Rethinking energy planning to mitigate the impacts of African hydropower

Rethinking energy planning to mitigate the impacts of1 African hydropower2 Angelo Carlino1,2,†, Rafael Schmitt 2,3,4, Anna Clark2, Andrea Castelletti∗1 3 1Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milano, 20133,4 Italy5 2Natural Capital Project, Stanford University, Stanford, CA 94305, United States of America6 3The Woods Institute for the Environment, Stanford University, Stanford, CA 94305, United States7 of America8 4Doerr School of Sustainability, Stanford University, Stanford, CA 94305, United States of America9 †Present Address: Carnegie Institution for Science, Department of Global Ecology, Stanford, USA10 Abstract11 Around 100 GW of new hydropower projects have been proposed in continental12 Africa to contribute to meeting future energy demand. Yet, the future expansion of13 hydropower on the continent faces obstacles due to the impacts of dams on rivers,14 greenhouse gas emissions from reservoirs, and increasingly competitive alternative15 renewable electricity technologies. Here, we propose an integrated approach to in-16 clude these considerations in energy planning. Compared to planning for least-cost17 energy systems, capacity expansion strategies balancing environmental and techno-18 economic objectives increase electricity prices and total discounted costs by at most19 1.4% and 0.2%, respectively, while reducing impacts on annual hydropower emis-20 sions and river fragmentation by at least 50%. Our results demonstrate that refining21 techno-economic analysis in light of global and local environmental objectives can22 help policymakers reduce the river fragmentation and greenhouse gas emissions23 associated with hydropower development at marginal increases in energy costs.24 25 ∗corresponding author, email: [email protected] 1 26 Developing countries are implementing new energy infrastructure to improve their living stan-27 dards1and meet the needs of their growing populations2. In this context, the deployment of28 low-carbon technologies is paramount to reducing greenhouse gas emissions and mitigating cli-29 mate change3. From a levelized cost of electricity perspective, hydroelectricity has traditionally30 been an inexpensive source of power that does not rely on fossil fuels4with the potential to31 expand substantially5, especially in the global South, where most of its remaining potential32 lies6. In Africa, more than 300 hydropower projects have been proposed, which could provide33 an additional power capacity of around 100 GW for the continent7. This expansion would more34 than triple the existing hydropower capacity of 40 GW which currently represents 15% of the35 total installed capacity3.36 Exploiting the continent’s hydropower potential would have many social, environmental, and37 climatic consequences8. Worldwide, only 23% of the rivers flow uninterrupted to the ocean,38 of which the majority are in the Arctic, with the remainder found in Southeast Asia and the39 Congo Basin9. In particular, out of 543 rivers longer than 500 km flowing uninterrupted to the40 ocean globally, 156 (29%) are located in the African continent9. Flow patterns and blocked41 ecological corridors result in fragmentation of the fluvial ecosystem10.Consequently,species42 habitats might be negatively a!ected11,12, leading to biodiversity loss 13,14. Dams also have43 a substantial impact on physical processes such as sediment delivery 15,16, promoting coastal44 erosion and subsidence17,18. Hydropower development has to strike an adequate compromise45 between safeguarding fluvial ecosystem services and building enough capacity to meet the46 energy system needs19,20.47 In addition, a growing body of research has documented the climate e!ects of reservoir hy-48 dropower. As organic matter decays in reservoirs, these artificial water bodies can become49 non-negligible sources of carbon dioxide and methane21,22. The main emissions pathways in50 reservoirs are methane bubbling, methane and carbon dioxide di!usion, and degassing when51 water is released from the turbines23,24.Sincetheseprocessesdependontheabundanceof52 biomass and the speed of biological processes regulated by temperature, emissions are higher53 for reservoirs at tropical latitudes, as confirmed by field measurements25. While carbon diox-54 ide emissions from reservoirs decline exponentially over time due to faster aerobic digestion,55 methane emissions increase and stabilize due to the long time scale of ebullition and degassing56 processes26.Thisimpliesthatinafewdecades,whenemissionsfromothersectorssuchas57 energy, industry, and agriculture have decreased, addressing reservoir emissions, in particular58 methane, will become essential to achieve net zero emissions26,27.59 In addition to ecological and climatic concerns, there are also economic challenges. Hydropower60 projects compete for capacity expansion with technologies whose cost e”ciency is rapidly in-61 creasing, especially solar photovoltaics (PV)28 and wind power29. Traditional hydropower is a62 mature technology whose costs are not expected to change substantially in the future. In con-63 trast, the costs of solar PV and wind are expected to continue their present declining trend30.64 For this reason, at least a third of the proposed hydroelectric projects in Africa are not expected65 2 to be cost-optimal31,32.66 Strategic hydropower planning can be one approach to solving challenging techno-ecological67 trade-o!s. It relies on multi-objective optimization to navigate the trade-o!s between environ-68 mental impacts and hydropower expansion33.Thismethodprovidesdecision-makerswithaset69 of Pareto-optimal dam portfolios minimizing environmental impacts for any given level of ad-70 ditional hydropower generation or capacity15,16,34–36. However, few studies have examined the71 role and cost-e!ectiveness of new hydropower in the context of future energy systems17,32,37.72 On the other hand, hydropower infrastructure sequencing within capacity expansion in energy73 and power system models is driven by techno-economic objectives. Environmental impacts have74 been examined a posteriori38 or considered in the optimization phase in the case of country-75 level power systems modelling39. Thus, there remains a fundamental disconnect between the76 two approaches.77 In the African continent, this disconnect may encourage risky infrastructure development, ul-78 timately hindering progress towards a sustainable and secure energy future. The ability of79 new hydropower to meet growing energy demand could have a substantial impact on rivers80 and associated ecosystem services and greenhouse gas emissions. To address these challenges,81 our research explores two questions. First, what are the emissions and impacts on ecosystems82 associated with the development of hundreds of hydropower projects planned until 2050 for83 the African continent? Second, we explore how the African energy system can benefit from84 strategic dam planning approaches to reduce these impacts. What would the ensuing economic85 consequences be for the energy sector?86 Here, we couple energy system modelling and strategic dam planning to assess the potential87 evolution of the hydropower sector on the African continent from 2020 to 2050 under di!erent88 climate and socio-economic scenarios. By integrating these approaches, we resolve their inconsis-89 tencies and reduce their respective drawbacks. As we consider the energy system and internalize90 previously overlooked climatic and environmental externalities, we design cost-e”cient path-91 ways for continental hydropower expansion that minimize environmental impacts while meeting92 the demands of the energy system.93 Our study demonstrates the value of integrating the techno-economic perspective with the94 strategic dam planning approach to select cost-e!ective low-impact hydropower