Joint strategic energy and river basin planning to reduce dam impacts on rivers in Myanmar

We developed and utilized a screening model for capacity expansion. The model explores pathways to expand electricity investments while optimizing costs, resulting in different energy futures for Myanmar [33]. We use the energy plans proposed in JICA's (Japan International Cooperation Agency) 'Myanmar Power Development Plan' [34] as a baseline, as it is the latest country-wide energy masterplan. Our model then assumes different electricity demand projections and generates viable pathways to meet different estimates of future energy demand with different technology mixes. The model is lumped, representing the country as a single load zone. This could be a limitation in settings where different load zones with different demand and supply profiles are present in a country. Because of its lumped nature, the model can also not account for the different spatial distribution of energy sources, demand centers, and existing transmission infrastructure. However, Myanmar can be considered a single load zone given its highly centralized transmission network (supplemental figure 8 which is available online at stacks.iop.org/ERL/16/054054/mmedia), and we account for the different spatial aspects using detailed auxiliary geospatial analyses (supplemental methods 3–5, supplemental results 3, supplemental figures 8–10).

Variable costs and benefits, e.g. electricity tariffs in Myanmar and international fuel prices, are included in the optimization of a least-cost generation mix. We assume average capacity factors for each generation technology. Similar to other nation-wide energy models [35], derivation of those data is based on a review of either Myanmar-specific or regional studies (supplemental tables 1 and 2). The capacity factors represent the maximum number of hours in a year that a particular technology can run, divided by the total number of hours in a year (8760), thus accounting for intermittency of wind and solar (supplemental table 1). A peak contribution factor further limits the contribution of wind and solar during pre-defined peak load hours, thus accounting for a possible mismatch between hours of peak renewable generation and hours of peak demand [33] (supplemental table 1). The model then makes year-by-year investment decisions that minimize the costs for expanding electricity generation while matching the projected demand for each year. The model also determines the dispatching (i.e. the actual generation) from each technology in each year.

Investment decisions are made based on cost considerations. The least-cost generation mix is determined by a linear optimization (parameters are defined in supplemental tables 3 and 4, parameter values in supplemental tables 1 and 2):

$$\begin{equation*}\mathop {\min }\limits_{\left( {{c_i}} \right)} .NPV{\text{ }}\mathop \sum \limits_{k = 1}^m \mathop \sum \limits_{i = 1}^n T{C_k}{\text{ }}\left( {{c_{i,k}}} \right){\text{ }}\end{equation*}$$

With:

$$\begin{align*}T{C_k} & = {\text{Capital}}{} {\text{cos}}{{\text{t}}_{i,k}}{*}\;{\text{Capacity}}{} ({c_{i,k}}) \\ & \quad + \sum\limits_{i = 1}^n \left[ O\& M{} {\text{Cos}}\;{{\text{t}}_{i,k}}{*}\;{\text{Capacity}}{} \left( {{c_{i,k}}} \right){}\right. \\ & \quad\quad\quad\quad \left.{*}\;{\text{Capacity}}{} {\text{facto}}{{\text{r}}_{i,k}} \right] \end{align*}$$

Subject to:

$$\begin{align*} & \sum\limits_{i = 1}^n {{\text{Capacity}}{} } ({c_{i,k}})\;{*}\;{\text{At}}\_{\text{Peak}}{} {\text{contributio}}{{\text{n}}_{i,k}} \\ & \quad \geqslant {\text{Peak}}{} {\text{deman}}{{\text{d}}_k}\;{*}\;\left( {1 + {\text{Reserve}}{} {\text{margin}}} \right)\end{align*}$$

$$\begin{align*} & \sum\limits_{i = 1}^n [{\text{Capacity}}{} ({c_{i,k}})\;{*}\;{\text{Capacity}}{} {\text{facto}}{{\text{r}}_{i,k}}\\ & \quad \quad {*}{} {\text{total}}{} {\text{time}}{} {\text{in}}{} {\text{period}}{} k] \geqslant {\text{Loa}}{{\text{d}}_k}\end{align*}$$

$$\begin{equation*}{\text{Total}}{} {\text{resource}}{} {\text{potenta}}{{\text{l}}_{i,k}} \geqslant {\text{Capacity}}{} ({c_{i,k}})\end{equation*}$$

