Climate change - the missing links
WJR Alexander, Professor Emeritus, Department of Civil and
Biosystems Engineering, University of Pretoria, South Africa
Introduction
Large areas of southern Africa are currently in the grip of a severe drought.
More than a million people are reported to be facing death by starvation within
the next nine months, with more than ten million at risk. It is only two years
since the widespread floods of February-March 2000 resulted in high loss of
life, and severe damage to the road and rail communication routes and water
supply systems. The floods were described as the worst humanitarian disaster
experienced in the subcontinent. Now the drought is in the same category.
There are many factors that contribute to the high socio-economic
consequences of these natural disasters, including poverty, uncontrollable
occupation of flood prone areas, and the situation in many regions where the
natural environment can no longer support the population that depends on it.
There are also fears that climate change may increase the magnitude and
frequency of these natural disasters.
More money and more effort involving more scientists have been spent on
climate change research in recent years than any other subject. The predicted
consequences are dire, yet it is still not possible to quantify the anticipated
changes to the degree required for the design and operation of measures to
counter the predicted effects. Nor does it appear that this information will
become available in the foreseeable future.
An alternative route that could be followed is to attempt to find links
between climatic driving mechanisms and the multi-decadal properties of rainfall
and river flow. If these can be found, they will provide the public and other
stakeholders with more confidence in the predictions, and more quantitative
information to counter them.
With this in mind, extensive numerical and graphical analyses were carried
out on representative, long, reliable South African records of rainfall and
river flow. In all, data from records at 170 sites were analysed, with a total
record length of 11804 years and an average record length of 69 years.
There were some confirmations, some new insights, and some surprises.
Dialogue on water and climate
At its 19th Session in Geneva in April 2002, the Intergovernmental Panel on
Climate Change (IPCC) adopted a proposal introduced by the Dialogue on Water and
Climate and the WMO-UNESCO World Climate Programme, to prepare a special report
on water and climate. The IPCC Special Report on Water and Climate is scheduled
for completion by 2005, in time for use as a leading document at the 4th World
Water Forum in 2006. The International Steering Committee of the Dialogue on
Water and Climate includes international organisations best known by their
acronyms. They are the WMO, FAO, IHP-UNESCO, IPCC, IUCN, IWA, World Bank and the
World Water Council. The International Secretariat has its office at Delft in
the Netherlands. The purpose of this presentation is to make a positive
contribution to the debate.
Terms used
In this presentation the term statistically significant refers to a 95% level
of statistical significance. The term discernible implies that it is visually
evident on the graphs, but has not yet reached a 95% level of statistical
significance. The term meaningful implies that it has a measurable effect that
should be accommodated in hydrological analyses.
Methodology
The emphasis in this presentation is on graphical rather than on mathematical
interpretation. Mathematical relationships are poor descriptors of processes
where instability and asymmetrical periodicity are present. The data used in the
analyses were the raw data provided by Department of Water Affairs and Forestry,
and by the South African Weather Service. The data were used directly and were
not smoothed or manipulated in any way. The calculations of the accumulated
departures from the record mean values were simple arithmetic exercises. Other
than the conventional serial correlation analyses, no mathematical procedures
were used to determine the presence and properties of the periodicity, or trends
in the data. Standard Monte Carlo simulation procedures were used to identify
the most likely shape and amplitude of the periodic changes.
Knowledge gaps
Most climate change scenarios are based on predicted changes relative to the
current, undisturbed mean values, which are assumed to be sensibly constant.
There are two basic problems with this approach. Firstly, it is not the mean
values that are important when developing methods to accommodate the postulated
increases in the occurrence and severity of floods, droughts, and changes in
rainfall patterns. It is not the mean rainfall that determines whether or not
maize can be grown economically, but the year-to-year variability about the
mean. It is not the mean river flow that determines the yield from a dam on a
river, but the duration and magnitude of sequences of below average flows. The
mean is only one of the three principal moments used in flood frequency
analyses. The other two are the variance and the skewness. Droughts are
multi-year phenomena. Their severity is a function of a combination of their
duration and the extent of the departure from the mean value.
