Fig. 123. Mean annual SAT recorded at meteorological station Fichtelberg (Germany) during 1891-2004 and its 10-year running average. Solid lines represent the mean warming rates for the whole observation period and for years 1980-2004.
increased by approximately 1.2-2K above the average level for 1950-1985 (Borchert, 2005). This temperature course is quantitatively similar to the global temperature change. However, while the twentieth century global warming rate coincides well with the one observed at the Fichtelberg weather station, the rate of the recent warming in the latter location is approximately four times higher (Figure 4, Chapter 1). The supposed cause of the global warming is the combined effect of the anthropogenic activity and the natural forcing (see Section 3.4, Chapter 3), and it is very enticing to represent an increased warming trend in the study area as man-made, caused by extensive industrial activity. On the other hand, investigations by Borchert (2005, 2006) have revealed significant correlation between recent climate warming and air pollution characteristics with increasing Sun activity, which were observed in Central Europe, represented by increasing sunspot number and flare intensities as well as by decreasing cosmic radiation4 (neutron rates) resulting in reduced cloudiness and corresponding increase of intensity and duration of sunshine. On the basis of the detected correlations, the author has proposed extraterrestrial and not anthropogenic causes for the recent temperature increase in Central Europe, when the transportation and concentration of air pollution may also be strongly affected by external effects. Further studies (including prolonged temperature monitoring) can make causal connections of the recent warming clearer.
The success of the monitoring experiments performed in the Czech Republic has inspired an establishment of the joint international monitoring project in the Czech Republic, Slovenia, and Portugal (Safanda et al., 2006). For all the three experiments, 100-200m deep holes were chosen. Because a thorough thermal equilibrium is an essential requirement to obtain an undisturbed temperature time series suitable for the climate study, only old boreholes that have already achieved thermal equilibrium were used for monitoring experiments. Repeatedly measured temperature-depth profiles revealed fingerprints of an appreciable warming in the uppermost parts of all holes. Temperature monitoring began in the years 2002 (the Czech Republic), 2003 (Slovenia), and 2005 (Portugal) in several depths from 2 cm to 40 m. The air temperatures at 2 m and 5 cm above the ground surface were also measured. Preliminary comparative results are expected to be available in the end of the year 2006.
All above-described monitoring experiments have proved that the present-day warming corresponding to the last one to several decades can be reasonably well extracted by precise temperature monitoring at shallow boreholes below the depth of penetration of the seasonal variations. Of course, the detection of the linear trends in the monitoring data sets and their interpretation represent only prelude to the precise analysis of this data. The geothermal inverse theory (e.g., ramp/step method, see Eqs. (2.14) and (2.15), Section 2.3.3) can be used to quantify more precisely the amount and the onset time of the warming trend. Anyhow, even preliminary studies have confirmed the applicability of this kind of temperature measurements for the GST reconstruction. The temperature monitoring in shallow boreholes of 30-50m depths may be an alternative to the temperature log inversion, routine method of detecting local recent climate changes
4Cosmic radiation (cosmic rays) is a naturally occurring ionizing radiation coming outside the Earth and filtering through atmosphere. A significant amount of these high-energy particles is discharged by the Sun. Scientists have argued that cosmic radiation can cause the changes in weather, e.g., can cause clouds to form in the upper atmosphere. The cosmic radiation shows an inverse relationship with the sunspot cycle. The reason is that the Sun's magnetic field is stronger during sunspot maximum and shields the Earth from cosmic rays.
and for the direct assessment of the present warming rate. Evidence obtained in the shallow subsurface by precise temperature monitoring can also satisfactorily complement meteorological observations.
Borehole temperature monitoring in the recent decade became one of the building blocks of the borehole research to help us to understand how the Earth's climate is changing. A heap of the monitoring experiments is performed all over the world. Except for the detection of the recent warming trends, numerous subsurface monitoring experiments were established for specific climatic applications. Of special interest are reports on the data from so far uncovered areas and attempts to separate the potential man-made components of the global warming from the natural climate variability. Below, we mention some of them.
