Trends and Future Projections of Extreme Precipitation Indices in Limpopo Province, South Africa

2. Study Area, Data and Methods

2.1. Study Area

The study area covers the Limpopo Province, South Africa’s northernmost province, which borders Mozambique, Zimbabwe, and Botswana (Figure 1a). The province is divided into five district municipalities (Figure 1b): Waterberg, Capricorn, Vhembe, Mopani, and Sekhukhune. The Limpopo Province is categorised as having a hot and semi-arid climate according to South Africa’s temperature and precipitation data classification [30]. According to the Köppen climatic classification, the province falls within the Cwa (summer rain with hot summers) and the BS (Semi-arid) climatic regions [31]. In terms of precipitation, the province is made up of three distinct climatic regions: the Lowveld region characterised by a semi-arid climate, the Middle and Highveld classified as semi-arid, and Escarpment that experiences a sub-humid climate [31]. Precipitation in the province occurs mostly between the months of October to April and ranges from 200 to 300 mm in the hot dry areas to 1500 mm in the high precipitation areas [32]. There is high annual precipitation variability in the province as well as relatively frequent extreme precipitation events [33,34].

2.2. Data and Methods

2.2.1. Historical Precipitation Analysis

Precipitation data from the South African Weather Service (SAWS) database were used for the historical precipitation analysis. The availability of precipitation data was considered in conjunction with the homogeneous rainfall districts (Figure 2). South Africa was divided into a total of 94 homogeneous rainfall districts by SAWS (then the South African Weather Bureau [35]). The delineation of these districts was mainly done according to the annual march of maximum rainfall, and the boundaries between the winter, whole year and summer rainfall regions. 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 [36]. However, due to the large number of stations per district and the relative homogeneity of rainfall within districts, the use of the district data sets is arguably more suitable for climate change studies than simply gridded data sets. Long-term daily rainfall stations with relatively long precipitation data sets, i.e. at least since 1921, is shown in Table 1 for the Limpopo Province. The names of the stations and positions are listed per rainfall district.

2.2.2. Future Projections of Precipitation

The spatial resolution of Global Climate Models (GCMs) grid squares is relatively low, especially when applied to produce South African provincial scale climate change projections. Therefore, to address the spatial scale limitations posed by the GCM fields, the Coordinated Regional Downscaling Experiment (CORDEX) dynamically downscaled simulations over the African domain to a grid spatial resolution of 0.44°x 0.44° was used. The Rossby Centre regional model (RCA4), forced across its lateral boundaries by the Global Circulation Models (GCMs) (Table 2) of the 5^th phase of the Coupled Model Intercomparison Project (CMIP5) [37] was used. Rossby Centre Regional Climate Model version 4 (RCA4) was the version used in the downscaling of CMIP5 simulations for CORDEX and with predefined regions, grids, experiment protocols, output variables and output format facilitating easier analysis of possible future regional climate changes, not only by the scientific community but also by end-user communities at regional and local levels [38]. The RCA4 is a coupled ocean-atmosphere Regional Climate Model (RCM) based on the Numerical Weather Prediction (NWP) model HARLAM [39]. For the climate change projections, outputs from the historical or reference period (1976-2005), and three projected time intervals, spanning the current climate from 2006 – 2035; the near-future, defined as the period starting from 2036 – 2065; and the distant future, spanning from 2066 – 2095 were considered. The CORDEX-Africa model simulations under the Representative Concentration Pathway (RCP) 4.5 and RCP8.5 scenarios across the selected time intervals were used for the climate change projections.

Multi-model ensembles, i.e., models produced by combining multiple model ensemble members, are often described as “ensembles of opportunity” [40]. This is attributed to the way they are created, which involves the combination of information from all participating models [41]. Multi-model ensembles are believed to increase the skill, reliability, and consistency of output [42], and are often found to out-perform single-models [43,44]. In this study, multi-model ensembles refer to a set of model simulations from eight different CORDEX-Africa models (Table 2). These model ensembles were created by using the Simple Multi-model Averaging (SMA) technique [45]. The SMA approach can be described as:

$${\left(P_{SMA}\right)}_t=\overline{P_{obs}}+{\sum }_{i=1}^N{\displaystyle \frac{{\left(P_{sim}\right)}_{i,t}-{\left(\overline{P_{sim}}\right)}_i}{N}}$$

(1)

where ${\left(P_{SMA}\right)}_t$ is the multi-model precipitation simulations from CORDEX-Africa models derived using SMA at time t, ${\left(P_{SMA}\right)}_{i,t}$ corresponds to the i^th model precipitation simulation for time t, ${\left(\overline{P_{sim}}\right)}_{i,t}$ is the time average of the i^th model precipitation simulation, $\overline{P_{obs}}$ corresponds to the observed average precipitation and N represents the number of models under consideration.

