Historical rainfall trends in South Africa: 1921–2015

Over recent decades, significant changes in the mean state of the climate have occurred over most parts of the globe, particularly for surface temperatures. In contrast to surface temperatures, which show almost consistent upward trends, trends in precipitation were variable across the globe (Donat et al., 2013). These variable trends are of interest to various socio-economic sectors such as agriculture, health, water and natural resource management. In addition to trends in seasonal and annual rainfall totals, of importance in terms of the impact of rainfall are trends in extreme precipitation events. Various recent studies have acknowledged tendencies of increased extreme precipitation over some parts of South Africa, such as increased annual frequency of very heavy rainfall events in the eastern parts of South Africa from 1906 to 1997 (Groisman et al., 2005), the increased intensity of high rainfall events in the 1961–1990 period compared to 1931–1960 over much of South Africa (Mason et al., 1999), and the increased extreme rainfall index values over eastern parts of South Africa from 1910 to 2004 (Kruger, 2006). These findings confirm those of the global study by Donat et al. (2013), which indicates a general increase of the contribution of extreme daily rainfalls to annual totals, and a strong recent upward trend of the intensity of daily rainfall. A thorough analysis of these trends is usually restricted by different factors, which may include low and uneven station density over the investigated area, the period of station record, as well as data quality and plausibility. Although South Africa has a relatively extensive station network and long data records, it is still often difficult to detect clear signals of long-term change, particularly those of rainfall which vary considerably in space and time (McKellar et al., 2014). These factors result in any trend calculation being sensitive to the specific location and period of observations. Therefore it is essential to update the analysis of historical rainfall trends on a regular basis. Here we did not only add additional recent data (5 years), but used an analysis period which is much longer than the most recent historical trend study by McKellar et al. (2014), i.e., 1921–2015 vs. 1961–2010, more than doubling the analysis period. Precipitation variables of interest usually include the mean annual precipitation, the onset and end of rainfall seasons, wet and dry spell duration periods and the occurrence of extreme heavy rainfall events (McKellar et al. 2014). This study intends to use an optimal balance between station density and length of records to derive conclusions about the general trends in annual and seasonal rainfall, as well as extreme rainfall events. For the latter, the analysis was based on extreme precipitation indices developed by the World Meteorological Organization (WMO) Expert Team on Climate Change Detection and Indices (ETCCDI), which is used globally to detect trends in relevant climate extremes (Donat et al., 2013).


INTRODUCTION
Over recent decades, significant changes in the mean state of the climate have occurred over most parts of the globe, particularly for surface temperatures.In contrast to surface temperatures, which show almost consistent upward trends, trends in precipitation were variable across the globe (Donat et al., 2013).These variable trends are of interest to various socio-economic sectors such as agriculture, health, water and natural resource management.In addition to trends in seasonal and annual rainfall totals, of importance in terms of the impact of rainfall are trends in extreme precipitation events.
Various recent studies have acknowledged tendencies of increased extreme precipitation over some parts of South Africa, such as increased annual frequency of very heavy rainfall events in the eastern parts of South Africa from 1906to 1997(Groisman et al., 2005), the increased intensity of high rainfall events in the 1961-1990 period compared to 1931-1960 over much of South Africa (Mason et al., 1999), and the increased extreme rainfall index values over eastern parts of South Africa from 1910 to 2004(Kruger, 2006)).These findings confirm those of the global study by Donat et al. (2013), which indicates a general increase of the contribution of extreme daily rainfalls to annual totals, and a strong recent upward trend of the intensity of daily rainfall.
A thorough analysis of these trends is usually restricted by different factors, which may include low and uneven station density over the investigated area, the period of station record, as well as data quality and plausibility.Although South Africa has a relatively extensive station network and long data records, it is still often difficult to detect clear signals of long-term change, particularly those of rainfall which vary considerably in space and time (McKellar et al., 2014).These factors result in any trend calculation being sensitive to the specific location and period of observations.Therefore it is essential to update the analysis of historical rainfall trends on a regular basis.Here we did not only add additional recent data (5 years), but used an analysis period which is much longer than the most recent historical trend study by McKellar et al. (2014McKellar et al. ( ), i.e., 1921McKellar et al. ( -2015McKellar et al. ( vs. 1961McKellar et al. ( -2010, more than doubling the analysis period. Precipitation variables of interest usually include the mean annual precipitation, the onset and end of rainfall seasons, wet and dry spell duration periods and the occurrence of extreme heavy rainfall events (McKellar et al. 2014).This study intends to use an optimal balance between station density and length of records to derive conclusions about the general trends in annual and seasonal rainfall, as well as extreme rainfall events.For the latter, the analysis was based on extreme precipitation indices developed by the World Meteorological Organization (WMO) Expert Team on Climate Change Detection and Indices (ETCCDI), which is used globally to detect trends in relevant climate extremes (Donat et al., 2013).

