Recently, considerable diplomatic and media attention has been dedicated to the ongoing construction of the Grand Ethiopian Renaissance Dam (GERD) on the Blue Nile1,2, which will be Africa’s largest hydropower plant with more than 5 GW of installed capacity. Since Ethiopia started its construction in 2011, GERD has been highly controversial because of the trade-offs between power generation, water availability and river ecology it may create across Ethiopia, Sudan and Egypt1,2,3,4,5.

Located near the Ethio-Sudanese border (Fig. 1a), GERD will substantially increase Ethiopian electricity generation, which will support economic development, electrification and industrialization2,6. The river flow control that GERD will allow stirs mixed feelings for Sudan: it could mitigate anomalously dry periods, improve flood control, reduce sedimentation and benefit hydropower and irrigation2,3,4,7,8,9, but also cause water shortages and negative environmental impacts8,10,11. Further downstream, Egypt, which already controls the Nile flow with Lake Nasser and the High Aswan Dam (HAD; Fig. 1a) to serve its own water needs12,13, is strongly opposed to GERD. Egypt primarily fears that GERD, in damming the Blue Nile which contributes up to 60% of overall Nile flow2, may negatively affect Lake Nasser refilling by holding water back during dry years2,5,14. Additionally, GERD’s control of river flow may lead to increased water abstraction in Sudan by facilitating dry-season irrigation2,5,12, and potentially reduce the flow reaching Egypt.

Fig. 1: Hydro, solar and wind power in the Blue Nile countries.
figure 1

a, The study area consists of Ethiopia, Sudan and Egypt, and includes all the current and future locations for hydropower, solar power and wind power generation considered here. Major lakes are indicated with ‘L.’ and rivers with hydropower projects are indicated in italics. Hydro, solar and wind power parks that are not yet operational (either in construction or in planning) are marked with an asterisk17,84. b, The median and IQ range of the Blue Nile’s monthly flow profile at GERD as modelled for this study, based on 26 years of simulation data (Methods). c, Monthly average capacity factors for solar PV panels and wind turbines at the considered locations in Ethiopia (left column) and Sudan (right column), based on Global Solar Atlas37 and Global Wind Atlas38 data (Methods). The letters and numbers refer to the existing or future power plants listed in a.

Source data

Central to the current geopolitical struggle that surrounds GERD are, therefore, the seemingly conflicting interests between countries, as Ethiopia seeks to increase power generation, Egypt is afraid of losing its hegemony over Nile flow control and the consequences for its agricultural areas, and Sudan is torn between the sides3,4,14. This struggle has numerous implications. In the short-term, disagreement persists over the first filling of GERD’s reservoir (74 billion m3 in volume, or 1.6 years of average Blue Nile discharge), whose timeframe will directly impact the flow that reaches Egypt4. In the long-term, GERD’s intended operational strategies and their consequences for downstream water availability are under discussion5,14.

Most political attention is currently being devoted to the filling time1,2, with Egypt heavily contesting Ethiopia’s proposed 4–7 years1. Consequently, scientific studies on GERD are also mostly centred on the filling time4,5,12,14,15,16,17,18. However, as reservoir impoundment will take years rather than decades2,14, any compromise on the filling time will allay concerns around GERD only for a limited period. As the GERD infrastructure and the downstream impacts of the post-filling operation may last beyond a century, it is of major importance that long-term strategies with benefits for all the involved parties be identified for GERD operation. This may help to address certain objections that the downstream countries currently have to GERD, and support moving the ongoing negotiations forward in the absence of basin-wide comprehensive water-sharing agreements.

This article presents a proposal to arrive at such strategies. Our results highlight the clear added value of linking the long-term operational scheme of GERD with the development of a solar and wind power infrastructure in Ethiopia and other countries in the Eastern Africa Power Pool (EAPP). Based on dedicated modelling of hydro, solar and wind power generation at a high spatiotemporal resolution, we identified multiple energy- and water-related benefits that this may bring across the region: better grid integration of variable renewable electricity (VRE), higher compliance with environmental flow requirements (EFRs), improved valorization of the GERD infrastructure, harmonization of the yearly GERD and HAD operation, and strategic diversification of Ethiopia’s power mix.

Long-term GERD operational strategies

The GERD reservoir storage capacity would allow near full control of the Blue Nile’s seasonal flow (Fig. 1b), which suggests that GERD was designed for year-round power generation that follows electricity demand2. Given that the Ethiopian electricity demand is relatively unseasonal19, Ethiopia’s preferred objective for the long-term GERD operation is thus presumably a near-constant outflow from month to month2,18,20,21,22. The corresponding benefits for Ethiopian power generation may conflict with downstream water needs for three reasons.

First, as the HAD has its own seasonally specific outflow to safeguard Egypt’s water needs and mitigate floods, Egypt is likely to demand a harmonized GERD–HAD operation2,5,12,14,15, which could be precluded if Ethiopia insists on a near-constant GERD outflow. Most probably, Egypt would prefer a discharge regime that resembles pre-GERD conditions; for example, the outflow of Uganda’s Nalubaale dam on the White Nile, at Lake Victoria’s outlet, is still governed by a colonial-era ‘agreed curve’ that resembles natural flow23. However, such an option would clearly be unfair to Ethiopia and out of line with its electricity needs. Second, although the potential value of GERD in guaranteeing a minimum dry-season flow for Sudan is undisputed, a near-constant GERD outflow would increase the dry-season flow by such an extent that water extraction for dry-season irrigation in Sudan may reach levels beyond those Egypt considers acceptable2,22,24. Third, a near-constant GERD outflow may endanger the ecological river integrity up to Lake Nasser18,24,25,26,27, as to safeguard river ecology typically requires dam outflow to mimic the natural discharge seasonality18,25.

From the Egyptian and Sudanese point of view, a long-term agreement on the GERD operation could probably benefit from the inclusion of a controlled seasonality of GERD outflow. If well-designed, this would allow for a coordinated Ethio-Egyptian GERD–HAD operation, allay Egyptian fears of lower water availability, reduce GERD’s environmental impact on Sudan’s river ecology and still protect Sudan from droughts and flooding, guarantee a minimum dry-season flow and benefit Sudanese hydropower generation compared with that of the present-day2,22. The principal challenge with controlled seasonality is that Ethiopia has no obvious reason to accept it over near-constant outflow, as it would still be out of line with power demand2,18,20. By itself, a controlled seasonality appears unreasonable for Ethiopia. This gap between Ethiopian energy objectives and Sudanese–Egyptian water objectives must be urgently addressed.

Here we propose a solution that would provide strong incentives to Ethiopia to operate GERD according to controlled seasonality to simultaneously benefit regional energy and water objectives. This solution lies in the complementary operation28 of GERD with a strongly expanded power-generating fleet of VRE units, namely solar photovoltaic (PV) and wind power, in Ethiopia and other EAPP countries. Using hydropower to compensate the inherent fluctuations of VRE on hourly and seasonal timescales28,29 would allow for a firm, reliable hydro–solar–wind power portfolio highly suited for load following and supporting a year-round reliable power supply for Ethiopia. Such a portfolio would naturally favour an increased hydropower dispatch during the wet season, when regional VRE potential drops considerably28 (Fig. 1c). The contribution of our study thus lies in proposing, first, a massive expansion of regional VRE generation, at a scale substantially beyond that foreseen by current policy, and, second, an explicit coupling of GERD operational strategies with regional cooperation on VRE generation to stimulate new discussions that could alleviate certain elements of the ongoing conflict around GERD.

Complementary GERD–VRE operation

We designed four scenarios (S1–S4) for the GERD operation and analysed their potential to harmonize regional energy and water objectives, using a high-fidelity dispatch model for hydropower with VRE28 (see Methods and Supplementary Note 1). Given GERD’s large storage capacity and dispatch flexibility, current literature usually assumes that its operation will strongly flatten out the Blue Nile’s flow seasonality by following hourly electricity demand2,18,20,21,22 (Fig. 2a). This situation, with near-constant outflow on monthly timescales, represents the reference scenario S1 (‘no VRE’).