Table 2. List of the Global Circulation Models (GCMs) used in the study.

Model name	Country	Resolution	Literature
CanESM2m	Canada	2.8° x 2.8°	[46]
CNRM-CM5	France	1.4° x 1.4°	[47]
CSIRO-Mk3	Australia	1.9° x 1.9°	[48]
IPSL-CM5A-MR	France	1.9° x 3.8°	[49]
MIROC5	Japan	1.4° x 1.4°	[50]
MPI-ESM-LR	Germany	1.9° x 1.9°	[51]
NorESMI-M	Norway	1.9° x 2.5°	[52]
GFDL-ESM2M	USA	2.0° x 2.5°	[53]

Table 2. List of the Global Circulation Models (GCMs) used in the study.

Model name	Country	Resolution	Literature
CanESM2m	Canada	2.8° x 2.8°	[46]
CNRM-CM5	France	1.4° x 1.4°	[47]
CSIRO-Mk3	Australia	1.9° x 1.9°	[48]
IPSL-CM5A-MR	France	1.9° x 3.8°	[49]
MIROC5	Japan	1.4° x 1.4°	[50]
MPI-ESM-LR	Germany	1.9° x 1.9°	[51]
NorESMI-M	Norway	1.9° x 2.5°	[52]
GFDL-ESM2M	USA	2.0° x 2.5°	[53]

2.2.3. Precipitation Extreme Indices

In this study, selected precipitation extreme indices from the original 27 core climate indices developed by the Expert Team (ET) on Climate Change Detection and Indices (ETCCDI) [54] were analysed. These precipitation indices were selected based on their relevance to the climatology and hydrology of the Limpopo Province and are presented in Table 3. Eight precipitation extreme indices were analysed to assess the projected changes in precipitation in the province. For the historical analysis trend magnitudes were determined with Sen’s slope and the statistical significance with the Mann-Kendall (MK) test at the 5% level of significance (p < 0.05).

The non-parametric MK monotonic trend test was used to detect the increasing or decreasing trends in the time series precipitation data [36,55]. The MK test is given by

$$S={\sum }_{k-1}^{n-1}{\sum }_{j-k+1}^n\mathrm{sgn}\left(x_j-x_k\right)$$

(2)

where x_j and x_k represent a precipitation variable, and k is the total number of data points available in the time series for analysis.

Sen’s non-parametric estimator method has been used for predicting the magnitude (true slope) of precipitation time series data. The Sen’s slope estimator method uses a linear model for trend analysis and is computed using the following:

$$Q={\displaystyle \frac{Yi-Yj}{Ni-i}}$$

(3)

where Q is the slope estimate; Yi and Yj are the values at times i and j, where i is greater than j; N’ is all data pairs for which i is greater than j. Positive values for Sen’s slope indicates an increasing trend, and a negative indicates a decreasing trend over a given time.

Since the station data selected are stations that did not undergo any moves throughout their recording periods, no bias correction was deemed necessary.

For the climate change projections, precipitation extreme indices were calculated from daily precipitation CORDEX ensemble data using the Climate Data Operator (CDO). For detailed information on the calculation of climate extreme indices, the reader is referred to the following website: http://etccdi.pacificclimate.org/list_27_indices.shtml. The climate change indices relating to precipitation were used based on both the historical datasets and the projections across the three-time intervals. The results were then inter-compared with the reference period to assess the projected change in precipitation extremes over time, in relation to the historical period. This comparison method overcomes biases in precipitation frequency and intensity, using raw model outputs for extreme indices like R20mm or Rx1day, which can lead to inaccurate projections Figure 3 presents a flowchart diagram depicting the data and methods used in this study.

3. Results and Discussion

3.1. Historical Trends

The extreme precipitation indices were assessed for each rainfall district using the station data specified in Table 1. However, the analysis presented in this section focuses on trends in the extreme precipitation indices for stations with very long time series (i.e., 1921 – 2021, as listed in Table 1) for ease of reference. Figure 4 displays the trends in annual total precipitation (prcptot) based on wet days (daily precipitation >= 1 mm). Out of the 15 stations with sufficient data, 10 exhibit negative trends, while five show positive trends. Among the stations with negative trends, six are statistically significant, whereas none of the stations with positive trends are significant. Therefore, a noticeable long-term drying trend is apparent in large parts of the province. The annual maximum 1-day precipitation, rx1day, is significant as it can signal a potential shift in the maximum daily precipitation expected annually or over specific return periods. Figure 5 illustrates the trends in rx1day, with most stations showing negative trends (12 stations decreasing vs. 3 stations increasing), indicating mostly a decrease in the highest daily precipitation per year, with only one station showing a significantly positive increase.

