Strategic siting and design of dams minimizes impacts on seasonal floodplain inundation

Dams and reservoirs aid economic development but also create significant negative impacts. Dams fragment rivers and reduce longitudinal connectivity on a network scale. However, dams may also alter discharge regimes and flood peaks, consequently reducing floodplain inundation and lateral channel floodplain connectivity, which impacts floodplain associated ecosystems. Strategic planning has emerged as a promising approach to find a balance between dam impacts and benefits. Yet, strategic planning has predominantly focused on longitudinal connectivity due to the difficulty of including the complex interactions between dam design and operations, hydrologic regime alteration, and the hydrodynamic processes controlling downstream flood extent. Here, we present how to reduce conflicts between hydropower development and loss of floodplain inundation extent by jointly optimizing siting and design of many dams in a data scarce basin. We deploy a coupled hydrological—hydraulic simulation model linked to a multiobjective optimization framework to find development options with the least trade-offs between power generation and downstream impacts on floodplains. Our results for the Pungwe Basin in Mozambique indicate that whilst portfolios of many small storage and run-of-river diversion hydropower plants might create less impacts on the downstream floodplains, installation of some large storage dams would be necessary to attain higher levels of hydropower generation.


Introduction
Hydropower is projected to play a significant role in economic development and decarbonization globally, especially in developing regions (Zarfl et al 2019). Additionally, dams and reservoirs provide water storage for other uses, e.g. irrigation and flood protection (Price and Probert 1997). However, dams pose one of the greatest threats to the physical functioning of rivers and the associated biodiversity and livelihoods (Beilfuss 2012, Moran et al 2018, Grill et al 2019, Zarfl et al 2019. An important environmental impact of hydropower is that dams interrupt the longitudinal connectivity of rivers and alter flow regimes, consequently reducing lateral connectivity between rivers and floodplains (Bunn and Arthington 2002, Grill et al 2019, Opperman and Galloway 2022. The effect is cumulative, with more dam storage causing greater flow alteration and consequent reduction in lateral river-floodplain connectivity (Poff and Hart 2002, Deitch et al 2013, Kibler and Tullos 2013, Poff et al 2016. The associated reduction in floodplain inundation extent and duration may disrupt ecosystem services or recession based agriculture that the floodplain may provide (Poff andHart 2002, Poff et al 2016). The extent of floodplain inundation is therefore an important indicator of the floodplains natural functioning and associated ecosystems and ecosystem value.
As our understanding of floods shifts from considering them a natural hazard to an integrated vision of both their impacts and their socio-ecologic benefits, the importance of preserving natural flood regimes for agriculture, biodiversity, and associated benefits such as fisheries has become apparent (Poff et al 1997, Price and Probert 1997, Kingsford 2000, Bunn and Arthington 2002, Lytle and Poff 2004, Beilfuss et al 2007, Richter et al 2010, Beilfuss 2012, Guiamel and Lee 2020. However, preservation of flood regimes may conflict with the power generation benefits afforded by bigger storage dams, which might be critical for renewable energy grids, and provide ancillary benefits such as for uses such as irrigation (Schmitt et al 2022). As millions of people globally rely on fisheries and floodplain agriculture (Richter et al 2010); and 472 million people are estimated to live alongside dam impacted river reaches (Richter et al 2010), including impacts of dams in decisions about individual dams and river basin development becomes increasingly important.
Environmental impacts of dams can be mitigated across scales, ranging from altering design and operations of single dams, to strategic site selection, a field that has received notable recent attention (Almeida et al 2022, Flecker et al 2022, Opperman et al 2023. Yet, many studies exploring multi-dam portfolio impacts on river flow regime (Piman et al 2016, Wang et al 2018, Shrestha et al 2020 have considered a limited ranges of alternative siting scenarios and lack the dynamic, non-linear hydrological and hydraulic process representation required to properly model impacts on floodplain inundation. Angarita et al (2018) integrates basic overbank flow routing and floodplain mass balance to simulate many (1000) siting scenarios but provides neither a full exploration of trade-off solutions, nor representation of the dynamic evolution of flow over a floodplain according to physics, nor opportunities to reduce conflicts by redesigning certain high-impact dams. Such effects have been demonstrated to be important for modelling floodplain inundation (Hunter et al 2008, Trigg et al 2009, Bates et al 2010a, Yamazaki et al 2011. Studies that fully integrate basin and floodplain hydrodynamics (Kuriqi et al 2020, Shin et al 2020, Trung et al 2020, Fleischmann et al 2021 have considered only limited siting scenarios. Conversely, studies that have used multi-objective optimisation for optimal dam siting (Ziv et al 2012, Schmitt et al 2018, design and operation (Geressu and Harou 2015), operation only (Suwal et al 2020), or all three aspects (Geressu et al 2020, Hurford et al 2020) lack hydrodynamic floodplain inundation modelling. To begin to fill these research gaps, this paper highlights opportunities to integrate the siting, design, and operation (SDO) of dam portfolios as an opportunity to alleviate trade-offs between dam benefits and their impacts on downstream floodplains.
In this study, we present the first simulationoptimization framework for dam SDO that integrates the outputs of a high-resolution floodplain inundation model into a multi-objective optimization framework for dam siting and design (figure 1). Thus, the optimization algorithm can choose between implementations of large storage and small storage dams (with reservoirs) or run of river diversion projects for each dam site. Operation rules are not optimized but they are modelled in accordance with the dam type for each site. Our study uses the Pungwe Basin in Mozambique and its ecologically sensitive floodplains as a case study. The results and methods presented herein fill a gap in representing dam impacts on downstream hydrology and floodplain hydraulics, while considering the cumulative implications of different dam designs on both environmental and energy performance from a strategic perspective. Our approach is applicable to data scarce basins, and thus most basins where any significant hydropower potential remains.

