According to the latest Intergovernmental Panel on Climate Change (IPCC) report, there is strong scientific evidence that climate change is mostly attributed to human activities (Trenberth et al., 2007). Climate change has resulted in rising temperature trends with associated changes in temperature extremes across the globe (Hansen et al., 2001; Alexander et al., 2006; Brohan et al., 2006; Caesar et al., 2006; Lugina et al., 2005; Smith and Reynolds, 2005). The rate of warming over the last 50 years is almost double that over the last 100 years (0.13 ± 0.03 °C vs 0.07 ± 0.02 °C per decade). Also, over 74% of the global land area sampled, a significant decrease in the annual occurrence of cold nights (i.e. extremely low minimum temperatures) is shown, while a significant increase in the annual occurrence of warm nights (i.e. extremely high minimum temperatures) took place over 73% of the area.
In the TAR (IPCC, 2001), global surface temperature trends were examined for three sub-periods: 1910–1945, 1946–1975 and 1976–2000. The first and third sub-periods had rising temperatures, while the second sub-period had relatively stable global mean temperatures (Trenberth et al., 2007). Therefore, since the instrumental period in temperature observations began, the global trend in temperatures was not near-constant, but consisted of periods of stronger and weaker trends.
The inclusion of as many regions as possible in assessments of the global trend is essential, as the trends are highly variable, not only on a temporal but also on a regional basis. Examples of the most recent trend studies over parts of southern Africa are by Kruger and Shongwe (2004) who, in their temperature trends study for 1960–2003 in South Africa, found in general positive trends in the annual mean, maximum and minimum temperatures, as well as increases in days and nights with high temperatures and decreases in days and nights with low temperatures. However, the observed warming was not consistent between the analysed weather stations, which indicated regional variations in temperature trends. New et al. (2006), reporting on a regional workshop on temperature trends, also indicated warming trends for most of the southern African subcontinent, with magnitudes similar to those found by Kruger and Shongwe (2004). The aim of this study is to update the state of South African trends of daily maximum and minimum temperatures extremes for the period 1962–2009, and longer where possible, which will be useful in the assessment of climate change in the region.
A common analysis period of 1962–2009 of daily maximum and minimum temperature data from the SAWS (South African Weather Service) climate database was selected, in order to obtain the longest possible period with a reasonable number of available climate stations to cover most regions in South Africa.
The data sets of the selected stations were subjected to quality control to remove any values which were possible erroneous. As a first step, the metadata files of the potential climate stations were scrutinized for large movements, inadequate exposure, as well as maintenance problems that could have resulted in inhomogeneities in their data sets.
The validity of anomalously high or low values were checked against the reports of the prevailing weather conditions around the time of observation, and also compared to the values measured at available neighbouring stations sharing the same climate regime. Consequently, any suspicious data were removed from the time series. Thereafter, the completeness of the data series was verified to be higher than 90%.
A total of 28 weather stations were accepted for analysis. The SAWS climate number, station name, location in terms of latitude and longitude, altitude, mean minimum and maximum temperature, as well as the available period of record, are presented for each station in Table I. The spatial distribution of the weather stations, as well as the provinces of South Africa for reference purposes, is shown in Figure 1.
Table I. List of weather stations utilized in the study
Mean Minimum Temperature ( °C)
Mean Maximum Temperature ( °C)
The period indicates the longest period of record, unique for each weather station, for which index trends could be calculated.
0 003 020
0 012 221
0 021 178
0 035 209
0 050 887
0 059 572
0 106 880
0 127 272
0 169 880
0 193 561
0 240 808
0 242 644
0 247 668
0 261 516
0 290 468
0 300 690
0 317 475
0 399 894
0 403 886
0 406 682
0 432 237
0 461 208
0 513 284
0 546 630
0 554 816
0 596 179
0 677 802
0 809 706
3.1. Trend analysis
The trend analysis of the extreme temperature indices were performed using the RClimDex software, which is available from the WMO/CLIVAR/JCOMM ETCCDI website http://cccma.seos.uvic.ca/ETCCDI. This software was used in 19 regional workshops worldwide, e.g. for southern Africa (New et al., 2006) and most recently Indonesia in 2009 (workshop report available on the ETCCDI website), in an attempt to address the scarcity of climate trend studies over some regions of the world. RClimDex is capable of computing 27 core indices. However, only the indices which could be relevant to South Africa were selected, and are shown in Table II. In particular, the threshold indices were not included, due to the highly variable climate of South Africa, mostly because of the topography, as can be deduced from the altitudes of the weather stations in Table I.
