Time of emergence (TOE) of GHG-forced precipitation change hot-spots



[1] The Time Of Emergence (TOE) of 14 greenhouse gas (GHG) - forced precipitation change hotspots (PSPOTs) is identified from the CMIP3 multi-model ensemble. The TOE is defined as the time in 21st century projections at which the magnitude of the ensemble mean precipitation change signal becomes greater than the uncertainty due to the inter-model spread and the internal model multi-decadal variability. Of the 14 PSPOTS identified, 6 have a TOE in the early decades of the 21st century (northern high latitudes, Mediterranean, and East Africa), 3 in the mid decades (East and South Asia, Caribbean) and 5 in the late decades or beyond (South Africa, Western United States, Amazon Basin, Southern Australia, Central America). The TOE is sensitive to the GHG emission scenario for some of the PSPOTS. The TOE has important implications for the predictability and detection of GHG-forced PSPOTS and for impact and adaptation studies.

1. Introduction

[2] The signature of anthropogenic greenhouse gas (GHG) forcing on surface temperature has been well established both at the global and continental scales. However, although precipitation trends for the last decades have been detected in many areas of the World, they have not been unambiguously attributed to anthropogenic GHG forcing [Hegerl et al., 2007]. This is mostly because of the pronounced variability of precipitation, which makes it difficult to extract GHG-forced signals from the underlying variability and uncertainty noise.

[3] On the other hand, continued and increasing GHG-forced global warming is expected to modify many features of the general circulation, which in turn would affect precipitation patterns across the globe. Examples of such features include changes in storm tracks, positioning of the Inter-Tropical Convergence Zone, characteristics of monsoon circulations, vertical atmospheric stability and atmospheric water vapor content. In fact, future climate projections for the 21st century based on ensembles of coupled Atmosphere-Ocean General Circulation Models (AOGCMs) show a number of well defined patterns of precipitation change [Giorgi et al., 2001; Giorgi and Bi, 2005; Christensen et al., 2007].

[4] The magnitude of GHG-forced precipitation changes tends to become greater as the GHG forcing increases, therefore these changes become more evident with time and with GHG emission intensity [Giorgi and Bi, 2005]. The question then is posed of when in the 21st century does the GHG-forced precipitation change signal emerge from the underlying variability and uncertainty noise. This is a question of high relevance in two respects. First, knowledge of what we can call Time Of Emergence (TOE) of GHG-forced precipitation change signals constitutes relevant information for impact and adaptation studies. Second, the TOE can provide valuable insights on the detection and predictability of regional changes in the hydrologic cycle at decadal to multi-decadal temporal scales, an issue which has been receiving increasing attention [Smith et al., 2007].

[5] Based on these premises, this paper presents an analysis of the CMIP3 ensemble of AOGCM projections [Meehl et al., 2007] aimed at identifying the TOE of prominent precipitation change hotspots (or PSPOTS) at the sub-continental scale and their dependence on the GHG emission scenario.

2. Data and Methods

[6] Identification of the precipitation change TOE is based on the analysis of available global climate change projections. We use here a subset of the CMIP3 ensemble of AOGCM simulations described by Meehl et al. [2007], which consists of simulations of 20th and 21st century climate by more than 20 AOGCMs from laboratories worldwide freely accessible at the PCMDI web site www-pcmdi.llnl.gov. The 20th century simulations use historical GHG concentration pathways and historical variations in solar, volcanic and anthropogenic aerosol forcing, while the 21st century simulations employ GHG concentrations and aerosol forcing from the Intergovernmental Panel on Climate Change [2000] SRES emission scenarios B1 (low GHG forcing), A1B (medium GHG forcing) and A2 (high GHG forcing).

[7] Table 1 shows the sub-set of models employed in this study, which includes only models that simulated all three emission scenarios in order to have an intercomparable set of models across scenarios. However, additional calculations (not shown) showed that use of the full set of available simulations does not change our basic conclusions. For some models and scenarios multiple realizations starting from different atmosphere and ocean initial conditions are available (Table 1). As given by Giorgi and Bi [2005], the AOGCM data are interpolated onto a common 1-degree grid including a 1-degree land mask.

Table 1. CMIP3 Models and Simulations Used in the Present Study

[8] Climate change projections are affected by a number of uncertainties whose effects compound throughout the subsequent steps of the projection process [Giorgi, 2005]. The most important sources of uncertainty are those associated with the use of different models, or model configurations, and different GHG emission scenarios [e.g., Déqué et al., 2005]. The contribution of internal multi-decadal variability may also be important, especially for high order precipitation statistics and at the regional scale [Giorgi and Francisco, 2000]. This multi-decadal variability can be assessed by analyzing different realizations with the same model and scenario. By definition, for a given GHG forcing scenario the ensemble average filters out the uncertainty, or noise, associated with model configuration and internal variability and allows us to extract the mean GHG-forced signal.