Case study
We develop the proposed SDO framework for the Pungwe Basin, spanning 31 151 km 2 (Alferes et al 2006), 95% of which is located in Mozambique and 5% in Zimbabwe (figure 2). The Pungwe Basin has both significant untapped hydropower potential, and an ecologically sensitive riparian corridor. For instance, in the Dinge Dinge Marshes located on the southern edge of Goronogosa National Park, the annual flood pulse generated from the Pungwe river and a large tributary, the Urema River, is critical to maintain green pastures for large ungulates (Beilfuss et al 2007, Desai et al 2019, Tinley 2020. The Pungwe Basin therefore faces conflicts between hydropower development and the integrity of floodplains of outstanding conservation and livelihood values typical for many African river basins, e.g. the Okavango, Zambezi, and Niger (Adams and Hughes 1986, Andersson et al 2006, Murray-Hudson et al 2006. Despite potential impacts, developing a large reservoir dam on the Pungwe River is a high priority for water management in central Mozambique, aimed at providing hydropower, irrigation, and mitigation of salt intrusion and floods in the lower Pungwe and its estuary (World Bank 2007). Six proposed large dam sites were identified in (Beilfuss et al 2007), including Bue Maria and Pavua (figure 2). The Pavua dam, with a proposed power capacity of 120 MW, was found to be economically and environmentally

Methods
Our approach consists of a hydrologic-hydraulic simulation model for the Upper Pungwe Basin, a 2D hydrodynamic floodplain model for the Dinge-Dinge Marshes, and a surrogate model that represents the numerically expensive 2D model as objective function within an optimization approach. As this modelling framework uses public, globally available datasets, it could be easily adapted to other rivers.

Upstream hydropower simulation model
Upstream hydrology is represented using a daily time step, 5 km resolution VIC model (Liang et al 1994) to calculate daily run-off time series (supplementary information S2 contains calibration and validation details). VIC does not include a routing routine, therefore we cascade its run-off predictions through the 1D-2D LISFLOOD-FP hydraulic solver which solves a simplified local inertial formulation of the shallow water equations (Bates et al 2010a), to create discharge time series at all locations along the river network (represented at 1 km resolution, see supplementary information S3 and S4). A simulation time period of 1985-1988 was used because this was the earliest time when satellite rainfall forcings are available to drive the VIC-LISFLOOD-FP model, and thus closest to the time period when the basin gauge record was available , see supplementary information S4). The first 2 years (October 1985-October 1987) constitute a model warm up period, with the core analysis undertaken in the final year (October 1987-October 1988. The year 1987-1988 was chosen because it represents an 'average year' in terms of its peak flow. The core analysis was limited to this relatively short period to minimise simulation time. See supplementary information S1 for more details.

