Climate change projections from a multi-model ensemble of CORDEX and CMIPs over Angola

Angola has been characterized as one of the most vulnerable regions to climate change. Climate change compounded by existing poverty, a legacy of conflict and other risk factors, currently impede development and are expected to become worse as climate change impacts increase. In this study we analyze the signal of climate change on temperature and rainfall variables for two time periods, 2020–2040 and 2040–2060. The analysis is based on multi-model ensemble of the Coupled Model Intercomparison Projects (CMIP5 and CMIP6) and the Coordinated Regional Downscaling Experiments (CORDEX). Our findings from the observed dataset indicate that the mean annual temperature over Angola has risen by an average of 1.4 °C since 1951, with a warming rate of approximately 0.2 [0.14–0.25] °C per decade. However, the rainfall pattern appears to be primarily influenced by natural variability. Projections of extreme temperature show an increase with the coldest nights projected to become warmer and the hottest days hotter. Rainfall projections suggest a change in the nature of the rainy season with increases in heavy precipitation events in the future. We investigated how droughts might change in all river basins of Angola, and we found an increased uncertainty about drought in the future. The changes in climate and increased variability demonstrate the need for adaptation measures that focuses on reducing risks in key sectors with a particular focus on adaptation of cities in Angola given a potential increase in mobility towards urban areas.


Introduction
Angola has been characterized as one of the most vulnerable countries to climate change, and more specifically, the south has been suffering periods of drought from 2012 to 2022 (Cain 2017, Limones et al 2020, Serrat-Capdevila et al 2022. Studies show that some changes in climate are already being observed in Africa and that some of these changes would not be possible without human influence (Bezner Kerr et al 2022). Increasing trends in hot extremes and extreme precipitation events have been found in many regions in Africa accompanied with overall decrease in annual precipitation (Gutiérrez et al 2021). Moreover, there has been a noticeable rise in the frequency of droughts across various regions worldwide, and it is projected that this trend will persist in the future, even if global warming is stabilized at 2 • C (Seneviratne et al 2021). These changes are expected to have important impacts on key sectors such as agriculture, health and water which plays an enabling role in the economy and livelihoods of Angola. Situated on the south-west coast of Africa, the Republic of Angola is bounded to the north and northeast by the Democratic Republic of Congo, to the east by Zambia, to the south by Namibia, and to the west by the Atlantic Ocean. In addition, it encompasses the exclave of Cabinda, which shares a border with the Republic of Congo to the north. According to the ND-GAIN index-an index that illustrates the relative resilience of countries comparative to one another-Angola is the 49th most vulnerable country and the 15th least ready country to deal with the impacts of climate change and other stressors (ND GAIN 2021). This score denotes high vulnerability and low readiness. The vulnerability to climate change that Angola faces varies regionally depending on the climatic variables, on the topography, and on socio-economic variables. Some regions are already experiencing frequent episodes of extreme weather events of concern such as heatwaves, droughts or floods, as well as coastal degradation (Meque et al 2022, Serrat-Capdevila et al 2022, Trisos et al 2022. These climate shocks, compounded by existing poverty, a legacy of conflict and other risk factors, currently impede development and are expected to become worse as climate change impacts increase. Due to its high vulnerability to climate change, investment efforts in development and adaptation in Angola require information on the range of possible futures to expect. A report from the World Bank found that sector development plans in Angola require further climate information to adequately mainstream climate change adaptation, and described research on climate trends as 'scarce' (World Bank 2019). Previous research has demonstrated a wide variety of possible futures for rainfall in Angola, with most models agreeing that temperatures are likely to increase. Pinto et al (2016) used two regional climate models and found that total precipitation is projected to decrease, accompanied by increases in dry days over Angola. Carvalho et al (2017) described regional model trends in temperature and total annual rainfall, as well as in 6 month droughts, but did not provide detailed information on changes to the timing of the rainy season or other climate extremes indices. This study performs the first comprehensive analysis and comparison of a set of climate models and scenarios with a high spatial resolution for Angola for the 21st century. This work looks at future temperature and precipitation as projected by the CORDEX, CMIP5 and CMIP6 model datasets. The main focus of the analysis is on several climate variables and indicators, as well as the evolution of the frequency and intensity of droughts throughout this century.