4.2.1 Emigrant Pass Observatory, Utah
Over a decade-long ground temperature monitoring has been performed at the Emigrant Pass Observatory (EPO), Utah (Bartlett et al., 2004, 2006; Davis et al., 2006). To better understand the GST-SAT coupling and to document the details of the penetration of the surface signal into the ground, a climate and ground temperature observatory was installed in arid NW Utah in 1994. The EPO (41.50°N, 113.68°W, 1750m a.s.l.) consists of a standard weather station situated on exposed granitic rock at the top of a 150 m deep borehole (GC-1) drilled in 1978. Results of its repeated temperature logging are presented in Figure 16 (Chapter 1). Inversion of the measured T-z profiles inferred surface temperature changes that are closely coherent with those observed at the nearby meteorological station 40 km to the northeast (Chisholm and Chapman, 1992). The EPO consists of an array of thermistor strings in the subsurface. Ground temperatures are monitored at several shallow depths from 2.5 cm to 1 m. Meteorological and shallow ground variables are recorded simultaneously. All data from the EPO since November 2004 are available and can be found on the web site http://thermal.gg.utah.edu/facilities/ epo/EPO_data. The file is automatically updated daily. The combined database gives an opportunity to observe the GST-SAT dependence in near real time and to test theoretical models of the GST-SAT interactions. It can also be used for the investigation of the energy balance in the Earth's surface, reconstruction of the climate change from borehole temperatures, and other geothermal studies.
Over decade-long continuous temperature monitoring has shown that the subsurface temperatures at all monitored depths are in general agreement with the air temperature (e.g., correlation coefficients are 0.97 and 0.87 for air-10cm and air-1m depth temperatures, respectively). Except for the surface air temperature that explains 94% of the GST variance, the GST variations are influenced by incident solar radiation that accounts for 1.3% of the GST variance and by snow cover. Daily averaged GST-SAT differences range between +14 and -10K. Observed differences occur due to the solar radiation effect in the summer and the insulating effect of snow cover in the winter. They are much lower on the annual scale and vary between only 2.3 and 2.5 K. In this scale of aggregation, ground temperatures are generally warmer than air temperatures. Much of the inter-annual variations in the GST-SAT difference occur due to the changes in solar radiation. It was shown that incident solar radiation is more important during the summer. On the long scale there is a linear relationship between the GST and SAT difference and solar radiation with a slope of 1.21K/100W/m and intercept of 2.47K (Bartlett et al., 2006; Davis et al., 2006). Because of its low thermal diffusivity, snow attenuates surface temperature variations in the winter, but its insulating effect has only minor influence on the annual GST-SAT coupling at the EPO site (accounts for only 0.5% of the annual GST variance). Using EPO monitoring results, Bartlett et al. (2004) have developed two-layered forward numerical model of snow-ground interactions. Model is based on three characteristics of snow cover: (1) the onset time, (2) duration of the snow cover, and (3) its thickness. These parameters are generally available from meteorological and remotely sensed data, and the authors have validated their model using the National Weather Service data from 23 sites over North America. Their calculations have verified the applicability of the developed model for the broad spectrum of snow conditions and have confirmed its suitability for the prediction of the GST changes in different environments. On the whole, the EPO observations have shown that the GST really tracks the air temperature on the timescales relevant to the climate change studies. The GST reconstructions generally assume that the GST-SAT difference is constant over long timescales and thus the transient temperature changes at the ground surface reproduce the transient SAT changes measured at weather stations. The EPO monitoring results have warranted this assumption over the past decade and thus have given a serious experimental support for the use of the GST histories as a valuable addition to the SAT measurements and multiproxy reconstructions in climate change research.
In numerous regions, e.g., high elevation sites, islands, flat northern environments, etc., the surface air temperature may represent a complex result of an interaction of some climatic variables. The detection of the real warming trends in such areas and their separation from an impact of the short-term changes of other climatic variables may be quite difficult. Climate monitoring in such locations can help to filter out disturbing effects and identify long-term climatic trends. One of such monitoring efforts is being carried out at the Faroe Islands (Denmark). It is a small group of islands that is situated in the stormiest part of the North Atlantic, midway between Scotland and Iceland. The Faroe Islands are located in a key region for understanding land-atmosphere-ocean interaction in the North Atlantic region. Under the influence of the warm ocean current of the Gulf Stream, the climate is relatively mild for the latitude. On the other hand, because these islands lie in the path of the majority of Atlantic depressions, they are cloudy (daily sunshine in the summer months averages only about 4h), wet (annual precipitations ranges between 1500 and 2500 mm), and windy throughout the year. The air surface temperatures in the region strongly depend on the wind speed and direction as well as on the cloud and snow cover (Cappelen and Laursen 1998; Humlum and Christiansen, 1998). The Climate Research Station was situated at the summit of Sornfelli mountain (799 m a.s.l.) on the main island Streymoy, and the monitoring was started in November 1999. It is expected that this experiment will provide meteorological data on the arctic climate environment on the Faroes that can then be placed in a wider North Atlantic as well as the Northern hemisphere perspective. Traditionally, meteorological stations in the area are located near sea level, which makes studies of the vertical climate change effects problematic. Data from Sornfelli borehole can allow the calculations of climatic altitudinal gradients, which can then be used for additional studies such as the interpretation of the vegetational zona-tion, soil development and present periglacial processes, and their relation to the past climatic conditions. Because of the cold, wet, and windy climate, as well as expected heavy icing problems, the measurements with standard meteorological equipment are difficult on Sornfelli, if not impossible. Thus, a special meteorological station was constructed, using a drum-shaped housing and internal heating. For the subsurface temperature monitoring, an 11.32m deep borehole was drilled and thermocouples were installed. The station was redesigned in the spring of 2004, and new instruments and data logging equipment were installed in June 2004. Meteorological data are logged each 30min. Similar to the EPO data, the Sornfelli monitoring results are regularly published in the web site www.metsupport.dk/data/sornfelli. On-line borehole data are updated every hour except at night local time.