Representation of storage dams in the river network
We assume three different possible dam designs: large and small dams with reservoir storage, and run-ofriver (RoR) dams without storage. We define a large dam as the largest possible structure at a location relative to the highest adjacent valley point, minus a freeboard of 20 m; and a small dam as a structure 50% of that height. Dam length is limited to 1000 m. Further details are presented in supplementary information S5. We represent large and small dams within LISFLOOD-FP as volumes physically defined by dam height (h), maximum dam volume (V), spillway width (b), spillway coefficient (cd), and spillway height (H). The maximum reservoir volume is derived from the dam height and the DEM (supplementary information S5). The rate of change in dam storage is calculated at each simulation timestep according to equation (1): where S is the volume of water stored (m 3 ), and I and Q H06 are inflows and releases (m 3 s −1 ) respectively. Releases are calculated using a generic operation rule commonly used to represent dam operation in global hydrologic modeling (Hanasaki et al 2006). Herein, we use the H06 rules (thus the notation Q H06 ), which adjusts releases from each dam every month based on irrigation water demand for that month and reservoir storage at the start of the hydrologic year. Specifically, the release Q H06,m,y in the m th month of the y th hydrologic year is defined in equation (2) below: where the k rls,y coefficient is proportional to the storage at the start of the hydrological year (S first,y ) divided by reservoir capacity (C) (precisely: k rls,y = S first,y 0.85C as in Hanasaki et al 2006); r ′ m,y is the provisional monthly release which is set to the regional irrigation demands; c is the ratio of dam capacity (C) to mean annual inflow volume (I mean ) and i are the daily inflows to the reservoir (use of daily inflows are a feature from Hanasaki et al 2008, but the rules are otherwise identical to, 2006).
For simplicity, equation (1) neglects reservoir evaporation, seepage and dead storage. Note that the H06 works to maintain an environmental flow of 50% at all times (Hanasaki et al 2006, Biemans et al 2011. supplementary information S6 contains further details on H06.

Calculation of power generation from storage dams
Power generation in storage dams is calculated from the turbined flows Q turbine . Q turbine is set equal to the release, Q H06 , but capped to the turbine capacity. In the absence of detailed designs, we set the turbine capacity to the mean flow rate in the undeveloped scenario, Q mean , assuming dams are designed to turbine a maximum of the average natural flow rate. Hydroelectric generation, E, is a function of the dams water level (assuming that the turbine(s) are located at the bottom of the structure, thus hydraulic head equals the time-varying dam water level relative to the bottom of the dam), the flow Q turbine , unit density of water, pg, and the turbine efficiency n, set to 0.7:

Representation of RoR plants
We represent RoR plants as diversion (offstream) types, diverting water off the river and through a headrace and/or penstock to generate hydropower further downstream, assuming no storage, and a constant head difference. Given the objective of maximising power generation, RoR are modelled here as diversions only, since diversions can generate greater hydraulic head, and therefore power generation, than onstream designs. Similarly to storage dams, the maximum turbine flow Q turbine is set to Q mean . An environmental flow is set at the 95th percentile of longterm average discharge over the 5 year baseline (no dam) simulation period, following common E-flow guidances (Penche 2004). The turbine flow used for power generation by the RoR at each timestep is therefore: Q turbine is then used with equation (5). Further RoR implementation details are given in supplementary information S5.

Downstream floodplain model
A 90 m resolution LISFLOOD-FP model was built to represent flood inundation in the Dinge Dinge Marshes using a pseudo 2D routing scheme, using common assumptions in flood modeling (Neal et al 2021) (supplementary information S7). The model receives an input flow hydrograph from the upstream hydropower simulation model described above, routes it through the wetland channels, and where water surface elevations exceed bank height, distributes the water over the floodplains as a function of overbank flow and elevation derived from the DEM according to the simplified St Venants equations described in Bates et al (2010a) and Neal et al (2012). This allows calculation of the inundated area at every timestep in response to a hydrograph altered by an upstream dam SDO combination. The maximum inundated area can then be identified and used as the environmental objective to be optimized.

Surrogate model
The high-resolution floodplain model just described takes approximately 3 h to simulate 1 year with a daily time step (or 9 h for the full time horizon). The optimization routine (next paragraph) evaluates floodplain inundation for many thousands of SDO combinations, making direct integration of the LISFLOOD-FP model into the optimisation routine computationally infeasible. This issue is addressed by creating a computationally-efficient surrogate model, applied similarly to Zischg et al (2018). Specifically, we generated 100 different SDO combinations, and for each, we generated a 3 year-long hydrograph using the upstream hydropower simulation model and calculated the associated daily flood extents using the LISFLOOD-FP. From this set of 100 hydrodynamic simulations, we built a polynomial regression relationship to predict A max (maximum inundated area) from Q p (river peak flow) and Q init (river flow 10 d before the peak). The polynomial regression relationship is described in supplementary information S8. Q p and Q init are readily calculated from the output hydrograph of the Upstream Hydropower Model, therefore estimation of A max with the surrogate model takes seconds (supplement information S7 and S8). The high-resolution floodplain model was used again to simulate selected Pareto Optimal portfolios and generate maps of lost flooded extent (see figure 4).