Data
The determination of rainfall trends was mainly based on the analysis of daily rainfall measurements from individual rainfall http://dx.doi.org/10.4314/wsa.v43i2.12Available on website http://www.wrc.org.zaISSN 1816-7950 (Online) = Water SA Vol.43 No. 2 April 2017 Published under a Creative Commons Attribution Licence or weather stations.At the manual stations, measurements were done at 08:00 SAST, which would indicate the rainfall for the previous day, throughout the study period.The rainfall totals from automatic weather stations were, similarly to manual stations, based on the period 08:00 SAST to 08:00 SAST on the following day.Only stations that were operational throughout the study period were selected, and for which the position has stayed more or less the same.In addition, at least 90% of daily values should have been available, which excludes any accumulated rainfall totals.Due to the substantial distances and the complex nature of rainfall in South Africa, mainly due to the regionally different rainfall-producing mechanisms or systems and the complex topography in some regions, it is difficult to perform proper quality control on rainfall data.However, those cases where zero rainfall was recorded for relatively long periods of time, when surrounding stations recorded significant rainfall amounts, were flagged and removed.
In addition to the data discussed above, the delineation of districts of homogeneous rainfall, as used by the South African Weather Service (SAWS), aided in the attempt to cover as much of the areas with unique rainfall characteristics as possible.South Africa was divided into a total of 94 rainfall districts by SAWS (then the South African Weather Bureau: SAWB, 1972), which are presented in Fig. 1a.The delineation of these districts was mainly done according to the annual march of maximum

Figure 1a
Rainfall districts for South Africa (SAWB, 1972) with provincial borders rainfall, and the boundaries between the winter, whole year and summer rainfall regions.SAWS updates the district rainfall totals on a monthly basis for the period 1921 -present, of which 90 of the 94 districts have rainfall totals available for all months of the period.A daily district rainfall total is calculated as the average of the daily values available in the district.Therefore, it follows that the particular stations used in the calculation of district rainfall values vary through time, more so for those districts with a relatively denser network of stations.Due to the number of stations used, this cannot be effectively shown on a map.
It should be noted that for many rainfall districts the rainfall cannot be considered to be homogeneous across the district, especially in regions with complex topography, e.g., the southwestern Cape and along the escarpment.Due to there being different rainfall stations available for analysis at different times of the study period, it is possible that there is a bias in the number of stations in wetter areas over a particular sub-period, while there can be an opposite bias in the number of stations in relatively drier areas in another period, affecting the trend results.For most districts the particular stations used varied through time as stations opened and closed in a particular district.Therefore, it is not advisable to use only district rainfall totals to, in this case, determine historical trends.Therefore, mainly the rainfall records of individual stations was used to determine countrywide trends, with results from the districts http://dx.doi.org/10.4314/wsa.v43i2.12Available on website http://www.wrc.org.zaISSN 1816-7950 (Online) = Water SA Vol.43 No. 2 April 2017 Published under a Creative Commons Attribution Licence used in regions without sufficiently long individual time series, as additional information about regions where some rainfall trends are likely but not conclusive.
The coverage of most of the rainfall districts with rainfall records of individual stations that remained operational throughout the analysis period should provide an assessment of the ability of the station coverage to capture the tendencies in rainfall across South Africa.A set of near-complete daily datasets was identified using an approach to cover as many rainfall districts as possible for the period 1921-2015, which spans the period of available district rainfall records.For each rainfall district a rainfall station was identified with the most complete record.A total of 60 rainfall districts could be covered with rainfall stations with at least 90% of daily rainfall records available over the period 1921-2015.The locations of the rainfall stations are presented in Fig. 1b.For the trend analysis of annual rainfall, both results of individual station data and district rainfall are presented, to provide an impression of the probable extent of significant trends, particularly in those areas where continuous individual station data are not available for the whole analysis period.These regions include the western interior of the Northern Cape, some coastal areas of the Western Cape, as well as parts of the Eastern Cape, northern KwaZulu-Natal and the extreme north-east.