Fig. 2: Load following with GERD and VRE.
figure 2

ad, Hourly (in Ethiopian time, UTC + 3) and seasonal load following under scenario S1, with only GERD and no VRE (a), scenario S2, with GERD supporting VRE deployment in Ethiopia (b), scenario S3, with GERD supporting VRE deployment in Ethiopia and Sudan (c) and scenario S4, with more ambitious VRE deployment in both countries that requires the deployment of GERD’s full range of flexibility (d). Simulations spanned 26 years of GERD operation (Methods); shown are sample 48 h time slices from the 12th simulation year (top row), as well as average monthly profiles (bottom row). e, Total annual followed load hour-by-hour under each scenario by GERD hydropower in combination with VRE. Numbers in brackets indicate the VRE capacity (GW) deployed in each scenario (solar + wind total in both countries; Methods).

Source data

Given the region’s seasonal hydro–solar–wind complementarity (Fig. 1b,c), we considered three other scenarios in which GERD is explicitly operated to support VRE uptake17,28. In scenario S2 (‘Ethiopian VRE’), substantial solar PV and wind power capacity is deployed in Ethiopia (Fig. 1a; see Methods and Supplementary Note 2) and GERD is operated to compensate VRE fluctuations relative to electricity demand. In scenario S3 (‘Power Pool’), GERD additionally compensates shortfalls in VRE generated in Sudan (Fig. 1a; see Methods and Supplementary Note 2), which leverages Sudan’s higher solar power yield and higher hydro–wind complementarity relative to that of Ethiopia (Fig. 1c). Last, in scenario S4 (‘Power Pool+’), more ambitious VRE deployment takes place in both countries, with a very high instantaneous VRE penetration becoming the norm, and requires GERD’s full flexibility potential.

Each scenario increases the potential for GERD to supply reliable power together with VRE (Fig. 2b–d). At hourly scales, increasing the VRE infeed requires GERD to ramp up and down more sharply (Fig. 2b–d). Complementary operation also requires a clear increase in reservoir outflow and hydropower generation during the wet season (Fig. 2a–d; see Supplementary Note 3). In S2, hydropower peaks in May and September (Fig. 2b); this bimodality disappears in S3 and S4, in which hydropower peaks in August, simultaneously with Blue Nile flow, thanks to the better hydro–VRE complementarity enabled by Ethio-Sudanese electricity exchange (Fig. 2c,d). In S4, the strongly increased VRE generation compared with that of S3 further reduces the need to dispatch GERD in the dry months. Here, GERD frequently ramps down to its minimum power output during episodes of substantial surplus VRE generation. This saves water for an even higher peak hydropower in August compared with that in S3 (Fig. 2d). The total load followed by flexible GERD operation, nominally 12.3 TWh yr–1 in S1 (comparable to Ethiopia’s current power generation30; see Fig. 6), thus increases substantially in combination with VRE, to reach 27.4 TWh yr–1 in S2, 29.7 TWh yr–1 in S3 and 41.2 TWh yr–1 in S4 (Fig. 2e). We note that these results are not sensitive to projected future changes in electricity demand profiles19 (Methods and Supplementary Note 4).

Additionally, GERD outflow reflects the natural seasonality better under each successive scenario (Fig. 3a–d). We assessed the extent to which natural seasonality is mimicked by comparing outflow with minimum EFRs24. Applying stringent EFR criteria (Methods), we calculated the outflow deficit compared with the EFRs, both in volume and in relative monthly flow (Fig. 3e; also see Supplementary Note 3 for a visualization of how EFRs compare to natural flow). Clearly, although the near-constant GERD outflow in S1 would be highly EFR deficient, the VRE scenarios S2–S4 improve the situation substantially, with scenario S4 nearly always in line with the EFRs. Notably, annual downstream releases are only marginally affected by the introduction of controlled seasonalities, and GERD mitigates anomalously dry and wet years in nearly equal measure between S1 and S4 (Supplementary Note 3), to Sudan’s benefit.

Fig. 3: GERD outflow, environmental flow deficit and hydroturbine use.
figure 3

ad, Median and IQ range of monthly average GERD outflow under scenarios S1 (a), S2 (b), S3 (c) and S4 (d), from simulations that spanned 26 yr. e, Deficits of GERD outflow compared with the minimum EFRs (Methods) by scenario. Shown are absolute EFR deficits per year (percentages indicate EFR deficit relative to annual average Blue Nile flow) (top) and the relative spread of EFR deficits during the months in which the EFRs are violated (percentages indicate the deficit relative to monthly EFR values (bottom), and medians and the IQ ranges are taken across all months during which EFRs are not met). f, Total spread of GERD hydroturbine use under flexible operation by scenario, calculated by first creating a histogram of the instantaneous power output required from GERD at each time step, and then binning this data into categories in which each increment represents one additional active turbine. Vertical lines indicate the maximum number of simultaneously active hydroturbines in each scenario.

Source data

The monthly average outflow hides the subdaily hydropeaking performed by GERD (Fig. 2a–d), which may require simultaneous deployment of many hydroturbines (Fig. 3f). In S1, at most 6 out of 16 turbines will be in use simultaneously, which echoes past allegations that GERD’s turbine capacity is substantially ‘oversized’4. Under VRE integration, the turbine usage spread increases considerably; in S4, all 16 turbines are necessary to ensure a proper operation, which provides justification for GERD’s design. We note that environmental impacts of this hydropeaking31 will remain limited, given that Sudan’s close-by Roseires reservoir (Fig. 1a) can easily smooth out such fluctuations and also preserve GERD outflow seasonality due to its tenfold smaller storage capacity7.

Benefits for coordinated GERD–HAD operation

Having established that the complementary GERD–VRE operation and Ethio-Sudanese electricity exchanges benefit VRE integration, EFR compliance and GERD infrastructure use, we next investigated the implications of GERD operation for the yearly refilling of HAD’s reservoir, Lake Nasser. HAD is a multipurpose, multiyear storage dam that protects Egypt from peak floods, provides year-round irrigation and navigation benefits and generates hydropower. Normally, HAD reservoir refilling happens during the August–November flood, after which lake levels stabilize and then decrease for several months; by the next August, they must again be low enough to enable buffering the next flood15,32.

Although GERD can mitigate extremely high flows, its presence will not fully preclude flood peaks from reaching HAD in scenarios S1–S4. The GERD reservoir can hold 1.6 years of average Blue Nile flow, but during relatively wet years with substantially above-average flow, GERD will not be able to hold the full flood peak (see the interquartile (IQ) ranges in Fig. 3a–d). For instance, although the median GERD outflow is unseasonal under S1, a flood peak still arrives at HAD once every three years on average in this scenario. For this reason, in the following, we assumed that HAD would have to keep following the historical practice of drawdown by August to buffer potential incoming floods, even if these may not arrive every year once GERD becomes fully operational.

Under this assumption, the specifics of GERD operation may substantially affect the HAD emptying–refilling cycle (Fig. 4a,b; see Methods and Supplementary Note 1). We compared typical GERD and HAD lake levels during GERD emptying (January–July; Fig. 4a) and refilling (July–December; Fig. 4b). The near-constant GERD outflow in S1 results in considerable delay of both the emptying and refilling of HAD compared with that of a pre-GERD state, whereas each of S2, S3 and S4 brings the result closer to pre-GERD conditions (Fig. 4a,b, in which pre-GERD would be a vertical line).

