10.3 Atmosphere and Surface
10.3.2 Water Cycle
10.3.2 Water Cycle
Detection and attribution studies of anthropogenic change in hydro-logic variables are challenged by the length and quality of observed data sets, and by the difficulty in simulating hydrologic variables in dynamical models. AR4 cautiously noted that the observed increase in atmospheric water vapour over oceans was consistent with warm-ing of SSTs attributed to anthropogenic influence, and that observed changes in the latitudinal distribution of precipitation, and increased incidence of drought, were suggestive of a possible human influence.
Many of the published studies cited in AR4, and some of the studies FAQ 10.1 (continued)
Overall, FAQ 10.1, Figure 1 shows that the pattern of observed temperature change is significantly different than the pattern of response to natural forcings alone. The simulated response to all forcings, including human-caused forcings, provides a good match to the observed changes at the surface. We cannot correctly simulate recent observed climate change without including the response to human-caused forcings, including greenhouse gases, stratospheric ozone, and aerosols. Natural causes of change are still at work in the climate system, but recent trends in temperature are largely attributable to human-caused forcing.
FAQ 10.1, Figure 1 | (Left) Time series of global and annual-averaged surface temperature change from 1860 to 2010. The top left panel shows results from two ensemble of climate models driven with just natural forcings, shown as thin blue and yellow lines; ensemble average temperature changes are thick blue and red lines.
Three different observed estimates are shown as black lines. The lower left panel shows simulations by the same models, but driven with both natural forcing and human-induced changes in greenhouse gases and aerosols. (Right) Spatial patterns of local surface temperature trends from 1951 to 2010. The upper panel shows the pattern of trends from a large ensemble of Coupled Model Intercomparison Project Phase 5 (CMIP5) simulations driven with just natural forcings. The bottom panel shows trends from a corresponding ensemble of simulations driven with natural + human forcings. The middle panel shows the pattern of observed trends from the Hadley Centre/Climatic Research Unit gridded surface temperature data set 4 (HadCRUT4) during this period.
1860 1880 1900 1920 1940 1960 1980 2000 Year
Temperature anomaly (°C)
Natural and Human forcing CMIP3
CMIP5 observations CMIP3 CMIP5 observations
1860 1880 1900 1920 1940 1960 1980 2000 Year
-0.5 0.0 0.5 1.0 1.5
Temperature anomaly (°C)
-0.5 0.0 0.5 1.0 1.5
Natural forcing
180 90W 0 90E 180
90S 45S 0 45N
90N Observed trend 1951-2010
180 90W 0 90E 180
90S 45S 0 45N
90N Natural and Human forcing
180 90W 0 90E 180
90S 45S 0 45N
90N Natural forcing
-2 -1 0 1 2
Trend (°C per period)
cited in this section, use less formal detection and attribution criteria than are often used for assessments of temperature change, owing to difficulties defining large-scale fingerprint patterns of hydrologic change in models and isolating those fingerprints in data. For example, correlations between observed hydrologic changes and the patterns of change in models forced by increasing GHGs can provide suggestive evidence towards attribution of change.
Since the publication of AR4, in situ hydrologic data sets have been reanalysed with more stringent quality control. Satellite-derived data records of worldwide water vapour and precipitation variations have
10
lengthened. Formal detection and attribution studies have been car-ried out with newer models that potentially offer better simulations of natural variability. Reviews of detection and attribution of trends in various components of the water cycle have been published by Stott et al. (2010) and Trenberth (2011b).
10.3.2.1 Changes in Atmospheric Water Vapour
In situ surface humidity measurements have been reprocessed since AR4 to create new gridded analyses for climatic research, as discussed in Chapter 2. The HadCRUH Surface Humidity data set (Willett et al., 2008) indicates significant increases in surface specific humidity between 1973 and 2003 averaged over the globe, the tropics, and the NH, with consistently larger trends in the tropics and in the NH during summer, and negative or non significant trends in relative humidity.
These results are consistent with the hypothesis that the distribution of relative humidity should remain roughly constant under climate change (see Section 2.5). Simulations of the response to historical anthropogenic and natural forcings robustly generate an increase in atmospheric humidity consistent with observations (Santer et al., 2007;
Willett et al., 2007; Figure 9.9). A recent cessation of the upward trend in specific humidity is observed over multiple continental areas in Had-CRUH and is also found in the European Centre for Medium range Weather Forecast (ECMWF) interim reanalysis of the global atmos-phere and surface conditions (ERA-Interim; Simmons et al. 2010). This change in the specific humidity trend is temporally correlated with a levelling off of global ocean temperatures following the 1997–1998 El Niño event (Simmons et al., 2010).