The inputs for each generation technology include lifetime (years), initial capital cost ($2019/kW), fixed operation and maintenance cost (O&M) ($2019/kW), variable O&M cost ($2019/kWh), carbon emissions (tCO₂/MWh), fuel costs ($2019/MWh), initial efficiency (%), capacity factor (%) and electricity tariff ($2019/kWh).

To derive realistic costs, we obtained cost estimates for planned and existing power plant projects in the region where available. Where such data were not available, we derived estimates from peer-reviewed regional and global case studies. By using learning rates for electricity generation technologies, we account for future cost declines and economies of scale (supplemental table 2). We apply a conservative range of capital cost but assume that solar PV, small-scale hydropower, and wind generator costs decrease according to technological learning rates from literature estimates [36].

The optimization framework determines the final energy mix for each scenario based on a set of constraints or preconditions (e.g. for the business-as-usual case, the planned hydropower and thermal capacity in the JICA [34] plan is assumed to be built). Finally, the model produces estimates of how much electricity from which source must go online each year to match demand, thus resulting in 'futures' with different mixes of energy generation. Our futures extend to the year 2030 to match the planning horizon in the JICA [34] and the National Energy Master Plan [35]. Based on the capacity factors, the model also determines how much electricity each technology generates (in GWh yr⁻¹) each year.

For hydropower, the model determines how many GW of hydropower must be installed each year and how much total installed hydropower is required by 2030, i.e. ${J_1}\left( F \right)$ (hydropower demand by 2030) described in the following section. However, the energy system model neither prescribes which dams to build to meet the hydropower demand nor considers the cumulative impacts of the resulting hydropower portfolio.

Using this model, we developed two alternative electricity scenarios ('futures') based on the data on costs and availability of NHRs and different projections for domestic electricity demand and export opportunities. In contrast to the BAU, which is based on extrapolating the unusually high growth rates of the mid-2010s (up to 13% pa) [35], the first electricity future is based on more realistic but still high assumptions of demand growth (8% pa) ('Myanmar Domestic Demand', MDD), similar to growth rates proposed in the Asian Development Bank's 2015 'Myanmar Energy Master Plan' [35]. The third future aims to maximize export revenues and meet even most optimistic scenarios of demand growth [37] through a strong boost of installed transmission capacity and export opportunities ('Regional Export Opportunities', REO). Background information and assumptions for each future are provided in the supplemental materials (supplemental methods 3).

The energy model does not consider project-specific transmission costs. However, to compare transmission costs of NHR portfolios with those of hydropower, we propose a novel geospatial analysis to estimate technology-specific distance (and thus transmission costs) from the existing high voltage grid. For solar PV, we also perform a spatial optimization to estimate land demand for different scenarios, and compare results to previous estimates of Myanmar's potential for low-impact solar on already degraded land [23] (supplemental methods 4 and 5).

As an additional measure of costs and externalities, the model includes social costs of carbon, for which we assume emission factors of 0.8 kg CO_2e/KWh for coal and 0.39 kg CO_2e/KWh for natural gas [38]. We then multiply these values with the annual generation from each technology [KWh yr⁻¹] for the year 2030 derived from the energy model, and social costs of carbon of 46 $/t_CO2 [39].

