Investigating the applicability of emergent constraints
Recent research on emergent constraints (ECs) has delivered promising results in narrowing down uncertainty in climate predictions. The method utilizes a measurable variable (predictor) from the recent historical past to obtain a constrained estimate of change in an entity of interest (predictand) at a potential future CO2 concentration (forcing) from multi-model projections. This procedure first critically depends on an accurate estimation of the predictor from observations and models and second on a robust relationship between inter-model variations in the predictor–predictand space. Here, we investigate issues related to these two themes in a carbon cycle case study using observed vegetation greening sensitivity to CO2 forcing as a predictor of change in photosynthesis (gross primary productivity, GPP) for a doubling of preindustrial CO2 concentration. Greening sensitivity is defined as changes in the annual maximum of green leaf area index (LAImax) per unit CO2 forcing realized through its radiative and fertilization effects. We first address the question of how to realistically characterize the predictor of a large area (e.g., greening sensitivity in the northern high-latitude region) from pixel-level data. This requires an investigation into uncertainties in the observational data source and an evaluation of the spatial and temporal variability in the predictor in both the data and model simulations. Second, the predictor–predictand relationship across the model ensemble depends on a strong coupling between the two variables, i.e., simultaneous changes in GPP and LAImax. This coupling depends in a complex manner on the magnitude (level), time rate of application (scenarios), and effects (radiative and/or fertilization) of CO2 forcing. We investigate how each one of these three aspects of forcing can affect the EC estimate of the predictand (ΔGPP). Our results show that uncertainties in the EC method primarily originate from a lack of predictor comparability between observations and models, the observational data source, and temporal variability of the predictor. The disagreement between models on the mechanistic behavior of the system under intensifying forcing limits the EC applicability. The discussed limitations and sources of uncertainty in the EC method go beyond carbon cycle research and are generally applicable in Earth system sciences.