projects. Ad-95 ditionally, these results underscore the importance of discovering near-optimal solutions40 in96 energy system models to reduce non-monetizable risks associated with infrastructure planning.97 3 Integrating environmental objectives in energy planning98 We combine an energy system model with a hydropower development simulation model to99 examine techno-economic, global climatic, and local environmental objectives. In particular,100 we rely on the OSeMOSYS-TEMBA open source energy system model for the African con-101 tinent41–44, modified to include specific existing and proposed hydropower plant data from102 the African Hydropower Atlas7.Weconsidereachofthe628existingandfuturehydropower103 projects using capacity factors obtained from a distributed hydrological model32,45 using the104 full expansion configuration under three climate and land-use change scenarios and we discuss105 the importance of this approximation in Supplementary Note 1. The model derives the devel-106 opment of cost-optimal energy systems from 2020 to 2050 sequencing more than 300 proposed107 hydropower projects. We consider climate impacts on water availability, land-use change, and108 socio-economic projections for final energy demands using the Inter-Sectoral Impact Model109 Comparison Project (ISIMIP2b)46, a subset of the Shared Socioeconomic Pathways (SSP)47.110 These scenarios, i.e. SSP1-2.6, SSP4-6.0, and SSP5-8.5, represent future narratives associated111 with ambitious sustainability targets, increasing inequalities, and development based on fossil112 fuels, respectively. While we model the full energy system including non-electricity sectors, we113 report results for the power capacity expansions and the generation trajectories, mostly focusing114 on hydropower.115 To address the impacts of hydropower development, in line with strategic dam planning, we116 introduce the ecosystems-emissions-generation (EEG) simulation model. Given a set of existing117 and future dams, the EEG simulation model computes the value of average annual hydropower118 generation, fluvial ecosystem fragmentation using the river fragmentation index38,48,49,and119 annual carbon dioxide equivalent emissions using a per-mass 100-year warming potential for120 methane equal to 34 times that of carbon dioxide25,26. The emissions are computed using the121 surface area of each reservoir and the emission factors based on the climatic zone26.TheEEG122 simulation model can be coupled with multi-objective optimization algorithms to sequence123 hydropower projects e!ectively balancing climatic and environmental considerations15,16.For124 more details about the models, see Methods.125 4 Impacts on river connectivity and hydropower emissions126 Using the EEG simulation model, we examine the impacts of cost-optimal hydropower sequenc-127 ing obtained from energy system modeling in terms of hydropower greenhouse gas emissions128 and average river fragmentation across African river basins where dams are present or under129 consideration. Fig. 1a) illustrates that at least 32 GW, i.e., 32% of the proposed capacity, is130 not cost-optimal due to competition from other sources, particularly solar PV and wind power.131 As a consequence, the competitiveness of other power sources is responsible for a reduction in132 the impacts associated with hydropower development compared to the full expansion scenarios,133 resulting in a decrease of at least 4% (-16% of the additional impact) in river fragmentation134 and 4.5 MtCO2-eq/y (-19% of the additional impact) in hydropower emissions.135 However, the impacts of cost-optimal hydropower development considering other renewable136 sources are still substantial compared with present conditions. The current average river frag-137 mentation index across the African river basins is around 26%, as shown in Fig. 1b). Under138 the cost-optimal energy system expansion scenarios, it would increase to at least 33% and up139 to 38% in the worst case. For reference, under the full expansion scenario, the river fragmenta-140 tion index could reach 42%, almost reaching the global average river fragmentation index, i.e.,141 43%49. Both fluvial ecosystem and hydropower emission impacts are larger under the SSP1-2.6142 scenario, intermediate under SSP5-8.5, and lowest under SSP4-6.0, as the former requires more143 hydropower to meet more ambitious emission reduction targets.144 Because the energy expansion model does not consider greenhouse gas emissions from reservoirs,145 the dam portfolio for SSP 1-2.6 results in the highest emissions from the hydropower sector146 reported in Fig. 1c). When considering only hydropower, SSP4-6.0 and SSP5-8.5 have lower147 reservoir emissions, because less hydropower is installed. Of course, the total energy system148 emissions of the latter are higher than the one observed in SSP1-2.6. As the energy system149 considers only fossil fuel emissions, it deploys more hydropower to reach ambitious emissions150 reduction targets.151 Under our modelling framework, the emissions associated with the hydropower portfolio of152 SSP1-2.6 would not align with the achievement of the 2.0 °C emission reduction target. Indeed,153 when the emissions constraints become stringent, any additional emissions in the power sector154 conflict with the demand and emissions from other sectors still relying on fossil fuels, as shown155 in Fig. S1. At present, emissions from existing hydropower, in the order of 50 MtCO2-eq/y,156 if accounted for, would result in a 3% increase in the total African energy system emissions.157 However, as the use of fossil fuels declines under SSP1-2.6, the full hydropower expansion158 scenario would result in an increase by 7% of total energy system emissions in 2050, if they159 were considered using CO2-eq units based on the global warming potentials of 100 years26.160 In this sense, the higher the reduction in fossil fuels and the following reduction in emissions,161 the higher the relevance of hydropower emissions for achieving net zero emissions. Yet, large162 uncertainty remains in the value of emissions50 and in the fraction attributable to hydropower163 generation rather than to other water uses.164 5 Reducing impacts via strategic dam planning165 To reduce the impacts of cost-optimal hydropower expansion, we adopt a multi-objective per-166 spective typical of strategic dam planning perspective and examine the trade-o!s between167 average annual hydropower generation, river fragmentation, and reservoir emissions.168 We compare the objectives of cost-optimal hydropower expansion using the energy system169 model with hydropower expansion portfolios minimizing conflicts between environmental, cli-170 mate, and generation objectives obtained from the EEG simulation model via multi-objective171 optimization in Fig. 2. The energy system scenarios remain suboptimal in this framing as their172 performance is always Pareto-dominated by the strategic dam planning approach. Indeed, with173 the same level of emissions and river ecosystem fragmentation, it would be possible to provide174 between 75 and 100 TWh of additional generation using Pareto-optimal portfolios discovered175 via this planning approach. Similarly, 10 MtCO2-eq/y fewer emissions could be produced for176 the same generation while reducing average river fragmentation by at least 5%.177 However, the potential for impact reduction promised by strategic dam planning does not yet178 consider the economic realities of energy system constraints and the competition from other179 power sources. For this reason, we integrate the two approaches. First, we select the subset180 of Pareto-optimal dam portfolios that outperform the best case achieved using energy system181 planning in all three objectives: emissions, ecosystem impacts, and generation. Using this cri-182 terion, the selected dam portfolios provide at least the same mean