3.3.4 Changes in Soil Moisture, Drought, Runoff and River Discharge
Historical records of soil moisture content measured in situ are available for only a few regions and often are very short (Robock et al., 2000). A rare 45-year record of soil moisture over agricultural areas of the Ukraine shows a large upward trend, which was stronger during the first half of the period (Robock et al., 2005). Among over 600 stations from a large variety of climates, including the former Soviet Union, China, Mongolia, India and the USA, Robock et al. (2000) showed an increasing long-term trend in surface (top 1 m) soil moisture content during summer for the stations with the longest records.
One method to examine long-term changes in soil moisture uses calculations based on formulae or LSMs. Since the in situ observational record and global estimates of remotely sensed soil moisture data are limited, global soil moisture variations during the 20th century have been estimated by LSM simulations. However, the results depend critically on the ‘forcings’ used, namely the radiation (clouds), precipitation, winds and other weather variables, which are not sufficiently reliable to determine trends. Consequently the estimates based on simulations disagree. Instead, the primary approach has been to calculate Palmer Drought Severity Index (PDSI; see Box 3.1) values from observed precipitation and temperature (e.g., Dai et al., 2004a). In some locations, much longer proxy extensions have been derived from earlier tree ring data (see Section 6.6.1; e.g., Cook et al., 1999). The longer instrumental-based PDSI estimations are used to look at trends and some recent extreme PDSI events in different regions are placed in a longer-term context (see specific cases in Section 3.8, `). As with LSM-based studies, the version of the PDSI used is crucial, and it can partly determine some aspects of the results found (Box 3.1).
Box 3.1: Drought Terminology and Determination
In general terms, drought is a ‘prolonged absence or marked deficiency of precipitation’, a ‘deficiency of precipitation that results in water shortage for some activity or for some group’ or a ‘period of abnormally dry weather sufficiently prolonged for the lack of precipitation to cause a serious hydrological imbalance’ (Heim, 2002). Drought has been defined in a number of ways. ‘Agricultural drought’ relates to moisture deficits in the topmost one metre or so of soil (the root zone) that impact crops, ‘meteorological drought’ is mainly a prolonged deficit of precipitation, and ‘hydrologic drought’ is related to below-normal streamflow, lake and groundwater levels.
Drought and its severity can be numerically defined using indices that integrate temperature, precipitation and other variables that affect evapotranspiration and soil moisture. Several indices in different countries assess precipitation deficits in various ways, such as the Standardized Precipitation Index. Other indices make use of additional weather variables. An example is the Keetch-Byrum Drought Index (Keetch and Byrum, 1988), which assesses the severity of drought in soils based on rainfall and temperature estimates to assess soil moisture deficiencies. However, the most commonly used index is the PDSI (Palmer, 1965; Heim, 2002) that uses precipitation, temperature and local available water content data to assess soil moisture. Although the PDSI is not an optimal index, since it does not include variables such as wind speed, solar radiation, cloudiness and water vapour, it is widely used and can be calculated across many climates as it requires only precipitation and temperature data for the calculation of potential evapotranspiration (PET) using Thornthwaite’s (1948) method. Because these data are readily available for most parts of the globe, the PDSI provides a measure of drought for comparison across many regions.
However, PET is considered to be more reliably calculated using Penman (1948) type approaches that incorporate the effects of wind, water vapour and solar and longwave radiation. In addition, there has been criticism of most Thornthwaite-based estimates of the PDSI because the empirical constants have not been re-computed for each climate (Alley, 1984). Hence, a self-calibrating version of the PDSI has recently been developed to ensure consistency with the climate at any location (Wells et al., 2004). Also, studies that compute changes or trends in the PDSI effectively remove influences of biases in the absolute values. As the effects of temperature anomalies on the PDSI are small compared to precipitation anomalies (Guttman, 1991), the PDSI is largely controlled by precipitation changes.
