1 Introduction

Climate change has significant impacts globally (IPCC 2013, 2021) including southern Africa (Engelbrecht et al. 2015; Maúre et al. 2018; Almazroui et al. 2020). In this southern Africa region lies the Okavango River Basin (ORB; Fig. 1) focused on in the present study, which, like most of the subcontinent, is affected by large climate variability which is likely to be worsened by climate change (Andersson et al. 2003; Wolski and Murray-Hudson 2008). The ORB, stretching from Angola through Namibia to Botswana, contains the world-famous Okavango Delta located in the latter country. The Delta is a world heritage and Ramsar site famous for its highly biodiverse and sensitive ecosystems (Murray-Hudson et al. 2006; UNESCO 2014) supporting economically vital ecotourism (Mbaiwa 2004, 2015). These ecosystems crucially depend on the highly seasonal ORB streamflow, which is also a critically important source of freshwater for the rural population, most of whom rely on rain-fed subsistence farming like over the bulk of southern Africa (McCarthy et al. 2003; Andersson et al. 2003, 2006; Kgathi et al. 2006; Murray-Hudson et al. 2006; Moses and Hambira 2018).

Fig. 1
figure 1

The Okavango River Basin (ORB), i.e., the blue polygon within the box 11–25°S, 14.5–28.5°E, with its location in southern Africa shown in the insert. “A” and “B” within the ORB denote the two main rivers, namely, the Cubango and Cuito, respectively, which merge to form the Okavango River which then feeds into the Okavango Delta through Mohembo hydrological station (orange circle). The ORB is also subdivided into the north catchment (red dashed polygon denoted by “NC”) and Ngamiland (black dashed polygon denoted by “Ngami”)

Full size image

The large climate variability impacts rain-fed subsistence farming and water availability (Tyson 1986; Reason et al. 2006; Murray-Hudson et al. 2006; Conway et al. 2015; Moses and Hambira 2018) as well as wildlife distribution and tourism (Mbaiwa 2004, 2015; Murray-Hudson et al. 2006). It is worth noting that clearing of vegetation over large areas for farming and potential oil mining in the ORB region (Richardson 2021) could cause widespread ecological disruption. Factors contributing to the mentioned large climate variability are complex, with the El Niño-Southern Oscillation (ENSO) in the tropical Pacific being the main interannual climate mode affecting the region (Lindesay 1988; Reason et al. 2000; Reason and Jagadheesha 2005; Blamey et al. 2018; Hart et al. 2018; Moses et al. 2022). Recurring climate extremes such as droughts, hot days and floods brought about by extreme rainfall events are common (e.g., Tyson 1986; Reason et al. 2006; Meque et al. 2022; Moses et al. 2023a, 2023b), and their intensities and frequencies are likely to increase over the region due to global warming (IPCC 2013, 2021). Changes in the climate are also associated with tropical diseases such as schistosomiasis (Gabaake et al. 2023).

Temperatures are generally high particularly in the Botswana part of the ORB, which may worsen the impacts of severe drought for example (Moses 2017; Moses and Gondwe 2019; Moses et al. 2022). Rainfall occurs mainly during October to March or April, with the Botswana/Namibian (Angolan) part of the basin having a semi-arid (subtropical) climate. Moses et al. (2023a) found significant drying and warming trends over the ORB and most of southern Africa particularly in the early summer (October–November) during the period 1981–2021. These trends are consistent with warming (Barros and Field 2014; Engelbrecht et al. 2015; Maúre et al. 2018; Meque et al. 2022) and early summer drying (IPCC 2021; Munday and Washington 2019; Wainwright et al. 2021) found over southern Africa. If these trends persist, they may have adverse impacts on rain-fed subsistence farming (Guilpart et al. 2017), water availability and ecosystems (Murray-Hudson et al. 2006). Decreases in rainfall combined with increases in temperature may increase soil moisture losses which may limit crop growth.

