Uncertainty partition challenges the predictability of vital details of climate change

Abstract

Decision makers and consultants are particularly interested in "detailed" information on future climate to prepare adaptation strategies and adjust design criteria. Projections of future climate at local spatial scales and fine temporal resolutions are subject to the same uncertainties as those at the global scale but the partition among uncertainty sources (emission scenarios, climate models, and internal climate variability) remains largely unquantified. At the local scale, the uncertainty of the mean and extremes of precipitation is shown to be irreducible for mid and end-of-century projections because it is almost entirely caused by internal climate variability (stochasticity). Conversely, projected changes in mean air temperature and other meteorological variables can be largely constrained, even at local scales, if more accurate emission scenarios can be developed. The results were obtained by applying a comprehensive stochastic downscaling technique to climate model outputs for three exemplary locations. In contrast with earlier studies, the three sources of uncertainty are considered as dependent and, therefore, non-additive. The evidence of the predominant role of internal climate variability leaves little room for uncertainty reduction in precipitation projections; however, the inference is not necessarily negative, because the uncertainty of historic observations is almost as large as that for future projections with direct implications for climate change adaptation measures.

Mapping (dis)agreement in hydrologic projections

“…Accounting for the internal climate variability implies considering different realizations of the same GCM. It has been demonstrated that slightly different initial conditions can lead to a substantially different GCM realization (Deser et al, 2012;Fatichi et al, 2016). Furthermore, land-use and soil parameters have been kept constant in the hydrologic model for both modelling periods, although it is very likely that land use will change in the future.…”

“…These results show that precipitation is the driving force for hydrologic models, and that the disagreement among GCMs regarding the change in precipitation introduces uncertainty in the hydrologic projection. Reducing this uncertainty, however, is not a trivial task (Knutti and Sedláček, 2013;Fatichi et al, 2016). While it may seem illogical that both the basins with agreement (Figs.…”

Identifying in situ climate refugia for plant species

“…Formal decision analysis can be valuable here, allowing land managers to prioritize conservation action (e.g. However, uncertainty about the complexities of future climate (Fatichi et al 2016) and the velocity of environmental change (Loarie et al 2009, Ohlemüller 2011) poses major challenges for conservation practitioners. comm.).…”

Designing ecological climate change impact assessments to reflect key climatic drivers

“…For example, parameters in a simple weather generator describe the probability of a wet day and the distribution of the amount of precipitation on days that are wet; in the most basic application under climate change, the change in these parameters can be calculated based on a GCM and applied to simulate future sequences (Wilks, 2012). Weather generators have the advantage of quickly sampling natural climate variability via many simulations (e.g., Fatichi et al, 2016) and can do so for multiple variables and at daily or subdaily temporal scales. However, an evaluation Year Minimum temperature (° C) RCP 8.5 RCP 4.5 Modeled historic Maurer Prism Fig.…”

“…For example, parameters in a simple weather generator describe the probability of a wet day and the distribution of the amount of precipitation on days that are wet; in the most basic application under climate change, the change in these parameters can be calculated based on a GCM and applied to simulate future sequences (Wilks, ). Weather generators have the advantage of quickly sampling natural climate variability via many simulations (e.g., Fatichi et al ., ) and can do so for multiple variables and at daily or subdaily temporal scales. However, an evaluation step is necessary because it should not be assumed that an ecological model which was parameterized using weather station data produces similar predictions based on output from a weather generator – like other downscaling and bias‐correction methods, a weather generator matches some, but not all, statistical aspects of the historical data.…”