From reviewing data on costs and learning rates of alternative renewable energy in the region, we find that advances in natural gas, small-scale hydropower, solar photovoltaics, wind, and battery storage all can provide reliable electricity supply at lower cost than large-scale hydropower and coal [40], which dominate regional power development plans [34]. By 2020 the initial capital cost of solar PV dipped below $1000/kW. As a global average, costs for large hydropower plants (>100 MW installed capacity) were between $1200/KW and $1700/kW over the past decade with an increasing trend [41]. By reviewing price data for 130 dams in the Mekong Basin, we found an average capital cost of 2023 ± 640 $/KW for large dams (>100 MW) and 2700 ± 1700 $/KW for all dams [42]. Wind and photovoltaic solar are now less costly than large hydropower, with capital costs of around $1100/kW and $1500/kW [43]. NHRs are also likely to see major cost reduction [36], in contrast to hydropower.

We use a separate river system model to predict how sediment budgets and associated ecosystem functions would be impacted by dams in the Salween and Irrawaddy. We then couple these models to the BORG multi-objective evolutionary algorithm (MOEA) [44] to identify dam portfolios, i.e. combinations of the potential dam sites (figure 1), with optimal trade-offs between three energy and river health objectives. Of the objectives, ${J_1}$ denotes the installed hydropower (MW), ${J_2}$ the sediment flux to the mouth of the Irrawaddy (Mt yr⁻¹) and${\text{ }}{J_3}$ the sediment flux to the mouth of the Salween (Mt yr⁻¹). The Borg MOEA then solves a decision problem of the form:

$$\begin{equation*}\mathop {\min }\limits_u \left( { - {J_1}\left( u \right),{\text{ }} - {J_2}\left( u \right),{\text{ }} - {J_3}\left( u \right)} \right)\end{equation*}$$

where u is a binary vector, consisting of $1 \ldots {N_{{\text{Sites}}}}$ decision variables indicating if a dam site is selected as part of a portfolio (existing dams in China and Myanmar are included in all portfolios). The results are Pareto-optimal (PO) portfolios, each creating an optimal trade-off between the three objectives.

3.3.1. J1: Installed capacity

The installed capacity provided by portfolio $P$, ${J_1}\left( P \right)$ [GWh yr⁻¹] is calculated as sum of the tabulated installed capacity [22], $C$, of all dams in portfolio $P$

$$\begin{equation*}{J_1}\left( P \right) = \mathop \sum \limits_{d \in P} C\left( d \right)\end{equation*}$$

where $d \in P$ are all dam sites included in portfolio $P$.

3.3.2. J2 and J3: sediment flux in the Irrawaddy and Salween

Sediment flux in the two rivers for each dam portfolio is determined using the CASCADE sediment routing model [20]. The spatial framework of the CASCADE model is defined by the river networks, which we derive from a 250 m DEM using a threshold of 500 km² for channel initialization. The networks are represented as directed graphs, i.e. sets of nodes and edges. Each node is assigned a specific sediment supply rate, based on estimates of the spatial variability in sediment yield.

CASCADE tracks sediment along the flow path from its source to the river mouth. Each of these paths, referred to as a sediment cascade, represents the sediment transport from a specific sediment source through the river network and finally to the river mouth. Let ${{\Omega }}$ denote the most downstream node of each river network, i.e. the river mouth, then the total sediment delivery to ${{\Omega }}$ is

$$\begin{equation*}{\Theta _{S,\Omega }} = {\text{ }}\sum\limits_{k \in {\Gamma _\Omega }} {\Theta _{S,\Omega }^k} \end{equation*}$$

where ${\Gamma _{{\Omega }}}$ is the set of all sediment cascades that are connected to ${{\Omega }}$, and $\Theta _{S,{{\Omega }}}^k$ is the sediment flux from source $k$ to the outlet of the river network.