While relatively more studies have considered rainfall and temperature characteristics for the recent past over the ORB region or over southern Africa, relatively limited research has been done on future characteristics of these variables over the former region. Some of the above-mentioned studies like Wainwright et al. (2021) considered future rainfall characteristics over southern Africa more generally, not necessarily specifically over the ORB region, based on the Coupled Model Intercomparison Project Phase 6 (CMIP6; Eyring et al. 2016). They found delays in the wet seasons of up to 2 weeks by the end of the twenty-first century across southern Africa. Others like Almazroui et al. (2020) also used CMIP6 models to examine future changes in rainfall and temperature over Africa as a whole and over eight subregions which included southwest Africa (southern Africa landmass west of 26°E longitude and south of 10°S latitude). These authors considered future changes in rainfall and temperature on annual scales and specifically for the boreal winter (December-February) and boreal summer (July–August). They found large spatial variability in rainfall, and significant warming which varied regionally, but changes in the variables over the ORB region were not significantly addressed. Lim Kam Sian et al. (2021) considered changes in rainfall in the twenty-first century over southern Africa using regional demarcations like those used by Almazroui et al. (2020). Focusing on December-March and May–October, Lim Kam Sian et al. (2021) found that rainfall trends over large regions cannot be generalised, hence recommended localised studies to provide more accurate findings.

The principal aim of this study is to analyse future (up to the end of the twenty-first century) changes in rainfall and temperature using CMIP6 models over the ORB region, focusing on the extended rainy season October–April (months during which most crops are grown), split into October–November (ON), December-February (DJF) and March–April (MA), as explained further in Section 2. Due to the massive improvements of the CMIP6 compared to previous CMIP projects (Eyring et al. 2016), CMIP6 models are expected to give more reliable results as shown by previous studies (Mauritsen et al. 2019; Moseid et al. 2020), hence these models are used in the present study. Given the vulnerability of the ORB region and the dependency of the local population on rain-fed subsistence farming, the results of the study may help improve the management of agricultural activities, water resources and the highly biodiverse ecosystems. Improved management is critical for sustainable use of the ORB resources.

2 Data and methods

The study area is the ORB and its selected two subregions, namely, the north catchment and Ngamiland (Fig. 1). The former subregion is where most of the ORB streamflow is generated and the latter subregion contains the world-famous Okavango Delta, where most of the streamflow terminates (Andersson et al. 2003; Wolski and Murray-Hudson 2008; Moses et al. 2023a, 2023b). Due to lack of high-resolution observational data in the study area, the 0.05° resolution Climate Hazards Infrared Rainfall with Station data (CHIRPS; Funk et al. 2015) version 2, and the 2-m air temperature (T2m) from 0.25° resolution ERA5 reanalyses (Copernicus Climate Change Service 2017; Hersbach et al 2020), were analysed to depict the annual cycles and climatologies of rainfall and temperature. These CHIRPS and ERA5 T2m data have been found to agree well with station data (Moses et al. 2022; Moses 2023). To be in line with the Intergovernmental Panel on Climate Change Sixth Assessment Report (IPCC 2021), which is based on CMIP6 models, ERA5 T2m and CHIRPS annual cycles and climatologies were analysed for the period 1985–2014, taken to be the historical period. The climatologies were analysed for ON, DJF and MA, based on previous analyses for rainfall over the ORB (Moses et al. 2022, 2023a) and for dry spells over southern Africa (Thoithi et al. 2021). Very little rain occurs in the ORB during May–September. Typically, December to February are the wettest months, hence the main growing season. Some years also experience significant falls in ON (onset of the rainy season) or March–April (end of the rainy season).

In addition, CHIRPS and ERA5 T2m data were used to evaluate the performance of the selected CMIP6 models (discussed later in this section) in reproducing the observed annual cycles of rainfall and temperature over the study area, and to evaluate the performance of these models in reproducing the observed interannual variability in rainfall and temperature changes during the historical period, focusing on ON, DJF and MA, for the same reasons as above. For the performance evaluation, model historical simulations were used. The metrics used included the bias, root-mean-square-error (RMSE) and correlations, which have also been used in previous studies (Almazroui et al. 2020; Ayugi et al. 2021; Xu et al. 2021; Jamal et al. 2023). Biases and RMSEs close to zero indicate that model data are close to the observations. A correlation of -1 represents an opposite relationship, 0 means no correlation, and 1 represents a perfect correlation.