Kamchatka is a peninsula in the Russian Far East comparable in size to Japan and surrounded by the Pacific Ocean and the Bering and Okhotskoe Seas. Human intervention to the atmospheric temperature and environment is expected to be small in Kamchatka because of very sparse population and small industrial activities. The recent climatic trends detected there probably reflect natural climate variability characteristic for the Northern Pacific region and not an anthropogenic influence. As a part of the 3-year joint Japanese-Czech-Russian research project "Reconstruction of the climatic changes from borehole temperature profiles and tree rings in the Kamchatka Peninsula" (2000-2002), precise temperature measurements were performed in a number of holes (Yamano et al., 2002). This project primarily concentrated on obtaining precise temperature-depth profiles in a number of boreholes, drilled more than 15 years ago, and on verification of the previous measurements in the region. Temperature logs were repeatedly measured in 12 boreholes during 2000-2002 at intervals of a few months to one year (for details see Section 3.1.2, Chapter 3). Data were used to propose a climate model of the last 100-150 years (Figure 97, Chapter 3). All temperature logs have shown a general turn to the warmer conditions since approximately 1950. The most detailed GST history was inferred for a suite of boreholes at Malki location (53.33°N, 153.47°E). Climatic history shows warm period with the maximum near 1850, cold conditions culminating between 1920 and 1950 and pronounced warming of 1.2-1.6K since then. Obtained results are in good agreement with the existing SAT series. Jones et al. (1999) have presented global patterns of the surface temperature change over the past 150 years' combined land and marine data on the 5° X 5° grid box basis. Figure 98 (Chapter 3) shows one box of this database, namely, an estimate of the SAT changes for southern part of the Kamchatka Peninsula. The temperature anomaly time series, available back to the beginning of the twentieth century, exhibits high interannual temperature variation that somewhat attenuated between 1960 and 1990. Warming trend of 0.007 K/year calculated for the interval 1890-1998 is insignificant. The slow temperature rise with warming rate of 0.026 K/year occurred after approximately 1960-1963. It was followed by the marked period of warmth during the last 10-15 years of the record. Similar warming trends were obtained for the Pacific Ocean at latitudes 40-60°N and in eastern Siberia (Rogers and Mosley-Thompson, 1995).
It was this warming that Budyko (1977) and other climatologists have interpreted as the start of a new large-scale climate warming.
Except of the numerous temperature loggings, a high-resolution temperature monitoring of 1mK accuracy was performed at several selected depth levels in four boreholes (Yamano et al., 2002; see also Bodri and Cermak, 2005b and the references therein). Temperatures in two unstable wells in Elizovo (E-1) and Yugozapadnaya (UZ) sites were monitored at 325 and 108 m depth, respectively. Such high depths were chosen because the scatter of data during repeated borehole logging has exceeded any explainable differences due to instrumental incorrectness and/or field of technical problems. Data loggers for temperature monitoring were installed at the depths of maximum temperature gradient. Temperatures have shown high degree of irregularity over all measured periods within up to several hundreds of degree, but because of relatively large monitoring depths they did not exhibit any significant linear trend that could be attributed to the recent warming. However, monitoring was not performed in vain, because an analysis of these deep microtemperature time series has helped to quantify the stochastic heterogeneity of the borehole temperature signal and provided valuable information on the fine-scale features of the heat transfer process in different geological environments (see, e.g., Bodri and Cermak, 2005b). Temperature loggers were also installed at four shallower depths 25, 30, 35, and 40 m in two boreholes with more stable temperature-depth profiles (Malki-2 and Malki-19) for 10-11 months. Similar to monitoring results described above short-term temperature variations observable in the upper 10-20m depth interval of Malki holes significantly decayed in comparison with the surface temperature variations. On the other hand, recorded temperature time series have not shown any significant linear trend. Temperature has remained almost constant (within 2mK). The reason is that the most recent warming that began 10-15 years ago still not penetrated to 25-40m depth and insignificant warming trend characterized for the most of the twentieth century appeared too weak to be archived in the subsurface. For the same period, temperature was monitored at 50 and 100 cm below the ground surface in the close vicinity of Malki-12 and Malki-19 boreholes with lower accuracy of only 0.1 K. Analysis of these time series have shown that heat transfer in the uppermost ground is pure conductive at Malki-12 location, while small non-conductive component was detected at Malki-19 hole during February to May 2002. This non-conductive disturbance can be related to the freezing/thawing of the soil around 50 cm depth.