annual generation of the183 SSP1-2.6 expansion scenario while reducing by at least 10% the value of greenhouse gas emis-184 sions and fluvial ecosystems fragmentation (in both cases associated with SSP4-6.0). Second,185 we run the energy system model to find the cost-optimal hydropower expansion constrained to186 develop only dams belonging to the selected Pareto-optimal portfolios.187 Integrating environmental and climatic objectives into energy system planning substantially188 reduces the impacts of cost-optimal hydropower expansion. When the hydropower expansion189 in the energy system model is constrained to the most e”cient dams, the average river frag-190 mentation index is only slightly higher than current levels, with an increase of at most 1%, i.e.,191 -94% of the additional impacts of the full expansion scenario. At the same time, the increase in192 hydropower emissions is limited to no more than 5 MtCO2-eq/y resulting in -80% of additional193 impacts compared to the full expansion scenario. Additionally, we run the energy system model194 constrained to develop dams that are among the ones reported in any selected Pareto-optimal195 portfolio. In this case (reported in Fig. S2), compared to the present conditions, the river frag-196 mentation index and hydropower emissions can increase by at most 3% and 8.5 MtCO2-eq/y,197 respectively. This would translate to -82% and -66% additional impacts compared to the full198 expansion scenario for river fragmentation and hydropower emissions, respectively.199 6 Exploring alternative dam portfolios200 The hydropower projects selected by each approach (energy system planning, strategic dam201 planning, and integration) are reported in Extended Data Figure 1. For completeness, we report202 results when considering all the dams in selected Pareto-optimal dam portfolios in Fig. S3.203 Energy system planning consistently selects 62 projects, and 31 more are selected in at least204 one scenario. On the other hand, among the selected Pareto-optimal portfolios, 146 projects are205 always selected, and 52 projects are selected in at least one scenario. This shows an important206 di!erence between the approaches. Strategic dam planning only considers installed hydropower207 capacity without consideration of energy systems needs. On the other hand, energy system208 planning requires lower hydropower expansion than what was previously anticipated by water209 resources planners as it takes advantage of alternative cost-competitive power sources.210 In addition, energy system planning tends to spread hydropower expansion across a larger211 area, while strategic dam planning concentrates dams on particular rivers. For example, more212 projects tend to be selected in the Congo, Zambezi, and upper Nile River Basin according213 to the strategic dam planning approach. This is because the energy system model considers214 transmission constraints and costs. The strategic dam planning approach instead prioritizes215 areas where adding more dams does not substantially increase existing river fragmentation,216 notably upstream of existing dams, and greenhouse gas emissions (e.g., in East Africa).217 In Fig. 3a), we report the number and total capacity of projects using a Venn diagram to218 examine relationships between the di!erent approaches. The selection of projects resulting219 after integration overlaps more with the one of energy system planning (63 out of 93 projects220 or 45 out of 60 GW) and less with selected Pareto-optimal portfolios (70 out of 146 projects,221 or 47 out of 65 GW). Focusing on the foregone projects, also shown in Fig. 3b), it can be222 seen that constraints associated with the energy system reduce substantially the number of223 projects selected by the strategic dam planning approach. Similar results are observed when224 the integration is performed constraining the energy system model to all the dams belonging225 to selected Pareto-optimal portfolios as reported in Fig. S4. This underscores the importance226 of considering alternative power sources to provide electricity to meet the demand. While the227 projects selected by strategic dam planning and then foregone after integration are distributed228 all over the continent, 16 out of 28 projects foregone after being selected by the energy system229 planning approach are located in West Africa. These few projects are of particular interest230 because, by avoiding them, impacts on greenhouse gas emissions and river fragmentation can231 be drastically reduced, mitigating the concerns associated with unintended consequences of232 hydropower expansion.233 7 The costs of including environmental and climatic objectives234 In Fig. 4 we report the change in installed capacity when constraining the energy system model235 to develop only dams among the ones belonging to all selected Pareto-optimal portfolios. We236 examine also the capacity changes when considering all dams in the selected Pareto-optimal237 portfolios (Fig. S5) and we report the di!erences in generation in Fig. S6. As hydropower’s238 role declines in terms of capacity and generation, other sources fill the gaps. For SSP4-6.0,239 in the short term and also in the long term, fossil fuels are important to guarantee demand240 satisfaction, but wind and solar are important as well. Yet, especially for SSP1-2.6, most of241 the power capacity additions are solar and wind, while in terms of generation, nuclear and242 geothermal also substantially contribute to closing the gap. Considering the changes in each243 river basin reported in Fig. 5, after integrating environmental and climatic objectives in energy244 system planning, the capacity selected in smaller river basins is reduced by around 50% in all245 scenarios. To a smaller extent, this also occurs in the Nile, Niger and Zambezi river basins.246 Yet, the cost-optimal capacity proposed by the energy system planning approach in the Congo247 River basin (including the construction of the Inga 3 project) is not sensitive to environmental248 and climatic objectives. When considering a less stringent constraint for integration, changes249 in additional hydropower capacity per basin are smaller, as Reported in Fig. S7. Yet, Niger,250 Zambezi, and in particular smaller river basins see the largest reduction.251 Given the changes in installed capacity and hydropower projects’ selection, we estimate the252 economic impacts on total discounted costs and electricity prices. If the changes in costs or253 prices are not substantial, the integration of energy system modelling and strategic dam plan-254 ning might provide a more desirable infrastructure plan for policymakers. Indeed, near-optimal255 solutions in energy system modelling can unveil promising system design alternatives40,51.We256 report in Table 1 the percentage increase in mean electricity prices and total discounted costs.257 Table S1 provides the change in absolute values, also reported in Fig. S8 and Fig. S9. The258 increase in mean electricity prices is at most 1.4% while the increase in total discounted costs259 is always below 0.2% and we can consider them to be not substantial given the uncertainties260 involved with our modelling frameworks. The maximum di!erence of total discounted costs,261 obtained under SSP1-2.6 constrained to projects always part of the selected Pareto-optimal262 portfolios, is around 5 billion USD in net present value. At current costs4,43, this is equivalent263 to the capital investment required to build a 3 GW solar project, a negligible amount given the264 scale of the transformation required for the transition to a low-carbon energy system. Similarly,265 compared to the annual 190 billion USD investment in clean energy in Africa suggested by the266 International Energy Agency3, the additional cost of developing hydropower sustainably and267 strategically seems negligible.268 8 Discussion269 Energy