Based on the results of the model performance evaluation and on previous studies (Xu et al. 2021; Jamal et al. 2023) that model projections without bias correction are highly uncertain, in the present study model projections were bias corrected using the mean bias correction method (Holland et al. 2010; Bruyère et al. 2014). This method was applied to model outputs for the historical period (1985–2014) and for the selected future periods, i.e., 2030–2059 (near term) and 2070–2099 (long term), so that the difference between projections for the future and historical periods is a result and signal for climate change in the seasons (ON, DJF, ON) of interest. The mean bias correction method was applied to rainfall and temperature simulations based on Eqs. 1 and 2, respectively.

$$\mathrm{Corr}\;\mathrm{sim}\;\mathrm{RN}=\mathrm{uncorrected}\;\mathrm{sim}\;\mathrm{RN}/(\mathrm{mean}\;\mathrm{observed}\;\mathrm{RN}/\mathrm{mean}\;\mathrm{sim}\;\mathrm{RN})$$
(1)
$$\mathrm{Corr}\;\mathrm{sim}\;\mathrm T=\mathrm{uncorrected}\;\mathrm{sim}\;\mathrm T-(\mathrm{mean}\;\mathrm{sim}\;\mathrm T-\mathrm{mean}\;\mathrm{observed}\;\mathrm T)$$
(2)

where corr, sim, RN and T denote corrected, simulated, rainfall and temperature, respectively.

The CMIP6 models that were used in this study were the CESM2-WACCM (Danabasoglu 2019) (hereafter, CESM), CMCC-CM2-SR5 (Lovato and Peano 2020) (hereafter, CMCC), MPI-ESM1-2-HR (Schupfner et al. 2019) (hereafter, MPI), and the average of these models (hereafter, model average). Selection of these models was based on the available computer resources used to perform the analyses, downloadability (dependent on internet connectivity) of historical and projected model outputs, as well as on relatively high model grid resolutions. The CESM and CMCC models have resolutions (longitude x latitude) of 1.25° × 0.94° each, whereas the MPI model has a resolution of 0.94° × 0.94°. These models were re-gridded to a common grid of lowest model resolution using linear re-gridding scheme. The first realization (r1i1p1f1) for each model was used in the analysis. The model projections were analysed under two shared socioeconomic pathways (SSPs; Gidden et al. 2019), namely, SSP245 and SSP585. SSP585 is the highest emission scenario and the SSP245 scenario represents a medium emission scenario (Van Vuuren et al. 2011; O’Neill et al. 2016).

Trends in rainfall and temperature changes were considered. They were tested for statistical significance at the 95% confidence level, using the Hamed and Rao (1998) and Yue and Wang (2002) tests. Both tests are modified from the nonparametric Mann–Kendall test (MKT) (Mann 1945; Kendall 1975), which is widely used because it exempts datasets from normal distribution requirements and is less sensitive to outliers. However, unlike the modified trend tests used in the present study, the original MKT does not take data autocorrelation into consideration.

3 Results and discussion

3.1 Observed climatology and model performance

Figures 2a-c show the observed rainfall climatology over the ORB in ON, DJF and MA, for the historical period. In all three seasons, the north catchment receives more rainfall than Ngamiland (see Fig. 1 for locations). This is consistent with Moses et al. (2023b) who found the former region to experience more extreme rainfall events than the latter region during their study period (1981–2021). The main contributors to high rainfall over the former region include the Angola Low and orographic uplift (Munday and Washington 2017; Moses et al. 2022). Figures 2a-c also show that most of the rainfall occurs in DJF, with the northern part of the north catchment receiving 7–8 mm/day, and the southern part of Ngamiland receiving 2–3 mm/day. Although MA experiences less rainfall than DJF, it generally experiences more rainfall than ON. The observed temperature climatology over the ORB in ON, DJF and MA, for the historical period is shown in Figs. 3a-c. Most parts of the ORB including Ngamiland and the southern parts of the north catchment, experience largest temperatures (seasonal mean of 26–28 °C) in ON, which is also relatively drier compared to the other two seasons as already mentioned. The southern parts of Ngamiland also experience the same range of temperatures in DJF. Overall, over the ORB, the lowest temperatures (16–18 °C) in all three seasons occur over the northern parts of the north catchment and the highest temperatures (26–28 °C) occur over Ngamiland. Occurrence of highest temperatures over Ngamiland is in harmony with previous studies (Moses and Gondwe 2019; Moses et al. 2022).