4.2.4 Livingston Island, Antarctic
At the polar and sub-polar environments there are large areas subjected to high energy transfer in the ground surface. To investigate the surface energy balance in such regions for prognostic research of climate change two shallow boreholes (1.1 and 2.4m) were drilled in the year 2000 in Livingston Island (South Shetlands, Antarctica; 62.65°S, 60.35°W). Temperature monitoring was performed in four depths during 2000 and 2001 (Ramos and Vieira, 2003; www.igme.es/internet/cnda). Temperature data were collected at 4-h intervals in several shallow depths. Because of quartzite bedrock setting, temperature time series were characterized by an absence of any trace of the phase change processes. Measurements have shown that the subsurface temperature regime is almost exclusively controlled by air temperature, which conductively penetrates to the depth with usual loss of amplitude and phase delay. The monitoring results at the Livingston Island's boreholes bottom have shown only seasonal variations. Future international project plans drilling and monitoring of two new 20 m deep boreholes in Livingston and Deception Islands.
Although shallow measurements described above can provide information about the surface ground temperature history only for the short time intervals comparable with the length of the time series, they are useful complements to the longer scale but less well-resolved GST histories inferred from the temperature logs measured in deeper holes. Current temperature monitoring experiments are performed in the single borehole sites. Once trends are detected and local characteristics and causes are identified, these results can be integrated into wider spatial scale network. These results can also be incorporated into other research fields or ecological issues, e.g., the environmental management.
4.3 Recent Climate Variability
For a better understanding of the nature of the climate change, attention is to be focused not only on the evolution of mean climate characteristics, but also on the changes in climate variability, and on climate extremes. The necessity of including the variability characteristics in the climate change studies has been demonstrated in several works (Katz and Brown, 1992; Wilks and Riha, 1996; Rebetez, 1996, 2001; Bodri, 2004; and the references therein). It can be demonstrated that the frequency of climatic extremes is more sensitive to the changes in variability rather than to the mean climate state (Katz and Brown, 1992). Increase or decrease in the frequency of extremes can be enormously large even for relatively small mean changes in climate (Katz, 1999). Rebetez (1996) has shown that climate variability is one of the most important characteristics in the human perception of climate. The potential response of the socio-economic fabrics of the global community to the changes in climate variability may be stronger than to the changes in climatic averages (Rebetez, 1996; Wilks and Riha, 1996), while these changes are completely obscured when examining only the evolution of mean characteristics.
In the everyday life, climate change and climate variability are often confused. In its exact mean climate (and any other real-valued random variable) variability refers to the spread of a data set. An assessing of variability generally includes two key components: (1) how spread out are the data values near the center, and (2) how spread out are its tails. The common definitions of the central value that best describes data are their mean, median, and mode. The common numerical measures of the spread are variance, standard deviation, range, average absolute deviation, etc. The changes that are greater than 4 standard deviations are generally referred as extreme events. When assessing variability, variations in the central (typical) state and the spread statistics of the climate should be detected on all temporal and spatial scales beyond that of individual weather events. An analysis of the climate and its variability from observed data is especially challenging in the case of a changing climate. An interaction between mean characteristics of climate and its variability and extremes depends on the statistical distribution of given climatic variable (Meehl et al., 2000). Possible influence of the changes in the mean and variability on climate is illustrated in Figure 124. The climatic temperatures
Fig. 124. Effect of the change in the mean and in the variance for the standard normal distribution of temperature. "Previous climate" curve corresponds to mean = 0 and variance = 1. "New climate" is calculated for the next cases: (A) mean temperature increases (mean = 1), (B) variance of temperature increases (variance = 2), and (C) both characteristics increase (mean = 1, variance = 2).