It is therefore not sufficient to postulate changes in mean conditions
arising from climate change and expect that action should be taken to counter
the effects. An essential step is to provide a quantifiable linkage between the
atmospheric processes on the one hand, and the resulting hydrometeorological
processes on the other.
The second and more difficult problem is to determine whether or not the mean
itself is sufficiently stable for use as a reference for possible future
changes. This analysis requires that the signal (behaviour of the mean), has to
be separated from the noise (nature and extent of the variability about the
mean). Furthermore, these properties have to be determined to a high degree of
assurance if they are to be plausible and useful. For example, it has been known
since 1978 that hydrologically meaningful 20-year periodicity is present in
South African river flow records, but it was only in 1995 that the length of the
records became long enough for the statistical analyses to reach the 95%
confidence level.
Databases
The South African rainfall database extends back to the middle of the 19th
century. A monthly district rainfall database was established based on records
from 1921 onwards. South Africa was divided into 93 districts and the average
rainfalls within each district for each month of record were determined. The
database was extended to March 2000. This is a particularly useful database, as
it overcomes problems that arise when single station, daily data are used for
analyses.
Scattered observations of high flood discharges date back to the
establishment of early mission stations in the first half of the 19th century.
Routine daily observations of river flow started in the period 1904 to 1910. The
severe droughts of the 1930s precipitated an appreciable expansion of the
Department of Water Affairs' hydrological data collection network. The
observations included river flow at gauging structures and into dams, as well as
concurrent daily observations of rainfall and open water surface evaporation (Symons
pan) at selected dams and other sites. These records are now long enough to
provide a basis for the determination of the multi-decadal properties of the
principal hydrological variables - annual open water surface evaporation,
rainfall, river flow and flood peak maxima.
The district rainfall data set from the 93 districts has a total length of
7254 years of data. The hydrological data sets were chosen on the basis of their
representativeness and uninterrupted, long records. Sixteen flow gauging
stations were used with 1020 years of data; 16 flood gauging stations with 1299
years of data; five miscellaneous sites with 365 years of data; and concurrent
rainfall and open water surface evaporation from 20 stations with 933 years of
record each. In total, records at 170 sites were analysed with a total record
length of 11804 years, and an average record length of 69 years.
Periodicity
In 1889 Hutchins reported the presence of 11-year periodicity in South
African rainfall and its linkage with the sunspot cycle. He noted that the
periodicity rose steeply to a maximum and then fell gradually to a minimum. He
also made several references to occasions when 'great floods ended a period of
drought'. In 1971 and subsequently, Tyson and his colleagues published a number
of papers in which they demonstrated the presence of an 18-year cyclicity in
South African rainfall. In 1978 Alexander demonstrated the presence of 20-year
periodicity in South African river flow and its linkage with the double sunspot
cycle. Subsequent analyses confirmed his conclusions. In 1986 Tyson published
his book on Climatic change and variability in southern Africa in which the
results of South African and international studies on rainfall periodicity were
summarised.
Hydrological data were not included in the analyses. He presented what he
described as an unambiguous 18-year oscillation in the rainfall over the
north-eastern summer region of South Africa. He reported that no evidence exists
to support the notion of steady progressive changes in rainfall. The conclusions
were based on mathematical analyses.
In 1993 Alexander confirmed the 20-year periodicity in river flow based on
graphical analyses and successfully predicted the imminent and sudden change
from the then prevailing severe drought to a period of well above average river
flow. The periodicity had still not reached a 95% level of statistical
significance at the time that the prediction was made. This has since been
achieved. There is now very clear graphical and statistical evidence of the
periodicity. The results are presented in this article. The analyses show that
91% of the rainfall and river flow sites have a discernible 19 to 21-year
periodicity. At 20% of the sites the periodicity has reached a 95% level of
statistical significance; and only 9% of the sites show no discernible
periodicity in this range.
