State-of-the-art global models underestimate impacts from climate extremes

Jacob Schewe, Simon N. Gosling, Christopher Reyer, Fang Zhao, Philippe Ciais, Joshua Elliott, Louis François, Veronika Huber, Heike K. Lotze, Sonia I. Seneviratne, Michelle T. H. van Vliet, Robert Vautard, Yoshihide Wada, Lutz Breuer, Matthias Büchner, David A. Carozza, Jinfeng Chang, Marta Coll, Delphine Deryng, Allard de Wit, Tyler D. Eddy, Christian Folberth, Katja Frieler, A. D. Friend, Dieter Gerten, Lukas Gudmundsson, Naota Hanasaki, Akihiko Ito, Nikolay Khabarov, Hyungjun Kim, Peter Lawrence, Catherine Morfopoulos, Christoph Müller, Hannes Müller Schmied, René Orth, Sebastian Ostberg, Yadu Pokhrel, Thomas A. M. Pugh, Gen Sakurai, Yusuke Satoh, Erwin Schmid, Tobias Stacke, Jeroen Steenbeek, Jörg Steinkamp, Qiuhong Tang, Hanqin Tian, Derek P. Tittensor, Jan Volkholz, Xuhui Wang, Lila WarszawskiNature Communications 2019 · DOI

TL;DR

State-of-the-art global climate impact models, when tested against the severe 2003 European heatwave and drought, systematically underestimate the severity of impacts across agriculture, terrestrial ecosystems, and human mortality. This suggests that future societal risks from extreme climate events may be greater than currently projected by these models.

What Problem Does This Paper Address?

Climate change projections often rely on global impact models that simulate how natural and human systems will respond to changing climate conditions. These models are crucial for integrated assessments of climate change and informing policy. However, their ability to accurately represent the impacts of extreme climate events – which cause disproportionately large damages – has not been systematically tested. This is a critical gap, as extreme events are projected to become more frequent and intense in the future. The paper addresses this by evaluating how well current state-of-the-art impact models capture the effects of a well-documented, severe extreme event: the 2003 European heatwave and drought (EHWD).

Methods

The study employs a multi-sectoral approach, evaluating an ensemble of global impact models across various domains: hydrology, agriculture, terrestrial ecosystems, energy (hydropower), and human health. The core methodology involves comparing model simulations driven by observational climate data for the 2003 EHWD against observed impacts. A key innovation is the focus on the extremeness of the event, using a metric of deviation from the historical average, adjusted for long-term trends and normalized by the historical standard deviation. This approach circumvents biases in baseline data or inter-annual variability and specifically tests the models' ability to capture anomalous events. For human health, statistical models were used to estimate excess mortality based on temperature-mortality relationships.

Datasets

  • Climate Forcing Data:
    • WATCH Forcing Data methodology applied to ERA-Interim reanalysis data (WFDEI) - primary dataset.
  • Global Soil Wetness Project phase 3 forcing data set (GSWP3) - for comparison in some sectors.
  • Princeton Global Forcing data set version 2 (PGFv2) - for comparison in some sectors.
  • Observational Data:
    • Water Resources: Global Runoff Data Centre (GRDC) monthly discharge data.
  • Agriculture: FAOSTAT data on crop yields (maize and wheat).
  • Terrestrial Ecosystems: MODIS GPP estimates derived from remotely sensed fAPAR.
  • Energy: U.S. Energy Information Administration (EIA) data on annual hydropower generation.
  • Human Health: Literature-based estimates of excess mortality for specific cities, compiled from various studies.

The paper does not introduce new benchmark datasets but rather uses existing observational data to evaluate model performance against a specific extreme event.

Key Insights and Ideas

  • Underestimation is the Norm for Severe Impacts: The most striking finding is the consistent underestimation of severe impacts in agriculture, terrestrial ecosystems, and human mortality. This suggests that our current models are not adequately prepared to predict the full extent of damage from extreme events.
  • "False Alarms" in Water Resources: While many sectors show underestimation, hydrological models sometimes overestimate impacts, predicting significant flow reductions where observations show only moderate anomalies. This highlights a different type of model bias.
  • Southern Europe Hit Hardest by Underestimation: The study reveals a geographical pattern, with southern Europe experiencing particularly severe underestimation of impacts across multiple sectors.
  • Model Ensembles Still Struggle: Even with an ensemble of models, the ability to accurately capture the magnitude of extreme events is limited. The spread across models is often large, indicating significant uncertainty.
  • Implications for Economic Assessments: The findings have direct implications for economic assessments of climate change, which often rely on these impact models. Underestimated impacts translate to underestimated economic damages and societal costs.