Fig. 124. Effect of the change in the mean and in the variance for the standard normal distribution of temperature. "Previous climate" curve corresponds to mean = 0 and variance = 1. "New climate" is calculated for the next cases: (A) mean temperature increases (mean = 1), (B) variance of temperature increases (variance = 2), and (C) both characteristics increase (mean = 1, variance = 2).
often have a normal distribution ("the bell curve"). The non-stationarity of this distribution implies changes in the mean temperature and/or its variance. Increase in the mean temperature gives more warm conditions and less cold weather (Figure 124, panel A). However, it does not produce any change in climate variability. In other words, the range between the warmest and coldest temperatures does not change. The change in extremes occurs only due to a shift in the distribution without a change in its shape. This means that for the real situations prediction of changes in extremes there is no need of additional study of variability. They could be predicted simply from the changes in longer term monthly, seasonal, or annual means that are generally available (e.g., global grid-ded data by Jones et al., 1999).
On the contrary, an increase in variability without change in average temperature (panel B) produces the change in the shape of the probability distribution resulting in the same increase in the probability of both warm and cold extremes as well as increase in the absolute value of these extremes. Prediction of changes in extremes needs detection of changes in meteorological variables (e.g., indices of extremes) such that determination from station data is not a trivial task. Panels A and B illustrate that the global warming is not equivalent to climate change, and significant climate change can occur without any global warming or cooling. Increase in both characteristics of temperature distribution (panel C) results in an asymmetric increase of the probability of extremes producing more frequent warm events with more extreme hot temperatures. Its influence on cold extremes is far less pronounced. Figure 124 illustrates typical case of the global warming. Obviously, other combinations of changes in the mean temperature and its variability would lead to different patterns of the probability of cold and warm events occurrence. For the climatic variables that, like precipitation, are not well approximated by the normal distribution situation may be far more complex. Consequently, even when changes in temperature extremes were detected, their attribution to the changes in the mean or variance (or both) needs specific analysis and/or some kind of "key test" that may provide an idea on the degree of confidence associated with obtained conclusions.
The fact is that Earth's climate is always changing. It varies on a broad range of timescales and over many orders of magnitude. Climate oscillates on the millennial timescales between ice ages and interglacials causing global scale rearrangements of ice cover and ocean circulation. The shorter scales of its variation embrace periods from centuries like the Medieval Warm Period and the Little Ice Age to decades, as indicated by the temperature changes in the twentieth century. Generally the shorter the timescale, the stronger is the impact expected on a local spatial scale and the longer the timescale, the more is impact on the global scale and resulting socio-economic consequences. For example, long-term climate variations may alter agricultural productivity, land and marine ecosystems together with the resources that supply these ecosystems, while seasonal to interannual climate variations can strongly affect agriculture, the abundance of water resources as well as the demand of energy. On the other hand, different-scale climate variability modes cannot be treated separately. Results of the recent investigations increasingly support that short- and long-term climate variability are intrinsically linked.
The climate system is quite complex and highly non-linear. Expected modes of its variability are also complex. Variability may be due to natural internal processes within the climate system (internal variability) and/or variations in natural or anthropogenic external forcing (external variability). An overview of the natural climate variability and its causal mechanisms was presented in the pioneering work by Mitchell (1976). It was partly this work that inspired the U.S. National Geophysical Data Center (NGDC) to design an interactive web site "Climate Timeline Tool: What is Variability?" that helps to assess the basic processes and causes of climate variability (www.ngdc.noaa.gov/paleo/ctl/about1.html). This graphic demonstrates also the interactions of variability over varying timescales.
Significant climate variations are occurring within the diurnal scale to the 100000 timescale corresponding to orbital forcing. All these variations have occurred before any anthropogenic influence on the climate system could be in operation. The natural variability of the climatic system itself is quite high. Evidence for such intrinsic variability has been found in observations and coupled general circulation models (Delworth and Mann, 2000). The past three to five decades have seen an increasing recognition that human activities may have substantial effect on the climate system. Recent climate variability may be intensified by the human influence. Because this is one of the great, still unresolved problems of climate science, changes in climate variability and in both weather and climate extremes have received increased interest in the recent decades. Numerous research programs, such as the international program "Climate Variability and Predictability" (CLIVAR; www.clivar.org), the Climate Variability and Trends Group of the NOAA (U.S. National Oceanic and Atmospheric Administration) Air Resources Laboratory (www.arl.noaa.gov/ss/climate), Climate Variability Working Group (CVWG; www.ccsm.ucar.edu/working_groups/Variability/index.html) together with the Intergovermental Panel of Climate Change (IPCC; www.ipcc.ch), were put on operation. The general objective of these efforts is to describe and understand the physical processes responsible for climate variability and predictability on various scales through the collection and analysis of observations and the development and application of models of the climate system. This goal can be achieved in wide cooperation with other relevant climate research and observing programs.