The anthropogenic water vapour fingerprint simulated by an ensemble of 22 climate models has been identified in lower tropospheric mois-ture content estimates derived from Special Sensor Microwave/Imager (SSM/I) data covering the period 1988–2006 (Santer et al., 2007).
Santer et al. (2009) find that detection of an anthropogenic response in column water vapour is insensitive to the set of models used. They rank models based on their ability to simulate the observed mean total column water vapour, and its annual cycle and variability associated with ENSO. They report no appreciable differences between the finger-prints or detection results derived from the best or worst performing models, and so conclude that attribution of water vapour changes to anthropogenic forcing is not sensitive to the choice of models used for the assessment.
In summary, an anthropogenic contribution to increases in specific humidity at and near the Earth’s surface is found with medium con-fidence. Evidence of a recent levelling off of the long-term surface atmospheric moistening trend over land needs to be better understood and simulated as a prerequisite to increased confidence in attribution studies of water vapour changes. Length and quality of observation-al humidity data sets, especiobservation-ally above the surface, continue to limit detection and attribution studies of atmospheric water vapour.
10.3.2.2 Changes in Precipitation
Analysis of CMIP5 model simulations yields clear global and region-al scregion-ale changes associated with anthropogenic forcing (e.g., Scheff and Frierson, 2012a, 2012b), with patterns broadly similar to those
identified from CMIP3 models (e.g., Polson et al., 2013). The AR4 con-cluded that ‘the latitudinal pattern of change in land precipitation and observed increases in heavy precipitation over the 20th century appear to be consistent with the anticipated response to anthropogenic forc-ing’. Detection and attribution of regional precipitation changes gen-erally focuses on continental areas using in situ data because observa-tional coverage over oceans is limited to a few island stations (Arkin et al., 2010; Liu et al., 2012; Noake et al., 2012) , although model-data comparisons over continents also illustrate large observational uncer-tainties (Tapiador, 2010; Noake et al., 2012; Balan Sarojini et al., 2012;
Polson et al., 2013). Available satellite data sets that could supplement oceanic studies are short and their long-term homogeneity is still unclear (Chapter 2); hence they have not yet been used for detection and attribution of changes. Continuing uncertainties in climate model simulations of precipitation make quantitative model/data compari-sons difficult (e.g., Stephens et al., 2010), which also limits confidence in detection and attribution. Furthermore, sparse observational cover-age of precipitation across much of the planet makes the fingerprint of precipitation change challenging to isolate in observational records (Balan Sarojini et al., 2012; Wan et al., 2013).
Considering just land regions with sufficient observations, the largest signal of differences between models with and without anthropogenic forcings is in the high latitudes of the NH, where increases in precip-itation are a robust feature of climate model simulations (Scheff and Frierson, 2012a, 2012b). Such increases have been observed (Figure 10.10) in several different observational data sets (Min et al., 2008a;
Noake et al., 2012; Polson et al., 2013), although high-latitude trends vary between data sets and with coverage (e.g., Polson et al., 2013).
Attribution of zonally averaged precipitation trends has been attempt-ed using different observational products and ensembles of forcattempt-ed simulations from both the CMIP3 and CMIP5 archives, for annu-al-averaged (Zhang et al., 2007; Min et al., 2008a) and season-spe-cific (Noake et al., 2012; Polson et al., 2013) results (Figure 10.11).
Zhang et al. (2007) identify the fingerprint of anthropogenic chang-es in observed annual zonal mean precipitation averaged over the periods 1925–1999 and 1950–1999, and separate the anthropogenic fingerprint from the influence of natural forcing. The fingerprint of external forcing is also detected in seasonal means for boreal spring in all data sets assessed by Noake et al. (2012), and in all but one data set assessed by Polson et al. (2013) (Figure 10.11), and in boreal winter in all but one data set (Noake et al., 2012), over the period 1951–1999 and to 2005. The fingerprint features increasing high-lati-tude precipitation, and decreasing precipitation trends in parts of the tropics that are reasonably robustly observed in all four data sets con-sidered albeit with large observational uncertainties north of 60°N (Figure 10.11). Detection of seasonal-average precipitation change is less convincing for June, July, August (JJA) and September, October, November (SON) and results vary with observation data set (Noake et al., 2012; Polson et al., 2013). Although Zhang et al. (2007) detect anthropogenic changes even if a separate fingerprint for natural forc-ings is considered, Polson et al. (2013) find that this result is sensi-tive to the data set used and that the fingerprints can be separated robustly only for the data set most closely constrained by station data.
