Sustained cost declines in solar PV and battery storage needed to eliminate coal generation in India

We undertook a multi-step process for gathering the required multi-year hourly wind and solar generation data for our site locations. First, proprietary wind and solar power production data normalized to plant capacity were obtained from project operators under Non Disclosure Agreement for six power plant sites (one each for wind and solar PV in each state) in the three states of Karnataka, Tamil Nadu, and Gujarat. In each state, the location of the wind and solar PV site was less than 100 km from each other to ensure that it could represent a potential co-located hybrid system. We used this data to create a typical year of generation of wind and solar power which provide us with the CFs shown in table 1. For more details on the process used to simulate a typical year of generation from our proprietary data, see Supplementary Information (SI) Note 1.

Second, we obtained 20 years of hourly wind and solar output based on Modern-Era Retrospective analysis for Research and Applications (MERRA-2) satellite reanalysis data for the period 2000–2019 (both years inclusive) for the same exact proprietary data site locations (see first step above) in Karnataka, Gujarat and Tamil Nadu. MERRA-2 has a horizontal resolution of 0.5^∘ by latitude (−90^∘−90^∘) and 0.625^∘ by longitude (−180^∘−179.375^∘) with 361×576 grid cells worldwide [22]. Solar irradiance data is converted into power output based on the Global Solar Energy Estimator model [23]. Wind speeds are converted into wind turbine power output using the Virtual Wind Farm model [24]. We use the same wind turbine (hub height and power curves) and solar panel (azimuth angle, tilt angle, and tracking selection) configurations as used by the developers at the sites in these states to maintain consistency with our proprietary data. For more details on this, see SI note 1.

In the third step, to account for the fact that wind and solar output from reanalysis data such as MERRA-2 can have issues with CFs not matching real world observed CFs, we scale the median of the 20 years of reanalysis data CFs to the observed CFs from the proprietary data (table 1) we have for these states (we scale the MERRA-2 data by the ratio between our proprietary data CF and the median CF of the 20 individual years of MERRA-2 data).

This approach, where we draw on the reanalysis data for incorporating the temporal and spatial characteristics of the resource over multiple years, but scale the data to real world observed data for CFs (which in our case were obtained at the plant level from industry developers) has been recently outlined in Tong et al [25]. The unscaled CFs for the raw MERRA-2 based dataset at the proprietary site locations our shown in tables S11, S12 and S13 in SI note 4.

In the absence of any publicly available multi-year wind resource data for India, previous studies [9–11] are based on a single year of national wind resource data published by the National Renewable Energy Laboratory [26]. To our knowledge, this is the first study on India to use multiple years of both wind and solar power production data. Our modeled hybrid plants have an expected lifetime of 20 years so we run our optimization model over the full 20 year period using the entire scaled MERRA-2 based wind and solar dataset.

We performed multiple checks using different available data sources for wind and solar generation to ensure robustness of our results. For more on this see supplementary note 4. Overall, we find that our results, i.e. suitability of solar for pairing with lithium-ion battery storage and the high cost of hybrid systems today compared to existing coal plants in India, are both robust to alternate data inputs.

We also find, consistent with existing literature on the subject [27, 28], that just a small portion of hours in the twenty year dataset are significantly influencing the design of the system and as a result, driving up costs. These results affirm once again the need for multi-year resource datasets for electricity planning. A detailed exploration of wind and solar resources in the three states is provided in SI note 5. We find high seasonality for wind generation with significant generation during the monsoon period (June–September) and limited output for the remaining parts of the year. Wind also has notable inter-annual variability in output. Solar output is found to be more consistent both across years and seasons.

We consider two types of stylized generation profiles for coal power as useful bounding cases for estimating the cost of replacing coal power generation with hybrid power plants. First, the target coal plant to be replaced is assumed to run as inflexible baseload, providing 100 MW of power every hour of the year, as shown in figure 1(a). While inflexible baseload operation may be a useful limiting case to understand the costs of such systems, coal plants in India are increasingly run in a flexible manner to incorporate higher shares of wind and solar generation in India. At times of high wind and solar production, coal generators often back down to lower levels of output, rising again to maximum output during the evening hours as demand rises and solar generation drops off. If the objective is to replace a dispatchable source of power such as flexible coal plants rather than firm inflexible baseload, grid operators may require hybrid power producers to similarly reduce output during hours where standalone wind or solar plants without battery storage are producing at full capacity. We therefore model a second stylized operational profile for coal power shown in figure 1(b). For the stylized flexible generation profiles, we incorporated data on the relative shares of already installed stand-alone (without energy storage) wind and solar PV capacity in each state. A detailed description of how we developed the stylized profiles for flexible generation is provided in note 1 of the SI. Note that while figure 1 only shows an average 24 h period, our optimization model uses the full time varying twenty year hourly target profile, which differs from the 24 h average profile in the case of flexible generation.