Using the PDSI, Dai et al. (2004a) found a large drying trend over NH land since the middle 1950s, with widespread drying over much of Eurasia, northern Africa, Canada and Alaska. In the SH, land surfaces were wet in the 1970s and relatively dry in the 1960s and 1990s, and there was a drying trend from 1974 to 1998 although trends over the entire 1948 to 2002 period were small. Overall patterns of trends in the PDSI are given in FAQ 3.2, Figure 1. Although the long-term (1901–2004) land-based precipitation trend shows a small increase (Figure 3.12), decreases in land precipitation in recent decades are the main cause for the drying trends, although large surface warming during the last two to three decades has likely contributed to the drying. Dai et al. (2004a) showed that globally, very dry areas (defined as land areas with a PDSI of less than –3.0) more than doubled (from ~12 to 30%) since the 1970s, with a large jump in the early 1980s due to an ENSO-related precipitation decrease over land and subsequent increases primarily due to surface warming. However, results are dependent on the version of the PDSI model used, since the empirical constants used in a global PDSI model may not be adequately adjusted for the local climate (see Box 3.1).
In Canada, the summer PDSI averaged for the entire country indicates dry conditions during the 1940s and 1950s, generally wet conditions from the 1960s to 1995, but much drier conditions after 1995 (Shabbar and Skinner, 2004) with a relationship between recent increasing summer droughts and the warming trend in SST. Groisman et al. (2007) found increased dryness based on the Keetch-Byrum forest-fire drought index in northern Eurasia, a finding supported by Dai et al. (2004a) using the PDSI. Long European records (van der Schrier et al., 2006) reveal no trend in areas affected by extreme PDSI values (thresholds of either ±2 or ±4) over the 20th century. Nevertheless, recently Europe has suffered prolonged drought, including the 2003 episode associated with the severe summer heat wave (see Section 3.8.4, Box 3.6).
Although there was no significant trend from 1880 to 1998 during summer (JJA) in eastern China, precipitation for 1990 to 1998 was the highest on record for any period of comparable length (Gong and Wang, 2000). Zou et al. (2005) found that for China as a whole there were no long-term trends in the percentage areas of droughts (defined as PDSI < –1.0) during 1951 to 2003. However, increases in drought areas were found in much of northern China (but not in northwest China; Zou et al., 2005), aggravated by warming and decreasing precipitation (Ma and Fu, 2003; Wang and Zhai, 2003), consistent with Dai et al. (2004a).
A severe drought affecting central and southwest Asia in recent years (see Section 3.8.4, Box 3.6) appears to be the worst since at least 1980 (Barlow et al., 2002). In the Sahel region of Africa, rainfall has recovered somewhat in recent years, after large decreasing rainfall trends from the late 1960s to the late 1980s (Dai et al., 2004b; see also Section 188.8.131.52 and Section 3.7.4, Figure 3.37). Large multi-year oscillations appear to be more frequent and extreme after the late 1960s than previously in the century. A severe drought affected Australia in 2002 and 2003; precipitation deficits were not as severe as during a few episodes earlier in the 20th century, but higher temperatures exacerbated the impacts (see Section 3.8.4, Box 3.6). There have been marked multi-year rainfall deficits and drought since the mid- to late-1990s in several parts of Australia, particularly the far southwest, parts of the southeast and along sections of the east coast.
A multi-decadal period of relative wetness characterised the latter portion of the 20th century in the continental USA, in terms of precipitation (Mauget, 2003a), streamflow (Groisman et al., 2004) and annual moisture surplus (precipitation minus potential evapotranspiration; McCabe and Wolock, 2002). Despite this overall national trend towards wetter conditions, a severe drought affected the western USA from 1999 to November 2004 (see Section 3.8.4, Box 3.6).
Available streamflow gauge records cover only about two- thirds of the global actively drained land areas and they often have gaps and vary in record length (Dai and Trenberth, 2002). Estimates of total continental river discharge are therefore often based on incomplete gauge records (e.g., Probst and Tardy, 1987, 1989; Guetter and Georgakakos, 1993), reconstructed streamflow time series (Labat et al., 2004) or methods to account for the runoff contribution from the unmonitored areas (Dai and Trenberth, 2002). These estimates show large decadal to multi-decadal variations in continental and global freshwater discharge (excluding groundwater; Guetter and Georgakakos, 1993; Labat et al., 2004).