Fig. 2
figure 2

Observed rainfall climatology (based on CHIRPS) over the ORB, for (a) ON, (b) DJF and (c) MA, 1985–2014. The ORB and its subdivisions are described in Fig. 1

Full size image
Fig. 3
figure 3

Observed temperature climatology (based on ERA5 T2m) over the ORB, for (a) ON, (b) DJF and (c) MA, 1985–2014. The ORB and its subdivisions are described in Fig. 1

Full size image

The performance of historical simulations from the CMIP6 models in reproducing the observed rainfall annual cycle for the historical period over the north catchment and Ngamiland is shown in Figs. 4a-b. Note that the climatological year is defined as starting from July and ending in June the following year to capture the full austral summer rainfall period (October–April) as a single season (Maruatona and Moses 2022). The figure shows that over both regions, all the models (CESM, CMCC, MPI and the model average) generally overestimate the observed rainfall annual cycle. The performance of the models is consistent with other studies over southern Africa (Lim Kam Sian et al. 2021), over Africa as a whole (Almazroui et al. 2020) and elsewhere (Ayugi et al. 2021; Liang-Liang et al. 2022). It is noted that rainfall values for the north catchment are larger than those for Ngamiland (Figs. 4a-b), consistent with Fig. 2. The wettest months over both regions are November-March and the driest ones are May–September (Figs. 4a-b), consistent with previous authors (Moses et al. 2022). Considerable amount of rainfall also occurs in October and April.

Fig. 4
figure 4

(a) Annual cycle of rainfall over the north catchment based on historical (1985–2014) simulations from the CMIP6 models (CESM, CMCC, MPI, model average) compared with observations (CHIRPS). (b) As in (a) but for Ngamiland. (c) Annual cycle of temperature over the north catchment based on historical simulations from the CMIP6 models compared with observations (ERA5 T2m). (d) As in (c) but for Ngamiland. Locations of the north catchment and Ngamiland are shown in Fig. 1

Full size image

Figures 4c-d show the performance of the models (historical simulations) in reproducing the observed temperature annual cycle for the historical period over the north catchment and Ngamiland. Over both regions, temperatures reach their minimums in July, increase sharply thereafter until they reach their peak in October before dropping again. As found for rainfall, the models generally overestimate the observed temperature annual cycle. Over Ngamiland, the exception is in January-March (December) when the MPI (CMCC) underestimate this annual cycle. Over the north catchment, the CESM and the model average generally overestimate this annual cycle except in July–September whereas the CMCC overestimates it in all the months. It is only the MPI that generally underestimates the observed temperature annual cycle except in October-December. Ayugi et al. (2021) suggested that model overestimations maybe related to equilibrium climate sensitivity.

Figure 5 shows the performance of the models (historical simulations) in reproducing the observed interannual variability in rainfall changes over the north catchment. Generally, the performance of the models is fair. It is worth noting that in ON during some years like 1985–1986 and 2006 (1989 and 1999), all the models including the model average underestimate (overestimate) the observed interannual variability in rainfall changes (Fig. 5a). Compared with the individual models, the model average resembles the observed interannual variability better (Figs. 5a-c). Previous authors noted that model average reduces biases by partly cancelling them out (Pincus et al. 2008; Ayugi et al. 2021). Figure 6 is as Fig. 5, but for Ngamiland. As found for the north catchment, the performance of the historical simulations from the models in reproducing the observed interannual variability in rainfall changes is also fair over Ngamiland. The individual models and the model average resemble the observed variability better in ON (Fig. 6a) than in DJF and MA (Figs. 6b-c).

Fig. 5
figure 5

(a) Changes in rainfall over the north catchment based on historical (ON 1985–2014) simulations from the CMIP6 models [CESM, CMCC, MPI, model average (aver)] compared with observations (CHIRPS). (b) As in (a) but for DJF. DJF 1985 means December 1985 to February 1986, and so on. (c) As in (a) but for MA. Location of the north catchment is shown in Fig. 1

Full size image
Fig. 6
figure 6

As in Fig. 5 but for Ngamiland, whose location is shown in Fig. 1

Full size image

The performance of the models (historical simulations) in reproducing the observed interannual variability in temperature changes over the north catchment is shown in Fig. 7. Generally, the models fairly reproduce the observed interannual variability. As found for rainfall, the model average shows a better resemblance of the observed interannual variability in temperature changes than the individual models. Also, as found for rainfall, the models including the model average overestimate (underestimate) the observed variability during some years such as DJF 2011–2013 (DJF 1987–1988), as can be seen mainly in Fig. 7b. Figure 8 is as Fig. 7 but for Ngamiland. The interannual variability in temperature changes is larger over Ngamiland than over the north catchment. As in the case for the north catchment, the individual models and the model average overestimate (underestimate) the observed interannual variability in temperature changes over Ngamiland during some years like 2011–2014 (1998–1999), which is clearer in Fig. 8b (Fig. 8c).