Table II. Temperature indices covered in the study
TX indicates daily maximum temperature and TN indicates daily minimum temperature. Percentiles are based on the 1971–2000 period.
Annual percentage of days when TX > 90th percentile
Annual percentage of days when TX < 10th percentile
Annual maximum value of TX
Annual minimum value of TX
Annual count of days with at least 6 consecutive days when TX > 90th percentile
Annual maximum value of TN
Annual minimum value of TN
Annual percentage of days when TN > 90th percentile
Annual percentage of days when TN < 10th percentile
Annual count of days with at least six consecutive days when TN < 10th percentile
The base line period of 1971–2000 was used in the analysis, initially for the estimation of threshold values for the identification of anomalous temperature values for quality control purposes, but also the thresholds for the percentile-based indices.
Linear trends were calculated with the least-squares method for each index and the correlation factors tested for significance with the t-test at the 95% level of confidence (Wilks, 2006). In addition, the error bars of the trends are included in the results, as provided by the outputs of the RClimDex software, also at the 95% confidence level.
3.2. Cluster analysis
Cluster analysis is often used in climatological studies to define regions with similar climatological characteristics, and was performed on the index trend values for this particular purpose. Of the different cluster analysis techniques the most widely applied method is the K-means method, as it is relatively simple to use and also allows reassignment of observations as the analysis proceeds from one number of clusters to the next. The K refers to the number of groups or clusters, which is specified in advance of the analysis. The K-means algorithm usually begins with a random partition of the n data vectors into the pre-specified number of groups. The algorithm proceeds then as follows:
1.Compute the vector means, i.e. x̄k, k = 1, …, K; for each cluster.
2.Calculate the Euclidian distances between the current data vector xi and each of the Kx̄k's.
3.If necessary the xi is reassigned to the group whose mean is closest. Repeat for all xi, i = 1…n.
4.Return to step 1.
The algorithm is iterated until a full cycle through all the data vectors produces no reassignments (Wilks, 2006). The K-means method was performed on the trend results for each index, together with the geographical coordinates of the weather stations. The groupings of stations according to the maximum number of clusters that could be resolved are depicted in the maps of the results, to aid in the more objective interpretation of the regional differences thereof.
3.2.1. Identification of thermal regimes
As an example of the application of cluster analysis, and to interpret the trend results presented in the following section in the context of the thermal regimes of South Africa, cluster analysis was applied to the annual means of the minimum and maximum temperatures, of which the result is shown in Figure 2. Six clusters, A–F, were resolved, with the stations grouped broadly according to the general characteristics of the mean temperatures in South Africa. These clusters can broadly be considered to be homogeneous groups of stations, which exhibit roughly similar annual thermal characteristics or regimes.
The clusters or groups of stations identified can be summarized as follows, with reference to Figure 2.
Coastal: cluster A represents the western and southern coastal region, which exhibits a generally mild Mediterranean climate, while cluster B represents the southeast and eastern coast, which has a subtropical climate with relatively higher temperatures than A.
Interior: cluster C represents the Lowveld, a region in the northeast at low altitude with relatively high mean temperatures throughout the year. Cluster D represents the dry western interior, which is characterized by temperature extremes with relatively high maximum temperatures in summer, usually exceeding 30 °C, but low temperatures in winter, when minimum temperatures often drop below freezing. The diurnal range in temperature of cluster D is also higher than elsewhere. The southeastern interior, represented by cluster E, exhibits relatively small diurnal ranges in temperature, most probably due to the moderating influence of the regular influx of oceanic air from the Indian Ocean. However, the northwestern part of the region on the plateau at high altitude, experiences very low temperatures in winter. If it was possible to identify a larger number of clusters, a separate group of stations around the escarpment would probably have been identified. Cluster F in the north has higher temperatures during winter than the larger part of cluster E, and is not as prone to temperature extremes as cluster D.