Fig. 4: Coordinated GERD–HAD operation.
figure 4

a,b, HAD water levels versus GERD water levels during the GERD’s emptying (a) and refilling (b) stages (m.a.s.l., metres above sea level). Numbers represent the median monthly values from simulations that spanned 26 yr (Methods). Each data point represents one month; arrows represent a forward direction in time from January to July (a) and July to December (b). c, Typical yearly GERD–HAD refilling schedules pre-GERD and under S1–S4. The colours reflect the emptying and refilling speeds of each dam’s reservoir (Methods), and squares indicate the contention period (see text and Methods). d, HAD water levels pre-GERD and under S1–S4; box plots show the median, IQ range and minimum/maximum values from full simulated time series. e, Mean monthly outflow curves for HAD, separated by turbined flow12 (top) and spilled flow (bottom) (Methods). f, Mean monthly HAD power-generation capacity factor.

Source data

Egypt’s concern would presumably not be a delay as such, but rather the resulting lag of HAD compared to that of GERD: prolonged periods could occur during which GERD water levels would be stable or rising while HAD would lag behind, at worst still emptying fast while GERD refills. We define these as ‘contention periods’; the longer these periods, the more likely Egypt’s fear of Ethiopia hoarding water will be provoked2,5,14 (see Methods and Supplementary Notes 5 and 6). We show typical GERD–HAD emptying–refilling calendars in Fig. 4c. Clearly, under S1, the lag in HAD refilling compared with GERD refilling is so substantial that four-month-long contention periods would be the norm. The VRE scenarios S2–S4 mitigate this, as they harmonize GERD–HAD refilling by mimicking the pre-GERD situation. In S3 and S4, HAD refilling would typically lag GERD’s by only one month, which reflects the natural delay between flood peak arrival at GERD and at HAD33. Thus, a high VRE integration could substantially allay Egyptian fears of GERD holding water back and also allow Ethiopia to fully benefit from increased year-round power generation, which leads to a regional win–win situation.

Given that GERD will reduce the Blue Nile flow’s interannual variability, how will HAD hydropower generation and flood protection be impacted? Compared with pre-GERD, HAD will receive more inflow during the dry years, and less during the wet years. This has skewed impacts on HAD water levels: more inflow in dry years would raise water levels, but less inflow in wet years would not have the same inverse effect, as very wet periods typically raise water levels to such an extent that spillways are activated15; mitigating very wet periods would primarily reduce spilling, instead of translating to lower water levels. Average Lake Nasser levels may thus be higher under S1–S4 compared with those pre-GERD (Fig. 4d), with extreme highs and extreme lows mitigated as compared with those pre-GERD.

By reducing extremely high flows, all the scenarios, S1–S4, lower the amount of spilled flow at HAD in favour of turbined flow (Fig. 4e). The effects on HAD power generation are positive but small, at roughly a one percentage point increase for S1–S4 compared with that pre-GERD. More substantial is the change in seasonality of the HAD power generation (Fig. 4f), with somewhat reduced (increased) output in October–November (December–March) under S1 compared with pre-GERD values. The scenarios S2–S4 incrementally bring the seasonality in power generation closer to the pre-GERD state, with S4 again the closest.

To summarize, a controlled seasonality in the GERD outflow in response to a high VRE penetration would help to harmonize the GERD–HAD filling timelines and safeguard all the power generation benefits of both GERD and HAD.

GERD–VRE contribution to the EAPP

Ethiopia currently exports substantial power in the EAPP17, and seeks to become Africa’s largest electricity exporter with GERD34. We investigated the implications of synergetic GERD–VRE operation for cross-border electricity trade, in this case between Ethiopia and Sudan, whose cooperation could trigger quadruple benefits with VRE integration, EFR compliance, GERD infrastructure use and the GERD–HAD cooperation discussed above.

To illustrate the potential of GERD–VRE to contribute to the EAPP objectives to leverage the power system interconnectivity to optimize the regional energy resource usage35, we considered a hypothetical Ethio-Sudanese power exchange agreement based on the GERD–VRE portfolio of S4 (Fig. 2d), with Ethiopia receiving 60% of instantaneously generated power and Sudan 40% (totalling 24.7 and 16.5 TWh yr–1, respectively; Fig. 2e). This can be considered a fair allocation as it reflects the share of what Ethiopia (GERD + VRE) and Sudan (VRE) would generate annually on average in this portfolio; power exchanges would then be necessary because VRE is frequently not available when needed (which requires GERD dispatch) or overproduced when not needed (which allows GERD to ramp down). Based on which power plants produce electricity where and when, the necessary import–export profiles between both countries will therefore show typical hourly and seasonal patterns (Fig. 5).

Fig. 5: Cross-border electricity exchanges between Ethiopia and Sudan enabled by joint GERD–VRE operation.
figure 5

a,b, The average monthly (a) and hourly (b) (in Ethiopian time, utc + 3) profiles by month of electricity exchanges (exports and imports) between Ethiopia and Sudan, assuming Ethiopia is entitled to 60% of the power generated by the GERD–VRE portfolio of scenario S4. The allocation of exports to the different resources (hydro, solar and wind) is to be interpreted as proportional to the respective average contribution of each resource to the GERD–VRE portfolio in the shown months and hours.

Source data

Seasonally, cross-border electricity trade would be relatively low during October–May as VRE across both countries makes up the portfolio’s bulk (Fig. 5a). Ethiopia would mostly export hydropower from GERD during the evening to help meet Sudanese peak demand, and import VRE during morning and midday hours when Sudan’s solar and wind farms produce large volumes of electricity (Fig. 5b). During June–September, when Sudanese VRE potential slumps to a minimum, Ethiopia would export large volumes of hydropower to Sudan (Fig. 5a), especially during mornings and evenings (Fig. 5b).

The total Ethiopian exports to Sudan would amount to 2.5 TWh yr–1 on average (15% of Sudan’s current electricity demand17), with 1.0 TWh yr–1 going the other way. The minimum net-transfer capacity between Ethiopia and Sudan needed to always guarantee transmission of such amounts is 2.4 GW, fully consistent with the current plans to increase Ethio-Sudanese transfer capacity to 3.2 GW by 202517. Further, in the long term, such seasonally dependent electricity trade under scenario S4 is likely to constitute a better business case for Ethiopia as compared with a case without seasonality in GERD output; the reason is that solar and wind power are expected to play strong roles in Sudan’s future power mix as the country has among the highest VRE resources in Africa35,36,37,38, but these leave a shortfall in the rainy season (cf. Fig. 1c) that could be filled by dispatchable plants like GERD. If Sudan pursues an exploitation of its VRE potential, the added value of GERD exports will be maximized by introducing a controlled seasonality. Given all of the above, we consider such a GERD–VRE power exchange agreement to be a realistic avenue to valorize an enhanced Ethio-Sudanese power system interconnectivity and displace fossil fuel-based power generation in Sudan, which again showcases the suitability of scenario S4 to foster regional win–win situations.

The spatiotemporal hydro–solar–wind complementarity observed here for Ethiopia and Sudan can be equally well unlocked through Ethiopian exchanges with other EAPP countries. Among Ethiopia’s direct neighbours in the EAPP, Djibouti, Kenya and South Sudan have excellent solar resources which can complement Ethiopian hydropower37; strong and complementary wind resources are found in Djibouti and specific locations in Kenya and South Sudan38. Ethiopia already exports power to Djibouti, and interconnections to Kenya17,35 and South Sudan39 are planned, which may thus all benefit integrated hydro–VRE operation. In this context, Egypt, which is currently the largest electricity consumer among EAPP countries, could benefit from reinforced Ethiopia–Sudan–Egypt interconnection corridors to leverage its own excellent solar PV and wind resources as another potential complement to Ethiopian hydropower35.

Prospects for Ethiopia’s electricity mix

Last, we consider Ethiopia’s overall electricity mix. Ethiopia’s current plans foresee a renewables-based future with hydropower as the principal source, but with important contributions from solar, wind, geothermal and biomass cogeneration plants17,20.