Fig. 7
figure 7

(a) Changes in temperature over the north catchment based on historical (ON 1985–2014) simulations from the CMIP6 models [CESM, CMCC, MPI, model average (aver)] compared with observations (ERA T2m). (b) As in (a) but for DJF. DJF 1985 means December 1985 to February 1986, and so on. (c) As in (a) but for MA. Location of the north catchment is shown in Fig. 1

Full size image
Fig. 8
figure 8

As in Fig. 7 but for Ngamiland, whose location is shown in Fig. 1

Full size image

The computed biases (Table 1), which are generally positive, suggest that historical simulations from the models generally overestimate the observed interannual variability in temperature changes (Figs. 7 and 8). There are only a few cases where the models underestimate (negative biases) the observed interannual variability in temperature changes, such as in MA when the MPI biases are -0.31 °C over the north catchment and -0.18 °C over Ngamiland (Table 1). Similarly, rainfall biases (Table 1) suggest that historical simulations from the models mainly overestimate the observed interannual variability in rainfall changes (Figs. 5 and 6). It is only in ON when the CESM and MPI underestimate the observed interannual variability in rainfall changes, with these models showing biases of -0.62 mm/day (over the north catchment) and -0.10 mm/day (over Ngamiland), respectively (Table 1). Although the models’ historical simulations generally overestimate the observed interannual variability in rainfall and temperature changes (Table 1), RMSEs (Table 2) suggest that overall, the performance of models (Figs. 5–8) is fair since these errors are generally low over both the north catchment and Ngamiland, with those for temperature (rainfall) ranging from 0.77 to 4.00 °C (0.79 to 3.29 mm/day) in all three seasons. Correlations between the observed and simulated (historical) interannual variability in rainfall and temperature changes (Figs. 5–8) were generally weak, except for those between the CESM temperature and ERA5 T2m in ON, which were positive and significant over both the north catchment and Ngamiland (0.35 over both regions).

Table 1 Biases of temperature (in °C) and rainfall (in mm/day) from historical simulations of CMIP6 models [CESM, CMCC, MPI, model average (aver)] compared with observations (ERA5 T2m and CHIRPS) over the north catchment and Ngamiland, in ON, DJF and MA, 1985–2014. The biases have been computed from seasonal fields
Full size table
Table 2 RMSEs of temperature (in °C) and rainfall (in mm/day) from historical simulations of CMIP6 models [CESM, CMCC, MPI, model average (aver)] compared with observations (ERA5 T2m and CHIRPS) over the north catchment and Ngamiland, in ON, DJF and MA, 1985–2014. The RMSEs have been computed from seasonal fields
Full size table

Based on the results of model performance evaluation, the model outputs were bias corrected using the method described in Section 2. The bias corrected model outputs were used in the next section to analyse future climate projections over the study area.

3.2 Future climate projections

3.2.1 Rainfall projections

Figure 9 shows bias corrected future projections (up to the end of the twenty-first century) of changes in rainfall under the SSP245 and SSP585 scenarios over the north catchment. Generally, all the models including the model average show a mixture of increases and reductions in average rainfall during the future periods [near term (2030–2059) and long term (2070–2099)]. For the near term period under the SSP585 (SSP245) scenario, the CESM projects average rainfall changes of 0.12 mm/day (0.06 mm/day), -0.11 mm/day (0.06 mm/day) and -0.02 mm/day (0.02 mm/day), in ON, DJF and MA, respectively, whereas for the long term period, it projects 0.18 mm/day (0.11 mm/day), -0.11 mm/day (-0.03 mm/day) and -0.02 mm/day (-0.03 mm/day). The CMCC under the SSP585 (SSP245) scenario, for the near term period projects average rainfall changes of -0.09 mm/day (-0.07 mm/day), 0.01 mm/day (-0.00 mm/day) and -0.01 mm/day (-0.01 mm/day), in ON, DJF and MA, respectively, whilst for the long term period, it projects -0.13 mm/day (-0.13 mm/day), 0.02 mm/day (-0.19 mm/day) and -0.00 mm/day (0.00 mm/day). The MPI under the SSP585 (SSP245) scenario, for the near term period, projects average rainfall changes of 0.12 mm/day (0.04 mm/day), 0.14 mm/day (-0.13 mm/day) and 0.06 mm/day (0.04 mm/day), in ON, DJF and MA, respectively, whereas for the long term period, it projects 0.20 mm/day (0.21 mm/day), 0.18 mm/day (0.12 mm/day) and 0.11 mm/day (0.11 mm/day). For the model average during the near term period under the SSP585 (SSP245) scenario, it projects average rainfall changes of 0.05 mm/day (0.01 mm/day), 0.01 mm/day (-0.02 mm/day) and 0.01 mm/day (0.02 mm/day), whilst for the long term period, it projects 0.08 mm/day (0.06 mm/day), 0.03 mm/day (0.02 mm/day) and 0.03 mm/day (0.02 mm/day). Overall, the models project larger variability in rainfall changes in DJF than in ON and MA.