system models consider hydropower as a cheap and carbon-neutral power source that270 should be developed where it is technically and economically feasible52. For this reason, more271 than 300 hydropower projects are under consideration in the African continent. Yet, envi-272 ronmental and climatic concerns coupled with increased cost-competitiveness of other power273 sources require a multi-objective planning approach. In particular, we examine two research274 questions: i) What are the impacts of African hydropower expansion in terms of emissions and275 fluvial ecosystem fragmentation? ii) What is the cost of reducing these impacts? We show that276 cost-optimal hydropower development increases hydropower emissions by 15-20 MtCO2/y and277 river fragmentation across the continent by 5-10%. By integrating energy system modelling278 and strategic dam planning, these impacts which represent already a reduction of 20-40% in279 hydropower emissions and 44-75% in river fragmentation impacts associated with the full ex-280 pansion of hydropower can be reduced by at least 50% and up to 75% and 86%, for hydropower281 emissions and river ecosystem fragmentation respectively. These climate and environmental282 benefits come with little impact on the economic performance of the energy system as we283 report a maximum increase of 1.3% in electricity prices and 0.2% in total discounted costs.284 In this work, we connect energy system planning with strategic dam planning to integrate285 techno-economic constraints and competition with other power sources with environmental and286 climatic consequences of hydropower using multi-objective optimization. Notwithstanding the287 uncertainty associated with the estimation of greenhouse gas emissions from reservoirs, these288 emissions can be substantial and would comprise part of the carbon budget associated with a289 particular warming target. Additional ecologic constraints might be imposed by, e.g., targets290 for riparian conservation and biodiversity. Thus, it is a major limitation that energy systems291 models are not aware of local, regional, or global impacts of hydropower generation; while292 approaches to reduce environmental impacts of hydropower do not, or only rudimentarily35,293 integrate energy systems considerations53.294 With our work, we reveal the substantial potential of integrating energy system modelling and295 strategic dam planning to reconcile energy, climate, and rivers. Additionally, we show that these296 benefits come only with a minor penalty in terms of costs. This allows us to maintain a balanced297 approach that guarantees meeting energy demand at low cost and developing hydropower where298 new infrastructure does not substantially degrade the value of ecosystem services or produce a299 long-term source of greenhouse gas emissions.300 While our results encourage the integration of strategic dam planning with energy system301 modelling, many of the assumptions made for this study should be further explored and refined.302 For what concerns the emissions from reservoirs, more data could reduce the uncertainty in303 greenhouse gas fluxes estimate. Second, energy system models including multiple greenhouse304 gases in the style of integrated assessment models could be used to avoid the need to convert305 emissions into carbon dioxide equivalent and to allow emissions from the hydropower sector to306 feed back into the other energy sectors for a full readjustment of the energy system. Third, we307 adopted river fragmentation across the continent as a proxy of ecosystem services but a large308 9 Regarding the greenhouse gas emissions objective, we sum the emissions for each project i520 GHGi, depending on its construction decision ui.Theemissionsforeachprojectiare de-521 scribed in detail in the section above.522 Depending on the decision variable of each project ui, the values of average annual generation523 GENiare summed to compute the value of the generation objective. To compute the value of524 the average generation, we compute the annual generation under the three SSP scenarios con-525 sidered. For each scenario, we compute the average of the monthly capacity factors CFSSP,m,i 526 obtained under the full expansion from the African Hydropower Atlas and then multiply it by527 the nominal capacity of the project i(CAPi)andthenumberofhoursinayeartopassfrom528 capacity to generation units. Finally, we find the average annual generation as the average of529 the annual generations under the number of SSP scenarios considered NSSP . Our approach ne-530 glects the cascading e!ects resulting from the di!erent dam configurations and the importance531 of reservoir operating policies, i.e., we do not consider the e!ects that upstream dams might532 have on the generation at downstream dam sites. We adopt this approach to keep the opti-533 mization problem tractable given the large number of hydropower projects considered (628).534 However, we estimate the errors stemming from this assumption using a network simulation535 model of the Zambezi River basin, one of the four largest African rivers. We obtain that the536 mean error estimation in the capacity factors due to network configurations is 4% while alterna-537 tive dam operations could influence capacity factors by 5.5%. These experiments are described538 in detail in the Supplementary Material (Supplementary Note 1).539 We solve the multi-objective optimization problem using Borg-MOEA79, an auto-adaptive540 many-objective evolutionary algorithm widely adopted in water resources engineering. Notwith-541 standing the full number of potential combinations (2330 ↔10100), the optimization converges542 in 106function evaluations as shown in Fig. S16. For our results, we used the output of a543 single seed. However, we provide evidence that the approximate set of the seed we used covers544 a fraction equal to 0.99 of the hypervolume of the reference set obtained running five seeds for545 the same number of iterations. We compare the two sets of solutions in Fig. S17.546 Constraining capacity expansion with selected dam portfolios To derive the dam547 portfolios for integration with the energy system model we select the subset of Pareto-optimal548 solutions that: a) provide more generation than all the hydropower expansion plans generated549 by the energy system model, b) reduce greenhouse gas emissions by 10% compared to the550 least emitting capacity expansion plan derived from the energy system, and c) reduce the river551 fragmentation index by 10% compared to the lowest value observed among the hydropower ca-552 pacity expansion plans optimized in the energy system. The selected dam portfolios are used to553 constrain the energy system model in two ways. Indeed, we run the three scenarios constrained554 to the hydropower projects that are part of the selected Pareto-optimal dam portfolios or to all555 dams belonging to the Pareto-optimal dam portfolios. By doing so, we examine the potential556 for impact reduction at di!erent costs associated with the integration of strategic dam planning557 and energy system planning.558 16 Data Availability The OSeMOSYS-TEMBA energy system model data is available on Zen-559 odo (https://zenodo.org/record/3521841) as its modified version including power plant560 data for more than 600 hydropower projects from the African Hydropower Atlas (https:561 //zenodo.org/record/7931050). The African Hydropower Atlas contains technical informa-562 tion about the existing and future hydropower projects and is available online (https://www.563 hydroshare.org/resource/5e8ebdc3bfd24207852539ecf219d915/). The GRAND database564 containing information about existing dams worldwide is available thanks to the Global Dam565 Watch project (https://www.globaldamwatch.org/grand). Emission Factors are available566 from the National Renewable Energy Laboratory (https://data.nrel.gov/submissions/567 