Observational datasets (historical reference)
A set of five gridded observational global mean surface air temperature datasets were used for the analysis of mean surface temperature, namely, NOAA Global Temperature Analysis v5 (NOAA at 5 degree × 5 degree resolution, Zhang et Rohde and Hausfather 2020). These datasets represent a combination of surface air-temperature fields over land and sea-surface temperature fields over the ocean. The period of analysis of this data set is from 1951 to 2020. Rainfall and temperature data from the European Centre for Medium range Weather Forecasts (ECMWF) ECMWF Reanalysis version 5 or simply ERA5 (Hersbach et al 2020) was also used. ERA5 has a spatial resolution of 31 km and is available from 1979 to present. Observational and reanalysis climate datasets are critical tools for studying climate trends and variations. However, observational and reanalysis datasets are not without their limitations and uncertainties. For example, some areas of the world have fewer weather stations or satellite coverage, which can lead to data gaps. Other sources of uncertainty include spatial and temporal resolution, sampling error, bias correction and data assimilation. For more details on uncertainty of observational climate datasets refer to (Zumwald et al 2020).

Climate models (future projections)
Available model simulations from the Coupled Model Intercomparison Project (CMIP5 (29 models) and CMIP6 (27 models), (Taylor et al 2012, Eyring et al 2016) and the Coordinated Regional Downscaling Experiments (CORDEX-Africa, Giorgi et al 2009, Gutowski et al 2016 were also used. CORDEX Africa project downscaled a set of CMIP5 global models across the full African continent at a grid resolution of around 0.44 • horizontally (50 km). The higher spatial resolution of the participating dynamical downscaling models enables the CORDEX ensemble to more realistically represent regions of complex topography and localized rainfall (Pinto et al 2016, Dosio et al 2019. We use a set of available models over Angola in the understanding that these models should provide an improved representation of the impact of the regional topography on the climate and change in the climate. The CMIP5 experiment utilized representative Here we include RCP 4.5 which represents a middle of the road scenario with strong mitigation action culminating in an increase in 4.5 W m −2 in equivalent radiative forcing by 2100, as well as RCP 8.5 which represents a very negative scenario with very little mitigation of emissions into the future. The 6th iteration of the CMIP experiment was developed in preparation for the IPCC 6th assessment report. While deploying a different approach to future emissions scenarios called shared socio-economic pathways (O'Neill et al 2014), the experiment designers did choose to align the Shared Socioeconomic Pathways (SSPs)s with the CMIP5 RCPs in terms of radiative forcing. This means that we can analyze largely equivalent scenarios across CMIP5 and CMIP6. In this case we analyze SSP5 8.5 and SSP2 4.5 as roughly equivalent to RCP 8.5 and RCP 4.5. Table S1 shows the full list of the combination of global climate models (GCMs) downscaled and the regional climate models (RCMs) used.

Climate indices
Future projections were analyzed for two time periods, 2020-2040 and 2040-2060, and compared to the current WMO climate normal period of 1981-2010. The two future horizons are the same used in the IPCC AR6 and are considered long enough to show future changes in rainfall and temperature when averaging over ensemble members of multiple models, and short enough to enable the time dependence of changes to be shown throughout the 21st century. These periods provide climate change information for the near (2030s) and mid future (2050s), thus supporting decision making processes at different planning horizons. Table 1 lists a collection of eight climate indices associated with temperature and precipitation, selected from the Expert Team on Sector-Specific Climate Indices (ET-SCI). These indices are specifically designed to characterize moderate extreme events with a recurrence interval of one year or less, as described in Klein Tank et al (2009) andZhang et al (2011). One of the indices, the maximum 5 d precipitation index (RX5day), represents the highest accumulation of 5 d precipitation and is often utilized to evaluate potential flood hazards. Heavy rain conditions over multiple consecutive days can contribute to flood conditions (Sillmann et al 2013). Simple daily intensity index (SDII) describes the daily precipitation amount averaged over all wet days in a year, and is an indication of how heavily rain falls. The number of wet days (RR1) represents days with precipitation above 1 mm. The consecutive dry-day index (CDD) is a measure of the longest continuous stretch of dry days within a year, where a dry day is defined as having less than 1 mm of rainfall. CDD is commonly referred to as a drought indicator since it is utilized to depict the tail or low end of the precipitation distribution (e.g. Sillmann et al 2013). Since drought is a multifaceted phenomenon that relies on numerous factors aside from insufficient precipitation, CDD can only offer insight into meteorological drought. Therefore, it should be evaluated in conjunction with other precipitation indices such as Standardized Precipitation Index (SPI) and Standardized Precipitation Evaporation Index (SPEI) for a more comprehensive understanding.