Zoom In Zoom Out Reset image size

We build on the framework described in Ziegler et al [13]. In this optimization problem, the design objectives are the capacity of the solar plant, wind plant and battery that must be built; the charging and discharging decision for the battery system at each time step; and the rate of charge or discharge. The objective is to minimize the levelized cost of energy (LCOE) of the hybrid system, which is simply the present value of all costs divided by the total energy sold by the system over its lifetime.

The optimization objective is based on the full 20 years of operation (175 200 h). The costs are composed of capital costs including installation costs of the battery system, wind plant, and solar PV plant. We ignore operating costs such as maintenance given that capital costs far outweigh operating costs for renewable energy [13, 27].

This optimization problem is constrained foremost by the target output profile of the coal plant that the hybrid system must match. Further, the operation of the battery is constrained to the maximum rate of charge or discharge, the energy level from one time step to the next, and the maximum and minimum energy level. These further bound the problem.

Our optimization problem is a Linear Programming problem which is therefore convex and quickly yields globally optimal solutions. Solar, wind and battery storage system capital costs are assumed as $700 kW⁻¹, $1100 kW⁻¹, and $400 kWh⁻¹ respectively [29–31]. Storage duration is assumed to be 4 h with a roundtrip efficiency of 75%. Plant lifetime is assumed to be 20 years. All model equations and associated parameters are summarized in SI note 1 (table S1).

To understand the relative value of each intermittent renewable technology, we model different mixes of wind and solar PV in a hybrid system with 4 h battery storage. Figure 3 shows the change in the levelized cost of the system providing baseload generation for different shares of wind and solar PV capacity in the total renewable energy capacity of the plant.

Zoom In Zoom Out Reset image size

Costs of hybrid systems which only incorporate wind energy with storage are significantly higher at the sites considered. This is due to the high seasonality of wind generation in India (also see SI notes 4–5), with high output during monsoon season and very limited output the rest of the year. As such, wind is a poor companion to short duration storage for providing baseload or flexible generation every hour of the year. As solar is added to the hybrid mix, costs fall. Hybrid systems which combine even small amounts of wind capacity are cheaper than a system with only solar PV and storage for each of the three sites considered. For example, table 2 shows that this is the case for Karnataka and Tamil Nadu for replacing baseload generation, with just 1% and 7% of the wind-solar capacity coming from wind power. We do note that really small sized wind plant capacities (e.g. 14 MW) may not be practical as part of a hybrid plant, even if they are optimal. Figure 3 however shows that the LCOE of hybrid power plants with 0%–25% wind share are relatively similar, indicating that there may not be large trade-offs in optimality vs practicality in the design of hybrid power plants. The equivalent analysis for flexible generation is shown in SI supplementary note 2 (figure S1).

The behavior of the battery storage system both in terms of a 24 h average across the 20 year period as well as the hourly SOC across the full 20 years is shown in SI note 2 (figures S2–S5).

Hybrid systems will also be allowed downtime, similar to coal power plants in India that often operate at 60%–80% CF for the year [34]. We define system availability here as the percentage of hours over the operational lifetime where the hybrid power plant is able to provide the required generation profile (baseload or flexible). While the results presented in figure 2 are for 100% availability systems, here we model the levelized costs of hybrid plants with lower than 100% availability, where the constraint of meeting the required generation profile is relaxed for some hours [13] in the 20 year period. A 100% system meets the generation required for all 175 200 h in 20 years. A 90% system would meet this requirement for 157 680 h. Results for these different scenarios are shown in figure 4 below for the case of baseload generation.