Fig. 9
figure 9

(a) Changes in rainfall over the north catchment based on CMIP6 model [CESM, CMCC, MPI, model average (aver)] projections for the future (ON 2030 to 2099), under SSP245 and SSP585 scenarios. CESM_ssp245 means CESM projection under the SSP245 scenario, and so on. Model aver ssp245 (ssp585) means model average under the SSP245 (SSP585) scenario. (b) As in (a) but for DJF. DJF 1930 means December 1930 to February 1931, and so on. (c) As in (a) but for MA. Location of the north catchment is shown in Fig. 1

Full size image

Bias corrected rainfall projections for Ngamiland are shown in Fig. 10. Under the SSP245 scenario in ON during the near term period, the CESM, CMCC, MPI and the model average project average rainfall changes of 0.15 mm/day, 0.44 mm/day, 0.00 mm/day and 0.20 mm/day, respectively, whereas for the long term period these models project 0.18 mm/day, 0.75 mm/day, 0.04 and 0.33 mm/day, respectively. The changes in DJF and MA rainfall (Figs. 10b-c) over Ngamiland are larger than those projected over the north catchment (Figs. 9b-c), whereas in ON, the changes are larger over the north catchment (Fig. 9a) than over Ngamiland (Fig. 10a). This difference in spatial distribution of rainfall changes over the two regions is consistent with Almazroui et al. (2020) who found large variability in rainfall changes over Africa during the twenty-first century. Over both Ngamiland and the north catchment, projected rainfall changes under the SSP245 scenario cannot be generalised to be larger or smaller than those projected under the SSP585 scenario.

Fig. 10
figure 10

As in Fig. 9 but for Ngamiland, whose location is shown in Fig. 1

Full size image

Some significant decreasing trends have been found in the projected rainfall changes over Ngamiland, but none were significant over the north catchment. During the near term period those significant trends are projected by the CESM, MPI and the model average, with trend magnitudes of -63.90 mm/decade (in DJF under SSP245), -28.20 mm/decade (in MA under SSP245) and -30.60 mm/decade (in MA under SSP585), respectively. During the long term period, it is only in DJF under the SSP585 scenario that the models project significant decreasing trends over Ngamiland, with the CESM, CMCC, MPI and the model average projecting -49.50 mm/decade, -87.20 mm/decade, -74.70 mm/decade and -85.50 mm/decade, respectively. The significant decreasing trends in the projected rainfall changes are consistent with previous studies (Almazroui et al. 2020; Lim Kam Sian et al. 2021). It is worth noting that for the nonsignificant trends over both Ngamiland and the north catchment, some were negative while others were positive.

3.2.2 Temperature projections

Figure 11 shows bias corrected future projections (up to the end of the twenty-first century) of changes in temperature under the SSP245 and SSP585 scenarios over the north catchment. Considering the projections for the long term period first, the figure shows that under the SSP585 (SSP245) scenario the CESM projects average warming values of 6.20 °C (3.54 °C), 5.90 °C (3.25 °C) and 5.37 °C (2.77 °C), in ON, DJF and MA, respectively. For the same long term period, the CMCC under the SSP585 (SSP245) scenario projects 6.04 °C (4.18 °C), 3.92 °C (2.09 °C) and 3.84 °C (2.18 °C), in ON, DJF and MA, respectively. The MPI projects smaller warming values than both the CESM and CMCC, i.e., under the SSP585 (SSP245) scenario it projects 3.84 °C (2.88 °C), 3.28 °C (2.00 °C) and 3.84 °C (2.16 °C), in ON, DJF and MA, respectively. For the same seasons, the model average under the SSP585 (SSP245) scenario projects average warming values of 5.36 °C (3.53 °C), 4.37 °C (2.44 °C) and 4.35 °C (2.37 °C). Paying attention to the near term period in Fig. 11, under the SSP585 (SSP245) scenario the CESM projects average warming values of 2.71 °C (2.10 °C), 2.51 °C (1.82 °C) and 2.03 °C (1.88 °C), in ON, DJF and MA, respectively, which are larger than those projected by the CMCC [2.65 °C (2.04 °C), 1.56 °C (1.00 °C), 1.70 °C (1.13 °C)], MPI [1.91 °C (1.33 °C), 1.52 °C (0.75 °C), 1.61 °C (1.17 °C)] and the model average [2.42 °C (1.82 °C), 1.86 °C (1.19 °C), 1.78 °C (1.39 °C)]. Overall, Fig. 11 shows that all the models project a larger warming during the long term period than during the near term period.