Clearly, Ethiopia stands before important policy choices. Given its rapidly increasing demand17, and assuming geothermal and biomass plant development goes ahead, Ethiopia may have to prioritize between hydropower and VRE to close the remaining supply–demand gap. Thus, it will have to weigh the consequences of the negative socioenvironmental effects of hydropower exploitation, often a subject of critique in Ethiopia25,40,41,42, and of the increased deployment of VRE resources, which requires substantial system flexibility and investment shifts. Here, based on the concept of hydro–VRE complementarity, we propose an alternative pathway to Ethiopia’s current policy plans that could benefit environmental sustainability, irrigation prospects and electricity generation costs. We note that the purpose of this proposal is primarily to stimulate debates around the wider potential benefits of hydro–VRE integration.

By 2030, Ethiopia’s power demand may reach 54.8 TWh yr–1 (ref. 17), and current forecasts17 for 2030 still foresee a major role for hydropower by then (Fig. 6). We show that applying a hydro–VRE complementary operation to Ethiopia’s existing hydropower fleet plus GERD can deliver the flexibility to support a much higher VRE penetration as compared to those of the forecasts (Fig. 6), which reduces the need for additional hydropower to meet 2030 demand and allows Ethiopia to remain a regional electricity exporter. We estimate that Ethiopia would need only around 32% of the additional hydropower (aside from GERD) currently planned by 2030 (1.8 GW instead of 5.4 GW) for such a scenario (Methods). For example, the potential Karadobi (1.5 GW) and Yeda I and II (0.3 GW) hydropower projects in the Blue Nile basin17, upstream of GERD, could be candidates to deliver the additional hydropower needed by 2030 for such a scenario with high solar and wind power penetration. A strong diversification of Ethiopia’s power-generating fleet can therefore help to reach 100% renewable electricity in 2030 while reducing sustainability concerns around additional hydropower25,42.

Fig. 6: Ethiopian power mix by 2030.
figure 6

a,b, The forecast (‘2030 forecast’) Ethiopian installed capacity (a) and power generation (b) mix compared with those of the 2017 situation (‘2017 historical’)17, as well as an additional scenario based on this study’s consideration of hydro–VRE complementarity (‘2030 hydro–VRE’), in which GERD, all existing Ethiopian hydropower and a selection of the additionally planned hydropower capacity by 2030 is flexibly deployed to aid VRE integration. Under both a and b, we show absolute as well as relative values. The dashed line in b is the forecast 2030 electricity demand in Ethiopia17; power generation does not equal demand because of exports to and imports from other EAPP countries, such as Sudan. The calculation details are explained in Methods.

Source data

Such a diversification also implies that future needs for irrigation and power generation in Ethiopia could be more readily decoupled. In multipurpose hydro projects, the dual objectives of irrigation and electricity generation are often at odds with one another43. A shift away from hydropower dominance to a balanced mix of hydropower and VRE means that future water infrastructure projects in Ethiopia could be designed to more fully benefit irrigation purposes in environmentally friendly ways, and so reduce the need to co-optimize irrigation and electricity needs, which may result in trade-offs.

Given hydropower’s comparatively high capital costs44, such a diversification strategy would avoid substantial upfront expenses for new hydro plants, which could be directly reallocated to finance the needed VRE. Regional VRE may also have higher chances of benefitting from international donors’ support than hydropower45, given that large hydropower projects have largely fallen out of favour with the international community in recent years (China being a notable exception) due to their environmental footprint25. With the levelized costs of both solar PV and wind power expected to continue their reducing trend and approach the typical range for hydropower in Ethiopia well before 203017, it is likely that a diversification towards more VRE would not endanger the low costs of power supply in Ethiopia30. We estimate that the proposed VRE plants in Ethiopia and Sudan, if constructed a few years from now, could achieve even lower levelized costs of electricity (LCOE) than those of GERD, which thus implies substantial financial long-term benefits of joint hydro–VRE operation as compared with a continued hydropower dominance (Methods and Supplementary Note 7). Moreover, such a diversification would be in line with the expectation that VRE-plus-storage may eventually become the backbone of power systems worldwide36,44.

Although both solar and wind power could synergize well with hydropower on their own, diurnal solar–wind46 and solar-load19 complementarity is substantial in the Ethiopian case (Fig. 2b–d). We therefore consider a continued development of joint hydro–solar–wind strategies for all existing and upcoming hydropower plants, similar to the GERD–VRE strategies presented here and emphasizing the value of all three resources, to be a no-regret near-term policy for Ethiopia’s power sector.


Operating GERD in complementarity with solar and wind power in Ethiopia and neighbouring countries marks a strong potential to alleviate certain elements of political tension around GERD between Ethiopia, Sudan and Egypt, and could provide useful input for consideration in the ongoing trilateral GERD negotiations. We found that such a complementary operation, which safeguards year-round power generation targets while mitigating GERD’s effects on downstream flow profiles, could provide fivefold benefits relative to the currently foreseen GERD operation: higher VRE penetration in the EAPP, compliance with EFRs, improved valorization of the GERD infrastructure, harmonization of yearly GERD–HAD refilling schedules and diversification of Ethiopia’s power mix, which theoretically reduces the need for new hydropower dams while safeguarding Ethiopia’s export objectives in the EAPP.

There are various other benefits that GERD may provide to downstream regions4. If a complementary GERD–VRE operation is to be a viable option, these other benefits must not be negatively affected. The most important such benefits include to reduce the sedimentation behind Sudanese dams and HAD, to protect downstream regions from both droughts and floods, and to enable a more efficient water storage than that of HAD. Benefits for sedimentation control are present in all the scenarios presented here, as these are directly related to the presence of GERD itself as a barrier behind which sediment may accumulate; operating GERD with a controlled seasonality does not change this. Flood and drought mitigation is also effective in all scenarios, as GERD will stabilize outflows around a certain median curve as compared with the natural situation, whether or not seasonality is present (Fig. 3a–d and Supplementary Note 3). Lastly, GERD has often been promoted as a more efficient means of storing water than HAD5, as it is located in a cooler and wetter zone, which thus leads to fewer evaporation losses than from HAD. We found that complementary GERD–VRE operation under S4 may further enhance this effect as compared with S1 (Supplementary Note 8). Operating GERD with a controlled seasonality, in complementarity with VRE, therefore does not harm any of the other beneficial aspects of GERD.

In our VRE scenarios, all the involved countries would benefit in a regional win–win situation. Ethiopia would capitalize on year-round reliable power generation to support electrification and industrialization with a robust mix of hydropower and VRE, making optimal use of GERD’s infrastructure and strengthening its position in the EAPP. Sudan would receive a more reliable Blue Nile flow each year, even during the dry season, and reduce its reliance on fossil fuels, and changes in ecological river conditions would be limited. Finally, Egypt’s concerns about GERD’s impact on water management would be mitigated because GERD outflow profiles would appear as those of a relatively small reservoir. Our analysis thus demonstrates a holistic viewpoint of the GERD in the context of the water–energy nexus, beyond hydropower and water politics only. It shows that transboundary water and renewable energy issues are not only linked, but can be harmonized with positive outcomes for all the involved parties.

Important questions remain as to the political viability of the proposed solutions. Even with all the above benefits present, some factors could pose risks to complementary GERD–VRE development, for instance, financial concerns and conflicts in fragile regions. On the financial side, our calculations (Supplementary Note 7) indicate that VRE may become cost-competitive with hydropower in the region within the next decade in terms of levelized costs. The necessary transmission grid reinforcement would entail further costs; however, this study only considered locations whose grid integration already forms part of the policy plans anyway (Methods), and our scenarios were shown to be in line with the planned strengthening of the Ethio-Sudanese cross-border transmission capacity as well. As far as conflict situations in the region are concerned, these may present barriers to the rollout of an integrated power grid; however, VRE development could be argued to be well in line with conflict adaptation strategies, as, for example, VRE plants are unaffected by potential conflict-induced fuel shortages and are more modulizable than big hydro and thermal plants47. Therefore, our VRE scenarios may be more conflict aware than the regional status quo for expanding electricity generation, dominated by large hydropower and thermal plants.