Let $\varsigma {\text{ }}$ be a specific source node in the river network. Then, $\Theta _{S,{{\Omega }}}^\varsigma$ is the sediment flux from that source to the river outlet. $\Theta _{S,{{\Omega }}}^\varsigma$ is derived from spatially distributed estimates of sediment yield in the river basins as

$$\begin{equation*}\Theta _{S,{{\Omega }}}^\varsigma = {\text{ }}Y_S^\varsigma {*}AD_d^\varsigma \end{equation*}$$

where $Y_S^\varsigma$ is the sediment yield at the location of $\varsigma$ (t/km²/yr) and $AD_d^\varsigma$ is the direct catchment area draining to $\varsigma$. Estimates regarding sediment yields from the different sub-basins of the Irrawaddy and Salween are derived from geochemical sediment budgets [45] and estimates of total load from both rivers [46] (supplemental figure 1 and supplemental methods 1).

If one or multiple dams are built between a sediment source and ${{\Omega }}$, the sediment flux is reduced according to the trap efficiencies, $TE$, of the dams. Assume that $D\left( P \right){\text{ }}$ is the set of all dams for portfolio $P.{\text{ }}$ Then, $D\left( {\varsigma ,{{\Omega }},{\text{P}}} \right)$ is the subset of dams located between sediment source $\varsigma$ and the river mouth ${{\Omega }}$ for portfolio $P$. The resulting modified sediment flux to the river mouth from source $\varsigma$ is

$$\begin{equation*}{\Theta _{S,\Omega }^\varsigma} ^{*}\left( P \right) = \left[ {\prod\limits_{k \in D\left( {\varsigma ,\Omega ,{\text{P}}} \right)} {\left( {1 - T{E_k}} \right)} } \right]\;{*}\;\Theta _{S,\Omega }^\varsigma .\end{equation*}$$

The modified sediment flux to the river mouth is finally calculated as the modified sediment flux from all sources for portfolio $P$

$$\begin{equation*}{J_2}\left( P \right) = {\sum\limits_{\varsigma \in {\Gamma _\Omega }} {\Theta _{S,\Omega }^\varsigma } ^{*}}\left( P \right).\end{equation*}$$

Trap efficiency is calculated using the Brown formula [47], for each dam $d$, where

$$\begin{equation*}T{E_d} = 1 - \frac{1}{{1 + 0.00021{*}\frac{{{V_d}}}{{A{D_d}}}}}\end{equation*}$$

${V_d}$ is the total storage (dead and live storage from tabulated data) of each dam [m³], and $A{D_d}$ is the drainage area [km²] derived from flow accumulation on digital elevation model. For dams without information regarding storage we estimated the storage from the DEM [20]. To reduce dimensionality in the plots, we report combined sediment delivery, ${J_2} + {J_3}$, from the Irrawaddy and Salween to the Gulf of Martaban. This is also justified as most of the sediment delivered by the Salween is redistributed by strong tidal currents and thus contributes, together with sediment from the Irrawaddy, to building shorelines and delta land in the Gulf of Martaban [48].

3.3.3. Selecting dam portfolios

Each dam portfolio generates a different level of hydropower. The selection of dam portfolios is based on the foreseen demand for hydropower under the different futures, $F,$ generated by the energy expansion model. We denote the hydropower demand for a specific future as ${J_1}\left( F \right)$. We then compare ${J_1}\left( F \right)$ to all PO dam portfolios derived from the optimization (figure 2). It is unlikely that there is any dam portfolio $P$ for which the installed capacity, ${J_1}\left( P \right)$, is exactly equal to ${J_1}\left( F \right)$. There are various ways of matching ${J_1}\left( F \right)$ and ${J_1}\left( P \right)$, the simplest being to select the portfolio for which installed capacity is closest to ${J_1}\left( F \right)$. In reality, dam portfolios could be chosen not only according to their proximity to ${J_1}\left( F \right)$ but also according to additional criteria not considered as objectives in the optimization, e.g. practicality of construction, proximity to existing grid transmission lines, or impacts on, e.g. fish migration or displacement of people.