The second test for periodicity is the observation that long droughts are
often broken by the sudden occurrence of a series of exceptionally high river
flows, confirming Hutchins' observations more than a century ago. This produces
a characteristic signature on the graphical plots of the accumulated departure
from the mean values of the whole record. These sudden reversals occurred during
the hydrological years beginning October 1912, 1932, 1953, 1972 and 1994. The
length of the periods between these reversals is 20, 21, 19 and 22 years
respectively. These characteristic sudden reversals of the slopes of the graphs
of accumulated departures from the record mean are present in 21 of the 28
records of annual river flow and annual flood peak maxima.
The third test for periodicity is the predictability of the sudden changes.
The 1994 reversal was successfully predicted by this writer in a paper written
two years previously during the then current severe drought, and published in
the South African Journal of Science four months before it occurred.
There can be no doubt about the presence of this periodicity in the rainfall
and river flow records and its probable linkage with solar activity. The signal
is very clear, and if concurrent periodicity in climatological processes is
found, this could be an important step in linking the climatological and
hydrometeorological processes.
Shape and amplitude of the periodicity
Monte Carlo simulation studies were undertaken to determine the shape and
amplitude of these periodic changes. The well-studied record of the annual flows
in the Vaal River at Vaal Dam was used for this purpose. Four scenarios were
tested. These were a constant mean with cyclical changes of the coefficient of
variation; a stepped mean with 10-year long periods successively above and below
the record mean and constant variance relative to the periodic mean; a 20-year
sinusoidal mean with constant variance relative to the changing mean; and a
wedge-shaped mean with an instant increase at the beginning of the sequence and
a uniform decline to the end of the sequence. The coefficient of variation was
assumed to be constant, relative to the changing mean.
The conclusion was that the wedge-shaped mean with an amplitude starting with
150% of the record mean and decreasing to 50% of the record mean over a period
of 20 years and then suddenly increasing to 150% of the record mean provided the
best fit. Graphical analyses from the other river flow sites confirmed this
conclusion.
Rainfall changes during the past century
The graphs of accumulated annual district rainfall for the period 1923 to
2000 were examined in order to determine whether or not there have been
meaningful changes in the annual rainfall during the past century. The study
showed that of the 87 districts without gaps in the record, 59 districts showed
a clearly discernible increase in annual rainfall, 25 districts showed no
discernible change, while only two districts showed a discernible decrease in
annual rainfall during this period. The change in the mean annual rainfall
during this period relative to the record mean ranged from - 2% to + 8%, with an
average of + 2%. These figures have to be doubled if the first half century
values are used as the baseline.
Evidence from other studies showed that there was also a long period of high
river flow during the early decades of the last century. The mid-century decades
were relatively quiet, followed by an increase in the past 50 years. If climatic
processes show the same pattern, this would also provide a valuable link between
climatic and hydrometeorological processes.
Southern oscillation index
The SOI data from 1881 to 1995 were also analysed. This 114-year long record
shows a distinct discontinuity in the mean value from 1974 onwards. It also
shows statistically significant serial correlations with frequencies of seven
and 14 years, as well as 21 years with a lower level of confidence. This was
despite a sudden change in the mean value that occurred in 1974. Normally, data
displaying this sudden change in the mean would be suspect and the possibility
of measurement errors would have to be investigated, or the cause of the
discontinuity sought. This was beyond the scope of the analyses.
Extreme floods
Observations of extreme floods that occurred during the past 150 years do not
show any progressive changes during this period. The severe floods during the
mid 1800s remain the highest on record at a number of sites. Other than at one
site, the severe floods of February to March 2000 were not the highest on
record. These severe floods are the consequence of infrequent, but not rare,
widespread, severe rainfall events including tropical cyclones. They do not form
part of the same statistical population as the rest of the series of annual
maxima. The properties of these extreme floods were described in detail in a
2002 paper by this author. They showed no evidence of changes in the flood
magnitude/frequency relationship that could be attributed to climate change. It
was also demonstrated that changes of the order postulated in climate change
scenarios would be undetectable against the background of natural variability.
