| 英文摘要: | Earth system models simulate prominent terrestrial carbon-cycle responses to anthropogenically forced changes in climate and atmospheric composition over the twenty-first century1, 2, 3, 4. The rate and magnitude of the forced climate change is routinely evaluated relative to unforced, or natural, variability using a multi-member ensemble of simulations5, 6, 7, 8. However, Earth system model carbon-cycle analyses do not account for unforced variability1, 2, 3, 4, 9. To investigate unforced terrestrial carbon-cycle variability, we analyse ensembles from the Coupled Model Intercomparison Project (CMIP5), focusing on the Community Climate System Model (CCSM4). The unforced variability of CCSM4 is comparable to that observed at the Harvard Forest eddy covariance flux tower site. Over the twenty-first century, unforced variability in land–atmosphere CO2 flux is larger than the forced response at decadal timescales in many areas of the world, precluding detection of the forced carbon-cycle change. Only after several decades does the forced carbon signal consistently emerge in CCSM4 and other models for the business-as-usual radiative forcing scenario (RCP8.5). Grid-cell variability in time of emergence is large, but decreases at regional scales. To attribute changes in the terrestrial carbon cycle to anthropogenic forcings, monitoring networks and model projections must consider the timescale at which the forced biogeochemical response emerges from the natural variability. The carbon cycle influences climate through the carbon-concentration response, which is the gain in carbon storage with higher atmospheric CO2 concentration, and the carbon–climate response, which is the loss in carbon storage with climate change1, 3. Previous carbon-cycle analyses have emphasized these responses at multi-decadal to centennial timescales and their multi-model uncertainty1, 2, 3, 4, 9. Although these analyses quantify long-term carbon-cycle–climate feedbacks, they do not identify decadal-scale unforced variability in the carbon cycle. Earth’s climate has unforced variability internal to the climate system, generally termed natural variability in the climate science literature, which is an important factor in detecting the change in climate from anthropogenic forcings. Natural variability manifests as interannual-to-decadal climate variability, seen in observations and an individual model realization, as well as ensemble variability within a model5, 6, 7, 8. To confidently detect and attribute changes in temperature to increases in greenhouse gases, for example, one can determine the time when the signal of the forced temperature change becomes large relative to its natural variability8, 10, 11, also known as the time of emergence. Despite its importance in determining when a climate signal can be detected, however, natural variability is not considered in analyses of the twenty-first century carbon cycle1, 2, 3, 4. In this work, we determine when changes in the forced carbon signal can be detected by incorporating analyses of natural variability in Earth system models (ESMs), quantified using a multi-member ensemble of simulations. We evaluated the magnitude, timing, and spatial dependence of variability in terrestrial carbon pools (total ecosystem carbon, the sum of vegetation and soil carbon) and net land–atmosphere CO2 fluxes (net ecosystem exchange, NEE) through the twenty-first century to determine when future changes in the carbon cycle were detectable, defined as the time when the forced signal emerged from the noise of natural variability. Analyses were completed using a six-member ensemble of the Community Climate System Model version 4.0 (CCSM4) simulations for Representative Concentration Pathway 8.5 (RCP8.5; ref. 7), which has a radiative forcing of 8.5 W m−2 at year 2100, with a CO2 concentration of about 936 ppm. The six-member CCSM4 ensemble has a 3.53 °C global surface temperature warming averaged for the last 20 years of the twenty-first century compared to the 1986–2005 reference period. We additionally analysed flux tower data and a seven-member ensemble of the CCSM4 historical twentieth century simulations from 1992 through 2004 (the time period when flux data are available) for Harvard Forest to compare observed variability to model variability. We also analysed two other CMIP5 models; these models included a terrestrial carbon cycle in their RCP8.5 simulations and four or more ensemble members. The CCSM4 has a prognostic terrestrial carbon cycle driven by the simulated climate change arising from the radiative forcings, CO2 concentration, nitrogen deposition, and land-use and land-cover change. The land surface in the CCSM4 is a sink for carbon in the absence of anthropogenic land-use and land-cover change, but release of carbon from these activities results in a small net source of carbon over the twenty-first century12, whereas other ESMs project a net carbon sink2, 4. This occurs because the model has low carbon-concentration uptake compared with other ESMs (ref. 3). Under RCP8.5, cumulative ecosystem carbon projections among CMIP5 models at the end of the twenty-first century, relative to 2005, range from approximately −184 to 500 Pg C, with CCSM4 projecting a change of −69 Pg C (ref. 2). The model ranks 12th for soil carbon and 13th for vegetation carbon skill among the18 CMIP5 models9. Natural variability was seen in both observational flux tower measurements and in CCSM4 simulations. Annual NEE at Harvard Forest in Massachusetts over the period 1992–2004 averaged −245 g C m−2 yr−1 (negative NEE indicates a carbon sink), and the sink increased annually at a rate of −15 g C m−2 yr−2 (r2 = 0.34; ref. 13). The detrended annual anomaly ranged from −180 to 145 g C m−2 yr−1 (Fig. 1a; mean absolute value of the anomalies was 62 g C m−2 yr−1). The seven-member CCSM4 twentieth-century ensemble simulated a modest sink for the same period at the corresponding model grid cell (Fig. 1b; 13-year ensemble mean, −19 g C m−2 yr−1). In any given year the forest was either a source or sink of carbon (37% and 63% of the time, respectively, across ensemble members). Despite the larger spatial scale of a model grid cell, the interannual variability within each individual ensemble member was similar to the Harvard Forest flux tower, on the order of ±100 g C m−2 yr−1. The ensemble range for any particular year was similar in magnitude, and ensemble variability over the period 2080–2099 was also comparable (Fig. 1c). Other eddy covariance flux tower sites show ranges of variability similar to the measured and simulated Harvard Forest. A synthesis of flux measurements found that interannual variability in NEE was 86 g C m−2 yr−1 in North American deciduous broadleaf forests and 44 g C m−2 yr−1 in evergreen needle-leaf forests14. When analyses were scaled to larger regions, ensemble variability decreased. For example, ensemble variability averaged for North America was ±40 g C m−2 yr−1 at the end of the twenty-first century (Fig. 1d), with each of the ensemble members providing an equally likely realization of the twenty-first century carbon–climate system.
|