Key Results

  • Agriculture: Most crop models underestimate the severity of yield losses, particularly for maize in France and Italy, where observed anomalies were as high as 3-4 standard deviations.
  • Terrestrial Ecosystems: Global vegetation models largely fail to capture the observed reduction in gross primary productivity (GPP) during the EHWD, with many models underestimating the anomaly or even showing positive anomalies.
  • Human Health: Statistical models for excess mortality underestimated observed deaths in several cities, notably Paris, where models predicted only half the observed excess mortality.
  • Water Resources: Hydrological models generally matched observed low river flows well in some basins (e.g., Rhine, Danube), but often overestimated moderate anomalies in others (e.g., Elbe, Oder).
  • Hydropower: In many countries with observed negative hydropower generation anomalies, models simulated substantial reductions in usable capacity, aligning with observations. However, some models overestimated negative anomalies where observations were less severe.

Limitations and Future Work

  • Single Event Focus: The study evaluates a single extreme event (2003 EHWD). While this event was severe and well-documented, its specific characteristics might not generalize perfectly to all future extreme events.
  • Sectoral Exclusions: The study did not include all potential impact sectors, such as livestock production, vector-borne diseases, or forest fires.
  • Data Availability: For some sectors, like nuclear power generation, suitable observational data for model comparison were unavailable.
  • Aggregate Metrics: The study primarily used large-scale, aggregate impact indicators (e.g., country-level crop yield). Smaller-scale characteristics of impacts, which could be critical for infrastructure or societal functioning, were not fully captured.
  • Human Adaptation: The models generally do not account for human adaptation, which can significantly alter the severity of impacts from extreme events. The paper suggests that incorporating adaptation is crucial for more accurate future projections.
  • Model Development Focus: The authors call for a shift in model development and evaluation from focusing on mean conditions to prioritizing the accurate representation of extreme events.
  • Future Work: The paper suggests that further evaluation of marine ecosystem models with higher-resolution climate data and longer simulation runs is needed. For human health, future work could explore incorporating urban heat island effects and air pollution. Accounting for cross-sectoral interactions and societal adaptation is also highlighted as a critical area for future research.

Summary updated

Abstract

Global impact models represent process-level understanding of how natural and human systems may be affected by climate change. Their projections are used in integrated assessments of climate change. Here we test, for the first time, systematically across many important systems, how well such impact models capture the impacts of extreme climate conditions. Using the 2003 European heat wave and drought as a historical analogue for comparable events in the future, we find that a majority of models underestimate the extremeness of impacts in important sectors such as agriculture, terrestrial ecosystems, and heat-related human mortality, while impacts on water resources and hydropower are overestimated in some river basins; and the spread across models is often large. This has important implications for economic assessments of climate change impacts that rely on these models. It also means that societal risks from future extreme events may be greater than previously thought.
BibTeX
@inproceedings{schewe2019state,
  author = {Jacob Schewe and Simon N. Gosling and Christopher Reyer and Fang Zhao and Philippe Ciais and Joshua Elliott and Louis François and Veronika Huber and Heike K. Lotze and Sonia I. Seneviratne and Michelle T. H. van Vliet and Robert Vautard and Yoshihide Wada and Lutz Breuer and Matthias Büchner and David A. Carozza and Jinfeng Chang and Marta Coll and Delphine Deryng and Allard de Wit and Tyler D. Eddy and Christian Folberth and Katja Frieler and A. D. Friend and Dieter Gerten and Lukas Gudmundsson and Naota Hanasaki and Akihiko Ito and Nikolay Khabarov and Hyungjun Kim and Peter Lawrence and Catherine Morfopoulos and Christoph Müller and Hannes Müller Schmied and René Orth and Sebastian Ostberg and Yadu Pokhrel and Thomas A. M. Pugh and Gen Sakurai and Yusuke Satoh and Erwin Schmid and Tobias Stacke and Jeroen Steenbeek and Jörg Steinkamp and Qiuhong Tang and Hanqin Tian and Derek P. Tittensor and Jan Volkholz and Xuhui Wang and Lila Warszawski},
  title = {{State-of-the-art global models underestimate impacts from climate extremes}},
  booktitle = {Nature Communications},
  year = {2019},
  doi = {10.1038/s41467-019-08745-6},
}
Figure