable of capturing the temporal and spatial variations in NPP estimation in global terrestrial ecosystems. The results show the estimated annual global NPP varied between 54.95 Pg C and 66.75 Pg C during the period 1961–2010, with a mean value of 61.46 Pg C yr−1, and
Acknowledgements
This study was financially supported by the National Basic Research Program of China (2013CB956602), the Programme of NCET (Z111021401), QianRen Program, and the Natural Sciences and Engineering Research Council of Canada (NSERC) Discover Grant. We also express our thanks to senior editor Isaac N. for critical comments on an earlier version of the manuscript.
References (83)
- et al.
A method of biophysical parameter retrieval at global scale by inversion of a vegetation reflectance model
Remote Sens. Environ.
(1999) Importance of leaf area index and forest type when estimating photosynthesis in boreal forests
Remote Sens. Environ.
(1993)- et al.
The value of the world’s ecosystem services and natural capital
Ecol. Econ.
(1998) - et al.
Spatial prediction of soil organic matter content integrating artificial neural network and ordinary kriging in Tibetan Plateau
Ecol. Indic.
(2014) - et al.
A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin
Remote Sens. Environ.
(2008) - et al.
Artificial neural networks as a tool in ecological modelling, an introduction
Ecol. Modell.
(1999) - et al.
Analysis of spatial and temporal patterns of net primary production and their climate controls in China from 1982 to 2010
Agric. For. Meteorol.
(2015) - et al.
Simulations of runoff and evapotranspiration in Chinese fir plantation ecosystems using artificial neural networks
Ecol. Modell.
(2012) - et al.
Comprehensive comparison of gap-filling techniques for eddy covariance net carbon fluxes
Agric. For. Meteorol.
(2007) - et al.
Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS data
Remote Sens. Environ.
(2002)