Streamflow records for the world’s major rivers show large decadal to multi-decadal variations, with small secular trends for most rivers (Cluis and Laberge, 2001; Lammers et al., 2001; Mauget, 2003b; Pekárová et al., 2003; Dai et al., 2004a). Increased streamflow during the latter half of the 20th century has been reported over regions with increased precipitation, such as many parts of the USA (Lins and Slack, 1999; Groisman et al., 2004) and southeastern South America (Genta et al., 1998). Decreased streamflow was reported over many Canadian river basins during the last 30 to 50 years (Zhang et al., 2001b), where precipitation has also decreased during the period. Déry and Wood (2005) also found decreases in river discharge into the Arctic and North Atlantic Oceans from high-latitude Canadian rivers, with potential implications for salinity levels in these oceans and possibly the North Atlantic THC. These changes are consistent with observed precipitation decreases in high-latitude Canada from 1963 to 2000. Further, Milly et al. (2002) showed significant trends towards more extreme flood events from streamflow measurements in 29 very large basins, but Kundzewicz et al. (2005) found both increases (in 27 cases) and decreases (in 31 cases) as well as no significant (at the 10% level) long-term changes in annual extreme flows for 137 of the 195 rivers examined worldwide. Recent extreme flood events in central Europe (on the Elbe and some adjacent catchments) are discussed in Section 3.8.4, Box 3.6.
Large changes and trends in seasonal streamflow rates for many of the world’s major rivers (Lammers et al., 2001; Cowell and Stoudt, 2002; Ye et al., 2003; Yang et al., 2004) should be interpreted with caution, since many of these streams have been affected by the construction of large dams and reservoirs that increase low flow and reduce peak flow. Nevertheless, there is evidence that the rapid warming since the 1970s has induced earlier snowmelt and associated peak streamflow in the western USA (Cayan et al., 2001) and New England, USA (Hodgkins et al., 2003) and earlier breakup of river ice in Russian Arctic rivers (Smith, 2000) and many Canadian rivers (Zhang et al., 2001b).
River discharges in the La Plata River Basin in southeastern South America exhibit large interannual variability. Consistent evidence linking the Paraná and Uruguay streamflows and ENSO has been found (Bischoff et al., 2000; Camilloni and Barros, 2000, 2003; Robertson et al., 2001a; Berri et al., 2002; Krepper et al., 2003), indicating that monthly and extreme flows during El Niño are generally larger than those observed during La Niña events. For the Paraguay River, most of the major discharges at the Pantanal wetland outlet occurred in the neutral phases of ENSO, but in the lower reaches of the river the major discharge events occurred during El Niño events (Barros et al., 2004). South Atlantic SST anomalies also modulate regional river discharges through effects on rainfall in southeastern South America (Camilloni and Barros, 2000). The Paraná River shows a positive trend in its annual mean discharge since the 1970s in accordance with the regional rainfall trends (García and Vargas, 1998; Barros et al., 2000a; Liebmann et al., 2004), as do the Paraguay and Uruguay Rivers since 1970 (Figure 3.14).
For 1935 to 1999 in the Lena River Basin in Siberia, Yang et al. (2002) found significant increases in temperature and streamflow and decreases in ice thickness during the cold season. Strong spring warming resulted in earlier snowmelt with a reduced maximum streamflow pulse in June. During the warm season, smaller streamflow increases are related to an observed increase in precipitation. Streamflow in the Yellow River Basin in China decreased significantly during the latter half of the 20th century, even after accounting for increased human water consumption (Yu et al., 2004a). Temperatures have increased over the basin, but precipitation has shown no change, suggesting an increase in evaporation.
In Africa from 1950 to 1995, Jury (2003) found that the Niger and Senegal Rivers show the effects of the Sahel drying trend with a decreasing trend in flow. The Zambezi also exhibits reduced flows, but rainfall over its catchment area appears to be stationary. Other major African rivers, including the Blue and White Nile, Congo and inflow into Lake Malawi show high variability, consistent with interannual variability of SSTs in the Atlantic, Indian and Pacific Oceans. A composite index of streamflow for these rivers shows that the five highest flow years occurred prior to 1979, and the five lowest flow years occurred after 1971.