Questions that could be addressed with the focused study of borehole temperature monitoring data include:
1. How does climate variability varied?
2. Are these changes consistent in the key regions?
3. In cases when reported variability changes appear to be contradictory, it should be examined where detected differences represent real regional variability or simply reflect the differences in quality of data and/or detection methods used.
The answers to these questions will come from the development of the monitoring network and the acquisition data having sufficient length and resolution to provide a base for variability studies. Results from intensive local investigations should be combined for the studies of regional variability change patterns. Future valuable outcome of such efforts may be monitoring time series database similar to already existing "Borehole Temperatures and Climate Reconstruction Database" initiated by the Geothermal Laboratory of the University of Michigan (www.geo.lsa.umich.edu/~climate). A systematic review and evaluation of existing data can produce a coherent and internally robust data that will serve as a base for the variability studies, revealing potential forcing mechanisms and modeling of not only a warmer but also more variable future world.
The Earth's climate system consists of a number of subsystems (atmosphere, hydrosphere, lithosphere, cryosphere, and biosphere) with their own characteristic times of operation from days to millennia. Each subsystem has its own internal variability mode, when some of its parameters change intensively over narrow range of timescales, while others remain constant over fairly long time. These ranges may overlap between subsystems. Due to these complex interactions climate varies on all timescales. For simplicity the vast range of the global variability is studied across the hierarchy of frequency domains with different scales of aggregation (such as intra- and interannual, and interdecadal to multimillennial). These studies have revealed specific features of the variability within distinct frequency bands, e.g., day-by-day versus interannual temperature variability. As about temperature variability, there is a vast amount of research works using surface meteorological observations, upper-air temperatures estimated from radiosondes, satellite-inferred tropospheric temperature trends, and other variables to detect its variability trends. Because we would like to connect these efforts with the possibility of the variability detection from the ground temperature monitoring (multiyear time series), further we will describe only results concerning the high-frequency variability detected from the SAT data that can serve as a background for a comparison with the results obtained from the subsurface temperature monitoring data.
Considerable insight into empirical climate variability changes over the last century was obtained from the details of the patterns of annual and seasonal surface temperature variations. Recent studies have detected not only the global scale warmth but also changes in the SAT variability. Most recent efforts significantly advanced our knowledge of the temporal and spatial patterns of climate variability. Results of investigations of the local and spatial patterns of the high-frequency climate variability were presented in numerous works (Karl et al., 1993, 1995, 1999; Balling, 1995; Liang et al., 1995; Kelly and Jones, 1999; Moberg et al., 2000; Grieser et al., 2002; Bodri and Cermak, 2003; Bodri, 2004; Braganza et al., 2004; Seidel and Lanzante, 2004). Most of the authors have used only the twentieth century data. This has helped to avoid bias due to progressively increasing number of measurements during the whole observational period. Given the number of techniques for variability detection in different works, results of the earlier studies have shown significant scatter. Thus, Parker et al. (1994) have compared interan-nual variability for the global data of seasonally accumulated surface air temperatures for two periods 1954-1973 and 1974-1993 and found small global increase of SAT variability. Especially noticeable increase was obtained for central North America. Jones et al. (1999) have worked with global data and have not detected any change in variability. Investigations by Grieser et al. (2002) based on the monthly averaged European temperatures have shown that at least in this region of the world the SAT variance has mainly decreased or remained constant during the last 100 years. Michaels et al. (1998) have examined monthly averaged SAT data for the 5° X 5° grid boxes around the world and have detected decrease in the intra-annual variability that prevailed over the past 50-100 years. The authors also have found general decrease in monthly temperature variability for the United States, some regions of the former Soviet Union and China. Mixed trends were detected for Australia.