Zoom In Zoom Out Reset image size

As the required availability of the system is reduced, levelized costs fall. Crucially, lowering the availability required by just 1% point from 100% to 99% leads to a roughly 40% reduction in levelized costs across the three states. Hybrid systems that can provide baseload generation 99% of the hours over 20 years are therefore notably cheaper than those that provide the generation 100% of the hours. This shows that a small share of hours in the 20 year period are driving the high cost of hybrid systems that meet the generation requirement 100% of the time. Tables S2–S4 in the SI note 2 show why this is the case. In the 99% system, the size of the battery storage system is much lower, reducing costs compared to the 100% system. A small share of hours with prolonged low output from wind and solar over the 20 year period requires bigger battery sizing for the 100% system, and therefore leads to much higher costs overall. By reducing the availability requirement even slightly, the hybrid system can rely more on wind and solar resources instead of the battery, reducing costs. Costs fall further as the availability required is further lowered as shown in figure 4. We also find that as the availability required is lowered, the share of wind in the wind-solar mix increases (figure 4, tables S2–S4). For example in Karnataka, the hybrid system providing baseload generation in 100% of hours has just 1% of wind in the wind-solar mix. However, a 80% system is a wind dominated hybrid plant with 51% wind capacity in the wind-solar mix. Similarly for Gujarat, the share of wind goes from 25% in the 100% system to 62% in the 80% case, with steady increases in the wind share for each level of lower availability. In fact as the availability required is lowered, wind substitutes for storage in the hybrid system, which is what drives the cost reductions. We find that in Gujarat, an optimal mix hybrid system that provides baseload 100 MW power in 80% of the hours over 20 years has a lower levelized cost than ₹ 3.0 kWh⁻¹ (figure 4), the target for replacing existing coal generation. We caution however that these results do not necessarily mean that hybrid plants at 80% CF are cheaper than 80% CF coal power plants. The issue is that the precise timing of their availability will influence system and therefore consumer costs for grid electricity. The results shown in figure 4 are based on modelling downtime when it optimal for the hybrid plant operator, which due to the relative costs of wind, solar and storage, is when renewable generation is lowest and the battery would be needed most to fulfil the constraint of providing the target 100 MW generation. Of course, it is precisely when renewable generation on the electricity grid drops to low levels that coal generation today would likely be running at high output. Therefore, it is unlikely that hybrid plants will be allowed to have downtime in those hours if they are replacing coal power on the grid, as intended here. If so, other resources such as diesel generation or demand response may be required which incur their own costs. As such, we caution that while levelized costs for such systems fall in all the three states as availability is lowered to more realistic operational availability of less than 100%, this does not necessarily imply that coal generation can be replaced at those lower costs. Nevertheless, the results shown here can shed some light on the influence of system availability on levelized costs and also highlight the opportunities for other low cost grid flexibility mechanisms such as demand response or bi-directional electric vehicle charging which could potentially cover the 1% or 10% of hours over twenty years at a lower cost than the cost difference between 100% availability and 99% or 90% availability hybrid systems.

Our results so far have focused on hybrid power plants at a single site, drawing on co-located (<100 km apart) wind and solar resources in a single location along with a battery storage system. However, wind resources in particular may have high spatial heterogeneity. Drawing from resources in multiple locations could lead to reduced costs, as dips in output in one location could be compensated by another, smoothing the output profile of the variable renewable resource with potentially lower dependence on the battery and an increased role for wind energy.

To model this, we draw on twenty years of hourly wind and solar data in four individual sites in the northern, southern, eastern, and western parts of each state. The exact locations for each of the states are shown in SI figure S6. We model the levelized costs of hybrid power plants at each of these individual site locations. Then, we consider a fifth case, by averaging the hourly wind and solar output across these four individual locations to what could potentially represent a state averaged renewable resource, and model a hybrid system based on this state averaged resource. Comparisons between the costs of hybrid systems which can draw output from multiple locations versus the costs of hybrid systems that are located only in individual locations can help shed some light on how geographical smoothing of resources influence the design and costs of hybrid power plants for providing baseload or flexible generation.

Table 3 shows the results for the state of Gujarat. Similar results for Karnataka and Tamil Nadu are shown in tables S5 and S6 in the SI.

We find that smoothing resources over a wider region in Gujarat leads to lower levelized costs, and savings of 16% over the lowest cost individual location (West) and savings of 28% over the most expensive individual location considered (East). We caution however that the levelized costs for hybrid power plants based on the state averaged resource do not necessarily fully represent the costs of replacing coal generation. This is because to this cost one must also add the cost of transmission, functioning real time electricity markets, and other soft and hard infrastructure required to integrate renewable resources from different corners of a state. We also note that even with state averaged resources, wind continues to have a minor role, with the least cost hybrid system solar dominated in each of the three states (table 3, SI tables S5 and S6). Finally, the costs of such hybrid plants based on state averaged resources nevertheless continue to far exceed the ₹ 3.0 kWh⁻¹ cost target required to compete with existing coal generation.