Analysis of droughts
For the analysis of droughts, we deploy two very widely used drought indices. The first is the SPI (McKee et al 1993) which is a drought index based just on rainfall deficit. The usefulness of SPI is that rainfall deficits (or excesses) are standardized relative to the baseline distribution, thus allowing us to compare across regions with varying baseline rainfall distributions. However, the SPI only uses precipitation data to characterize droughts. The index assumes that precipitation is the most significant factor for the user and that the variability of precipitation surpasses the variability of other variables such as temperature, evapotranspiration, wind speed, and relative humidity that contribute to droughts. It also assumes that there are no trends in these other variables. However, as a result, SPI may not consider the impact of rising temperatures (i.e. global warming) on drought characteristics (Abiodun et al 2021). To overcome this limitation we used the SPEI (Vicente-Serrano et al 2010). This index is based on climate water balance (CWB), which is the difference between precipitation (P) and potential evapotranspiration (PET) (i.e. CWB = P-PET), and so comes closer to identifying hydrological drought. The PET data (for calculating CWB) were obtained from the maximum and minimum temperature data using the Hargreaves method (Hargreaves and Samani 1982). Here we apply a 12 month accumulation period ending at the end of October in order to capture the annual time scale rainfall deficits. This study concentrates on droughts that are classified as being at least moderate droughts (i.e. ⩽−1.0), and thus, a threshold of −1.0 was applied to identify droughts in the 12 month scale SPEI and SPI datasets. Both the SPEI and SPI indices have been employed for identifying, monitoring, and project droughts in Africa (e.g. Meque and Abiodun 2014), as well as in other regions of the world (e.g. Spinoni et al 2020).

Water availability (WA)
In order to evaluate water resources implications of changing climate, we use a simple WA index, which is defined as follows: This was done with WA values calculated on a monthly basis, and then summed up to annual values. WA is therefore a simple index of relevance to water resources that can be obtained without use of sophisticated hydrological models, only from climate data. The index considers the 'surplus' of rainfall (P) over PET and expresses the water that is effectively available in the environment to generate groundwater recharge and surface runoff.
When analyzing the results of regional climate models ensembles, it is important to consider the consensus amongst models and the statistical significance of future changes, i.e. how the change compares with natural variability. There are many different methods to define the robustness of the climate change signal. Here we use the same method used in the IPCC AR5 which is based on the models' agreement in terms of both significance and sign of the projected change. The projected change is considered robust if the ensemble mean change is significant and at least 70% of the models agree on the direction of change. The change is considered uncertain, if the change is not statistically significant and less than 70% of models agree on the sign of change. Although in many studies higher thresholds are used (e.g. 80% for the agreement in sign in the IPCC methodology), here a lower threshold is used to make it applicable to small ensembles (such as CORDEX) for which, otherwise, robustness may be too strongly dependent on the results of a single model.