4. Results and discussions
The index trend results are presented in Appendices A and B, for the maximum and minimum temperature indices, respectively. These results, as well as their delineation with cluster analysis, are shown in the maps of Figures 3–12. The clusters with more noteworthy results, which are highlighted in the discussions either because of their spatial extent or magnitudes of trends, are identified with capital letters in the relevant maps. In the discussions of the results reference is made to the thermal regimes over South Africa, as identified in Figure 2, as it is important to note the extent of how these regimes were affected by the observed trends.
4.1. Maximum temperature-related trends for 1962–2009
The trends in the TX90P index (percentage of days per year when the maximum temperature is greater than the 90th percentile of the 1971–2000 base period) is shown in Figure 3. Most of the weather stations, 22 out of 28 (79%), experienced increases in warm extremes which are statistically significant at the 5% level, with the strongest trends observed along the southern coastline (cluster A with a mean of 2.97% per decade). Other regions which exhibit relatively strong positive trends are the larger part of the Northern Cape province (cluster B with a mean of 2.26% per decade), and parts of the central and northern interior (cluster C with a mean of 1.87% per decade). In summary, it is the western half and the northern interior, mostly with relatively warmer thermal regimes, mostly in clusters D and C in Figure 2, which exhibited the strongest trends. In other parts of the country the trends were decidedly weaker.
TX10P (percentage of days per year when the maximum temperature is less than the 10th percentile) shows, as can be expected, some similarities in the spatial distribution of the results of TX90P but with trends of the opposite sign: 18 of the stations (64%) have pairings of significantly positive and negative trends for TX90P and TX10P, respectively. A total of 19 weather stations (68%) show significantly negative trends for TX10P, as shown in Figure 4. The strongest trends are found in the west in cluster A with a mean of − 1.57% per decade, which coincides more or less with clusters A and B in Figure 3.
Results for TXx (absolute annual maximum of the daily maximum temperature) shown in Figure 5, indicate significantly positive trends for only four weather stations (14%), which are located in the western, northwestern and central interior in clusters A, B and C. The highest trend of 0.80 °C per decade is found at Vredendal in the Western Cape province (the only member of cluster A). Clusters A, B and C mostly cover the region in South Africa with the highest summer temperatures (clusters D and F in Figure 2). In general, the results show very small trend values, which indicate that in the larger part of the country no significant trend in TXx is evident.
For TXn (absolute annual minimum of the maximum temperature) similar results than TXx was found, in which the number of significant trends, in this case increases at nine stations (32%), and are relatively small compared to the other maximum temperature indices. The strongest trends of 0.70 and 0.67 °C per decade are found in the northeast in the Lowveld (cluster C in Figure 2). The other significant trends are widespread, and do not show higher concentrations in particular regions.
Warm spell duration index (WSDI—annual count of days with at least six consecutive days when maximum temperature is greater than the 90th percentile) shows only four stations (14%) with significant increases, with the highest trend in the northern part of the Northern Cape at 6.74 d per decade, as indicated by the sole member of cluster A in Figure 7. Due to the small number of significant results, no regional trends can be deduced, except that the four stations are all located in the northern half of the country, where extreme temperatures are in general more prevalent than elsewhere in the country (cluster D in Figure 2). For most of the weather stations along the coast, no trends in WSDI could be calculated, due to the small number of years where cases of six or more consecutive days occurred when the daily maximum temperatures were greater than the 90th percentile.
4.2. Minimum temperature-related indices for 1962–2009
TN90P (percentage of days per year when the minimum temperature is greater than the 90th percentile) shows 13 weather stations (46%) with significantly positive trends, as shown in Figure 8. While half of the significant results are found in the southwest and west in cluster A, which has a mean trend of 1.05% per decade, stations with significant trends are also well represented in the northeast in cluster B, with a mean trend of 1.09% per decade. The strongest positive trends are therefore mostly confined to the regions with thermal regimes exhibiting relatively higher temperatures (clusters C and D in Figure 2), but also the western and southern coastal areas, represented by cluster A in Figure 2. Three stations indicate negative trends, situated in North West, Free State and southeastern Eastern Cape provinces, but only statistically significant at Kimberley in cluster C.