The trends in r10mm and r20mm are presented in Figure 6 and Figure 7, respectively. Most stations (11 out of 15) exhibited negative trends in r10mm, with seven being significantly negative. This reflects the overall decrease in precipitation across most areas of the province. In the case of r20mm, the number of stations showing significantly positive trends has risen from one to three, although the majority still display negative trends. In climate regions which become drier, it is expected that the number of days with precipitation above lower thresholds will decrease. However, if the precipitation climate is becoming more extreme, the number of days with precipitation exceeding higher thresholds can either increase or decrease.

The SDII, the Simple Daily Intensity Index, defined as the annual mean daily precipitation intensity is shown in Figure 8. The trends in SDII indicate that precipitation is becoming more extreme (eight exhibit significant positive trends).

The data in Figure 9 shows the trend in the annual maximum number of consecutive days with precipitation equal to or greater than 1 mm (CWD). The results of the CWD trend confirm a drying climate in the province. Out of the 15 stations, 12 show negative trends with nine being statistically significant. Consistent with previous findings, the trends in CDD index defined as the maximum number of consecutive days with less than 1 mm of precipitation, also indicate a drying climate, with 11 stations showing longer annual maximum dry spell lengths (six stations with statistically significant trends) (Figure 10). It is worth noting that stations in the east, where negative trends are observed (though not statistically significant), may suggest that the drying trend is less pronounced near the eastern escarpment compared to the rest of the province. The Limpopo Province falls in the summer rainfall region of South Africa and therefore the longest dry period is usually experienced in winter. During this time, very little precipitation is received, with almost no precipitation occurring in the central and western parts of the province.

The r95p index is defined as the annual total precipitation from daily precipitation greater than the 95th percentile (Figure 11). In a drying climate, as indicated by historical precipitation trends across the province, a decrease would be expected if precipitation is not becoming more extreme overall. However, this is not the case, as there are clear signals of more extreme precipitation, especially in the analysis of r95p (Figure 11). Out of the 15 stations, nine stations exhibit positive trends (seven of which are significant), while six stations show negative trends (two of which are significant).

3.2. Projected Trends

3.2.1. Historical and Projected CORDEX Precipitation

The average historical CORDEX ensemble mean annual precipitation (MAP) for the reference period (1976-2005) is presented in Figure 12. The MAP varied from a minimum of 300 mm to a maximum of 1500 mm for some parts of the province with mean MAP of 740 mm during the reference period. As shown in Figure 11, very high annual total precipitation (> 1200 mm) was observed for the southern (south-eastern Sekhukhune), central (south-eastern Capricorn, western Mopane and central Waterberg), as well as the northern (central Vhembe) parts of the province. Projected changes in MAP (mm) for Limpopo Province with reference to the baseline period (1976-2005) for the three projection time periods under RCP4.5 and RCP8.5 scenarios are depicted in Figure 13. Overall, the results indicate that MAP (mm) is projected mostly to decrease over the province for all the projection periods under both RCPs 4.5 and 8.5. Under the RCP 4.5 scenario, a maximum decrease of 210 mm (2006-2035) to 335 mm (2066-2095) was noted, with mean projected decline of 17 mm (2006-2035), 36 mm (2036-2065), and 54 mm (2065-2099) for the province (Figure 13a, c, and e). These projected highest reductions in MAP are more pronounced for the areas which receive higher annual precipitation (Figure 12), including the southern (south-eastern Sekhukhune), central (south-eastern Capricorn, western Mopane and central Waterberg), and northern (central Vhembe) parts of the province. However, MAP is projected to increase slightly (40 mm) for the south-western part (western Waterberg) of the province under RCP 4.5 scenario.

For the RCP 8.5 scenario, MAP is also projected to decrease with higher magnitude in the near to far future periods over most parts of the province compared to RCP 4.5 (Figure 13). A maximum projected decline in MAP of 350 mm (2036-2065) to 510 mm (2066-2095) were observed under this RCP, with mean projected decreases of 19 mm (2006-2035), 54 mm (2036-2065), and 85 mm (2065-2099) for the province (Figure 13b, d, and f). Similarly, highest reductions in MAP are more pronounced for the southern (south-eastern Sekhukhune), central (south-eastern Capricorn, western Mopane and central Waterberg) parts of the province (Figure 13d and f).