We note here that the ongoing GERD negotiations are taking place in the context of a wider uncertainty and disagreement on water allocation by country9. Beyond GERD, the consensual development of further water infrastructure in the Blue Nile basin for irrigation and other water storage purposes in the future will be strongly contingent on whether or not countries can arrive at comprehensive future water-sharing agreements. This topic is outside the principal scope of this study, which focuses primarily on the regional disagreements that surround future GERD operation.

Although the presented conclusions are robust enough to suggest a roadmap for Ethiopia’s and Sudan’s power sector and their contribution to the EAPP, various remaining uncertainties warrant further research, for example, around the precise locations of future VRE plants, irrigation water extraction in Sudan, initial GERD filling time, Ethio-Egyptian willingness for compromise, Ethiopia’s strategic objectives for further hydropower development to remain the main EAPP feeder, future river damming interventions to serve non-energy purposes, such as irrigation, future negotiations on water-sharing agreements among Nile riparian countries, and future climate change. On the last of these, we consider complementary hydro–VRE development to provide a potential sixth benefit: power sector diversification may lessen hydropower shocks due to increased interannual variability in Nile River flow caused by climate change48,49,50,51.


GERD and other hydropower plants in Ethiopia

The flexible dispatch of hydropower plants was modelled using the REVUB (Renewable Electricity Variability, Upscaling and Balancing) software introduced in previous work28; all the equations and modelling principles are explained therein and summarized hereafter. The REVUB model serves to derive hydropower reservoir operation rules under certain flexibility requirements determined by the VRE infeed and electricity demand at hourly resolution across multiyear time periods, while enforcing minimum flow rules and ensuring a seasonal lake-level stability within safe ranges.

In brief, the model starts from an initial reservoir state and marches forward in time, dispatching hydropower as necessary for each time step to follow a certain target load together with the VRE and recalculating the reservoir state at each next time step depending on released water, river inflow, precipitation gains and evaporation losses. After a simulation, the model verifies whether lake-level safe ranges are followed, and resimulates for a higher target load, iterating until the highest target load is identified with which lake-level criteria can be adhered to. This target load is denoted ‘followed load’ in Fig. 2.

As input data, the REVUB model requires various technical data on hydropower plants: rated capacity (MW), maximum reservoir volume (m3), maximum area (m2), maximum hydraulic head (m), reservoir bathymetry (volume–area–head relationships), live/dead storage ratio, minimum required outflow (for example, based on minimum stable load), efficiency and maximal ramping rates (MW min–1), as well as time series of reservoir inflow (m3 s–1), lake surface precipitation (kg m–2 s–1) and potential evaporation (kg m–2 s–1), normalized VRE power generation profiles (that is, hourly capacity factors of solar PV and wind; see ‘Solar PV and wind power generation’) and a normalized target load curve (see ‘Load profile’).

REVUB simulations were carried out for GERD and for all existing (as of May 2020) hydropower reservoirs in Ethiopia and covered monthly-resolution reservoir inflow during a 26 yr period (following the hydrometeorological conditions of 1980–2005), with typical hourly resolution profiles for solar PV, wind power and load applied to each simulation year (see ‘Solar PV and wind power generation’ and ‘Load profile’). Existing run-of-river hydropower plants were also simulated, assuming that the instantaneous outflow equals the instantaneous inflow at all times. In all the cases, hydropower generation Phydro(t) (MW) was calculated as:

$${P}_{{\rm{hydro}}}(t)={Q}_{{\rm{out}},{\rm{turbined}}}(t)\eta g\rho h(t)\times 1{0}^{-6}$$

where Qout,turbined(t) is turbined water release (m3 s–1), η the hydroturbine efficiency (%), g the gravitational acceleration (9.81 m s–2), ρ the density of water (= 1,000 kg m–3), h(t) the instantaneous hydraulic head (m) and the factor 10−6 converts from W to MW.

For all the reservoirs, bias-corrected precipitation was obtained from the EWEMBI dataset52 and the potential evaporation flux was taken from the ensemble mean of ten historical simulations from regional climate models driven by different global climate models, available through the Coordinated Regional Climate Downscaling Experiment—Africa framework initiative. The maximum turbined flow \({Q}_{{\rm{out}},{\rm{turbined}}}^{{\rm{max}}}\) was assumed to be the outflow that would generate peak power at maximum hydraulic head, the maximum ramping rates were assumed to be 2.6% of full capacity per minute28,53,54 and the efficiency η was assumed to be 95% in all cases28,29. Note that, in practice, η will vary in function of the variations in the load on the turbine, and partial loads below certain thresholds should therefore be avoided to ensure that turbines keep operating in the high-efficiency range. In our modelling, this is assured throughout by enforcing suitable minimum stable loads, obtained from ref. 17, for all plants such that turbines will always run in their high-efficiency range, both for single- and multi-unit plants.

Plant-specific data and sources are summarized as:

  1. (1)

    GERD: reservoir inflow from a SWAT+ (Soil and Water Assessment Tool, revision 55) simulation that covers Africa’s major rivers for the period 1980–2016, described in previous research28,55. The period 1980–2005 was extracted and bias corrected to the long-term average value of 47.2 × 109 m3 yr–1 (ref. 18). The rated capacity used was 6,450 MW (ref. 1), with a maximum reservoir volume of 7.4 × 1010 m3 with 80%/20% live/dead storage (ref. 18), maximum area of 1.9 × 109 m2 (ref. 18), maximum head of 133 m (ref. 18) and full reservoir bathymetry curves provided by S.L. and H.K. For the data in Fig. 3f, GERD was assumed to have 16 hydroturbines of equal rated capacity56. We note the recent uncertainty in media reports around the number of hydroturbines and the total installed capacity for GERD: some note that the former has been brought down from 16 to 13 and the total capacity reduced to 5,150 MW accordingly57, whereas others confirm the former but refute the latter58 and yet others cite a capacity between 5,150 and 6,450 MW (ref. 59). For this study, we opted for the most recently cited numbers in publicly available scientific1 and technical17 studies, notwithstanding the change these numbers may be subject to as GERD construction works near completion. We note here that re-running the simulations with an installed capacity of 5,150 MW instead of 6,450 MW, and with 13 instead of 16 turbines, does not change any of the principal conclusions.

  2. (2)

    Fincha: reservoir inflow from 1980 to 2005 was taken from ref. 17, with a rated capacity of 134 MW (ref. 60), maximum volume of 6.5 × 108 m3 (ref. 60), maximum area of 2.9 × 108 m2 (ref. 60), volume–area relationship using the approach of Kaveh et al.61, assuming a reservoir depth equal to the dam height of 20 m (ref. 60), constant head of 555 m (ref. 62) and the ratio between live/dead storage assumed to be 80%/20% (inspired by the GERD, Melka Wakena and Gibe I values).

  3. (3)

    Fincha–Amerti–Neshe: reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 95 MW (ref. 41), maximum volume of 4.5 × 108 m3 (ref. 63), maximum area of 2.9 × 107 m2 (ref. 63), volume-area relationship using the approach of Kaveh et al.61, assuming a reservoir depth equal to the dam height of 38 m (ref. 41), constant head of 579 m, estimated from the elevation difference between the dam outlet and the powerhouse on Google Earth, and the live/dead storage ratio assumed to be 80%/20% (inspired by the GERD, Melka Wakena and Gibe I values).