[9] Our procedure for calculating the PSPOTS TOE thus consists of the following steps. First, based on ensemble average changes of precipitation, the most prominent regional PSPOTS are identified (see next section). We are here mostly interested in the sub-continental scale, for which current AOGCMs can be expected to be relatively skillful [Christensen et al., 2007]. The PSPOTS are based on ensemble mean changes in 20-year average precipitation, a standard length period sufficient to filter out interannual variability but retain multi-decadal variability.

[10] The mean precipitation changes thus defined measure the forced signal for a given GHG forcing scenario. In order to measure the uncertainty, or noise, due to inter-model spread and internal multi-decadal variability we proceed as follows. We first separate the calculations of the variance due to inter-model configuration and to internal variability. For the inter-model configuration, the variance is calculated using the individual model changes after ensemble averaging over the different realizations for each model. Concerning internal model variability, for each model that completed multiple realizations we calculate the difference between the change in each realization (compared to the model ensemble average reference period, see below) and the ensemble average change. This defines an anomaly in the change due to internal multi-decadal variability. The variance is then approximated by adding the square of these anomalies over all models with multiple realizations and dividing by the total number of anomalies. In this way the information is compounded across all models performing multiple realizations, although it is most affected by the models with a larger number of realizations. The inter-model and internal variability variances are then added and the uncertainty is measured by the root mean square of this total variance, i.e., by the total uncertainty STD. Although we only present here the total STD values, we stress that the contribution of inter-model spread to the precipitation change STD is substantially larger (generally about one order of magnitude) than that of internal multi-decadal variability.

[11] After having defined the measures of signal and noise, for each PSPOT we calculate the running temporal average of 20-year ensemble mean change of precipitation and associated total STD of the 20-year changes. This results in a yearly time series of mean 20-year changes and associated uncertainty STD, where for each year of the time series the running average is taken over the previous 20 years. The reference period for the calculation of the changes is 1980–1999, thus for example the change value at year 2045 is the difference between the periods 2026–2045 and 1980–1999. Once the time series of ensemble mean changes and corresponding total STD are calculated, the TOE is defined as the time at which the magnitude of the mean change becomes greater than that of the STD and remains so thereafter, i.e., when the signal-to-noise ratio becomes greater than one. Note that we take as a measure of noise the total STD defined above which implies, within the limits of this definition, a likelihood of “emergence” of about 84%. More stringent thresholds could obviously be assumed, which would push the TOE later in time.

3. Identification of PSPOTS

[12] The identification of PSPOTS is here based on changes in mean precipitation for the 6-month periods April–September (AS) and October–March (OM). By using two 6-month periods we consider the entire annual precipitation and all rainy seasons for the different regions. However, essentially the same PSPOTS appear when considering the December–January–February and June–July–August values [Christensen et al., 2007]. Figures 1a and 1b show the CMIP3 ensembles average OM and AS precipitation change for 2081–2100 with respect to 1980–1999 in the A1B scenario. Based on Figures 1a and 1b, 14 PSPOTS are subjectively identified as defined in Table 2 and shown by the boxes in Figures 1a and 1b. Most of them encompass at least one sub-area with a change of magnitude exceeding 20% over land and are essentially the same when using the B1 and A2 scenarios (not shown).

Figure 1.

Ensemble average precipitation change in (a) October–March (OM) and (b) April–September (AS), for 2081–2100 with respect to 1980–1999 in the A1B scenario. The boxes with white contours indicate the PSPOTS (land only) described in Table 2. Units are % of 1980–1999 value.

Table 2. PSPOTS From Figure 1, Their Latitudinal and Longitudinal Extent (Land Only), and Their TOE for the B1, A1B and A2 IPCC Emission Scenariosa
  • a

    AS is April–September, OM is October–March.