Analysis methodology
RClimDex is a software package developed under the auspices of the WMO ETCCDI that provides for the calculation of 27 core indices of temperature and rainfall extremes for the detection of changes in the climate (Zhang and Yang, 2004).The relevant 11 precipitation indices utilised in this study, defined by the ETCCDI, are listed in Table 1.The index results are provided at annual and monthly time scales.In addition to the trends from annual values presented, seasonal index values of prcptot, the total rainfall from daily values ≥ 1 mm, and r20mm, the total number of days with rainfall ≥ 20mm, were also determined to obtain an impression of seasonal tendencies in total rainfall and high rainfall events.The base period, from which the percentiles for the relevant indices were calculated, was defined as 1981-2010, which can be considered to be the present general norm for similar trend studies.The statistical significance of the linear trends of the indices was evaluated by the 2-sided t-test at the 5% level, and assuming Gaussian non-correlation.

Annual total rainfall in wet days
The results in the 1921-2015 trends of the annual rainfall in wet days are presented for the individual rainfall stations and district rainfall in Figs 2a and 2b, respectively.The large conformity between the individual and district rainfall results is apparent, but with some discrepancies.For instance, from the district rainfall results, District 33 has a significant negative trend, while the single long-term station has a non-significant positive trend.However, the most evident result is a positive trend in annual rainfall totals over the southern interior, mostly higher than 5 mm per decade, and probably also to the north into the western interior, as indicated by the district rainfall.
Drier conditions are evident in the extreme north and possibly parts of the north-east, as indicated by the district rainfall results.The results from the district rainfall analysis show, in general, larger footprints of significantly positive and negative trends, which indicate that the positive and negative trends could be more widespread than indicated by the individual stations.However, due to the possible inhomogeneities which might exist in the district rainfall time series and number of stations in the same region with similar trends, one can only infer with certainty that the southern interior became wetter and the extreme north drier.

Seasonal rainfall in wet days
Figure 3 presents the trends in seasonal rainfall totals in wet days.It is clear that the positive trends in annual rainfall totals in the southern and western interior are reflected mostly in the summer rainfall trends, which is the main rainfall season for these particular regions, with significant upward trends of more than 5 mm per decade.
It is also evident that trends in total precipitation for autumn were declining for almost the whole country.However, these decreases are mostly non-significant.The downward trend in annual rainfall in the north-eastern interior is not well defined in the results of the seasonal rainfall, except for 2 stations in Limpopo Province in autumn, with declines of more than 5 mm per decade.
For winter, spatially coherent, but weak, mostly nonsignificant trends were found, with mostly upward trends in the western half of the country and downward trends in the eastern half.

High and extreme daily rainfall events
The r95p and r99p indices are calculated to determine trends in daily rainfall extremes, although other indices such as r10mm, r20mm, r25mm and rx1day also provide indications of trends in high and extreme daily rainfall events.The r95p and r99p indices measure the total rainfall per year from days with rainfall above the 95 th and 99 th percentile daily rainfall total.Generally significant increases in the rainfall, with very high daily rainfall totals in the southern interior, but with significant increases only from some isolated stations in the central and north-eastern interior, are indicated by r95p in Fig. 4. R99p represents the extreme top percentile of wet days.In a typical base period of 30 years (e.g.1981-2010) one might find, e.g., 1 500 wet days, but this figure will largely depend on the particular climate.For dry climates it will be much fewer than 1 500 and for wetter climates significantly more.R99p calculates the total rainfall for a year from days with rainfall > 99th percentile, i.e., in the above case from the 15 wettest days.For most years the total will be 0, but r99p is still useful as an indication of more extreme daily rainfall conditions that may have occurred, especially outside the base period, and whether these occurrences happened earlier or later during the period analysed.However, it can be noted that most stations in the country show increases in extreme rainfall, although not statistically significant, as shown by both r95p and r99p in Figs 4 and 5.
The results of the rx1day index, presented in Fig. 6, indicate whether there are trends in the annual maximum daily rainfall amounts.Results from rx1day were similar to r95p and r99p, with the most spatially coherent results some increases in the southern interior.Similar results are also displayed by the r10mm (Fig. 7), r20mm (Fig. 8) and r25mm (Fig. 9) indices which indicate the number of days per year above 10 mm, 20 mm and 25 mm, respectively.These results all indicate that the increases in the annual rainfall observed in the south and some central parts of the country are at least partially the result of increases in individual high or extreme rainfall events.