  4. (4)

    Tekeze: reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 300 MW, maximum volume of 9.2 × 109 m3, with 57%/43% live/dead storage, and maximum area of 1.5 × 108 m2, all from Tekeze Inauguration Bulletin64 and Basheer et al.65, maximum head of 163 m (ref. 60) and bathymetry using the approach of Kaveh et al.61, approximating the reservoir depth with the maximum head.

  5. (5)

    Melka Wakena: reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 153 MW (ref. 66), maximum volume of 7.6 × 108 m3 (ref. 66), with 79%/21% live/dead storage, constant head of 297 m (ref. 66), maximum area of 8.2 × 107 m2 (ref. 67) and bathymetry from Bosona68.

  6. (6)

    Genale Dawa III: reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 254 MW (ref. 69), maximum volume of 4.9 × 109 m3 (ref. 70), with 53%/47% live/dead storage, maximum area of 9.8 × 107 m2 (ref. 70), maximum head of 188 m (ref. 70) and bathymetry using the approach of Kaveh et al.61 assuming a reservoir depth equal to the dam height of 110 m (ref. 69).

  7. (7)

    Gibe I: reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 200 MW (ref. 71), maximum head of 110 m (ref. 71), maximum volume of 8.4 × 108 m3 (ref. 72), with 78%/22% live/dead storage, maximum area of 6.0 × 107 m2 (ref. 72) and bathymetry using the approach of Kaveh et al. 61 assuming a reservoir depth equal to the dam height of 40 m (ref. 71).

  8. (8)

    Gibe II (run-of-river): throughflow (diverted from Omo river via a tunnel through a mountain ridge, and rejoining the Omo river downstream) assumed to be equal to the Gibe I outflow (just upstream) as simulated, minus a minimum guaranteed runoff of 2 m3 s–1 into the Omo river73, with a rated capacity of 420 MW (ref. 74) and head of 505 m (ref. 74).

  9. (9)

    Gibe III: Reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 1,870 MW (ref. 75), maximum volume of 1.4 × 1010 m3 (ref. 75) and maximum area of 2.0 × 108 m2 (ref. 75), maximum head of 220 m (ref. 76) and bathymetry using the approach from Kaveh et al.61 approximating reservoir depth with maximum head and live/dead storage ratio assumed to be 80%/20% (inspired by the GERD, Melka Wakena and Gibe I values).

  10. (10)

    Tana Beles (run-of-river): throughflow (diverted from Lake Tana) assumed constant at 77 m3 s–1 (ref. 77) with a head of 311 m (ref. 77) and rated capacity of 460 MW (ref. 60).

  11. (11)

    Tis Abay I + II (run-of-river): river flow from 1980 to 2005 from ref. 17, with a rated capacity of 84.4 MW (ref. 78) and head of 53 m (ref. 60).

  12. (12)

    Awash I (Koka): reservoir inflow from 1980 to 2005 from ref. 17, with a rated capacity of 43 MW (ref. 60), maximum volume of 1.9 × 109 m3 (ref. 60), maximum area of 1.8 × 108 m2 (ref. 60), maximum head of 42 m (ref. 60), bathymetry using the approach from Kaveh et al.61 assuming the reservoir depth equal to the maximum head and a live/dead storage ratio assumed to be 80%/20% (inspired by the GERD, Melka Wakena and Gibe I values).

  13. (13)

    Awash II + III (run-of-river): throughflow assumed to be equal to Awash I (Koka) outflow (just upstream) as simulated, with a rated capacity of 64 MW (ref. 60 and head of 60 m (ref. 79).

A modelling flowchart showing the various steps taken in the modelling of Ethiopian hydropower plants, and the most important outputs of these simulations for the purposes of this study, is given in Supplementary Note 1.

HAD operation in Egypt

HAD inflow \({Q}_{{\rm{in}}}^{{\rm{HAD}}}(t)\) was estimated as the sum of the inflow from the Blue Nile and White Nile, which merge in Khartoum, and the Atbara, which joins a little further north (Fig. 1a), based on the assumption that little to no water is added to Nile flow after these rivers join the main branch in the desert33 and that the bulk of evaporation happens on Lake Nasser itself. The Blue Nile contribution was taken as equal to the GERD outflow, given that the historical average Blue Nile flows measured just downstream of GERD (Roseires/El Deim gauge) and at Khartoum are within 1% of each other33, which reflects the net effect of water gained from runoff and the Rahad and Dinder tributaries being compensated by water withdrawn for irrigation and other losses between the Ethio-Sudanese border and Khartoum. The Atbara contribution was taken from the same SWAT+ simulation used for GERD (see above), bias corrected to the average observed flow at the Atbara’s mouth reported in Sutcliffe and Parks33. For the White Nile contribution, instead of using the SWAT+ simulation results, we directly applied the average monthly observed flow at Khartoum (Mogren gauge) in Sutcliffe and Parks33 to all the simulation years, given that South Sudan’s Sudd wetland has complicated seasonal effects on water retention, losses and flow lags, which many hydrological models, including SWAT+, do not accurately reproduce80,81. We note that this approach neglects any interannual variability in the White Nile flow, but this is inconsequential for the assessment of typical GERD–HAD emptying–refilling schedules, because the seasonality of Nile flow, which drives the HAD emptying–refilling cycle, is almost fully determined by the Blue Nile and the Atbara. This approach also has the advantage of allowing us to more easily distinguish the effects of changes in the Eastern Nile sub-basin (Blue Nile + Atbara) from effects in the Equatorial Nile sub-basin (White Nile), whose hydroclimate is relatively uncorrelated to that in the Eastern Nile49, and thus to better discern their impact. A sensitivity study on the effect of including interannual variability in the White Nile flow was carried out and is described in Supplementary Note 5.

We assumed a one-month delay between these rivers joining in Sudan and their combined flow arriving at HAD, given typical water travel times from southern to northern Sudan33. We neglected any potential lag introduced by operation of the Roseires and Merowe reservoirs in Sudan (Fig. 1a), as both these reservoirs have a much smaller capacity than their respective inflow, mostly letting water pass unimpeded during the wet season7. Lastly, the substantial evaporation losses and near-negligible precipitation gains on Lake Nasser were estimated from the same EWEMBI and CORDEX datasets as used for the Ethiopian reservoirs (see above, ‘GERD and other hydropower plants in Ethiopia’).

Given that HAD is primarily a flood protection and irrigation dam, with hydropower generation as a secondary function (it provides less than 10% of Egyptian power demand), we did not use REVUB to model a flexible HAD operation in synergy with Egyptian solar and/or wind power. Instead, HAD operational rules were defined based on logarithmic-exponential release rules28,82 as a function of Lake Nasser volume VLN(t) aimed to keep lake levels stable on multiannual time scales, corrected by a seasonal factor fseasonal to reflect standard monthly variations in the HAD outflow (see below). The release from HAD at each time step t, denoted \({Q}_{{\rm{out}}}^{{\rm{HAD}}}(t)\), is thus given by:


The first term \({Q}_{{\rm{out}},{\rm{turbined}}}^{{\rm{HAD}}}(t)\) represents the standard turbined outflow, given28,82 by:

$$\begin{array}{l}{Q}_{{\rm{out}},{\rm{turbined}}}^{HAD}(t)=\\\left\{\begin{array}{ll}{Q}_{{\rm{claimed}}}\left[{d}_{{\rm{min}}}+\mathrm{ln}\,\left(\kappa {\left[\frac{{V}_{{\rm{LN}}}(t)}{{V}_{{\rm{LN}}}^{{\rm{max}}}}\right]}^{\phi }+1\right)\right]{f}_{{\rm{seasonal}}},&\,\text{for}\,{V}_{{\rm{LN}}}(t)/{V}_{{\rm{LN}}}^{{\rm{max}}}<{f}_{{\rm{opt}}}\\ {Q}_{{\rm{claimed}}}\left[\exp \left(\gamma {\left[\frac{{V}_{{\rm{LN}}}(t)}{{V}_{{\rm{LN}}}^{{\rm{max}}}}-{f}_{{\rm{opt}}}\right]}^{2}\right)\right]{f}_{{\rm{seasonal}}},&\,\text{for}\,{f}_{{\rm{opt}}}\le {V}_{{\rm{LN}}}(t)/{V}_{{\rm{LN}}}^{{\rm{max}}}<{f}_{{\rm{spill}}}\\ {Q}_{{\rm{out}},{\rm{turbined}}}^{{\rm{max}}}&\,\text{for}\,{V}_{{\rm{LN}}}(t)/{V}_{{\rm{LN}}}^{{\rm{max}}}\ge {f}_{{\rm{spill}}}\end{array}\right\}\end{array}$$