Optimisation model
The simulation model described above is coupled to the Non-Dominated-Sorting-Genetic-Algorithm (NSGAII) evolutionary algorithm (Deb et al 2002) to search for combinations of dam sites and designs which perform best for conflicting power generation and environmental objectives. This section details features of the optimisation problem definition. Details on the NSGAII algorithm can be found in supplementary information S8.

Objective functions
The optimisation model aims to maximise hydropower generation and floodplain inundation by changing dam SDO throughout the basin. For power generation, we consider two possible objectives: mean annual power generation, which characterizes total power generation from all dams over the simulation period (equation (9)), and firm power generation (defined as 90th percentile of daily hydropower generation) which indicates how much power can be reliably generated across all dams (equation (10)), where d represents the dth dam in a portfolio. Whilst power generation could be limited by hydropower demand, reliable records of regional power demand are absent, therefore power generation objectives are maximised without constraint.
To characterize the impact of SDO combinations on floodplain inundation, we consider two formulations of the objective. The first one is the change in downstream peak flow (equation (11)), which is the normalised version of the 7 d mean peak flow from the Indicators of Hydrological Alteration (IHA, Richter et al 1996), and can be calculated from the output of the upstream hydropower simulation model. This acts as a proxy of floodplain inundation. f downstream = Q 7 day peak, dams Q 7 day peak, no dam (11) The second formulation is the maximum inundated area (equation (12)), calculated using the (surrogate) downstream floodplain model.
While the first approach represents dam impact on floodplains only in terms of hydrology, the second approach also accounts for how those changes in hydrology translate to changes in flood inundation and thus the process driving floodplain productivity.

Decision variables
The location and type of dams or RoR plants in each portfolio are the decision variables that can be varied by the genetic algorithm. They are represented by a vector of 17 components, one for each of the 17 potential hydropower sites shown in figure 2. Each component assumes a discrete value from 0 to 3 representing a different hydropower design option at each site. A value of 0 represents the 'no build' decision. A value of 1 (or 2) represents the decision to build a large (or small) dam, and a value of 3 represents the decision to build a RoR plant.

Baseline simulation
Prior to running the simulation-optimisation, a baseline simulation with no dams is executed from October 1985 to October 1988, and propagated through the high-resolution Downstream Floodplain Model (LISFLOOD-FP). Simulated peak flow for the analysis period (October 1987-October 1988 was 512 m 3 s −1 , resulting in a maximum inundated area of 181 km 2 . These numbers represent the natural baseline conditions in terms of floodplain inundation. The baseline hydropower generation is zero.

Optimisation experiments
Three optimisation experiments are presented in this work (table 1) that differ in their definitions of the power and environmental objectives. Experiment 1 uses a hydrologic alteration indicator as environmental objective, which does not require floodplain modelling. Experiments 2 and 3 instead use downstream flood extent as the environmental objective, and thus require floodplain modelling (through the surrogate model). Experiments 2 and 3 differ in the formulation of the power objective. By comparing the results of experiment 1 and 3 we can assess the value of representing dam impacts on flood extents explicitly, rather than through a proxy of hydrologic alteration. Comparing experiment 2 and experiment 3 enables an understanding of the sensitivity of optimization results to using either mean annual generation or firm power as objectives. Full details of the workflow implemented to conduct these experiments are included in supplementary information S1.

Results
Here we present the trade-off surfaces for the three experiments outlined above, highlighting portfolios of interest and the composition of the portfolios on the trade-off surface. The composition of the dams in the optimised portfolios are examined, for instance which dam types (small, large or RoR) or sites tend to be built under specified objectives, and which portfolio layouts demonstrate clear increases in benefit (either in power generation or downstream impact). We also compare how the choice of power generation objective or environmental objective affects which portfolios are on the final trade-off surface. Moreover, we demonstrate the value added by modelling floodplain hydraulics.