High daily rainfall per season
The seasonal trends in r20mm show, as with the seasonal rainfall totals, that the significant annual increases in rainfall in the south and some central parts are at least partly due to significant increases in high daily rainfall episodes during summer, the  September to October -spring).Shaded symbols indicate significant trends at the 5% level.High and extreme daily rainfall events region's main rainfall season.These increases are in general higher than 0.1 days per decade and for several stations in the central interior (in the Free State, North West and Eastern Cape provinces) higher than 0.2 days per decade.
Apart from summer, the other provinces do not show spatially coherent statistically significant results, except for 4 stations in the north-eastern interior which show significant increases during spring.Also, it seems that during winter there is an increase in high daily rainfall totals in the south-western half of the country, and mostly decreases in the north-east, but these results are mostly statistically non-significant.

Intensity of daily rainfall
From the results for the high and extreme rainfall indices it is apparent that there are regional tendencies for daily rainfall episodes to become more intense or extreme.The Simple Daily Intensity Index (SDII)) was used to determine the annual average amount of rainfall that is received on a day with rainfall.Significant increases might indicate that the risks of high rainfall of short duration became more prevalent, which in turn indicates a higher probability of related disasters such as flash floods.
From Fig. 11 it is evident that general increases occurred in the daily intensity of rainfall over most of the country, but with some decreases evident in the extreme eastern parts of the country.The results from district rainfall trends (not shown) supported the significant increases in SDII for most districts from the south-western Cape to the central interior.Decreases in SDII were displayed by some districts in the far eastern interior, indicating that in that region some places experienced fewer days per year with relatively high rainfall.
The above results correlate with the general finding that rainfall in many parts of the southern and central interior experienced increases in annual and particularly summer rainfall, while in the extreme east decreases are evident in both annual rainfall and the frequency of high daily rainfall totals.

Intensities of continuous rainfall episodes
The rx5day index determines the highest annual amount of rainfall received in a continuous 5-day episode, which provides an indication of the trend in the intensity of continuous rainfall episodes.The trend results, presented in Fig. 12, indicate that the maximum annual amount of continuous rainfall in a       Most stations in the country show increases, although not statistically significant.Significant changes elsewhere are isolated and therefore no conclusions can be made therefrom in a regional sense.

Longest annual wet spell
The Continuous Wet Days (CWD) index indicates the maximum number of continuous wet days per year.For the south-western Cape the longest annual wet spell usually occurs during winter, while in the remainder of the country it usually falls in summer.
Figure 13, which presents the trends in CWD, predominantly shows a decrease in the largest part of the north-eastern half of the country, with some stations indicating statistically significant decreases in excess of 0.2 days per decade.This result corresponds to the decrease in rainfall observed in some of the eastern parts of the country.
For the south-western half of the country the trends are mostly non-significant, except for 3 stations in the Eastern Cape interior, which show significantly positive trends of between 0.1 and 0.2 days per decade.

Longest annual dry spell
The Continuous Dry Days (CDD) index defines the length in days of the longest annual period with no significant rain.The definition of the index indicates that this period should fall in the winter months over most of the country, but in the summer months over the south-west where most rainfall is received during winter.Therefore, this mean is essentially a measure of the dry season duration (i.e. in the order of 100 days), rather than a dry spell in a synoptic meteorological sense (i.e. in the order of 10 days).While most of the north-eastern half of the country indicated a decrease in the longest annual period of wet days, and the south-western half mostly an increase, the opposite applies for the annual maximum period of dry days.However, it is mostly the stations along the escarpment which show significant decreases in CDD.This could indicate that there might be a historical long-term increase in the annual period when there is sufficient influx of moisture from the ocean over the adjacent interior to produce rainfall along the escarpment, as this is the main source of rainfall for these areas.