Wind, solar PV, and lithium-ion battery storage have all seen rapid cost declines in recent years [29, 41] which provide evidence towards future lower costs for hybrid systems. For instance, global weighted installation costs of wind and solar PV have fallen 24% and 79% respectively between 2010–2019 [29]. Meanwhile, lithium-ion battery pack prices have fallen 89% between 2010–2020 [42].

We examine the future cost targets these technologies (wind, solar PV, battery storage) would have to hit, to enable levelized generation costs that can economically displace coal generation and achieve the goals of the Paris Agreement. To understand various scenarios of displacement, we consider different cost scenarios for coal generation. Recent governmental directives have called for the introduction of pollution control technologies that could increase marginal costs by up to 25% [33]. Current cost of coal generation in India also does not account for the health or climate externalities imposed by coal power, aside from a small ₹ 400/ton cess on coal production. We therefore consider cases where a carbon tax as low as $20/tCO₂ to as high as $60/tCO₂ is imposed, decreasing the relative competitiveness of coal versus alternate generation technologies. For the levelized cost of new coal generation we assume costs of roughly ₹ 4 kWh⁻¹ [17]. We consider baseload hybrid systems with 99% availability, i.e. hybrid systems that provide baseload 100 MW power for 99% of the hours over a twenty year period. We focus here on 99% and not 100% systems as it clearly conceivable that in the future other low cost grid flexibility mechanisms (e.g. bi-directional electric vehicle charging) could handle the lean 1% of hours, which as shown previously in figure 4 significantly increase the cost of hybrid power plants.

Figure 6 shows the levelized costs of hybrid systems for different capital costs of wind, solar PV and battery storage, in the three states of Karnataka, Gujarat, and Tamil Nadu. All costs are assuming a lower cost of capital of 2.5% and shown for providing a baseload generation profile (note that as shown in figure 2), costs for both flexible and baseload generation are not too dissimilar). Note that while figure 6 only shows the cost of battery storage capacity, total storage system costs including balance of system costs, power costs, and controls are normally twice that of storage capacity costs alone [30, 31]. So a $75 kWh⁻¹ capacity cost translates to a storage system cost of $150 kWh⁻¹ in our model. Levelized costs of hybrid systems today, previously shown in figure 2, are also shown here in the upper left quadrant in each row (state) in figure 6 and correspond to installation costs of $700 kW⁻¹ and $1100 kW⁻¹ for solar PV and wind respectively and battery storage capacity costs of $200 kWh⁻¹ (or total storage system costs of $400 kWh⁻¹) [29–31]. At these levels, levelized costs for 99% availability systems are in the range of ₹ 6–8 kWh⁻¹ as discussed previously in figure 4. This would require a $60/ton carbon tax to displace potential new coal generation.

Zoom In Zoom Out Reset image size

Figure 6 shows that to displace existing coal with 99% availability hybrid systems across all the three states, without any policy intervention, would require cost targets of at least $250 kW⁻¹ for solar PV along with battery storage capacity costs of $100 kWh⁻¹. This corresponds to a 50% cost reduction for energy storage and more than a 60% cost reduction for solar PV. If the objective is to displace coal power by 2040 in line with the goals of the Paris Agreement [15, 16], then this cost reduction must be achieved in the next two decades, consistent with an annual year on year cost decline of roughly 6%. An annual decline of 6% will also mean that by 2030, hybrid systems will be cheaper than new coal generation in all three states without any carbon taxes or economic incentives for clean energy.

Note that the results shown in figure 6 correspond to a 2.5% cost of capital. However, if we consider higher capital costs which are more common in financing of clean energy projects, even more ambitious cost declines in the next two decades would be necessary for hybrid systems that are cost competitive with coal generation in India.

Figure 6 also shows that the relative resources of solar and wind influence the nature of cost declines for each state. For example, in Gujarat which has the highest share of wind generation in an optimal mix hybrid system, cost reductions are achieved through a fall in costs of any of the three technologies, illustrated by the pattern shown in figure 6(b). However, for Karnataka and Tamil Nadu, cost declines are mostly influenced by the capital cost of solar PV and storage as in these states wind is a small share of the optimal hybrid system, even in cases where wind capital costs are lower than solar PV (figures 6(a) and (c)).