Predictability of hydrometeorological changes
The occurrence of the reversal in 1994, completed three full cycles from 1932
to 1994 and raised the level of statistical significance in the analyses. An
interesting question is why the statistically significant and hydrologically
meaningful properties of rainfall and river flow have become so clearly apparent
in these studies, but have not been reported previously with the same degree of
assurance. The main reason for this, is the reliance on mathematical analyses
which obscure the sudden reversals from droughts to floods and are unable to
accommodate instability and asymmetrical periodicity. Another and equally
important reason is that the smoothing of the data prior to mathematical
analyses reported by many authors, obscures these very important
characteristics.
These reversals are more apparent in river flow than in rainfall. The
hydrological processes in moderate to low rainfall regions amplify the
periodicity, which may not be apparent in climates with higher rainfall, and
less variable river flow. This proposition was confirmed in the analysis of data
from two fundamentally different processes. These were the annual flows in the
Zambezi River at Victoria Falls, and the annual average water levels in a large
dolomitic compartment in an arid region of South Africa, that have been observed
since 1923.
The Zambezi River is the largest river in Africa south of the equator. The
period used for the analyses was from 1907 to 1994. The annual flow for the
period 1945 to 1980 was appreciably higher than during the earlier period 1907
to 1945. It decreased again from 1980 to 1994 but this was succeeded by high
flows from 1998 to 2000. This is a very large catchment with a high rainfall and
large swamps within the river basin. These result in a strong base flow that is
not present in South African rivers. This in turn results in a statistically
significant serial correlation for all years up to 11 years, but 20-year
periodicity is not present.
These properties are very similar to those of the water levels in the
Wondergat sinkhole in a large dolomitic compartment in a semiarid area some 800
km to the south. The difference is that the turning point of the period of low
water levels occurred in 1963, but otherwise the behaviour of the long period
means is remarkably similar. The interpretation of the characteristics of these
two records is that both of them have a large storage component, and that this
masks the longer term periodicity if this is present.
None of the other rivers discussed in this presentation has a meaningful
perennial flow, with the result that they have little year-to-year 'memory'. Of
the 32 river flow and flood gauging sites, only 8 have statistically significant
serial correlations of one year, and none have longer serial correlations less
than 18 years. In contrast, the Zambezi River has statistically significant
serial correlations up to and including 11 years, but no discernible serial
correlation for longer periods. Similarly, and surprisingly, the water levels in
the dolomitic compartment have statistically significant serial correlations
from one to five years, but no serial correlation for longer periods. In both
cases, the multiyear memory inhibited the identification of the 19 to 21-year
periodicity, if it was present.
However, both sites show an appreciable decrease in the mean conditions up to
the middle of the last century, followed by a steep increase in the following 30
years through to 1980. Thereafter they both show a decrease through to the end
of the records in 1993 and 1994. This was immediately prior to the 1994
reversal, which would have negated this decrease.
Summary and conclusions
Table 1 is a summary of the analyses. It provides valuable insight into the
relationships between multi-decadal properties of the hydrometeorological
variables, as well as the possibility of providing a linkage with climatological
processes.
A number of sites had to be omitted from some of the analyses due to gaps in
the records or the records not being long enough for serial correlation
analyses. However, they provided confirmation of the results, but with a smaller
degree of assurance.
The most important observation is the presence of 19 to 21-year periodicity
in 90% of the rainfall and river flow records. A surprising observation is that
the characteristic reversal marking the end of the periods is not visually
present in the district rainfall records, but increasingly discernible in the
records of annual river flow and annual flood peak maxima.
Conversely, the increase in mean annual rainfall during the second half of
the last century is present in 57 of the 82 district rainfall records; no change
is discernible in 20 of the records, and a decrease is observable in 5 of the
records. This is less apparent in the river flow records.
These multidecadal properties of rainfall and river flow in South Africa, and
possibly in other regions with moderate to low rainfall, will have to be
accommodated in climate change scenarios before it will be possible to develop
and apply mitigating measures.
A detailed report is in preparation.
More Information
Professor Alexander: Email alexwjr@iafrica.com.
|