The analysis also finds that model simulated precipitation variability is smaller than observed variability in the tropics (Zhang et al., 2007;
10
1950 1970 1990 2010
-0.25 0.00 0.25
Global
0.00
1950 1970 1990 2010
-0.25 0.25
60N-90N
1950 1970 1990 2010
-0.25 0.00 0.25
30N-60N Land masked by Obs
1950 1970 1990 2010
-0.25 0.00 0.25
30S-30N
AllNat Obs
Figure 10.10 | Global and zonal average changes in annual mean precipitation (mm day–1) over areas of land where there are observations, expressed relative to the base-line period of 1961–1990, simulated by CMIP5 models forced with both anthropogenic and natural forcings (red lines) and natural forcings only (blue lines) for the global mean and for three latitude bands. Multi-model means are shown in thick solid lines. Observa-tions (gridded values derived from Global Historical Climatology Network station data, updated from Zhang et al. (2007) are shown as a black solid line. An 11-year smoothing is applied to both simulations and observations. Green stars show statistically significant changes at 5% level (p value <0.05) between the ensemble of runs with both anthropo-genic and natural forcings (red lines) and the ensemble of runs with just natural forcings (blue lines) using a two-sample two-tailed t-test for the last 30 years of the time series.
(From Balan Sarojini et al., 2012.) Results for the Climate Research Unit (CRU) TS3.1 data set are shown in Figure 10.A.2.
Polson et al., 2013) which is addressed by increasing the estimate of variance from models (Figure 10.11).
Another detection and attribution study focussed on precipitation in the NH high latitudes and found an attributable human influence (Min et al., 2008a). Both Min et al. (2008a) and Zhang et al. (2007) find that the observed changes are significantly larger than the model simulated changes. However, Noake et al. (2012) and Polson et al. (2013) find that the difference between models and observations decreases if changes
are expressed as a percentage of climatological precipitation and that the observed and simulated changes are largely consistent between CMIP5 models and observations given data uncertainty. Use of addi-tional data sets illustrates remaining observaaddi-tional uncertainty in high latitudes of the NH (Figure 10.11). Regional-scale attribution of pre-cipitation change is still problematic although regional climate models have yielded simulations consistent with observed wintertime changes for northern Europe (Bhend and von Storch, 2008; Tapiador, 2010).
Precipitation change over ocean has been attributed to human influ-ence by Fyfe et al. (2012) for the high-latitude SH in austral summer, where zonally averaged precipitation has declined around 45°S and increased around 60°S since 1957, consistent with CMIP5 historical simulations, with the magnitude of the half-century trend outside the range of simulated natural variability. Confidence in this attribution result, despite limitations in precipitation observations, is enhanced by its consistency with trends in large-scale sea level pressure data (see Section 10.3.3).
In summary, there is medium confidence that human influence has con-tributed to large-scale changes in precipitation patterns over land. The expected anthropogenic fingerprints of change in zonal mean precip-itation—reductions in low latitudes and increases in NH mid to high latitudes—have been detected in annual and some seasonal data.
Observational uncertainties including limited global coverage and large natural variability, in addition to challenges in precipitation modeling, limit confidence in assessment of climatic changes in precipitation.
10.3.2.3 Changes in Surface Hydrologic Variables
This subsection assesses recent research on detection and attribu-tion of long-term changes in continental surface hydrologic variables, including soil moisture, evapotranspiration and streamflow. Stream-flows are often subject to large non-climatic human influence, such as diversions and land use changes, that must be accounted for in order to attribute detected hydrologic changes to climate change. Cryospher-ic aspects of surface hydrology are discussed in Section 10.5; extremes in surface hydrology (such as drought) and precipitation are covered in Section 10.6.1. The variables discussed here are subject to large mod-eling uncertainties (Chapter 9) and observational challenges (Chapter 2), which in combination place severe limits on climate change detec-tion and attribudetec-tion.
Direct observational records of soil moisture and surface fluxes tend to be sparse and/or short, thus limiting recent assessments of change in these variables (Jung et al., 2010). Assimilated land surface data sets and new satellite observations (Chapter 2) are promising tools, but assessment of past and future climate change of these variables (Hoekema and Sridhar, 2011) is still generally carried out on derived quantities such as the Palmer Drought Severity Index, as discussed more fully in Section 10.6.1. Recent observations (Jung et al., 2010) show regional trends towards drier soils. An optimal detection analysis of reconstructed evapotranspiration identifies the effects of anthro-pogenic forcing on evapotranspiration, with the Centre National de Recherches Météorologiques (CNRM)-CM5 model simulating chang-es consistent with those chang-estimated to have occurred (Douville et al., 2013).