River basin scale analysis
A more detailed regional analysis is done over eight regions in Angola. Here, we initially focused on homogeneous climate regions, attempting to define regions that are characterized by similar climate variability, reflecting similar drivers of climate, and consequently, likely similar patterns of changing climate. Due to limitations, observational data could not be used for that purpose, nor could then be used for identification of the best 'surrogate' gridded dataset. We thus worked with several different rainfall gridded datasets attempting identification of climate regions that would be robust, i.e. similar across most of them. That turned out to be impossible-regions differed significantly between datasets and could not be defensibly aggregated. This is likely because differences between rainfall gridded datasets can often be very large, especially in regions where station networks are sparse (Dosio et al 2021) such as Angola. We therefore divided the country into regions corresponding to the main river basins of Angola, with some modification in the north-west of the country. Such defined regions are relevant from the water resources point of view, but also from an administrative point of view, and by nature capture main differences in climate across the country. The following are the eight basins, along with their respective areas: Congo (271 752 km 2 ), Cunene (94 113 km 2 ), Zambezi (256 014 km 2 ), Cuvelai (54 181 km 2 ), Okavango (150 199 km 2 ), South Coast (173 355 km 2 ), Cuanza (127 042 km 2 ), and North Coast (134 043 km 2 ). In the time-oriented analyses presented in this study, spatial climate data of various nature were averaged over these eight regions shown in figure 1(b).

Results and discussion
Angola is dominated by three climate zones based on the Köppen classification: hot and humid tropical zone in the north; temperate tropical climate over the central plateau; and a desert climate in the south-west. Precipitation levels exhibit a marked southwest-northeast gradient and are characterized by strong seasonality. The north-east is the wettest region in Angola, and precipitation decreases towards the south and west (figure 2(a)). Precipitation variability is influenced by the seasonal interplay between subtropical high-pressure systems and the migration of easterly flows associated with the Intertropical Convergence Zone (ITCZ) (Munday and Washington 2017). A secondary convergence zone, the Congo Air Boundary  Global mean surface air temperature between 2011 and 2020 was 1.09 • C higher than it was during the pre-industrial era. Furthermore, over the past half-century, the rate at which the global mean surface air temperature has risen is unprecedented in at least the last 2000 years (Gulev et al 2021). Over Angola, warming is evident from multiple datasets over the period 1951-2020 ( figure 3). There is a strong consensus that the recent observed warming is robust throughout the country. On average, the mean annual temperature has increased by 1.4 [1.01-1.78] • C at a rate of 0.2 [0.14-0.25] • C per decade since 1951. According to the climate report from the IPCC (2021), the frequency and intensity of hot extremes have increased and those of cold extremes have decreased globally since 1950, including over Angola.
There is high model agreement that temperatures are projected to increase in Angola in the future (figure 4). Beyond increases in average temperatures, the IPCC (2021) AR6 summary for policy makers states that 'It is virtually certain that there will be more frequent hot and fewer cold temperature extremes over most land areas on daily and seasonal timescales as global mean temperatures increase. It is very likely that heat waves will occur with a higher frequency and duration' . Figure 4 shows the projected mean changes in temperature indices over Angola for the near future (2020-2040) and the mid future (2040-2060). These changes are compared to the recent past (1981-2010) under two representative concentration scenarios, RCP45 and RCP85 respectively. Climate projections for the near future and mid future under RCP2.6 are very similar to RCP45 (not shown). For this reason, the focus of this study is on projections based on RCP45 and RCP85 scenario.
Mean temperature is projected to increase everywhere in the country independent of scenario ( figure 4(a)). Under RCP4.5 mean temperatures are projected to increase between 1 • C and 2 • C from near to mid-term future. Under RCP8.5 mean temperatures are projected to increase between 1 and 3 • C from the near-to mid-term future. As mean surface temperatures increase, extremes are also projected to increase in many parts of Angola. The number of days in which temperature exceeds 30 • C are projected to increase, on average, by between 60 and 90 d in the south of the country over the near future and mid future ( figure 4(b)). The implication is that regions that are already hot, will become even hotter and have more days exceeding critical thresholds. The coldest (TNn) and hottest (TXx) day of a year, simulated in the CORDEX ensemble, are shown in (figures 4(c) and (d)). The spatial patterns of change in TNn and TXx are similar in magnitude. The minimum amount of projected change is around 1 • C for the near future under RCP45 scenario. The minimum amount of projected change is around 1.5 • C for the near future under RCP45. Under RCP85, the TXx and TNn are projected to increase up to 3 • C for the mid future. This difference suggests that smaller temperature increases are more likely under a low emissions scenario and larger temperature increases are more likely under a high emissions scenario.