TN10P (percentage of days per year when the minimum temperature is lesser than the 10th percentile) shows 17 weather stations (61%) with significantly negative trends, as shown in Figure 9. The strongest negative trends are found in the north and east (cluster A with a mean of − 2.07% per decade), the southwestern Cape (cluster B with a mean of − 1.67% per decade), as well as most of the Northern Cape province (cluster C with a mean of − 1.28% per decade).
As with the comparison of results of TX90P and TX10P, TN90P and TN10P show similarities in trend results, but with opposite signs. Eleven weather stations (39%) show pairings of significantly positive and negative trends for TN90P and TN10P, respectively. This number is markedly lower than the results for the comparison of TX90P and TX10P (64%), which indicates that the positive shifts in the overall distributions of daily maximum temperatures are more pronounced than the daily minimum temperatures. However, the regions that exhibit significant changes in both the left and right tails of the maximum and minimum temperature distributions coincide more or less, and cover the western half of the country, as well as the northern and northeastern interior. Again, as with most of the previous indices discussed, the regions where these changes occurred coincide with the thermal regimes which exhibit in general higher temperatures.
TNx (absolute annual maximum of the daily minimum temperature) shown in Figure 10, shows significantly positive trends for only four weather stations (14%). Two of the weather stations with significant trends are situated in the west of the country in cluster A, while the other two are in the east in clusters B and C, respectively. There are no apparent regional trends as in most of the country the trends are almost non-existent.
Trends in TNn (absolute annual minimum of the daily minimum temperature) shown in Figure 11, show ten stations (36%) with statistically significant increases. All five of the stations in cluster A in the northeast indicate significantly positive trends, with a mean value of 0.54 °C per decade. Other clusters in which significantly positive trends are found are B and C. In summary, significant positive trends in TNn are evident in the northeast and east, and the southern parts of the western interior. However, while most of the significantly positive results coincide with clusters C and D in Figure 2, other thermal regimes also experienced some significant trends in TNn.
CSDI (annual count of days with at least six consecutive days when the minimum temperature is lesser than the 10th percentile), shown in Figure 12, shows five stations (18%) with significantly negative trends. The strongest negative trend of − 4.08 d per decade was found in cluster A, in which all three stations showed significantly negative trends with a mean of − 3.77 d per decade. It therefore seems that significant decreases in cold spells are confined to the northeast. Cluster B indicates an extensive region with positive trends, although the results were only significant for one station. For many of the climate stations in the south, southeast and east of the country, the trends could not be calculated due to the insufficient number of available years with occurrences where the minimum temperature was lower than the 10th percentile for six consecutive days.
4.3. Summary and integration of trend results for 1962–2009
To obtain a characterization of the general trends in extreme temperatures in South Africa for the common period of 1962–2009, the trend results of related indices were compared and integrated to obtain a condensed view or summary of the regional trends and their relative differences over South Africa.
4.3.1. Maximum temperature
As discussed in the previous section, if there are significant trends in the daily maximum temperature, there will most likely be a general agreement between the trend results of TX90P and TX10P, albeit of opposite signs. Such trends would indicate a long-term shift in the statistical distribution of maximum temperatures, provided that the variance of the annual maximum temperatures stays near-constant throughout the relevant period. The results shown in Figures 3 and 4 confirm this to a large degree, and also indicate a general increase in maximum temperatures over South Africa.
Closer inspection of the results of TX90P and TX10P reveal that the western half and the northeastern interior of the country experienced relatively stronger increases in daily maximum temperatures than elsewhere. The stronger increases in warm extremes broadly cover the regions in Figure 2 indicated by clusters A and D in the west, and C in the northeast.