Fig. 11
figure 11

(a) Changes in temperature over the north catchment based on CMIP6 model [CESM, CMCC, MPI, model average (aver)] projections for the future (ON 2030 to 2099) under SSP245 and SSP585 scenarios. CESM_ssp245 means CESM projection under the SSP245 scenario, and so on. Model aver ssp245 (ssp585) means model average under the SSP245 (SSP585) scenario. (b) As in (a) but for DJF. DJF 1930 means December 1930 to February 1931, and so on. (c) As in (a) but for MA. Location of the north catchment is shown in Fig. 1

Full size image

Figure 12 shows bias corrected temperature projections for Ngamiland. For the long term period under the SSP585 (SSP245) scenario, the CESM projects average warming values of 6.71 °C (3.76 °C), 7.02 °C (3.64 °C) and 6.06 °C (2.84 °C), in ON, DJF and MA, respectively, which are overall larger than those projected by the CMCC [6.23 °C (4.36 °C), 4.05 °C (2.67 °C), 3.95 °C (2.36 °C)], MPI [4.42 °C (2.64 °C), 3.83 °C (2.22 °C), 4.41 °C (2.72 °C)] and the model average [5.78 °C (3.59 °C), 4.96 °C (2.84 °C), 4.81 °C (2.64 °C)]. For the near term period the CESM again projects larger warming than the other models, but with smaller warming values during this period than during the long term period as in the case for the north catchment. Overall, the projected average warming over Ngamiland (Fig. 12) is larger than that projected over the north catchment (Fig. 11). Over both regions, unsurprisingly, the warming projected under the SSP585 scenario is larger than that projected under the SSP245 scenario. This is consistent with Almazroui et al. (2020) who, over Africa, projected mean annual temperature increases during the near term (long term) period of 1.5 °C (2.3 °C) and 1.8 °C (4.4 °C) under the SSP245 and SSP585 scenarios, respectively. The SSP245 and SSP585 represent medium and highest emission scenarios, respectively (Van Vuuren et al. 2011; O’Neill et al. 2016).

Fig. 12
figure 12

As in Fig. 11 but for Ngamiland, whose location is shown in Fig. 1

Full size image

Significant warming trends in the projected temperature changes have been found in all three seasons, with the strongest ones over the north catchment occurring in ON particularly during the near term period. The trend for this ON season over the north catchment under the SSP245 (SSP585) scenario projected by the model average is 0.56 °C/decade (0.67 °C/decade), which does not differ much with trend magnitudes projected by the individual models [CESM 0.73 °C/decade (0.48 °C/decade), CMCC 0.79 °C/decade (0.59 °C/decade), MPI 0.30 °C/decade (0.87 °C/decade)]. The projected warming trends are stronger during the long term period than during the near term period particularly under the SSP585 scenario. During the long term period under the SSP585 scenario over the north catchment, the model average projects trends of 1.20 °C/decade, 0.98 °C/decade and 0.85 °C/decade, in ON, DJF and MA, respectively, which are similar to those projected by the individual models [CESM (1.02 °C/decade, 1.09 °C/decade, 0.86 °C/decade), CMCC (0.87 °C/decade, 0.70 °C/decade, 0.54 °C/decade), MPI (1.51 °C/decade, 1.06 °C/decade, 1.00 °C/decade)].