NEU-OM50 N – 70 N10.5 W – 40.5 E<2020<2020<2020
MED-AS30 N – 48 N10.5 W – 38.5 E206120352034
MED-OM25 N – 43 N10.5 W – 40.5 E203520312038
NAS-OM35 N – 70 N85.5 E – 140.5 E<2020<2020<2020
CHN-AS10 N – 50 N100.5 E – 140.5 E204820482061
IND-AS5 N – 33 N64.5 E – 100.5 E207220542066
EAF-OM5 S – 12 N27.5 E – 52.5 E204620352029
SAF-AS35 S – 12 S9.5 E – 40.5 E>21002046>2100
NAM-OM40 N – 70 N170.5 W – 49.5 W<2020<2020<2020
WUS-AS30 N – 50 N125.5 W – 112.5 W>2100>21002093
CAM-OM15 N – 35 N121.5 W – 97.5 W>2100>21002067
CAR-AS10 N – 25 N97.5 W – 64.5 W>210020492077
AMZ-AS23 S – 2 S58.5 W – 35.5 W>2100>2100>2100
WAU-AS40 S – 27 S113.5 E – 154.5 E>2100>2100>2100

[13] The PSPOTS in Figure 1 are broadly consistent with what found in previous analyses [Giorgi and Bi, 2005], although they are more targeted to the specific GHG-induced precipitation change patterns. They include three European PSPOTS, i.e., a positive (increasing precipitation) one in Northern Europe during the cold season (NEU-OM) and negative ones (decreasing precipitation) over the southern Mediterranean in the cold season (MED-OM) and the Mediterranean/West European region in the warm season (MED-AS). These PSPOTS are consistent with a latitude-seasonal oscillation of the European climate change signal identified by Giorgi and Coppola [2007]. Other high-latitude cold season positive PSPOTS include Northern Asia (NAS-OM) and Northern North America (NAM-OM). They mostly result from the northward migration of mid-latitude storm tracks [Christensen et al., 2007].

[14] Figure 1 also shows positive PSPOTS corresponding to the intensification of the Indian (IND-AS) and China (CHN-AS) monsoon rain-belts, although the projected intensification of these monsoon rain systems must be taken with care due to the lack or crude representation of black carbon effects in the models [Meehl et al., 2008]. In Africa we find two PSPOTS, a positive one for the East Africa highlands short rainy season (EAF-OM) and a negative one over southern Africa in the cold/dry season (SAF-AS). In the Americas we identify negative (dry) PSPOTS over the western U.S. in summer (WUS-AS) [Diffenbaugh et al., 2008], Central America (CAM-OM) [Rauscher et al., 2008], the Caribbean region (CAR-AS) and the eastern portion of the Amazon basin (AMZ-AS) [Christensen et al., 2007]. Finally, a negative PSPOT is identified over Southern Australia (SAU-AS).

[15] Note that, although the PSPOTS are identified from the precipitation change signal, they broadly correspond to sub-continental scale climate zones, since they are related to shifts and/or changes in intensity of large scale circulation features which tend to define large scale climate zones. Of course, substantial climate variability may occur at smaller spatial scales due to regional and local processes, but this is not captured by the CMIP3 models. In the remainder of this paper, all calculations refer to averages over the regions of Figure 1 and Table 2 for 20-year running periods and including only land points in the reference 1-degree land mask grid mentioned above. Note that because the different models have different resolutions and land masks, this adds a certain element of uncertainty especially in areas of complex coastlines, which is however not important in view of the fact that we are considering broad regions.

4. Results

[16] Figure 2 shows examples of the temporal evolution of the 20 year running mean precipitation change and total STD for six PSPOTS and all three scenarios. The TOE is identified as the year at which the change and STD lines cross and the magnitude of the change remains greater than that of the STD thereafter. In some cases the STD is always greater than the change within the 21st century, i.e., the TOE possibly occurs after 2100, while in others the magnitude of the change is always grater than that of the STD, i.e., the TOE occurs in the first 20 years of the century. Table 2 reports the TOE for all regions and scenarios.

Figure 2.

Time evolution of 20-year mean change (solid lines) and inter-model STD (dashed lines) over 6 PSPOTS for the A2 (red), A1B (blue) and B1 (green) emission scenarios. The vertical bars indicate the TOE (see text), except when this occurs earlier than 2020 (left pointing arrows) or eventually later than 2100 (right pointing arrows). For regions where the mean change is negative, the STD is also given a negative value for easier comparison.

[17] The case NEU-OM is representative of high latitude regions (i.e., also NAS-OM and NAM-OM) and shows a continuous increase of the positive change signal in all scenarios, while the STD remains relatively constant. The change is greater than the STD early in the century and remains so thereafter so that the TOE is less than or about 2020 for all scenarios. Over the MED-AS PSPOT we find a strong decrease of precipitation throughout the century, especially in the A2 and A1B scenarios, while the STD magnitude increases in time much less markedly. Thus in these scenarios the TOE occurs early in the century. For the B1 scenario the change is smaller and the TOE occurs around 2061. We note however that, since global mean warming is close across the three scenarios in the early decades of the 21st century, this larger TOE in the B1 scenario over the MED-AS region may also be influenced by sampling uncertainty. A similar trend is found for the MED-OM, where however the TOE occurs early in the century for all scenarios (Table 2).