SUMMARY AND CONCLUSIONS
The most prominent finding from the rainfall trend analysis for the period 1921-2015 is the increase in rainfall over the west of South Africa, particularly in the southern interior, but also decreases in rainfall in some places in the far north-east.Although both this study and that of McKellar et al. (2014) agree on the identified increases in the west, this study and that of Kruger (2006) do not extend the increases to the south-western Cape.Kruger (2006) even indicated some significant decreases    along the southern Cape coast.Decreases in the east are more pronounced in the study by McKellar et al. (2014) which could indicate that decreases in the rainfall in this particular area have become more pronounced and significant during recent decades, especially in autumn, since the latter study focuses on more recent decades.
The second important observation, apart from the general trends in the rainfall, is that most of South Africa underwent increases in the intensity of daily rainfall, confirming the results of McKellar et al. (2014).However, all of the relevant regional studies show that very high daily rainfall totals have mostly increased in the southern and south-eastern interior with some variations displayed in the spatial extent of the significant trends.This regional increase in rainfall intensity is also shown on a global scale, where most regions examined by Donat et al. (2013) showed an increase in the R95p index, particularly since the early 1930s.We have shown that the daily intensity of rainfall has increased significantly for many stations in South Africa, which is also reflected by the Donat et al. study, where it is observed that since the mid-1990s there has been a strong upward global trend.
Thirdly, most places in the east showed significant decreases in long spells of continuous rainfall while in the southern interior the longest annual dry spells, which mostly occur during winter months, have shortened.
The results of the historical rainfall trends analysis broadly confirm those of projected changes (DEA, 2013), with an increase in summer rainfall in the western interior and decrease in the east.However, the projected increase in spring rainfall over South Africa is not evident in the historical trend results.

Figure 1b
Figure 1bLocations of rainfall stations used in the trend analysis of individual stations, with rainfall districts represented.Rainfall district borders are superimposed.

Figure 2a
Figure 2aTrends in total annual rainfall in wet days for individual stations for the period 1921-2015.Shaded symbols indicate significant trends at the 5% level.

Figure 2b
Figure 2bSignificant trends in total annual rainfall in wet days for district rainfall totals for the period 1921-2015.Blue-shaded districts indicate significantly positive and red-shaded districts significantly negative trends at the 5% level.Yellow-shaded districts indicate insufficient data for the period.

Figure 4
Figure 4 Trends in r95p, the annual precipitation from daily precipitation > 95th percentile, for the period 1921-2015.Shaded symbols indicate statistical significance at the 5% level.

Figure 5
Figure 5Trends in r99p, the annual total precipitation from daily precipitation > 99th percentile, for the period 1921-2015.Shaded symbols indicate statistical significance at the 5% level.

Figure 6
Figure 6Trends in rx1day, the annual maximum 1-day precipitation, for the period 1921-2015.Shaded symbols indicate statistical significance at the 5% level.

Figure 7
Figure 7Trends in r10mm, the annual count of days when precipitation ≥ 10 mm, for the period 1921-2015.Shaded symbols indicate statistical significance at the 5% level.

Figure 8
Figure 8 Trends in r20mm, the annual count of days when precipitation ≥ 20 mm, for the period 1921-2015.Shaded symbols indicate statistical significance at the 5% level.

Figure 9
Figure 9Trends in r25mm, the annual count of days when precipitation ≥ 25 mm, for the period 1921-2015.Shaded symbols indicate statistical significance at the 5% level.

Figure 10
Figure 10 Seasonal trends in r20mm, the annual count of days when precipitation ≥ 20 mm, for the period 1921-2015 (DJF: December to February -summer, MAM: March to May -autumn, JJA: June to August -winter, SON: September to October -spring).Shaded symbols indicate significant trends at the 5% level.

Figure 12
Figure 12Trends in rx5day, the annual maximum consecutive 5-day precipitation, for the period 1921-2015.Shaded symbols indicate significant trends at the 5% level.

Figure 13
Figure 13Trends in CWD, the annual maximum length of wet days, for the period 1921-2015.Shaded symbols indicate significant trends at the 5% level.

Figure 14
Figure 14Trends in CDD, the annual maximum length of dry days, for the period 1921-2015.Shaded symbols indicate significant trends at the 5% level.

TABLE 1 List of WMO ETCCDI indices used in this study Index Description Unit prcptot
Annual total precipitation in wet days, i.e., days with