171). The Shared Socioeconomic Pathways report socioeconomic projections using Integrated568 Assessment Models and are available online (https://tntcat.iiasa.ac.at/SspDb/dsd?Action=569 htmlpage&page=welcome). Figs. 3, S3, S4, and Extended Data Fig. 1 incorporate data from570 the HydroSHEDS version 1 database which is ©World Wildlife Fund, Inc. (2006-2022) and571 has been used herein under license. WWF has not evaluated the data as altered and incor-572 porated within these figures, and therefore gives no warranty regarding its accuracy, com-573 pleteness, currency or suitability for any particular purpose. Portions of the HydroSHEDS574 v1 database incorporate data which are the intellectual property rights of ©USGS (2006-575 2008), NASA (2000-2005), ESRI (1992-1998), CIAT (2004-2006), UNEP-WCMC (1993), WWF576 (2004), Commonwealth of Australia (2007), and Her Royal Majesty and the British Crown and577 are used under license. The HydroSHEDS v1 database and more information are available at578 https://www.hydrosheds.org70.579 Code Availability All the data and processing scripts to reproduce the results and figures580 are available on Zenodo (https://doi.org/10.5281/zenodo.8360437).581 Acknowledgments A.Car. acknowledges support from the European Union’s Horizon 2020582 research and innovation programme GEOCEP under the Marie Sklodowska-Curie grant agree-583 ment No. 870245. A.Car. was partially funded by the European Union’s Horizon 2020 Research584 and Innovation Actions programme under the project SOS-WATER (Grant Agreement ID:585 101059264). A.Cas. was partially funded by the European Union’s Horizon 2020 research and586 innovation programme under the GoNEXUS project (Grant agreement ID: 101003722).587 Author Contributions Statement Conceptualization: A.Car., R.S., A.Cas. Funding ac-588 quisition: A.Cas. Data curation: A.Car., A.Cl. Formal analyses: A.Car., A.Cl. Methodology:589 A.Car., R.S., A.Cl., A.Cas. Visualization: A.Car., R.S., A.Cl. Writing (original draft prepara-590 tion): A.Car. Writing (review and editing): A.Car., R.S., A.Cl., A.Cas.591 Competing Interests Statement The authors declare no Competing Interests.592 17 SSP1-2.6 SSP4-6.0 SSP5-8.5 Increase in mean electricity prices [%] 1.4 (0.7) 0.5 (0.8) 0.4 (0.1) Increase in total discounted costs [%] 0.2 (0.1) 0.2 (0.1) 0.2 (0.1) Table 1: Electricity prices and total discounted costs increase associated with integrating environmental and climatic objectives. Percentage increase of mean electricity and total discounted costs when moving from cost-optimal to constrained energy system development. Values reported are for the energy system development constrained to projects that are always in the selected subset of Pareto-optimal portfolios. Values in parentheses are for the projects in at least one of the selected Pareto-optimal portfolios. 18 Figure 1: Impacts of cost-optimal hydropower expansion. a Additional hydropower capacity installed according to energy system planning. b,c River fragmentation index (panel b) and annual carbon dioxide equivalent hydropower emissions (panel c). The dotted line and the dashed report the capacity, river fragmentation index, and hydropower emissions for the existing and full expansion scenarios respectively. 19 Figure 2: Multi-objective integration of energy system optimization and strategic dam planning approach. The values of average annual generation, annual carbon dioxide equivalent hydropower emissions, and river fragmentation index are reported for the dam portfolios analyzed. First, we analyse emissions, hydropower generation, and river fragmentation for three portfolios of dams generated by the energy systems model, which does not consider those broader environmental impacts of dams. Results are shown for three di!erent SSPs (box i). Second, we perform an independent optimization to derive dam portfolios that minimize conflicts between generation and environmental objectives, without considering energy systems aspects. We then select a subset of these dam portfolios (box ii) and we constrain the energy systems optimization to dams included in those scenarios. As a result, we find dam portfolios (box iii) that optimize environmental objectives and are compatible with the needs of the continental energy system by 2050. 20 Figure 3: Analysis of foregone hydropower projects. a The sets of projects selected under the three approaches are shown using a Venn Diagram to examine intersecting and complementary sets. For each set, we report the number of projects and their total total capacity (in GW units) between parentheses. bThe location of these projects is reported in the map. The color classification reports the approach that would select them. 21 Figure 4: Change in installed capacity after the integration of environmental and climatic objectives. Di!erences in installed capacity between energy system planning and integration for each scenario considered. a-c Di!erences between energy system planning and energy system planning constrained to always Pareto-optimal dams under SSP1-2.6 (a), SSP46.0 (b) and SSP5-8.5 (c). 22 Figure 5: Changes in installed hydropower capacity by basin for integration of environmental and climatic objectives. a-c New installed hydropower capacity in the four major river basins and aggregated for smaller river basins for SSP1-2.6 (a), SSP4-6.0 (b), and SSP5-8.5 (c) under the energy system planning approach and after integration of energy system planning and strategic dam planning for each scenario considered. 23 References593 [1] Kikstra, J. S., Mastrucci, A., Min, J., Riahi, K. & Rao, N. D. Decent living gaps and594 energy needs around the world. Environmental Research Letters 16, 095006 (2021).595 [2] Akintande, O. J., Olubusoye, O. E., Adenikinju, A. F. & Olanrewaju, B. T. Modeling596 the determinants of renewable energy consumption: Evidence from the five most populous597 nations in Africa. Energy 206, 117992 (2020).598 [3] IEA. Africa Energy Outlook 2022: World Energy Outlook Special Report (2022).599 [4] IRENA. Renewable Power Generation Costs in 2022 (2023).600 [5] Zarfl, C., Lumsdon, A. E., Berlekamp, J., Tydecks, L. & Tockner, K. A global boom in601 hydropower dam construction. Aquatic Sciences 77, 161–170 (2015).602 [6] Gernaat, D. E., Bogaart, P. W., Vuuren, D. P. v., Biemans, H. & Niessink, R. High-603 resolution assessment of global technical and economic hydropower potential. Nature En-604 ergy 2, 821–828 (2017).605 [7] Sterl, S. et al. A spatiotemporal atlas of hydropower in Africa for energy modelling pur-606 poses. Open Research Europe 1, 29 (2022).607 [8] Llamosas, C. & Sovacool, B. K. The future of hydropower? A systematic review of the608 drivers, benefits and governance dynamics of transboundary dams. Renewable and Sus-609 tainable Energy Reviews 137, 110495 (2021).610 [9] Grill, G. et al. Mapping the world’s free-flowing rivers. Nature 569, 215–221 (2019).611 [10] Anderson, E. P. et al. Fragmentation of Andes-to-Amazon connectivity by hydropower612 dams. Science Advances 4, eaao1642 (2018).613 [11] Zarfl, C. et al. Future large hydropower dams impact global freshwater megafauna. Sci-614 entific Reports 9, 18531 (2019).615 [12] Barbarossa, V. et al. Impacts of current and future large dams on the geographic range616 connectivity of freshwater fish worldwide. Proceedings of the National Academy of Sciences617 117, 3648–3655 (2020).618 [13] Dias, M. S. et al. Anthropogenic stressors and riverine fish extinctions. Ecological indicators619 79, 37–46 (2017).620 [14] Tickner, D. et al. Bending the curve of global freshwater biodiversity loss: an emergency621 recovery plan. BioScience 70, 330–342 (2020).622 [15] Schmitt, R. J., Bizzi, S., Castelletti, A. & Kondolf, G. Improved trade-o!s of hydropower623 and sand connectivity by strategic dam planning in the Mekong. Nature Sustainability 1,624 96–104 (2018).625 24 [16] Schmitt, R. J., Bizzi, S., Castelletti, A., Opperman, J. & Kondolf, G. M. Planning dam626 portfolios for low sediment trapping shows limits for sustainable hydropower in the Mekong.627 Science Advances 5(2019).628 [17] Schmitt, R. J. et al. Strategic basin and delta planning increases the resilience of the629 Mekong Delta under future uncertainty. Proceedings of the National Academy of Sciences630 118, e2026127118 (2021).631 [18] Kondolf, G. et al. Save the Mekong Delta from drowning. Science 376, 583–585 (2022).632 [19] Winemiller, K. O. et al. Balancing hydropower and biodiversity in the Amazon, Congo,633 and Mekong. Science 351, 128–129 (2016).634 [20] Chowdhury, A. K. et al. Hydropower expansion in eco-sensitive river basins under global635 energy-economic change. Nature Sustainability 7, 213–222 (2024).636 [21] Hertwich, E. G. Addressing biogenic greenhouse gas emissions from hydropower in LCA.637 Environmental science & technology 47, 9604–9611 (2013).638 [22] Deemer, B. R. et al. Greenhouse gas emissions from reservoir water surfaces: a new global639 synthesis. BioScience 66, 949–964 (2016).640 [23] Prairie, Y. T. et al. Greenhouse gas emissions from freshwater reservoirs: what does the641 atmosphere see? Ecosystems 21, 1058–1071 (2018).642 [24] Calamita, E. et al. Unaccounted CO2 leaks downstream of a large tropical hydroelectric643 reservoir. Proceedings of the National Academy of Sciences 118, e2026004118 (2021).644 [25] Harrison, J. A., Prairie, Y. T., Mercier-Blais, S. & Soued, C. Year-2020 global distribution645 and pathways of reservoir methane and carbon dioxide emissions according to the green-646 house gas from reservoirs (G-res) model. Global Biogeochemical Cycles 35, e2020GB006888647 (2021).648 [26] Soued, C., Harrison, J. A., Mercier-Blais, S. & Prairie, Y. T. Reservoir CO2 and CH4649 emissions and their climate impact over the period 1900–2060. Nature Geoscience 15,650 700–705 (2022).651 [27] Ou, Y. et al. Role of non-CO2 greenhouse gas emissions in limiting global warming. One652 Earth 5, 1312–1315 (2022).653 [28] Haegel, N. M. et al. Terawatt-scale photovoltaics: Transform global energy. Science 364,654 836–838 (2019).655 [29] Veers, P. et al. Grand challenges in the science of wind energy. Science 366, eaau2027656 (2019).657 25 Contents Supplementary Notes 1 - 2 Supplementary Figures 1 - 25 Supplementary Table 1 Supplementary References 2 Supplementary Notes Supplementary Note 1: Estimating the errors in capacity factors deriving from neglecting the cascading e↵ect of di↵erent network configurations Our study investigates how to incorporate considerations regarding the environmental and climatic impacts of hydropower within energy systems planning. Energy system modeling relies on linear programming. Therefore, reservoir operation and the dependence of capacity factors on upstream planning decisions (and thus the detailed generation for each portfolio) cannot be integrated. This is because both of these components, crucial in the context of a water systems simulation model, necessarily introduce non-linearities that cannot be treated within the current energy system modeling framework1,2. Instead of explicitly embedding a multitude of network dam simulation models in the energy system model, we consider each of the 628 existing and future hydropower projects using capacity factors obtained from a distributed hydrological model assuming the full expansion configuraration. Thus, we do not consider how upstream dam operation might impact the hydropower generation of downstream dams in each scenario. However, such an integration would require running a network simulation model for all the river basins and each dam portfolio. Such an integration is incompatible with the mixed integer linear programming (MILP) problem formulation of the energy systems optimization (because cascading e↵ects of upstream dams on downstream dams are typically highly non-linear). Here, we provide evidence that the error introduced by this implementation might be small and showcase how cascading dam operation can be handled for future more in-depth studies of individual basins. Given the nonlinearities discussed above, hydropower in energy system models is to our knowledge always represented by using deterministic capacity factors or generation profiles that can be obtained from the inflow assuming stylized reservoir operating rules3–13. Another approach is to optimize hydropower decisions within the energy system model, often assuming that the reservoir head is fixed, to avoid non-linearities2, and aggregating hydropower reservoirs to reduce computational burden5. As an example, in14, a network simulation is connected to a power system model to examine hydropower planning and operations in China. An iterative approach is designed to update the hydraulic head of each hydropower plant which is kept fixed during each optimization. However, no proof of convergence of this mechanism is provided. In addition to that, it is not clear if hydropower expansion is modelled and how projects are selected during the optimization process. They focus only on the electricity sector using a lower technological detail (only coal and nuclear as dispatchable generators), lower temporal resolution (reference days only for selected years), and coarse spatial detail (5 nodes). 3 In15, a network simulation model is used to represent high-resolution reservoir operating strategies in conjunction with hourly power system modelling. This study considers four reservoirs (existing and in construction) but there is no optimization of hydropower expansion. In16,a network dam simulation is integrated with a power system model, yet no capacity expansion is considered. In dam network simulation models, reservoir operations play a significant role. Indeed, the multipurpose nature of hydropower projects often involves selecting a compromise reservoir operating policy. Depending on the selected tradeo↵,thereleasedecisionsandtheplant’spower output can substantially change17,18, ultimately a↵ecting the capacity factors of the power plants. As a consequence, even though the generation of each hydropower plant is a↵ected by the changes in the reservoir network topology, the assumptions made on reservoir operations for possibly many objectives (hydropower, irrigation, flood control) would introduce additional uncertainty. To estimate the error in capacity factors associated with the omission of cascading e↵ects of dams, we performed a network simulation of the Zambezi River basin water system to derive the capacity factors of each large hydropower plant (100 MW) across di↵erent network configurations and operating policies. This experiment highlights the sensitivity of the capacity factors (for each dam) to the configuration and operation of upstream dams. We chose the Zambezi River basin because it is home to some of the continent’s largest reservoirs. Thus, the sensitivity of capacity factors in this basin would be stronger than in many other basins. We think that this experiment is useful to support the common practice of using fixed capacity factors in energy systems modeling. This numerical experiment is based on a previously published water resources system model17,19. The Zambezi River Basin is not representative of the diversity of conditions across river basins in the African continent. However, the high-resolution model available in this case provides a first ground to test the validity of our assumptions and the numerical consequences in terms of impacts on the obtained results. Similar detailed simulation studies could be implemented in the future for other river basins. From those studies, we could learn how di↵erent dam portfolios would impact the capacity factors of individual dams in those basins. In our network simulation of the Zambezi River basin, we consider 3 new hydropower projects, namely Devil’s Gorge (1240 MW), Batoka