More uniform results exist for the high-frequency temperature variability. An analysis by Karl et al. (1995) using SAT data of 1910-1990 observational period has revealed that day-by-day to interannual variability has generally decreased in the Northern hemisphere. Balling (1998), analyzing daily and monthly variability of historical temperature records, has found its overall decrease from 1897 to 1996. Collins et al. (2000) have identified reduced day-by-day variability trends for Australia. Karl et al. (1993), Easterling et al. (1997), and New et al. (2000) have shown that the land surface warming observed over the last 50 years has been accompanied by relatively stronger increases in daily minimum temperatures than in daily maximum temperatures (see also www.ncdc.noaa.gov/oa/climate/mxmntr/mxmntr.html). Thus, the difference that is called the diurnal temperature range (DTR) and represents effective measure of the daily temperature variability has decreased in recent years. Easterling et al. (1997) have revealed a decrease of the DTR from 1950 to 1993 for ~4100 stations in both the Northern and Southern hemispheres. A study by Karl et al. (1993) states: "Since 1950 all of the increase of temperature across the U.S.A. is due to an increase in the minimum temperature (about 0.75K/100 years) with no change in the daily maximum temperature. This caused a decrease in the diurnal temperature range". Subsequently, similar decrease in daily SAT variability has been observed at other locations and as stronger as one goes towards the Polar Regions. A study by Braganza et al. (2004) has detected strong negative trend of ~0.4K in the DTR over global land areas (gridded SAT data) for the last 50 years. The last 50-year period was chosen by most of the researchers because it has the largest and most consistent data coverage. A study by Braganza et al. (2004) detected that the increase in daily minimum temperature over this period was ~0.9 K, while the maximum temperature had risen by only ~0.6K. It now appears that most of the observed global surface warming of recent decades is occurring at night.
The studies of the correlation between changes in mean SAT value and its variability are sparse and not as unanimous as the results of the DTR change. Braganza et al. (2004) have shown that observed clear DTR decrease is not spatially uniform. The correlation of the DTR with the mean temperature over all observations of the 1901-2000 period is not significant and equals to only -0.24. Griffiths et al. (2005) have revealed significant location-dependent trends in the DTR in the majority of stations across the broad Asia-Pacific region, as well as the correlation between mean temperature and the frequency of extreme temperature events. This correlation appears stronger in the less populated/urbanized regions. Vincent and Mekis (2006) have examined trends in the mean temperature and the DTR for Canada and have shown that at least for the period 1950-2003 there is significant decrease in the DTR as well as a decrease in the variance of the daily mean temperature. Both trends were location dependent.
Observed reductions in daily temperature variability over the last century are large; they unlikely occur due to natural climate variability alone. Numerous attempts were undertaken to capture the correlation between changes in the SAT mean and variability through numerical modeling of the effects induced by humans. The majority of the climate model simulations associated with the build-up of greenhouse gases predicts not only climate warming but also a general decrease in the climate variability (e.g., Karl et al., 1999; McGuffie et al., 1999). Dai et al. (2001) and Stone and Weaver (2002, 2003) have shown that anthropogenic forcing by greenhouse gases and sulfate aerosols in General Circulation Models (GCMs) caused small but detectable decrease of 0.2 K/100 years in the global DTR over the twentieth century. The 50-years DTR trends of similar amount were simulated by Braganza et al. (2004). Modeling results have also corroborated that expected DTR decrease is not spatially uniform. Possible reasons for the DTR decrease are: (1) urban heat island effect (see previous Section), (2) an increase in cloudiness, and (3) anthropogenic greenhouse gases and sulfate aerosol emissions. Verdecchia et al. (1994), Stone and Weaver (2002), and Braganza et al. (2004) have shown that the main controlling factors for the DTR are clouds and soil moisture. Because of the number of atmospheric and surface boundary conditions affecting the maximum and minimum temperature, the linkages of the observed changes in the DTR to large-scale anthropogenic climate forcings still remain tentative. Further studies for more sure detection of the temperature variability (including measurement of underground temperature) are indispensable.
Detection of the temperature variability does not represent an easy task. Notwithstanding that all variability measures are based on the difference from some reference point, e.g., long-term mean or previous discrete value, there are many ways to define temperature variability. It may be calculation of the change of the magnitude of the DTR, frequency of occurrence of temperature extremes, the difference of the mean temperature from one day to the next, change of the standard deviation of temperature between two adjacent time periods, etc. Results are clearly dependent on the statistics chosen. Thus, for example, the latter technique may cause confounding of the high- and low-frequency variability and is insensitive to the position of large positive and negative departures from the mean within given interval (Karl et al., 1995). For example, two time series 0, 0, 0, 1, 1, 1 and 0, 1, 0, 1, 0, 1 have identical standard deviations, but significantly differing variability. The DTR is highly sensitive to small changes in maximum and minimum temperatures, etc. In addition, because all variance statistics are dependent on the reference level, e.g., mean, the uncertainties in the rate of change of the mean may confound detection of the changes in variance.