Projected warming trends during the near term and long term periods over Ngamiland are similar to those found over the north catchment, although they are overall larger over the former region than over the latter region. For example, over Ngamiland during the long term period under the SSP585 scenario, the model average projects trends of 1.11 °C/decade, 1.07 °C/decade and 0.90 °C/decade, in ON, DJF and MA, respectively, which are overall larger than those projected over the north catchment (already mentioned). Over both regions, generally the models project stronger trends in temperature changes under the SSP585 scenario than under the SSP245 scenario. The warming trends found here are consistent with those found by other studies over southern Africa (Engelbrecht et al. 2015; Maúre et al. 2018; Almazroui et al. 2020).

4 Conclusions

This study has examined changes in rainfall and temperature over the Okavango River Basin (ORB) region and its subregions (north catchment, Ngamiland) during future periods [2030–2059 (near term period) and 2070–2099 (long term period)] relative to the historical period (1985–2014). CMIP6 model (CESM, CMCC, MPI, and the model average) projections under the SSP245 and SSP585 emission scenarios have been used, focusing on the extended rainy season October–April split into October–November (ON), December-February (DJF) and March–April (MA).

It has been found that historical simulations from the individual models and the model average generally overestimate the observed rainfall and temperature annual cycles during the historical period (1985–2014), as well as the observed interannual variability in the changes of these variables. Future changes in rainfall and temperature have therefore been analysed using bias corrected (method described in Section 2) model projections. All the models show a mixture of reductions and increases in average rainfall during both future periods (near term and long term). For example, for the near term period under the SSP585 (SSP245) scenario over the north catchment, the model average projects an average rainfall change of 0.01 mm/day (-0.02 mm/day) in DJF, in the same season the CESM and MPI project -0.11 mm/day (0.06 mm/day) and 0.14 mm/day (-0.13 mm/day), respectively, and the CMCC projects -0.09 mm/day (-0.07 mm/day) in ON. Future changes in DJF and MA rainfall over Ngamiland are larger than those over the north catchment, whereas in ON, it is the reverse. Over both regions, projected rainfall changes under the SSP245 scenario are not necessarily larger or smaller than those projected under the SSP585 scenario.

The models show some significant decreasing trends in future rainfall over Ngamiland, but none are significant over the north catchment. During the near term period, those significant trends are projected by the CESM (63.90 mm/decade in DJF under SSP245), MPI (-28.20 mm/decade in MA under SSP245) and the model average (-30.60 mm/decade in MA under SSP585). During the long term period, those significant trends occur only in DJF under the SSP585 scenario and are projected by the CESM (-49.50 mm/decade), CMCC (-87.20 mm/decade), MPI (-74.70 mm/decade) and the model average (-85.50 mm/decade).

For temperature, over the north catchment under the SSP585 (SSP245) scenario during the long term period, the model average projects average warming values of 5.36 °C (3.53 °C), 4.37 °C (2.44 °C) and 4.35 °C (2.37 °C), in ON, DJF and MA, respectively, whereas during the near term period, it projects 2.42 °C (1.82 °C), 1.86 °C (1.19 °C), 1.78 °C (1.39 °C). The individual models project average warming values that are similar to those projected by the model average, with those projected by the CESM being generally the largest during both the long term period [6.20 °C (3.54 °C), 5.90 °C (3.25 °C) and 5.37 °C (2.77 °C)] and the near term period [2.71 °C (2.10 °C), 2.51 °C (1.82 °C) and 2.03 °C (1.88 °C)]. Overall, the projected average warming is larger over Ngamiland than over the north catchment. Over both regions, the projected average temperature changes are larger during the long term period than during the near term period. As expected, the average warming projected under the SSP585 scenario is larger than that projected under the SSP245 scenario.

There are significant warming trends in the projected temperature changes in all three seasons. Over the north catchment, particularly during the near term period, the strongest ones occur in ON. These warming trends under the SSP245 (SSP585) scenario projected by the model average, CESM, CMCC and the MPI are 0.56 °C/decade (0.67 °C/decade), 0.73 °C/decade (0.48 °C/decade), 0.79 °C/decade (0.59 °C/decade) and 0.30 °C/decade (0.87 °C/decade), respectively. Also, during the long term period the warming trends projected by the individual models are similar, with those projected by the model average being 1.20 °C/decade, 0.98 °C/decade and 0.85 °C/decade, in ON, DJF and MA, respectively, under the SSP585 scenario over the north catchment. The results of this study may help with the management of agricultural activities, water resources and the highly biodiverse and sensitive ecosystems in the ORB region.