Zoom In Zoom Out Reset image size

All of the dam portfolios create Pareto optimal trade-offs between the three independent objectives. However, even Pareto optimal portfolios can be inequitable, in the sense that impacts may be concentrated in one river, an aspect that was less relevant in previous studies focusing on single basins [19, 20, 49]. To identify portfolios with a relatively equitable distribution of impacts between the Salween and Irrawaddy, we identified dam portfolios that would result in less than a 50% reduction in sediment transport on either river. For the portfolio selection we then select first all PO dam portfolios in a $\pm$10% range around ${J_1}\left( F \right)$. Of those candidate portfolios we then considered only those with less than 50% sediment trapping in either river. From those candidate portfolios we selected the portfolio with an installed capacity as close as possible to ${J_1}\left( F \right)$. If there was no portfolio with less than 50% trapping in the $\pm$10% range, we selected the portfolio with ${J_1}\left( F \right)$ closest to ${J_1}\left( P \right)$. For selecting a dam portfolio for the REO future we added a last step and preferred portfolios preferentially including dam sites from the MDD portfolio such that the MDD portfolio could be expanded into the REO portfolio at a future time if there should be increasing opportunities for energy export.

Compared to the BAU with its focus on large hydro (+38 800 MW) and fossil fuel (coal: +12 200 MW, natural gas: +1100 MW) (figure 3(a), top), the alternative energy futures rely on a wider mix of biomass, solar, wind as well as small and large hydro. The MDD future would require 9400 MW of solar and wind and 2300 MW of Biomass, complemented by 6700 MW of hydropower, very little coal and some natural gas (+285 and +1164 MW) (figure 3(a), center). For the REO future the energy model predicts a massive expansion of solar (+66 900 MW). Hydropower is expanding, too, but much less (10 300 MW) than under the BAU (figure 3(a), bottom). For all futures, year by year expansion and resulting generation mixes are shown in supplemental figures 5–7.

Zoom In Zoom Out Reset image size

The MDD and REO futures not only require less hydropower, but also yield cheaper energy thanks to their reliance on less-costly solar and wind sources. The cost estimate for the BAU pathway is $11.7 billion USD, with an average system cost of $40.4/MWh, over a 15 year time horizon, assuming the system would generate at least 289 TWh over that period. Unit costs would be higher if generation was lower, e.g. because optimistic assumptions about demand growth do not materialize. For the MDD future, the model results in lower-bound costs of $8.1 billion USD. By 2030, the average system cost of electricity in this scenario is $37.3/MWh if the system generates at least 216 TWh over a 15 year period. For the REO future, lower-bound costs are $8.4 billion USD, with average system cost of $38.4/MWh if the system generates at least 218 TWh over a 15 year period. Thus, energy costs for the BAU are 8%–9% higher than under the alternative futures. Moreover, the BAU would require 37% higher up-front capital investment than the MDD and REO futures, because the alternative futures can take advantage of more flexible investments over time and dropping prices for NHRs.

Costs for the BAU would further increase when considering externalities of rivers and global climate. Social costs of carbon would be $3.6 billion for the BAU ($3.4 billion from coal, $260 million from gas), and $385 million from the MDD and REO ($155 million from coal, $230 million from gas). This does not yet include greenhouse gas emissions from hydropower reservoirs, which can be substantial but would require a dam-by-dam assessment because of the great variation in relative emissions [50].

In terms of transmission, we find that Myanmar has significantly more solar than hydropower resources in close proximity to its highly centralized high voltage grid, thus indicating that solar PV has very likely a transmission cost advantage over hydropower. This is because most hydropower potential lies in more remote parts of the country while the best solar resources are in the center of the country, close to existing transmission lines and demand centers (supplemental results 3, supplemental figures 8 and 9).

The hydropower demand for each future could be matched by different dam portfolios, i.e. different sets of potential dam sites that provide a similar generation but that might vary widely in their cumulative impacts. Using the river basin models for the Salween and Irrawaddy, we identified many PO dam portfolios covering hydropower expansion from the current level to the full development of all hydropower resources (figure 3(b)). Being PO, each portfolio minimizes the trade-offs between installed hydropower and impacts on sediment budgets.