[18] Over the EAF-OM region (Figure 2) the positive precipitation change signal increases rapidly in all scenarios. The STD also increases with time, but remains lower than the signal for most of the century, so that the TOE occurs in the early 21st century decades. By contrast, over the SAF-AS region (Table 2) the STD has a magnitude greater than or similar to the change in the A2 and B1 scenarios, yielding a TOE > 2100, while the TOE occurs at about 2045 for the A1B scenario.

[19] The two Asia monsoon PSPOTS (Figure 2 and Table 2) show a somewhat similar behavior. The change signal is lower than the STD in the early decades of the century, but it grows more rapidly, becoming greater than the STD in the mid-century decades, when the TOE occurs. Over these two regions the TOE shows a substantial sensitivity to the emission scenario. Finally, in the American continent, the CAR-AS region (Figure 2) also shows a scenario-dependent TOE (similarly to the WUS-AS and CAM-OM, see Table 2) while for the AMZ-AS (Figure 2), as well as the WAU-AS (Table 2) regions, the STD is large and always greater than the signal.

[20] Figure 3 summarizes the results by clustering the different regions into those with a TOE mostly (i.e., at least in 2 out of three scenarios) in the early, mid and late (or eventually beyond) decades of the 21st century, respectively.

Figure 3.

Summary depiction of the TOE for the different regions of Figure 1. Red, yellow and light blue colors indicate early (up to 2040), mid (2040–2080) and late (beyond 2080) 21st century decade TOE, respectively. Boxes with solid (dashed) contours indicate PSPOTS with a positive (negative) precipitation change.

5. Summary and Discussion

[21] In this work we calculated the TOE of prominent GHG-forced precipitation change PSPOTS from the CMIP3 model ensemble. We do so by comparing the ensemble average precipitation change, which measures the GHG-forced signal, with the corresponding total STD of the change, which measures the uncertainty due to inter-model spread and internal multi-decadal variability. We find that for nine PSPOTS the TOE occurs in the early to mid-decades of the 21st century. Although the signal is large also for the other PSPOTS, it is still smaller than the total STD and thus the TOE does not appear during the 21st century.

[22] The primary implication of the TOE is on the detection and predictability of the GHG-forced PSPOTS. Our results show that, based on the CMIP3 model ensemble projections, only the NEU-OM, MED-AS, MED-OM, NAS-OM, EAF-AS and NAM-OM PSPOTS have a high likelihood (about 84 %) to be predicted, or detected, within the first few decades of the 21st century by current multi-AOGCM ensembles. This result is little sensitive to the underlying emission scenario. This likelihood of detection/predictability moves to the mid 21st century decades for the CAR-AS and the Indian and China monsoon PSPOTS, where we also find a greater TOE dependence on the emission scenario. The other GHG-forced PSPOTs do not have a high likelihood of detection or predictability within the 21st century.

[23] A second implication of our analysis is for impacts and adaptation studies. Our results provide indications on the time scales involved in the emergence of mean precipitation change signals and associated risks. This can provide useful information for the design and implementation of adaptation measures.

[24] We emphasize that this paper deals specifically with the identification and emergence of GHG-forced PSPOTS and not short term PSPOTS that might be related to initial conditions of the slow components of the climate system (e.g., the oceans). In fact, ocean-forced PSPOTS, even if modulated by the GHG forcing, might have completely different patterns compared to GHG-forced PSPOTS. Moreover, although the largest currently available multi-model ensemble, the CMIP3 is still relatively small and does not allow a full estimate of the inter-model and internal variability contributions to the multi-decadal uncertainty. Similarly, our definition of PSPOT is based on the mean precipitation change signal and not the change in other precipitation statistics, which might in fact show stronger responses. The availability of larger ensembles planned for the next generation model projections should allow a more robust estimate of the TOE of relevant GHG-forced PSPOTS worldwide.


[25] We acknowledge the international modeling groups for providing their data for analysis, PCMDI for collecting and archiving the model data, WGCM and their Coupled Model Intercomparison Project (CMIP) for organizing the model data analysis activity, and the IPCC WG1 TSU for technical support. The CMIP Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science, U.S. Department of Energy. We also thank two anonymous reviewers fro their useful comments which lead to important improvements in the paper.