3.2.2. Extreme Precipitation Indices

The highest one-day precipitation amount per time period (rx1day) projections are presented in Figure 14 for the Limpopo Province under both RCP 4.5 and 8.5 scenarios. The maximum rx1day is projected to increase under RCP 4.5, with a mean increase of 43 mm for the period 2066-2095 (Figure14e) compared to the reference period. For RCP 8.5, rx1day projections ranged from a minimum of 20 mm to a maximum of 90 mm (Figure14b, d, and f). As shown in Figure14, the north and eastern parts of the province are projected to receive the highest 1-day precipitation under both RCP4.5 and 8.5 scenarios for all periods including some central parts (south-eastern Capricorn, western Mopane and central Waterberg) in 2036-2065 (Figure14d) and 2066-2095 (Figure14f).

The projected changes in the number of heavy precipitation days index per time period (r10mm) under RCP 4.5 and 8.5 for Limpopo province are presented in Figure 15. The r10 mm is projected to decrease for most parts of the province for all projection periods under both RCP4.5 and 8.5 scenarios. These results indicate that areas which receive higher amount of precipitation in the province are projected to have a maximum decline in r10mm (750 days) over the far future period (2066-2095) under RCP 8.5 (Figure15f). However, a slight increase in r10mm (40 – 70 days) is projected for some areas in the south-western parts of the province (south and western Waterberg) as shown in Figure 15.

The projected changes in the number of heavy precipitation days index per time period (r20mm) under RCP 4.5 and 8.5 scenarios for Limpopo province are presented in Figure 16. As shown in Figure 16, r20mm is also projected to decrease for most parts of the province for all projection periods under both RCP4.5 and 8.5 scenarios. Some areas which receive higher amount of precipitation in the province are projected to have a maximum decline in r20mm (260 days) for the near and far future periods under RCP 8.5. However, a slight increase in r20mm (2 – 20 days) is projected for some areas in the south-western parts of the province (south and western Waterberg and Sekhukhune) as shown in Figure 16.

Simple daily intensity index (SDII) values for the Limpopo Province for the three projection periods are shown in Figure17. As shown in Figure 17, the northern (Vhembe), central (south-eastern Capricorn and western Mopane), western (central Waterberg), and the southern (south-eastern Sekhukhune) parts of the province have higher SDII values. The SDII values ranged from a minimum of 2.68 to a maximum of 9.74 for the province, with mean values ranging from 4.15 to 4.29 for all three projection periods under both RCPs 4.5 and 8.5 scenarios (Figure 17). There is no significant difference in SDII between the reference period and the three projection periods under both RCPs 4.5 and 8.5, however, mean SDII for the province is projected to decrease slightly (<0.1) under both RCP4.5 and 8.5 for all three projection periods (Figure 17).

Projected changes in the number of consecutive wet days index (CWD) per time period with reference to the baseline period (1976-2005) for the Limpopo Province under RCP4.5 and RCP8.5 scenarios are shown in Figure 18. The projections indicate a decrease in CWD for most parts of the province as shown in Figure 18. A maximum decline in the number of CWD (70 days) is noticed for the southern and northern parts of the province under RCP 8.5 for the far future period (Figure 18).

The projected changes in consecutive dry days index per time period (CDD) are shown in Figure 19 for Limpopo Province. CDD is projected to increase for most parts of the Limpopo Province except for the western (parts of Waterberg district) and the north-eastern (Vhembe and Mopani) parts of the province under both RCPs 4.5 and 8.5 (Figure 19). The projected CDD for the province is likely to increase by a maximum of 45 to 85 days under both RCP4.5 and RCP8.5 compared to the reference period.

The projections for the very wet days with respect to the 95^th percentile of reference period (R95p) under RCP4.5 and RCP8.5 for Limpopo Province are shown in Figure 20. The percentage of wet days where the daily precipitation amount is greater than the 95^th percentile of the daily precipitation amount at wet days for the reference period (R95p) varied between 4 to 10 under RCP 4.5 as shown in Figure 20a, c, and e. Similarly, the range of R95p values for Limpopo province under RCP8.5 were between 3 and 11 for all projection periods (Figure 20b, d, and f). The highest percentage increase in R95p is noticed for some areas of the northern (north-western Vhembe), western (western Waterberg), and southern (southern Sekhukhune) parts of the province.

