On the uncertainty of phenological responses to climate change, and implications for a terrestrial biosphere model

[摘要] Phenology, the timing of recurring life cycle events, controls numerous landsurface feedbacks to the climate system through the regulation of exchangesof carbon, water and energy between the biosphere and atmosphere.

Terrestrialbiosphere models, however, are known to have systematic errors in thesimulation of spring phenology, which potentially could propagate touncertainty in modeled responses to future climate change. Here, we used theHarvard Forest phenology record to investigate and characterize sources ofuncertainty in predicting phenology, and the subsequent impacts on modelforecasts of carbon and water cycling. Using a model-data fusion approach, wecombined information from 20 yr of phenological observations of 11 NorthAmerican woody species, with 12 leaf bud-burst models that varied incomplexity.

Akaike's Information Criterion indicated support for spring warming modelswith photoperiod limitations and, to a lesser extent, models that includedchilling requirements.

We assessed three different sources of uncertainty in phenological forecasts:parameter uncertainty, model uncertainty, and driver uncertainty. The latterwas characterized running the models to 2099 using 2 different IPCC climatescenarios (A1fi vs. B1, i.e. high CO₂ emissions vs. low CO₂ emissionsscenario). Parameter uncertainty was the smallest (average 95% ConfidenceInterval – CI: 2.4 days century⁻¹ for scenario B1 and4.5 days century⁻¹ for A1fi), whereas driver uncertainty was thelargest (up to 8.4 days century⁻¹ in the simulated trends). Theuncertainty related to model structure is also large and the predictedbud-burst trends as well as the shape of the smoothed projections variedamong models (±7.7 days century⁻¹ for A1fi,±3.6 days century⁻¹ for B1). The forecast sensitivity of bud-burstto temperature (i.e. days bud-burst advanced per degree of warming) variedbetween 2.2 days °C⁻¹ and 5.2 days °C⁻¹ dependingon model structure.

We quantified the impact of uncertainties in bud-burst forecasts on simulatedphotosynthetic CO₂ uptake and evapotranspiration (ET) using aprocess-based terrestrial biosphere model. Uncertainty in phenology modelstructure led to uncertainty in the description of forest seasonality, whichaccumulated to uncertainty in annual model estimates of gross primaryproductivity (GPP) and ET of 9.6% and 2.9%, respectively. A sensitivityanalysis shows that a variation of ±10 days in bud-burst dates led to avariation of ±5.0% for annual GPP and about ±2.0% for ET.

For phenology models, differences among future climate scenarios (i.e.driver) represent the largest source of uncertainty, followed byuncertainties related to model structure, and finally, related to modelparameterization. The uncertainties we have quantified will affect thedescription of the seasonality of ecosystem processes and in particular thesimulation of carbon uptake by forest ecosystems, with a larger impact ofuncertainties related to phenology model structure, followed by uncertaintiesrelated to phenological model parameterization.

[学科分类] 地球化学与岩石