Moberg et al. (2000) have compared the properties of eight statistical measures of the day-by-day variability using European series of daily averaged surface air temperatures for the period 1880-1998. Two techniques were found to be most powerful for the detection of variability in temperature time series: (1) the intramonthly standard deviation of daily temperature anomalies and (2) suggested in the work by Karl et al. (1995) the mean of a series of values defined by the absolute value of the difference in temperature between two adjacent discrete time periods. The quality of observed data is a vital factor for both methods. For example, the latter procedure is very sensitive to the homogeneity of the temperature series; thus, it can be applied as the diagnostic tool for detection of the changes in the measurement techniques or other inhomogeneities in the temperature time series used. Applying both methods, Moberg et al. (2000) have revealed different behavior of daily variability trends in different parts of Europe. Variability has decreased by 5-10% in the northeast of Europe, has shown change of 0% to -5% in the northwest, and has increased by 5% to the southwest. On a longer timescale, day-by-day temperature variability in winter, spring, and autumn in northern Europe has decreased over the last approximately two centuries. The larger variability in northern Europe before twentieth century can be mainly attributed to a higher frequency of winter extremes.
General analysis of the climatic temperature change/variability includes decomposition of the observed time series into following significant components:
1. linear trend
2. harmonic components
3. extreme events
4. noise (stationary or non-stationary).
Robust estimation of each component in the presence of other components is not a trivial task. It can be performed by various statistical methods, from the linear regression and spectral analysis to far more complicated, e.g., the Generalized Additive Model (GAM; Vislocky and Fritsch, 1995; Grieser et al., 2002). Independently of their capacity/performance all these techniques should answer the following questions:
1. Is there significant linear trend in measured records?
2. Are there significant harmonic components?
3. If so, has any observed cycle changed, e.g., how has changed an amplitude of the annual cycle?
4. Are there extreme events that cannot be explained by the statistical properties of the record?
While many time series can be described in terms of two basic classes of components: trends and periodicity, climatic time series contain significant intrinsic stochastic component. Thus, the last but not least question should be:
5. When all significant deterministic components were removed, what is the structure of the remainder stochastic noise?
To meet these requirements a flexible stepwise strategy has to be used. Below we present an example of the detection of variability changes in the 8-year-long time series of the GST monitored at station Prague-Sporilov.
While borehole GST reconstructions capture low-frequency variability only, temperature monitoring data can be complementary to these long-term trends detecting short-term variability. Details of the monitoring experiment at Prague-Sporilov site are described in the previous section. The site is located on the top of a low hill in the campus of the Geophysical Institute of the Czech Academy of Sciences on the rim of large urban agglomeration. The temperature has been monitored since 1994 (Cermak et al., 2000) at a number of selected depth/elevation levels below/above the surface. Figure 125 shows results of 8-year temperature monitoring. These data refer to the temperature measurements obtained by zero-depth thermistor sensor installed on the top of a few millimeters of the rotten organic relics upon the compact soil ground. The individual measurements were taken at 15-min intervals and then averaged to 6-h regular grid; the precision of the individual readings is better than 0.01 K. The early years suffered by several data gaps; an uninterrupted continuous record exists only for the period 1998-2001. There were no changes on the observational procedure or in the equipment installation
during the whole experiment. The estimates of variability are thus not influenced by any data inhomogeneity problems, which otherwise may seriously bias the results (Moberg et al., 2000).
The record of the natural internal variability can be reconstructed by removing estimates of the response to the periodic external forcing (Jones and Hegerl, 1998). The actual character of changes in the temperature variability may be distorted by the annual temperature variations when the slope of the annual cycle is steep in the spring and autumn seasons (Karl et al., 1995; Moberg et al., 2000). To minimize the potential influence of the annual cycle, the measured data, before being processed, should be converted into non-periodic temperature anomalies. Figure 126 shows how this pre-processing works. The measured temperatures were expressed as T[, where Y = 1,... ,8 corresponds to years from 1994 to 2001, and index L = 0, 1,..., 1460 means the serial number of the corresponding 6-h long interval within the respective year. The mean annual cycle contained 1461 points from 0 to 365 days at 6-h intervals and was calculated by averaging 8-year values of TL = -^Zy=1T/. The reference temperature was then obtained from this cycle using the mean value, first four harmonics of the Fourier analysis, and the daily wave (wave number 365). Little, if any, additional variance could be explained when higher order harmonics are used. To obtain the temperature anomaly the reference temperature was removed from measured temperature (Figure 127). As seen, temporal oscillations of obtained signal are erratic and do not exhibit apparent regularity, trend, or cyclic pattern.
The first insight in the variability of this record can be gained using its probability distribution. Figure 128 presents the comparison of the probability distribution of the Sporilov temperature anomalies record with the normalized standard distribution. Both distributions generally coincide. Prominent feature of the temperature anomaly record is the prevalence of extremes in warm seasons of the year while extremes are relatively rarer
Was this article helpful?