The results for TXx and TXn do not broadly agree with those of the previously mentioned indices, with relatively small percentages of weather stations with significant trends, and little spatial coherence between the results. However, it is worth mentioning that most of the significant results are in the western half of the country, which relates to the relatively strong warming there, as indicated by the results for TX90P and TX10P.
Trends in the WSDI would indicate possible changes in the number of occurrences of extended periods with high temperatures, such as heat waves, and would not necessarily coincide with those of the other maximum temperature indices, apart from possibly confirming a general increase in maximum temperatures. The results show that the only regions where there are signs of significant increases in heat waves are the extreme northern parts of the western and northeastern interior, which are in thermal regimes prone to extremes in maximum temperatures, especially during the warm summer months.
4.3.2. Minimum temperature
The trend results of TN90P and TN10P indicate a general increase in daily minimum temperatures across South Africa, except for parts of the central interior. Relatively stronger increases in daily minimum temperatures were observed in the west, northeast and east of the country, which coincides with the results of TX90P and TX10P, and ultimately the warmer thermal regimes in the country, represented by clusters C and D in Figure 2.
The results of TNx and TNn are in some agreement with the regional results of TN90P and TN10P.
Trends in CSDI would indicate changes in the frequencies of cold spells, when the daily minimum temperatures are much lower than usual for an extended period of time. The results of CSDI do not necessarily have to reflect the results of the other minimum temperature indices, apart from possibly confirming the general trends in the daily minimum temperature. The regions where significant decreases in cold spells are detected are in the northeastern interior and northern part of the western interior, i.e. parts of larger regions of relatively strong positive trends of TN10P.
In summary, the trend results for most of the minimum temperature indices indicate that the western half, as well as the northeastern interior of South Africa experienced relatively stronger increases in minimum temperatures than elsewhere in the country.
4.4. Comparisons between trends of 1962–2009 and longer periods
All of the weather stations, except one, have record lengths which are longer than the common analysis period of 1962–2009, as indicated in Table I. Analyses of the index trends for longer periods can provide information on the persistence of the observed trends, especially those that are relatively strong and statistically significant for 1962–2009.
It should be considered that the results of trend analysis of variables, with magnitudes which are cyclical in nature, depend heavily on the analysis period. In climate analyses, this is particularly relevant to cyclical behaviour which is near-decadal, where non-existent long-term trends might be inferred if the analysis period spans a small number of decades, from a period with relatively high (low) values to a period with relatively low (high) values in the cycle. In such cases, erroneous or exaggerated long-term trends will be indicated by the trend analysis. An indication of persistence of trend over periods which are much longer than the common 1962–2009 analysis period will increase confidence in the results found so far.
Appendixes A and B present the results of trend analysis for both the common period of 1962–2009, as well as for the extended periods P, which are unique for the weather stations utilized in the study. Of interest is a comparison of the trend results for those weather stations, where P is much longer than the 1962–2009 period. These are Cape Agulhas, Mossel Bay and Port Elizabeth on the South Coast, Vanwyksvlei and Kimberley in the Northern Cape, Vryburg and Marico in the North West, and Musina in the Limpopo province.
For the maximum temperature indices in Appendix A it is observed that in the south, for Cape Agulhas and Mossel Bay, the increase in warm extremes are noticeably weaker over the longer term, with the trends for TX90P and TX10P non-significant for Mossel Bay over the longer period of 1920–2009. Closer inspection of the results reveal an opposite trend from the 1920s to the 1960s, which offsets the increase in warm extremes since the 1960s to 2009. For Cape Agulhas, the differences in trends are less pronounced, but here TX90P shows almost no trend from 1911 to 1965, as shown by the RClimDex output presented in Figure 13. For Port Elizabeth to the east, the differences in trend between the shorter and longer term are much smaller.
The stations in the Northern Cape show similar results than those in the south, with weaker trends over the longer analysis periods. For Kimberley, almost no trends were observed for TX90P and TX10P from 1911 to 1975, after which strong increases in warm extremes were observed. Similar weak trends in TX90P and TX10P were observed for Vanwyksvlei, but for the shorter period from 1932 to 1975.