Here, Qclaimed is Egypt’s claimed share of Nile waters at 55.5 × 109 m3 yr–1 (see our note in the Acknowledgments on colonial-era water allocations between Nile countries), dmin = 26.9% is the fraction of yearly average inflow required as minimum stable outflow, \({V}_{{\rm{LN}}}^{{\rm{max}}}\) is Lake Nasser’s storage capacity of 1.47 × 1011 m3 (with a 79%/21% live/dead storage ratio60), \({Q}_{{\rm{out}},{\rm{turbined}}}^{{\rm{max}}}\) is the maximum turbine throughflow of 3,125 m3 s–1, fopt = 75% is the optimal filling fraction, determined as the cross-over point between the conservation zone below 175 m.a.s.l. lake level elevation and the flood buffering zone above 175 m.a.s.l. and fspill = 80% is the filling fraction at which spilling starts, that is, whenever Lake Nasser’s water level exceeds 178 m.a.s.l. All these data were directly taken from Moussa15, unless otherwise indicated; Lake Nasser bathymetry was also taken from Moussa15. The seasonal correction factor fseasonal, which reflects the usual operational choice of increasing outflow in the months that precede the peak Nile flow to make room to buffer the annual flood12,15, was derived from the monthly outflow curve in Abd Ellah12, who in turn obtained it from Egypt’s Water Resources Research Institute, which falls under the Ministry of Water Resources and Irrigation. Tabulated values of fseasonal are given in Supplementary Note 6; a sensitivity test of our results to changes in fseasonal is also described therein. Further, κ, ϕ and γ are constants, with κ and ϕ given by28,82:

$$\kappa ={f}_{{\rm{opt}}}^{-\phi }\left[\exp (1-{d}_{{\rm{min}}})-1\right]\,\text{;}\,\phi =\alpha {\tau }_{{\rm{fill}}}^{1/2}$$

Here, τfill is the HAD’s filling time (reservoir volume divided by average annual inflow). We used α = 2/3 and γ = 10 based on the generalized reservoir operation rules28,82 and calculated τfill = 1.57 based on the average inflow as simulated.

The second term \({Q}_{{\rm{out}},{\rm{spill}}}^{{\rm{HAD}}}(t)\) represents the non-turbined outflow diverted via spillways at high lake levels. It is defined as:

$$\begin{array}{l}{Q}_{{\rm{out}},{\rm{spill}}}^{{\rm{HAD}}}(t)=\\\left\{\begin{array}{ll}0&\,\text{for}\,{V}_{{\rm{LN}}}(t)/{V}_{{\rm{LN}}}^{{\rm{max}}}<{f}_{{\rm{spill}}}\\ \,\text{max}\,\left(0;\,\text{min}\,\left[{Q}_{{\rm{in}}}^{{\rm{HAD}}}(t)-{Q}_{{\rm{out}},{\rm{turbined}}}^{{\rm{max}}};{Q}_{{\rm{max}},{\rm{spill}}}^{{\rm{HAD}}}\right]\right)&\,\text{for}\,{V}_{{\rm{LN}}}(t)/{V}_{{\rm{LN}}}^{{\rm{max}}}\ge {f}_{{\rm{spill}}}\end{array}\right\}\end{array}$$

where \({Q}_{{\rm{max}},{\rm{spill}}}^{{\rm{HAD}}}\) represents the maximum allowed spilled flow rate, taken to be 5,000 m3 s–1 (ref. 15). HAD power generation was calculated using equation (1).

A modelling flowchart showing the various steps taken in the modelling of HAD, and the most important outputs of these simulations, is given in Supplementary Note 1.


Minimum seasonal EFRs used for the analysis presented in Fig. 3e were based on the adapted Tessmann method, which represents the most stringent EFR analysed in Jägermeyr et al.24. In this method, each month m is characterized by a low, intermediate or high flow regime, based on the mean natural flow 〈Qnatm in that month:

$$\,\text{Flow regime in month}\,m:\left\{\begin{array}{ll}{\rm{low}},&\,\text{for}\,{\langle {Q}_{{\rm{nat}}}\rangle }_{m}\le 0.4\times {\langle {Q}_{{\rm{nat}}}\rangle }_{y}\\ {\rm{intermediate}},&\,\text{for}\,0.4\times {\langle {Q}_{{\rm{nat}}}\rangle }_{y}<{\langle {Q}_{{\rm{nat}}}\rangle }_{m}\le {\langle {Q}_{{\rm{nat}}}\rangle }_{y}\\ {\rm{high}},&\,\text{for}\,{\langle {Q}_{{\rm{nat}}}\rangle }_{m}>{\langle {Q}_{{\rm{nat}}}\rangle }_{y}\end{array}\right\}$$

where 〈Qnaty is the annual mean of natural flow Qnat. Subsequently, the minimum environmental flow Qenv,m is defined by category for each month m, depending on the prevailing flow regime in that month:

$${Q}_{{\rm{env}},m}=\left\{\begin{array}{ll}0.80\times {\langle {Q}_{{\rm{nat}}}\rangle }_{m},&\,\text{in the low flow regime}\\ 0.40\times {\langle {Q}_{{\rm{nat}}}\rangle }_{y},&\,\text{in the intermediate flow regime}\\ 0.40\times {\langle {Q}_{{\rm{nat}}}\rangle }_{m},&\,\text{in the high flow regime}\end{array}\right\}$$

We note that the outcomes of this EFR definition are in line with independent EFR assessments for the Blue Nile in other literature18,26. In addition to the minimum EFRs, there may also be maximum EFRs during dry months when river ecology may depend critically on low flows83. Although we do not have quantitative measures of maximum EFRs for the Blue Nile, scenarios S2–S4 bring dry-season outflows closer to the natural flow than does S1 (Fig. 3a–d), which means that harmonized GERD–VRE operation may also increase compliance with the maximum EFRs.

GERD–HAD emptying–refilling speed

The typical emptying–refilling speeds in Fig. 4c were defined through the relative change in median monthly water levels as compared to each previous month (medians taken for each of the 12 months of the year across the full 26 yr simulation). With zm the median lake level elevation in month m, and Δz = max(z) – min(z)the range spanned by zm across all m, the emptying-refilling speed \({v}_{m}^{{\rm{fill}}}\) for month m was defined as:

$${v}_{m}^{{\rm{fill}}}=({z}_{m}-{z}_{m-1})/{{\Delta }}z$$

Subsequently, each value of \({v}_{m}^{{\rm{fill}}}\) was allocated to one of five symmetrical categories: fast emptying (\({v}_{m}^{{\rm{fill}}}<-30 \%\)), slow emptying (\(-30 \% \le {v}_{m}^{{\rm{fill}}}<-10 \%\)), stable (\(-10 \% \le {v}_{m}^{\rm{fill}}<10 \%\)), slow refilling (\(10 \% \le {v}_{m}^{{\rm{fill}}}<30 \%\)) and fast refilling (\({v}_{m}^{{\rm{fill}}}>30 \%\)). ‘Contention periods’ were defined as any sequence of months in which \({v}_{m}^{{\rm{fill}}}\) was categorized as stable or refilling for GERD, but at least one category behind for HAD. Note that the cutoffs of these ranges can be defined in multiple ways without changing the generality of the conclusions drawn from Fig. 4c.