Trade-offs between power generation and downstream flood extent
Panel A of figure 3 shows the optimized trade-offs between the maximum inundated area and mean annual power generation objectives (experiment 2). Also indicated is the composition of each portfolio in terms of the dominant dam design.
Distinct clusters of portfolios are observed, labeled A1 and A2. Cluster A1 contains only RoR and small storage dams, providing 200 GWh yr −1 of power generation with negligible reduction of inundated area. Cluster A2 adds further small storage, but no large dams, and increases power generation by 100 GWh yr −1 to a total of 523 GWh yr −1 , and a further floodplain loss of approximately 10 km 2 from Number of iterations 50 Figure 3. Large storage dams lock in ecologic-economic trade-offs but are needed to achieve high levels of hydropower generation. Optimised portfolios for mean annual power generation optimised against maximum flooded area (panel A, maps to experiment 2); firm power generation optimized against maximum flooded area (panel B, maps to experiment 3); for firm power generation optimised against flow regime alteration (panel C, maps to experiment 1); and firm power generation optimised against maximum flooded area colour-coded according to flow regime alteration (panel D, maps to experiment 2). In panels (A)-(C), the colour (red, yellow, green and black) indicates which dam type is dominant (we classify a dam type as dominant if the number of dams of that type in the portfolio is three times-or more-than the combined number of all other types. If no dam type meets this criterion, the portfolio is classified as 'mixed'). The small blue rectangles labelled 1-4 highlight portfolios of interest which are discussed in further detail in main text. These portfolios can be regarded as similar between trade off surfaces-their compositions are similar, and represent similar points of inflection between trade off surfaces.
baseline. The layout of portfolio 1 and the associated reduction in flooded area is shown in figure 4, panel 1. The addition of the large Pavua dam at portfolio 2 of panel B of figure 3 (moving from portfolio 1-2) results in a step change in environmental impact, causing a floodplain loss of 10.7 km 2 (a 21 km 2 reduction from baseline), whilst generating no more power than portfolios of small dams with a similar cumulative generation. However, the addition of the Pavua Dam in portfolio 2 is necessary to achieve greater power generation, since all portfolios which generate more power than portfolio 2 (528 GWh yr −1 ) contain the Pavua Dam. Yet, for portfolios including Pavua dam (i.e. right from portfolio 3) an extra 280 GWh yr −1 of generation can be added with only a small additional floodplain loss (2 km 2 , see figure 4, panel 3 for the floodplain loss for portfolio 3 and its portfolio layout). Thus, once Pavua dam is developed, significant extra power generation can be added by developing small dams and RoR upstream of the Pavua Dam (see figure 4). Above 900 GWh yr −1 , most portfolios include mainly large dam sites, resulting in a notable reduction in flooded area. Portfolio 4 represents the full built-out of the system with a generation of more than 1000 GWh yr −1 and a reduction in flooded area to 130 km 2 . figure 3 shows the trade-off surface between the firm power objective, and the maximum inundated area objective. The configuration of the trade-off surface is similar to that of panel A (which uses a mean power objective), with a few differences. Firstly, from portfolios 1-2 in figure 3, an impact of 10 km 2 can be observed, which is attributed to the addition of the Pavua Dam (similarly to panel (A) with the mean power objective). However, the marginal gain is only 100 GWh yr −1 for the firm power objective after addition of the Pavua Dam (panel (B)), compared to around 280 GWh yr −1 for the mean annual power objective (panel (A)), indicating that maximising the firm power objective requires acceptance of greater environmental impact. The relatively greater trade-offs are because firm power necessarily is increased by adding reservoir storage to retain wet season discharges, consequently increasing flood attenuation and reducing the flooded area. Furthermore, there are far fewer RoR dominated portfolios for the firm objective, since greater storage is needed to generate firm power.

Differences between using a peak flow change and max flood objective
Panel (D) of figure 3 shows trade-offs between firm power and maximum inundated area, with peak flows changes indicated by the colors of each portfolio (i.e. the same as panel (A) but colour-coded according to changes in peak flows). At portfolio 4 in panel D of figure 3, the flooded area has been reduced to 130 km 2 (72% of baseline) resulting from an equivalent change in peak flow of 70% of baseline. When moving from portfolio 4-5 (which means setting most sites as large dams), the firm power generation increases by around 380 GWh yr −1 , and the peak flow reduces to 36% of baseline. However, observed flooding is still 130 km 2 (70% of baseline), indicating that changes in peak flow beneath 70% cause no further floodplain loss, because the floodplain is decoupled from the channels for lower discharges. Thus, around 380 GWh yr −1 can be added with little extra impact in terms of inundated area by increasing the number of large dams. This contrasts with Panel C, which optimises for peak flow change (experiment 1), and shows the increase in power generation from portfolio 4-5 results in a significantly worse outcome. This demonstrates the information that can be gained from considering inundation explicitly, since this effect is obscured when using only a peak flow change objective.