In the North West province of Vryburg, warm extremes decreased from 1920 to the mid 1960s, after which a significant increase of said extremes occurred. For Marico, the trend results show the opposite, especially for TX90P where the trend is stronger over the longer period of 1936–2009. A relatively weak increase in warm extremes occurred from the early 1980s to late 1990s, compared to the whole period of 1936–2009.
In the extreme north, Musina shows similar results than most of the longer term weather stations, with a small decrease in warm extremes over the period 1934–1965, and a strong increase thereafter.
Trends in minimum temperatures over the longer periods, as shown in Appendix B, indicate trends in TN90P along the South Coast comparable to that of 1962–2009, but stronger negative trends in TN10P. In the Northern Cape, the longer period analyses indicate accelerated decreases in cold extremes since the mid-1960s, similar to the results for the maximum temperature indices, which indicated increases in warm extremes. The weather stations in the North West province show similar trends for the shorter and longer periods. The indices for Musina in the extreme north show, similar to most of the longer term weather stations, an accelerating decrease in cold extremes since the mid-1960s.
The study provided an updated analysis of the daily maximum and minimum temperature trends of relevant extreme temperature indices over South Africa, for the period 1962–2009. While the maximum temperature indices show general increases in warm extremes, the minimum temperature indices show general decreases in cold extremes. This indicates that South Africa experienced general warming over the analysis period [also see Kruger et al. (2011) for an update on the trends in annual mean temperatures over South Africa].
Most of the results indicate relatively stronger increases in warm extremes and decreases in cold extremes in the western, northeastern and extreme eastern parts of the country. The results obtained are in general agreement with those of recent temperature trend studies for the region, which show a general warming trend, but with relatively weaker trends in the central parts of South Africa (Kruger and Shongwe, 2004; New et al., 2006). The parts of South Africa that experienced relatively stronger warming can be summarized as in Figure 14.
The regions in South Africa with relatively warmer thermal regimes, and which are more prone to hot daily extremes, i.e. the Lowveld in the northeast of the country, the east coast, and the dry western interior as indicated by clusters B, C and D in Figure 2, experienced the strongest increases in warm extremes.
It is envisaged that a persistence in the strong warming observed, particularly in the Northern Cape and parts of the Western Cape, both in the west where the interior can be described as semi-arid with highly variable precipitation, will have a negative effect on the biodiversity, due to habitat loss, and agriculture, due to likely increases in evaporation and consequent heat stress to livestock. Biodiversity has already been affected by rising temperatures in the drier regions of the northern and Western Cape provinces, as evidenced by Foden et al. (2007).
This remarkable differential warming over South Africa can most likely be attributed to possible changes in the atmospheric circulation over the subcontinent. Over the western parts it may include possible changes in the strengths of cold fronts moving over the subcontinent from the west, or weaker ridging by the quasi-stationary Atlantic Ocean high pressure system from the south or southeast, especially during the austral summer. In the east, weaker ridging by the Indian Ocean high pressure system might reduce the frequency or strength of the influx of cooler maritime air from the east. It is recommended that the possible changes be investigated with regional model studies and/or the analysis of long-term reanalysis data, both of which falls beyond the scope of the present study.
The analyses of longer time series than the common study period of 1962–2009, indicate that for most of the longer term stations which show relatively large differences between trends over the longer term and 1962–2009, the frequencies of warm extremes have accelerated since around the mid-1960s. This finding is in agreement with the mean global temperature trend, where increased warming is evident since the latter part of the 20th century, particularly from the mid-1960s (Hansen et al., 2001; Lugina et al., 2005; Smith and Reynolds, 2005; Brohan et al., 2006).
Table Appendix A. Maximum temperature index trends for the period 1962–2009
Maximum temperature indices
TXx TXx (P)
TXn TXn (P)
TX90P TX90P (P)
TX10P TX10 (P)
WSDI WSDI (P)
The values in brackets indicate the trends calculated for year P to 2009.
indicates that the trend is statistically significant at the 5% level.
N/A indicates that the trend could not be calculated due to too many zeros in the time series.