Gorge (1600 MW), and Mphanda Nkuwa (1300 MW), 4 existing projects, namely Kariba (1830 MW), Cahora Bassa (2075 MW), Itezhi Tezhi (120 MW), and Kafue Gorge Upper (900 MW), and 1 project under construction, Kafue Gorge 4 Lower (750 MW). Given three potential new dams, we examine 8 (23) potential dam configurations under di↵erent assumptions regarding the operation of both existing and future reservoirs. In particular, we consider four reservoir operating policies that maximize the objectives of either (1) hydropower, (2) environment, (3) irrigation, or (4) try to strike a compromise. For more details on the water system model, please refer to17. Figures 18 - 25 show the capacity factors of each hydropower reservoir. The four panels in each figure the 4 potential reservoir operations strategies while the line colors are used to indicate one of the 8 network configurations examined and listed in the legend. In the legend, we report the number of reservoirs and we list the new projects included between parentheses. We use a two-letter code to refer to the new projects (MN: Mphanda Nkuwa, BG: Batoka Gorge, DG: Devil’s Gorge). While the solid line describes the mean capacity factors over 20 years (1986-2005), the area shows maximum and minimum monthly capacity factors to show the e↵ect of natural variability. For each power plant, the capacity factor varies depending on the dam configuration and the 4 operating policies. Across network configurations, the coefficient of variation of monthly capacity factor is on average 0.1, or equivalently, we measure a signal-to-noise ratio (average divided by standard deviation) equal to 10. Capacity factors are on average contained in a range of variation of 0.08. As a consequence, if we assume to be in the center of this range, our assumption of constant capacity factors across network configurations results in an average error of 0.04. Across policies, the coefficient of variation is on average 0.12 (signal-to-noise ratio equal to 8.3), and capacity factors are contained within an interval of 0.11 on average. These results show the lower influence of networking e↵ects on capacity factors when compared to di↵erent reservoir operating policies and natural variability. The code of the HBV models is available in the open-source repository https://doi.org/10. 5281/zenodo.5726941. The Zambezi River Basin reservoir operations simulation model contains sensitive hydrologic data, along with hydropower plant characteristics from the Zambezi River Authority (ZRA), Zambia Electricity Supply Corporation (ZESCO) and Hidroel´ectrica de Cahora Bassa (HCB), thus it cannot be made public. The historical hydrologic data on the Zambezi River basin are from the Zambezi River Authority (ZRA) and were collected during the DAFNE project (http://dafne-project.eu/). They are protected by a nondisclosure agreement with ZRA. However, the climate model data used for the temperature and precipitation projections are freely available at the following website: http://www.csag.uct.ac.za/cordex-africa/. 5 Supplementary Note 2: Optimization problem formulation for the energy system model OSeMOSYS-TEMBA nomenclature Below, we report tables reporting the nomenclature adopted for describing the equations of the OSeMOSYS-TEMBA model. Symbol Set rREGION t TECHNOLOGY f FUEL mMODEOFOPERATION y YEAR lTIMESLICE eEMISSION Symbol Variable AFC AnnualFixedCost AVC AnnualVariableCost ACC AnnualCapitalCost DSV DiscountedSalvageValue DTEP DiscountedTechnologyEmissionsPenalty NC NewCapacity ROA RateOfActivity SV SalvageValue 6 Symbol Parameter ⇢DiscountRate FC FixedCost OL OperationalLife RC ResidualCapacity VC VariableCost YS YearSplit CC CapitalCost CF CapacityFactor CTAU CapacityToActivityunit AF AvailabilityFactor OAR OutputActivityRatio SAD SpecifiedAnnualDemand SDP SpecifiedDemandProfile IAR InputActivityRatio AAD AccumulatedAnnualDemand TAMaC TotalAnnualMaxCapacity TAMiC TotalAnnualMinCapacity TTAAUL TotalTechnologyAnnualActivityUpperLimit TTAALL TotalTechnologyAnnualActivityLowerLimit TTMPAUL TotalTechnologyModelPeriodActivityUpperLimit TTMPALL TotalTechnologyModelPeriodActivityLowerLimit EAR EmissionActivityRatio EP EmissionsPenalty AEE AnnualExogenousEmission AEL AnnualEmissionLimit MPEL ModelPeriodEmissionLimit Here is reported a list of the optional variables and parameters used to include the African Hydropower Atlas in the OSeMOSYS-TEMBA model, and for the robust scenario analysis. Symbol Description COTU CapacityOfOneTechnologyUnit (parameter) NNTU NumberOfNewTechnologyUnits (integer variable) 7 OSeMOSYS-TEMBA model equations The full name of the sets, variables, and parameters used in the model is reported in the nomenclature. min uX r,t,y "AF Cr,t,y +AV Cr,t,y (1 + ⇢)(yy0+0.5) +ACCr,t,y (1 + ⇢)(yy0)DSVr,t,y DTEPr,t,y#(1) s.t. AFCr,t,y =FCr,t,y ⇤X yy:yyy<OLr,t && yyy0⇥NCr,t,y⇤+RCr,t,y (2) AV Cr,t,y =X mX l VC r,t,m,y ⇤ROAr,l,t,m,y ⇤YS l,y (3) ACCr,t,y =CCr,t,y ⇤NCr,t,y (4) DSVr,t,y =SVr,t,y (1 + ⇢)yendy0(5) SVr,t,y =8 < : 0,y+OLr,t 1yend CCr,t,y ⇤NCr,t,y ⇤(1+⇢)(yendy) (1+⇢)(OLr,t 1), else (6) u="NCr,t,y,ROA r,l,t,m,y #(7) OSeMOSYS-TEMBA model formulates a linear programming problem whose objective function to minimize is the sum of various annual components of cost summed over the years, the technologies and the regions considered. As reported in (1), the total costs are composed by: annual fixed costs, described in (2), and annual variable costs, described in (3), discounted at mid-year as these costs occur during all over the year; additionally we have also annual capital costs, discounted at the beginning of each year and described in (4), discounted salvage value, described by (5) and (6), and discounted emissions penalty by technology, whose computation is described later in the text in (20). Finally, in (7), the decision variables with respect to which the total costs are minimized are reported. These are the new capacity to be installed in year yfor technology tin region r (NCr,t,y) and the rate of activity in time-slice l(i.e. time step associated with season and day night conditions) during year yfor technology tin region rwith mode of operation m(for technologies that operate in multiple directions such as transmission lines, pumped-storage hydro) (ROAr,l,t,m,y). Additional constraints are imposed so that generation from each technology is constrained by the installed capacity of the technology in a specific year, its capacity factor and its availability factor, that take into account for planned maintenance of technologies. This is described in (8) 8 and (9). X m ROAr,l,t,m,y  (X yy:yyy<OLr,t && yyy0⇥NCr,t,y⇤+RCr,t,y)⇤CFr,t,l,y ⇤CT AUr,t (8) X mX l ROAr,l,t,m,y ⇤YS l,y  X l(X yy:yyy<OLr,t && yyy0⇥NCr,t,y⇤+RCr,t,y)⇤CFr,t,l,y ⇤AFr,t,y ⇤CT AUr,t (9) Energy balances are formulated at the time-slice level in (10) and at the annual level (11). In their simplest terms, these equations ensure that enough energy is generated to meet demand from other technologies and pre-specified final energy demands, defined at the annual or timeslice level. X mX t ROAr,l,t,m,y ⇤OARr,t,f,m,y ⇤YS l,y  SADr,f,y ⇤SDPr,f,l,y +X mX t ROAr,l,t,m,y ⇤IARr,t,f,m,y ⇤YS l,y (10) X mX tX l ROAr,l,t,m,y ⇤OARr,t,f,m,y ⇤YS l,y  X mX tX l ROAr,l,t,m,y ⇤IARr,t,f,m,y ⇤YS l,y +AADr,f,y (11) The following constraints, eqs. (12) to (15), ensure that the capacity and the new capacity installed remains between predefined maximum and minimum capacity and capacity investment. X yy:yyy<OLr,t && yyy0⇥NCr,t,y ⇤+RCr,t,y TAMaCr,t,y (12) X yy:yyy<OLr,t && yyy0⇥NCr,t,y ⇤+RCr,t,y TAMiCr,t,y (13) NCr,t,y T AMaCIr,t,y (14) NCr,t,y T AMiCIr,t,y (15) 9 Annual and whole horizon (or model period) activity limits are enforced for each technology using eqs. (16) to (19). X l ROAr,l,t,m,y ⇤YS l,y TT AAULr,t,y (16) X l ROAr,l,t,m,y ⇤YS l,y TT AALLr,t,y (17) X yX l ROAr,l,t,m,y ⇤YS l,y TT MPAULr,t,y (18) X yX l ROAr,l,t,m,y ⇤YS l,y TT MPALLr,t,y (19) The discounted emissions penalty for each technology are computed in (20), while annual and model period emission limits are constrained using (21) and (22). X eX lX m EARr,t,e,m,y ⇤ROAr,l,t,m,y ⇤YS l,y ⇤EPr,e,y ⇤1 (1 + ⇢)(yy0+0.5) =DTEPr,t,y (20) X lX mX t EARr,t,e,m,y ⇤ROAr,l,t,m,y ⇤YS l,y +AEEr,e,y AELr,e,y (21) X lX mX tX y EARr,t,e,m,y ⇤ROAr,l,t,m,y ⇤YS l,y +AEEr,e,y MPELr,e,y (22) 10 Additional constraints to include data from the African Hydropower Atlas in OSeMOSYS-TEMBA To enforce that the new capacity built for a specific hydropower project is aligned to the nominal capacity reported in the African Hydropower Atlas, we adopt a set of built-in variables, parameters and constraints available in the standard OSeMOSYS model framework. We use the optional variable NNTUr,t,y, defining how many new units of technology tare built in year yin region r, the optional parameter COTUr,t,y, describing the minimum amount of capacity that has to be added when building technology tin year yin region r, and we add the additional constraint that relates these to the decision variable NCr,t,y using (23). COTUr,t,y ⇤NNTUr,t,y =NCr,t,y (23) It should be noted that the variable NNTUr,t,y is defined as an integer variable and constrained to be 0 or 1, as building one technology unit for the hydropower project examined would result in its realization. Furthermore, the parameter T AMaC, set equal to the hydropower plant nominal capacity for each hydropower project, ensures that the project is built only once during the model period considered. As a result, we are now formulating a mixed-integer linear programming problem, whose computational complexity is notoriously higher than the one of a linear program. Solving the mixed integer linear programming problem We rely on the commercial solver CPLEX to solve the mixed integer linear programming problem formulated above. We build the linea programming model file using the software GLPK which reads the model objectives, constraints, parameters, sets, and variables from the AMPL script. The abstract model is then populated with the input data for each scenario from an Excel spreadsheet. We solve the problem using a server with 32 CPUs and 150 GB of RAM. With the above computational infrastructure, the time needed to solve this problem varies depending on the scenario examined between 18 and 48 hours. 11 Supplementary Figure 7 |Changes in installed hydropower capacity by basin for integration of environmental and climatic objectives after integration based on all dams in selected Paretooptimal portfolios. a-c New installed hydropower capacity in the four major river basins and aggregated for smaller river basins for SSP1-2.6 (a), SSP4-6.0 (b), and SSP5-8.5 (c) under the energy system planning approach and after integration of energy system planning and strategic dam planning for each scenario considered. 18 Supplementary Figure 8 |Mean electricity prices (averaged over seasons and countries) over the modeling period. Each panel examines di↵erent SSP scenarios and the di↵erent type of lines report the trajectory for energy system planning, integration using the intersection of dams in the selected Pareto-optimal dam portfolios, and integration using the union of the dams in the selected Pareto-optimal portfolios. 19 Supplementary Figure 9 |Total discounted costs of the energy system under the scenarios examined. The panels show the results across di↵erent SSP scenarios, while the bar in each panel shows the total discounted cost of each scenario. Annotated on each bar is the percentage change compared to the energy system planning scenario. 20 Supplementary Figure 10 |Strength of the correlation between di↵erent socioeconomic variables (GDP, GDP per capita, population) and electricity demand for continental African countries for which data is available (n=47). The data is now reported using a log-log scale. For each of the coupling examined between these variables, we report the R2to show the explanatory power of each variable with respect to indicators of electricity consumption. Additionally, we report also the p-value of the F-test (one-sided), i.e., the probability of the observed F-statistic value for each regression, which indicates the significance of the R2. The highest significance and the highest R2are observed between GDP and electricity consumption. 21 Supplementary Figure 11 |Clockwise from top left: log-log plots of the ratio of dam-toreach discharge values to dam generation capacity for snapping to reaches of 1) no minimum Strahler Order, 2) Strahler Order of 1, 3) Strahler Order of 2, 4) Strahler Order of 3. 22 Supplementary Figure 12 |Final log-log plot of the ratio of dam-to-reach discharge values to dam generation capacity for snapping to reaches using the piecewise snapping algorithm defined above. 23 Supplementary Figure 13 |An example of network simplification, where communicating reaches not separated by a dam are merged. 24 Supplementary Figure 14 |An example of a graph network representation of a river network. River reaches are represented as nodes and the flow of water as directed edges. 25 Supplementary Figure 15 |Comparison of di↵erent methods to estimate the surface area of the reservoirs based on data available in the GRAND Database for African dammed reservoirs (n=62). The multiple log-log regression based on upstream catchment area and reservoir volume yields the best results and it is selected for estimation of carbon emissions from reservoirs and improves estimation compared to previously adopted methods22. The R2and the p-value of the F-test (one-sided) representing the significance of the regression model are reported for the two models we build. The highest significance and explained variance are observed for the model which we use to estimate the surface areas when data is not available. 26 Supplementary Figure 16 |Convergence of the optimization algorithm. Improvements, archive size, restarts, and new improvements are reported against the number of function evaluations for five random seeds. The lines are used to describe the evolution of these indicators for each seed. 27 [14] Liu, Z. & He, X. Balancing-oriented hydropower operation makes the clean energy transition more a↵ordable and simultaneously boosts water security. Nature Water 1, 778–789 (2023). [15] Gonzalez, J. M. et al. Designing diversified renewable energy systems to balance multisector performance. Nature Sustainability 6, 415–427 (2023). [16] Stevanato, N., Rocco, M. V., Giuliani, M., Castelletti, A. & Colombo, E. Advancing the representation of reservoir hydropower in energy systems modelling: The case of Zambesi River Basin. Plos one 16, e0259876 (2021). [17] Arnold, W., Salazar, J. Z., Carlino, A., Giuliani, M. & Castelletti, A. Operations eclipse sequencing in multipurpose dam planning. Earth’s Future 11, e2022EF003186 (2023). [18] Tangi, M., Bizzi, S., Schmitt, R. & Castelletti, A. Balancing sediment connectivity and energy production via optimized reservoir sediment management strategies. Water Resources Research e2022WR034033 (2023). [19] Giuliani, M., Lamontagne, J. R., Hejazi, M. I., Reed, P. M. & Castelletti, A. Unintended consequences of climate change mitigation for African river basins. Nature Climate Change 12, 187–192 (2022). [20] Harrison, J. A., Prairie, Y. T., Mercier-Blais, S. & Soued, C. Year-2020 global distribution and pathways of reservoir methane and carbon dioxide emissions according to the greenhouse gas from reservoirs (G-res) model. Global Biogeochemical Cycles 35, e2020GB006888 (2021). [21] Soued, C., Harrison, J. A., Mercier-Blais, S. & Prairie, Y. T. Reservoir CO2 and CH4 emissions and their climate impact over the period 1900–2060. Nature Geoscience 15, 700–705 (2022). [22] Li, M. & He, N. Carbon intensity of global existing and future hydropower reservoirs. Renewable and Sustainable Energy Reviews 162, 112433 (2022). 34