Analysis of the temporal evolution of mean temperature averaged over hydrological basins for Angola (see figure 2(b)) are shown in figure 5. These graphs represent data for three different model experiments and generations, the CORDEX and Coupled Model Intercomparison Projects (CMIP5 and CMIP6). All of the three ensembles show a consistent upward trend in mean temperature for all the river basins. The increases in temperature are higher under the higher emission scenario RCP85/SSP585. The current  projected changes in climate in combination with existing vulnerabilities have the potential to significantly impact key sectors in different ways across Angola. Increases in temperature are likely to affect agriculture systems that are not adapted throughout the country. Extreme high temperatures can have a negative impact on livestock and will require changes to livestock management (Bezner Kerr et al 2022). Ongoing high temperatures can require an increased need for irrigation in areas where evaporation reduces WA (Caretta et al 2022). In the absence of risks reduction measures, extreme heat events can cause health impacts, especially among those who already have underlying health conditions such as cardiovascular and respiratory diseases. The elderly, pregnant women and young children are also particularly vulnerable to these extreme events. Climate effects on crops, income, or disease prevalence can also affect nutritional outcomes, especially in children (O'Neill et al 2022). Vulnerable people living in urban areas have greater exposure to extreme heat, because urban areas are hotter than rural areas (the urban heat island effect). Without adaptation, there can be increased costs for service provision to urban areas, including meeting demand for cooling, sanitation, and supply of safe potable water. Transport can be affected by extreme heat (softened asphalt and expanded rails) as well as extreme flooding events.
Rainfall exhibits greater spatial and temporal variability than temperature. The climate system's decadal and longer-term variability arises from both internal and external natural processes. Major modes of variability such as ENSO, IOD, and SAM can significantly influence global and regional atmospheric circulation patterns on decadal, multi-decadal, and centennial timescales. These large-scale natural variability modes imply that a decade may be hotter or colder, and drier or wetter, than the previous decade without any alterations to external influences on the climate system. However, external forcings on the climate system, such as solar cycles, volcanic eruptions, biosphere processes, and more recently, human emissions of GHGs and aerosols (i.e. particulates in the atmosphere), can also induce variability and change on long timescales. Figure 6 shows the areas that were wetter (green) and drier (brown) during each of the last four decades. The colors represent the difference between the total mean annual rainfall for each decade and the total mean annual rainfall over the 1981-2020 (see figure 2(a)) period at each grid cell. This demonstrates significant rainfall variability on decadal time scales. For many locations, there is a distinct contrast in observed rainfall between certain decades and their preceding decades. For instance, the 1980s were much drier than the average in most locations across Angola, whereas the 2000s were much wetter.
In Angola, natural variability on interannual, decadal and multi-decadal time scales, as experienced in the past, is expected to continue to be the dominant influence on future rainfall. Figure 7 shows the projected changes in annual total precipitation on days with at least 1 mm of precipitation (PRCPTOT). The colors depict the ensemble mean anomalies from the CORDEX models for the mid and far future periods under RCP45 and RCP85. The hatching on the maps, shown over the land surface area, indicate regions where the changes are uncertain, i.e. the model agreement in the direction of change is low (see methods section for  The reductions in rainfall are due to a northward shift and delayed breakdown of the moisture convergence zone, the CAB (Howard and Washington 2020).
While there is widespread agreement regarding the direction of change in temperature indices regardless of the region under consideration, projected changes in mean annual total precipitation are less uniform. A further analysis was performed for each basin to obtain a more detailed regional picture of the extent of model agreement on projected changes in the rainfall. Figure 9 shows the average of annual total mean precipitation change for each basin. Rainfall projections for the near future do not indicate any significant or systematic changes in total annual rainfall, although there is some propensity towards drier future in the southern regions of Cuvelai, Okavambo and Zambezi. Figure 10 shows the spatial pattern of future anomalies from the multimodel mean over near-and mid-future for selected precipitation indices. There is an overall increase in five day maximum rainfall (RX5day) throughout the future time horizons and for both scenarios. Increases in annual one day maximum rainfall (RX1day) have similar patterns and directions of change as in RX5day. RX5day is projected to increase between 5%-10% in the near term and 10%-15% in some regions in Angola for the period 2040-2060 under RCP8.5. Rainfall intensity or SDII is projected to increase everywhere in Angola for both periods and scenarios. Increases in magnitude of wet extremes are associated with increased convective rainfall intensity (e.g. thunderstorms) which is expected to increase with global temperature increases. On the other hand, dry extremes are projected to increase. The increases of CDD are higher and statistically significant in the southern regions of Angola for the period 2040-2060 under RCP85. The number of wet days or rain days is also projected to decrease everywhere in the country. Increases in CDD are combined with increases in and RX5day, RX1day, SDII and decreases of RR1 indicating an intensification of both wet and dry seasons in these regions. Relatively high/low increases are more likely under a higher/lower GHG emissions scenario. This overall trend towards more extreme precipitation, decreases in total precipitation and increases in dry conditions is consistent with the scientific understanding of the impacts of global warming on the Earth's climate system (e.g. IPCC 2021).