10
−40 −20 0 20 40 60
−6
−4
−2 0 2 4 6
Latitude
Precipitation trend (% per decade)
JJA (f)
−40 −20 0 20 40 60
−6
−4
−2 0 2 4 6
Latitude
Precipitation trend (% per decade)
DJF (g)
−40 −20 0 20 40 60
−6
−4
−2 0 2 4 6
Latitude
Precipitation trend (% per decade)
ANNUAL (e)
Obs(V) Obs(Z) Obs(C) Obs(G) MM ZZ C Z V G
−4
−2 0 2 4 6 8 10
Scaling factor
(a)
1−sig 2−sig
CZVG CZVG CZVG CZVG Observation dataset
(b)
DJF MAM JJA SON
ALLANT NAT
Z Z (c)
HH (d)
Figure 10.11 | Detection and attribution results for zonal land precipitation trends in the second half of the 20th century. (Top left) Scaling factors for precipitation changes. (Top right and bottom) Zonally averaged precipitation changes over continents from models and observations. (a) Crosses show the best-guess scaling factor derived from multi-model means. Thick bars show the 5 to 95% uncertainty range derived from model-simulated variability, and thin bars show the uncertainty range if doubling the multi-model variance.
Red bars indicate scaling factors for the estimated response to all forcings, blue bars for natural-only forcing and brown bars for anthropogenic-only forcing. Labels on the x-axis identify results from four different observational data sets (Z is Zhang et al. (2007), C is Climate Research Unit (CRU), V is Variability Analyses of Surface Climate Observations (Vas-ClimO), G is Global Precipitation Climatology Centre (GPCC), H is Hadley Centre gridded data set of temperature and precipitation extremes (HadEX)). (a) Detection and attribution results for annual averages, both single fingerprint (“1-sig”; 1950–1999) and two fingerprint results (“2-sig”; Z, C, G (1951–2005), V (1952–2000)). (b) Scaling factors resulting from single-fingerprint analyses for seasonally averaged precipitation (Z, C, G (1951–2005), V (1952–2000); the latter in pink as not designed for long-term homogeneity) for four different seasons. (c) Scaling factors for spatial pattern of Arctic precipitation trends (1951–1999). (d) Scaling factors for changes in large-scale intense precipitation (1951–1999).
(e) Thick solid lines show observed zonally and annually averaged trends (% per decade) for four different observed data sets. Corresponding results from individual simulations from 33 different climate models are shown as thin solid lines, with the multimodel mean shown as a red dashed line. Model results are masked to match the spatial and temporal coverage of the GPCC data set (denoted G in the seasonal scaling factor panel). Grey shading indicates latitude bands within which >75% of simulations yield positive or negative trends. (f, g) Like (e) but showing zonally averaged precipitation changes for (f) June, July, August (JJA) and (g) December, January, February (DJF) seasons. Scaling factors (c) and (d) adapted from Min et al. (2008a) and Min et al. (2011), respectively; other results adapted from Zhang et al. (2007) and Polson et al. (2013).
Trends towards earlier timing of snowmelt-driven streamflows in west-ern North America since 1950 have been demonstrated to be differ-ent from natural variability (Hidalgo et al., 2009). Similarly, internal variability associated with natural decade-scale fluctuations could not account for recent observed declines of northern Rocky Mountain streamflow (St Jacques et al., 2010). Statistical analyses of stream-flows demonstrate regionally varying changes that are consistent with changes expected from increasing temperature, in Scandinavia (Wilson et al., 2010), Europe (Stahl et al., 2010) and the USA (Krakauer and Fung, 2008; Wang and Hejazi, 2011). Observed increases in Arctic river
discharge, which could be a good integrator for monitoring changes in precipitation in high latitudes, are found to be explainable only if model simulations include anthropogenic forcings (Min et al., 2008a).
Barnett et al. (2008) analysed changes in the surface hydrology of the western USA, considering snow pack (measured as snow water equivalent), the seasonal timing of streamflow in major rivers, and average January to March daily minimum temperature over the region, the two hydrological variables they studied being closely related to temperature. Observed changes were compared with the output of a
10
regional hydrologic model forced by the Parallel Climate Model (PCM) and Model for Interdisciplinary Research On Climate (MIROC) climate models. They derived a fingerprint of anthropogenic changes from the two climate models and found that the observations, when projected onto the fingerprint of anthropogenic changes, show a positive signal strength consistent with the model simulations that falls outside the range expected from internal variability as estimated from 1600 years of downscaled climate model data. They conclude that there is a detectable and attributable anthropogenic signature on the hydrology of this region.
In summary, there is medium confidence that human influence on climate has affected stream flow and evapotranspiration in limited regions of middle and high latitudes of the NH. Detection and attribu-tion studies have been applied only to limited regions and using a few models. Observational uncertainties are large and in the case of evap-otranspiration depend on reconstructions using land surface models.
10.3.3 Atmospheric Circulation and Patterns of Variability