Discussion and conclusions
Hydropower reservoirs provide benefits which need to be balanced with numerous potential negative impacts. An understudied impact is reduction of annual floodplain inundation, which can affect agricultural yields and ecosystem health in the world's major floodplains. Strategic planning and optimisation allows identification of dam portfolios which can balance these needs, yet such models often lack the detailed representation of hydrologic and hydraulic processes needed to characterize river-floodplain interactions and associated ecosystem benefits.
Herein we presented a simulation-optimisation framework capable of identifying dam portfolios that optimize conflicting power generation and floodplain inundation objectives, integrating publicly available information to estimate impacts of site, design, and operation on the environmental and hydropower objectives, with explicit optimization of site and design. The framework incorporates hydrodynamic modelling throughout the model chain, allowing better estimates of indicators of seasonal floodplain inundation within an optimisation framework compared to previous work.
Our results indicate that high benefit/low impact dam portfolios need to rely on diverse designs (ROR, large storage dam, or small storage dams). Portfolios of RoR plants alone cannot meet hydropower objectives, whilst large dams produce major downstream impacts. Small reservoirs can, to an extent, mitigate these impacts, highlighting how both dam design and siting can effectively mitigate environmental-economic trade-offs. Results also highlight that certain large dams act as backbones of the power system and are irreplaceable if a major increase in hydropower is required. However, if the impact of those dams is acceptable or otherwise mitigatable, extra power can be added with little extra impact by supplementing large storage dams with smaller ROR projects, highlighting the importance of diverse infrastructure portfolios. The results further demonstrate that small storage dams are more favourable compared to RoR when firm power generation is prioritized. Our results also highlight that including a process-related indicator for floodplain inundation is critical to measure downstream environmental impacts, as a simple peak flow change indicator would not accurately represent the non-linear relationship between dam releases and flood extent.
Further extension of this work includes investigating important aspects of river-floodplain connectivity which affect floodplain productivity, e.g. timing and duration of channel-wetland connection, as well as hydropower impacts on longitudinal connectivity and processes such as sediment transport or fish migration. The latter would address the issue that RoR dams have no environmental impact in the current framework. The Pungwe River Basin has suffered serious flooding in recent years from cyclones likely exacerbated by climate change, therefore inclusion of flood mitigation objectives into the optimisation framework would allow representation of multiple societal benefits from dams. Further important extension of this work would be to run the optimisation framework over multiple climate scenarios, including future climate change, to ensure suggested portfolio design are robust to future climate uncertainty. For all of the above, optimizing not only design and siting but also operations could allow exploration of further opportunities to mitigate such additional conflicts (Hurford et al 2014, 2020, Geressu and Harou 2015, Geressu et al 2020. Moreover,integrated optimisation of operations might highlight further possibilities for environmental flow management that mitigate trade-offs, for instance by allowing for artificial flood events to mimic natural seasonal inundation patterns.
This study attempts to include hydraulics into a framework for strategic optimization of dam portfolios, considering both dam siting and design. Results demonstrate that inclusion of hydraulic processes into such frameworks reveals extra information compared to the hydrologic indices more commonly used to represent dam impacts on downstream ecosystems. Given that our framework uses publicly available global datasets, and the ability to deploy high resolution hydrodynamic flood models at global and continental scales with relative ease (Sampson et al 2015, Bates et al 2020, replication of the study anywhere globally is entirely possible. This is of particular importance because future dam development is likely to take place in data scarce low income countries (Zarfl et al 2015(Zarfl et al , 2019, which also contain some of the world's most sensitive floodplains, e.g. the Okavango, Amazon, and Mekong (Murray-Hudson et al 2006, Andersson et al 2011, Zarfl et al 2015, Winemiller et al 2016. Deployment of this framework in such regions would allow better planning of hydropower portfolios to balance power generation objectives against preservation of floodplain functioning.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// zenodo.org/record/8070818.