Our analysis shows that the rainy season is likely to become more concentrated in the middle months, with less rain at the beginning and the end. This has implications for agriculture management practices, which will need to adjust to avoid reductions in crop yields. This should also include management options focused on an increased frequency of extreme rainfall events. Post-harvest processing and storage will also need to consider the possibility of more frequent extreme events. Rule curves for hydroelectric power generation will need to adjust to changes in the timing of rainfall throughout the season. These should also adjust to account for an increasing variability in rainfall, with greater amounts coming in extreme events, to avoid the risk of downstream flooding. Management should also account for greater amounts of evaporation with increasing temperatures. Extreme rainfall events can foster breeding of mosquitoes, which can increase the risk of vector-borne diseases such as malaria. Storage of water during dry spells and droughts can also increase breeding habitat, with similar results if preventative measures are not taken. Flood events can increase the prevalence of water-borne diseases, such as cholera. Flood events can also disrupt food supply chains. Without adaptation, any of these outcomes can reduce food access or increase diseases, which can ultimately increase malnutrition. Mental health can also be impacted by crisis events. Climate-related shocks can be a push factor for people to move, temporarily or permanently, to urban areas. Urban areas are vulnerable to extreme rainfall events, storm surges, and extreme temperature events, which can affect urban infrastructure and water quality. Figure 11 shows the annual mean of WA (see methods section) over each basin. Increased temperature and thus increased PET, combined with projections of weak change in rainfall, is likely to lead to lower WA in the future. Under RCP45 the projected changes in WA are uncertain in many basins. Under RCP85 this  uncertainty is reduced and the projected reductions are seen over Zambezi, Okavango and Cuvelai. Only in the North Coast region is there a likelihood of increased WA, but a drier future is also possible there.  Figure 12 shows the projected changes in the frequency of droughts in sets of ten years (decades) where SPI or SPEI is lower than −1.0 (see methods). An SPI or SPEI of less than −1.0 is only a moderate drought condition, and would normally be expected around 2-3 times per decade. These plots are divided into 10 year segments extending from the past into the future. Each segment has three sub-plots that are violin plots representing the ensemble distribution of values from each ensemble. Overall, rainfall-driven droughts (SPI) do not show a strong change in the future-as indicated by the tendency of the individual ensembles. Projections by individual models, however, encompass possible increases in frequency of drought years and these should be considered as one of the possible futures. SPEI also does not show a strong change in the future-as indicated by the tendency of the individual ensembles. CORDEX projections, on the other hand, indicate a strong increase in drought, particularly in the south of Angola and in the Cuanza region. In every region, projections by individual models encompass increases in drought events. These droughts, again, should be considered as one of the possible futures.
Our results demonstrate that there is a possibility of an increase in short-term and multi-year droughts which would require improvements in water management for agriculture. Occasional droughts coupled with high temperatures can increase the risk of fires, which may affect those who rely on the forests for fuel. Extreme hazards such as heat, floods, and fires, can also impact electricity power distribution infrastructure.