4. Discussion and Conclusions

The analysis of long-term daily precipitation provides a comprehensive overview of historical trends in precipitation in the Limpopo Province. Most extreme indices indicate a positive trend based on data from individual stations. For example, annual precipitation amounts from very wet days, which are in the top 5% of daily rain totals historically, have increased significantly for most stations with long-term data since 1921. However, the province has experienced a general drying trend, leading to longer periods without precipitation each year, less days with significant or heavy precipitation, and shorter maximum wet spells per year. Overall, the climate in the Limpopo Province has become drier to varying degrees, with increased sub-seasonal extremes in daily precipitation indices.

The province is expected to experience a decrease in precipitation across the three projection time periods under both RCP4.5 and RCP8.5. Reductions are more notable in the south-eastern parts of Sekhukhune district municipality, the central (south-eastern Capricorn, western Mopane, and central Waterberg), and northern (central Vhembe) parts of the province. Increased precipitation is also expected in the south-western part (western Waterberg) of the province under the RCP 4.5 scenario. Localised spatial variations are observed across the analysed precipitation extreme indices over the projected periods. The numbers of heavy (and very heavy) precipitation days are likely to decrease for most parts of the province for all projection periods under both RCP scenarios. The maximum number of consecutive wet days (CWD) is projected to decline in the southern and northern parts and increase in the southwestern parts of the province. Consecutive dry days and the maximum number of consecutive dry days (CDD periods) are projected to significantly increase during the analysed periods under both RCP4.5 and RCP8.5 scenarios.

Other studies on climate indices using various climate change scenarios based on GCM and RCM ensembles show a consistent trend towards hotter and drier conditions across all subregions in Africa [56,57]. These findings align with previous studies that have also projected changes in daily precipitation patterns in southern Africa [26,58] and extreme climate events in East Africa [59]. Future trends in extreme precipitation events across the South America region also suggest an increase in severity, frequency, and duration [60,61]. The Reboita et al. study [61] investigated the impact of key atmospheric teleconnection patterns, such as ENSO, on a region in South America with a Köppen climatic classification similar to the Limpopo province. Their study demonstrated how these global drivers alter local circulation, leading to extreme weather events such as droughts and floods in the Gran Chaco region of central South America [62]. Future trends in extreme precipitation events across the South America region also suggest an increase in severity, frequency, and duration [60,61] as well as in China [63].

This study uses the RCA4 regional model driven by various GCMs. Relying on a single RCM is a limitation as it overlooks model structural uncertainty and the differences in how various RCMs like CCLM (COSMO-CLM) and RACMO2 replicate regional physics. A multi-RCM ensemble would have provided more dependable results, and future research should take this into consideration. The RCA4 data used in this study, like most RCMs, have systematic errors (biases) in their output. Bias correction was not applied to the daily precipitation used in the analysis of precipitation indices in this study. The difference between the projection periods for RCP 4.5 and RCP 8.5 relative to the reference period was used to overcome the biases. In addition, multi-model ensembles of the individual RCMs were used, which are believed to increase the skill, reliability, and consistency of output and are often found to outperform single-models [64,65]. Furthermore, Teng et al. [66] noted that bias correction methods cannot overcome limitations of the RCM in simulating precipitation sequence, it can introduce additional uncertainty to change signals in high precipitation amounts. Therefore it was suggested that future climate change impact studies need to take this into account when deciding whether to use raw or bias corrected RCM data, and projections derived from bias corrected climate input should therefore be interpreted with caution and should be combined with other process-based approaches.

Climate change is expected to significantly impact annual variations in precipitation in the Limpopo Province, leading to more frequent droughts and floods. Projections suggest that precipitation will decrease, while extreme precipitation events will become more common and severe. These changes are likely to have adverse effects on water resources, agriculture, and ecosystems. Potential consequences include reduced water quality, increased risk of flooding and drought, decreased water availability, crop damage, soil erosion, loss of soil fertility, reduced hydropower potential, migration, higher energy demand, declining wildlife populations, loss of biodiversity, decreased tourism revenue, increased poverty and food insecurity, disruption of settlements, and increased costs for disaster response. These factors will particularly affect the development of several rural communities in the province that heavily rely on natural resources. Adaptation measures should focus on reducing the vulnerability of rural communities by promoting agricultural resilience through crop diversification and selection, implementing water resource management and security initiatives, establishing early warning systems, and encouraging livelihood diversification. Mitigation efforts should prioritize promoting renewable energy, reducing deforestation, and implementing sustainable land use practices.

The findings of this study can help support provincial initiatives to effectively plan and manage water resources in response to the changing climate. Additionally, the findings may also contribute towards enhancing the ability of farmers, catchment management agencies, and extension services in the province in developing sustainable adaptation strategies to mitigate the effects of climate change.