Solar PV and wind power generation

The locations for VRE power plants considered in this study (Fig. 1a) were based on the already existing on-grid projects, plus those with priority index 1–3 (out of 6) in ref. 17 (for Ethiopia) and those indicated as ‘in pipeline’ in ref. 84 (for Sudan). The geographical coordinates for the locations of planned VRE plants were determined based on project name, which reflect towns or cities; they are given in tabular format in Supplementary Note 2. We note that, according to current plans for transmission network expansion17,85, all the locations for solar PV and wind power generation considered in this study are to be connected to the respective national grids of Ethiopia and Sudan in the future.

The hourly resolution solar PV yield (one curve per month) per unit installed capacity for ground-mounted large-scale crystalline silicon PV systems was derived from the Global Solar Atlas for each set of solar PV plant coordinates37. Similarly, hourly resolution wind speed at 100 m height (one curve per month) was derived from the Global Wind Atlas for each set of wind power plant coordinates38 before being converted into hourly wind turbine yield per unit installed capacity, based on the characteristics of Vestas V100-1.8 wind turbines with a 100 m hub height and cut-in, rated and cut-out wind speed of 3, 12 and 20 m s–1, respectively86.

These databases do not contain day-to-day variability within a month or full interannual variability, and thus have less temporal detail than, for example, global reanalysis datasets, whose applicability for VRE resource assessment from hourly to multiannual timescales has been well-documented28,46. However, reanalysis datasets typically have a relatively coarse spatial resolution, in the order of tens of kilometres. For Ethiopia, in whose highlands solar PV and (especially) wind power potential vary strongly at the kilometre scale, the state-of-the-art reanalysis products are therefore mostly unsuitable to discern high-potential VRE sites17,87. Our choice of dataset thus represents a compromise between achieving the necessary spatial resolution and retaining adequate detail on hourly and seasonal profiles.

Load profile

The hourly electricity demand curve used as the input (target load) for the REVUB simulations was based on the actual Ethiopian grid load from 2013 available in Toktarova et al.19. As such data for Sudan was not available, we assumed the hourly profile of the Sudanese grid load to follow that of Ethiopia.

We performed an additional sensitivity analysis to establish that our conclusions will not change under future shifts in load profile shapes reflecting the combined effect of increased electricity access, higher use of cooling and other household appliances, increased industrial activity and energy efficiency gains to be expected in the coming years19. The results of this analysis are provided in Supplementary Note 4.


The modelling differences between the four scenarios developed for this study are described in detail below:

  • S1 (No VRE): GERD performs load following without any VRE contribution.

  • S2 (Ethiopian VRE): GERD performs load following in combination with VRE from Ethiopian solar PV and wind power plants. We assumed the capacity mix of solar PV and wind power to be 50/50, based on current projections, which foresee roughly comparable capacities for both resources in Ethiopia by 203017; further, we assumed solar PV and wind power capacity to be equally divided over the selected locations (Fig. 1a). Surplus VRE generation compared with the followed load was constrained to occur at most 10% of the time28.

  • S3 (Power Pool): equal to S2, except that Sudanese solar PV and wind power plants were added to the VRE mix, assuming an equal division of capacity of either resource across both countries and adequate cross-border transmission capacity.

  • S4 (Power Pool+): equal to S3, but relaxing the constraint on surplus VRE from 10 to 35% of the time, such that GERD’s flexibility potential is fully utilized. The latter is defined as GERD’s full range of flexibility being deployed, but without turbine capacity being exhausted more than 1% of the time (to avoid frequent loss of spinning reserves)28.

Ethiopian power mix

The historical (2017) and forecast (2030) capacity mix (Fig. 6a) was taken from ref. 17. We estimated historical and forecast power generation (Fig. 6b) by first calculating the hydro, solar PV and wind power generation using the capacity factors calculated in this study (for future hydropower plants aside from GERD, we applied the average capacity factor across the existing fleet), and subsequently assuming biomass power plants would run flexibly at 90% efficiency, but only during the cropping period October–May17, assuming geothermal power plants would run as baseload at a 90% efficiency17 and that gasoil capacity would be solely for back up, not normally being dispatched17.

The hydro-VRE (2030) option was then calculated as follows. First, we took biomass, geothermal and gas capacity and power generation to be equal to the forecast values. Second, we assumed GERD to be operated according to scenario S4, and all the other existing Ethiopian hydropower plants according to the settings of scenario S2. Third, having run the corresponding REVUB simulations, we summed up the cumulative capacity and generation from hydropower (existing + GERD) plus the Ethiopian VRE supported by this hydropower fleet. Fourth, we calculated the deficit between the expected 2030 demand (assumed to follow the same relative profile as the historical load; see ‘Load profile’) and the sum of hydropower (existing + GERD), VRE (supported by existing hydropower + GERD), biomass and geothermal power generation, both in terms of peak power (GW) and total power generation (TWh yr–1), correcting for the exports to and imports from Sudan (Fig. 5). Fifth, we estimated the additional hydropower and VRE capacity that would be necessary to compensate the identified peak power and power generation deficits, assuming that such additional hydropower plants would have similar capacity factors and could deliver similar flexibility for VRE integration as the existing fleet’s average. This additional hydropower and VRE capacity and generation was then added to the results from the third step.

As far as overall VRE potential in Ethiopia and Sudan is concerned, our study remains well within these limits. Ethiopia’s overall theoretical potential for wind power is estimated at 1,350 GW (ref. 17) and that for solar PV power at 5.85 kWh m2 day–1 (ref. 88). In our hydro-VRE scenario of Fig. 6, the installed wind power capacity in Ethiopia is 4.84 GW, less than 1% of Ethiopia’s total potential. Further, the average daily solar power generation needed in Ethiopia in that scenario is 24 GWh day–1, which is roughly equivalent to the irradiation received on only 4 km2; with typical solar panel efficiencies of 20%, this would require a surface area of only about 20 km2 covered by solar PV panels. The resource characteristics in Ethiopia would thus be by far sufficient for scenarios S2–S4.

In Sudan, the overall wind power potential appears not to have been calculated, but it is clear from, for example, the Global Wind Atlas38 that it is even higher than that of Ethiopia; this also applies to its solar PV potential, which reaches 6.32 kWh m–2 day–1 (ref. 88). Therefore, Sudan’s contributions to scenarios S3 and S4, with the same amounts of installed wind and solar PV power capacity as that of Ethiopia (because of the proposed 50/50 capacity split between the countries; see ‘Scenarios’), is also well within the limits of the resource characteristics.


The LCOE is equal to the sum of a power plant’s costs over its lifetime, divided by the total electricity output cumulatively generated over its lifetime:

$$\,\text{LCOE}\,=\frac{{\sum }_{y}\frac{{I}_{y}+{M}_{y}+{F}_{y}}{{(1+r)}^{y}}}{{\sum }_{y}\frac{{E}_{y}}{{(1+r)}^{y}}}$$

where y represents the year of the plant’s lifetime (0 ≤ y ≤ Y, with Y the plant’s lifetime), Iy are the initial (overnight) costs related to construction of the plant in each year y, My are the operational and maintenance costs in each year y, Fy are the fuel costs in each year y (these are zero for renewable resources, such as hydro and VRE), Ey is the total electricity generated by the plant in each year y and r is the discount rate.

To assess the financial viability of the proposed hydro–solar–wind mixes, we estimated the LCOE of GERD and the proposed VRE plants in Ethiopia and Sudan based on cost estimates for 2020, 2025 and 2030 as reported in the literature2,17, and based on plant capacity factors as elaborated in this study. Results and calculation details are given in Supplementary Note 7.