Bayesian Time of Emergence: Addressing uncertainty and spurious trends in precipitation change using a CMIP5 ensemble
Abstract
Climate model projections of precipitation change can be highly uncertain at the local scale. In many regions, climate models disagree not only on the magnitude but also the direction of precipitation change, impeding adaptation decisions. Additionally, unlike temperature, many precipitation projections do not monotonically increase or decrease; a decadal drying trend does not necessarily result in long-term drying. These two factors challenge estimates of the time in which the anthropogenic precipitation signal becomes detectably different from historical variability, known as the Time of Emergence (ToE). Common approaches for estimating ToE use ensembles of GCMs where the climate signal is estimated using the multi-model average. However, averaging over precipitation can provide a biased estimate of ToE due to the "flaw of averages": averaging across projections before calculating ToE is different than aggregating ToE estimates from individual projections. Here, we revisit the ToE framework for rainfall change by asking 1) when will we know if a location is wetting or drying? and 2) what does a near-term statistically significant change in rainfall indicate about long-term trends? To answer these questions, we develop a probabilistic Bayesian model averaging approach that uses an ensemble of climate models and accounts for correlation in rainfall change over time. We identify regions where the Bayesian approach provides substantially earlier estimates of ToE compared to the traditional estimates of ToE. We also identify regions which may never have definitive emergence of the wetting/drying signal despite extreme projected changes at the end of the 21 st century. Our results demonstrate that the most populated areas are often in regions with the greatest uncertainty in direction of rainfall change. While there is uncertainty across models in the direction of change of precipitation at the local scale, this probabilistic framework provides a more realistic representation of when we will know if a location is getting wetter or drier, and how near-term observations can reduce uncertainty in long-term outcomes.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMGC011..03L
- Keywords:
-
- 3305 Climate change and variability;
- ATMOSPHERIC PROCESSES;
- 1616 Climate variability;
- GLOBAL CHANGE;
- 1631 Land/atmosphere interactions;
- GLOBAL CHANGE;
- 1803 Anthropogenic effects;
- HYDROLOGY