Conclusions
The climate across Angola includes arid, temperate and tropical conditions. Most of the region receives the majority of its rainfall in the warm summer months (Nov-Mar). Temperature and rainfall vary on annual, decadal and multi-decadal timescales. The year-to-year differences are influenced by large-scale seasonal atmospheric patterns as well as the changing conditions over the cool South Atlantic Ocean. Rainfall has displayed significant decadal fluctuations over the last four decades. Specifically, portions of southern Angola experienced exceptional dryness during the 1980s and 1990s, whereas they were abnormally wet during the 2000s and 2010s. Conversely, other regions of Angola witnessed the inverse pattern. Climate change has already affected the entire country, and the most conspicuous impact of climate change has been an increase in temperatures. Mean annual temperatures have increased by approximately 1.4 [1.01-1.78] • C since 1951. The current rate of warming is about 0.2 [0.14-0.25] • C per decade.
Compared to temperature trends, rainfall trends over the previous four decades are less apparent, and there are significant fluctuations in both the magnitude and direction of rainfall changes across all regions. The changes also depend on the time period over which they are considered, which reflects the role of the decadal variability mentioned earlier.
Future projections of temperature change show significant increases across the region. Under moderate GHG emission scenarios, projected increases in average annual temperatures range from 1 to 2 • C by mid-term future (i.e. by 2050s). That range reflects uncertainty of climate projections. Under a high emission scenario, changes are projected to reach as much as 3.5 • C. Extreme temperature events will also continue to increase; the coldest nights will be warmer and the hottest days will be hotter. The annual number of very hot days (days with daily maximum temperature above 30 • C) is projected to rise dramatically in the future and with high certainty. The actual values depend on the pace of changes in GHGs emissions, there will be greater impacts on all temperature-related indices if GHG emissions are not curbed quickly enough.
Rainfall projections suggest a change in the nature of the rainy season, with less rain falling at the beginning (during September-October-November) and the end (March-April-May) of the rainy season. In contrast, models project an increase in rainfall in the middle of the season, during December-January-February. These changes represent a shortening but intensification of the wet season, but with little overall effect on total annual rainfall. These seasonal differences in rainfall are likely to be stronger in the western part of the country.
The severity of heavy precipitation events is projected to increase in the future during the wet season. The rainfall events are, however, likely to be fewer, with longer dry periods with no rain in between. The projected changes are consistent with the physics of atmospheric processes: a warmer atmosphere can hold more water, and therefore can 'dump' more precipitation in single, intense events.
Even if changes in rainfall are small, increased temperatures will likely increase evaporative demand of the atmosphere (PET). In the future, rainfall surplus (i.e. an excess of rainfall over the evaporative demand) is likely to reduce. This reduction would mean that groundwater recharge and surface runoff, and thus usable water resources, will also likely be reduced in the future. Only in the North Coast region, where increases in rainfall during the wet season are expected to be strongest, is there a likelihood of increased WA, but a drier future is also possible in that region. Overall, rainfall-driven droughts do not show a strong and consistent change in the future, but scenarios encompass both an increase and a decrease of droughts. Increase in short term and multi-year droughts should thus be considered as a plausible future.
The change in hazards exposure studied demonstrate the need for adaptation measures that focuses on reducing risks in key sectors, such as agriculture and health and with a particular focus on adaptation of cities in Angola given a potential increase in mobility towards urban areas.

Future research and implications for planning and management
This paper presents a comprehensive study on future climate projections over Angola and sets the foundation for further research on the uncertainty of climate impacts on specific sectors and systems, as well as on the Government's future development plans and investments. Future research to evaluate the vulnerability of specific sectoral systems to future climates may include the following.
Basin simulations with hydrological models beyond the current use of P-PET in this paper-continuing to use a stochastic approach to capture uncertainty from climate projections-to better quantify hydrological changes with a focus on hydrological extreme events: floods and hydrological droughts. Then, using these hydrological datasets for future climates, do stress-test simulations on existing and potential sectoral systems to determine their specific vulnerability to climate change. For example, simulating hydropower production in the Kwanza or Cunene River Basins may provide insights on future constraints and potential needs to adjust reservoir operations and energy production. The same could be assessed for irrigated agriculture and other water uses such as water resources systems that supply urban centers. Integrating uncertain climate change projections into water resources planning and design is not a trivial task but clear methodologies have been designed and tested to effectively do so in a practical way, such as the Decision Tree Framework (Ray and Brown 2015).

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary information files).