Each data-point in figure 3(b) thus indicates the trade-offs between dam impacts and benefits for a Pareto optimal dam portfolio. Square markers in figure 3(b) identify dam portfolios that meet the 50% threshold (i.e. not more than 50% reduction in the sediment budget of either river). The 50% threshold is only met for dam portfolios with installed capacity not exceeding 37 000 MW. Even with optimized dam portfolios, installing hydropower beyond that threshold would lead to major impacts on the Salween river (figure 4), because such portfolios require some major mainstem dams on the Salween. For instance, the portfolio selected to meet the BAU demand would reduce sediment flux from both rivers to around 300 Mt yr⁻¹, around 50% reduction in combined sediment export from both rivers. However, most of that reduction would be in the Salween, where mainstem dams would reduce the sediment budget by more than >90% (figure 4(a)).

Zoom In Zoom Out Reset image size

The worst performing PO portfolio meeting the BAU demand could reduce the total sediment budget to as little as 200 Mt yr⁻¹, resulting from an around 80% reduction in sediment transport in the Salween (figure 4(a)) and 60% reduction in the Irrawaddy (figure 4(b)). The BAU future would also lead to a major reduction in river connectivity and associated impacts on biodiversity (supplemental figure 2). By contrast, the MDD and REO futures would maintain sediment transport above 500 and 450 Mt yr⁻¹, respectively, and could match demand with dam portfolios that do not have disproportionate impacts on either river (figures 3(b) and 4). It should also be noted that dam development proceeds often in a non-strategic, project by project manner [20] and impacts of all futures on rivers could be much worse if dams where not chosen from the PO portfolios but just site-by-site. Thus, our estimates of impacts, even for the BAU future, can be regarded as best-case estimates.

Under the BAU future, even optimized dam portfolios include a significant number of mainstem dams (figure 5(a)), especially on the Salween. The MDD future omits all mainstem dams from the hydropower portfolio (figure 5(b)). Thus, the Salween and Irrawaddy remain free-flowing and hydropower demand is met from projects in tributaries and in already disconnected parts of the river network upstream of existing dams.

Zoom In Zoom Out Reset image size

The MDD future also avoids most dams in politically unstable tribal areas in Northern Myanmar, where dam development might have repercussions on human rights [51] and where hydropower projects have led to armed conflict in the past [37] (figure 5(b)). Compared to the BAU, the MDD future assumes a more modest domestic growth and fewer energy exports. With that regard, the REO future provides opportunities to satisfy most optimistic future export opportunities and would allow Myanmar to play an important role for the decarbonization of the region.

Under the REO future, the portfolio from the MDD future would be expanded to include most dams in the hydropower cascades of the Nmai and Myitnge rivers (supplemental figure 1 for river locations) but most mainstem dams could still be avoided (figure 5). Supplemental figure 4 shows a probabilistic analysis of all dam portfolios in the $\pm$10% range around the prescribed hydropower demand (supplemental results 2). This analysis indicates that the presented dam portfolios are representative for the respective generation levels. For example, dam portfolios for the MDD and REO futures have a very low probability to ever include major mainstem dams, whereas mainstem dams have a high probability to be included in portfolios meeting the BAU demand.

Lastly, we note that we do not have project specific data on total or marginal transmission costs (i.e. in $ or $/MW) for each hydropower dam. However, our geospatial analyses shows that the dam portfolios selected for the MDD future would require less transmission length than the BAU (supplemental results 3, supplemental figures 10 and 11). In terms, of required transmission capacity (km times required transmission rating in MW) the MDD future has significantly lower requirements than the BAU future (supplemental figures 10 and 11). Thus, there is strong evidence that the MDD and REO future would likely incur lower transmission costs than the BAU, first because alternatives include more NHRs that are located closer to the existing transmission grid (supplemental figures 8 and 9), and second, because included dams are closer to the grid (supplemental figures 10 and 11).