2.1. Storage requirements in extreme weather conditions
To study and compare the implications of the selected heating technologies on the UK whole energy system, we designed and simulated three “core” architectures (see Table 1 and Methods for further details), in which the main heat supply is provided by consumer heat pumps (HP scenario), district heating (DH scenario), or hydrogen boilers (H2 scenario). Since our aim was to find renewables-based zero-emissions energy systems that are resilient to extreme weather events – defined as prolonged periods of high energy demand and low generation at the same time – we simulated those three scenarios using a 3-year time window centred on a “stress year”, with 2010 having one of the most stressful meteorology periods between January and February – when the national average wind speed and temperature were 3 m/s and 2 oC, respectively – and then in December of the same year, with the same low wind speed and an average temperature of 1 oC. Storage is a crucial component for balancing meteorology-driven variations in renewable supply and demands from hourly to inter-annual temporal scales. As shown in Fig. 1, hourly storage levels in all simulated scenarios exhibit a similar annual seasonality, as stores are discharged during winter and recharged in summer. The heat storage in DH (Fig. 1A), for example, is heavily used in December and January. In winter, the average heat storage output is 15 GWth and its sum reaches 28 TWh, while the summer output averages around 2 GWth. In particular, the storage has a peak output of 154 GWth, with a large discharge of 16 TWhth, lowering the storage level down to 88 GWhth within 16 days (from 2010-12-17 to 2011-01-03). In the same period, the heat demand reaches 44 TWh, with a peak of 186 GWth. This event is the most demanding for storage, as in the previous winter the storage level is down to 7 TWhth, the peak storage output is 137 GWth, while the heat demand peaks at 145 GWth (i.e. in 2010, heat peak demand was 40 GWth higher). The output from hydrogen (Fig. 1B) and electricity (Fig. 1C) storage, in the H2 and HP scenarios respectively, shows a similar pattern to the heat storage in DH. This might be expected as demands and renewable generation are the same in the core scenarios. The storage dynamics during the end of 2010 shows that, given the renewable and interconnection capacities, the total system storage must be of the order of tens of TWh to cover supply deficits during extreme meteorology. In general, the capacities of the different storage types depend on their locations in the energy system, the other system components, and the parameters optimised in the system design. In ESTIMO, the stores are connected at the transmission level and their sizes were manually designed.
2.2. The impact of different energy system components on storage need
To explore the interplay between system components, such as renewable capacity or interconnectors, on storage needs, we created eight zero emission variants of the DH scenario (Table 1 and Methods) in which we offset heat storage size with wind and solar capacities (DH GenLo, DH GenHi, DH PVLo), or interconnectors (DH LinHi), changed the building heat loss parameter (DH InsLo, DH InsHi), and increased the ambient temperature to take into account climate change (DH CChLo, DH CChHi). After running simulations of these variants between 2009 and 2011, we found that a 50% reduction of the storage size of the DH core architecture can be balanced by either a 33% increase of wind and solar generation (DH GenHi) or a 200% increase of the UK interconnector capacity (DH LinHi). On the other hand, a 33% reduction of renewable generation is balanced by a 100% increase of the storage size (DH GenLo) (Fig. 2). This trend indicates a non-linear relationship between storage size and renewable supply, as a fractional decrease in renewable would require a greater increase in storage. Likewise, in the scenario with low solar PV (DH PVLo), the higher offshore wind capacity allows less storage, as its minimum level is 3 TWh higher than in the core DH architecture. These results suggest that 1 GW of installed offshore wind and solar capacity, when not critically low, are functionally equivalent at the margin to about 200 GWh of storage; with an impact of 2.5 times that of solar PV, offshore wind generation which would significantly contribute towards a reduction of storage needs.
In addition to storage size, we analysed how storage output changed with varying renewable capacities, building insulation, and climate change. From our simulations of 2010, we found that the annual heat storage output reaches 86 TWhth in the DH GenLo scenario and 58 TWhth in the DH GenHi scenario, corresponding to a change of + 24% and − 19%, respectively, compared to the 70 TWhth of the core DH architecture, suggesting that each GW of difference in the renewable capacity changes the delivered storage heat by about 3 TWhth. The scenarios DH InsHi and DH InsLo have a lesser impact on heat storage, just 5 TWhth (i.e. ~6% of the DH core scenario), meaning that a change of 1% in building insulation corresponds to roughly 1% difference in heat storage output. Compared to the core DH architecture, we found that storage is also used less in the DH CChLo and DH CChHi scenarios, where the total heat storage output is 20% and 36% smaller (56 and 45 TWhth) respectively. Therefore, every additional degree in the average ambient temperature reduces storage output by about 10%. The largest difference in terms of peak storage output compared to the 154 GWth of the core DH scenario can be found in the DH CChLo and DH CChHi scenarios (143 and 133 GWth, respectively), corresponding to 3% peak reduction per degree Celsius.
Finally, we analysed the DH derived scenarios in terms of storage dynamics, measured as the fraction of hours that the storage is discharged during the simulation. With respect to the core DH architecture, which shows an output fraction of 15%, the high generation DH GenHi scenario has a reduced fraction of 13%, corresponding to a more variable storage level, while the low generation DH GenLo shows a higher fraction of 19%, denoted by a flatter storage pattern (Supplementary Fig. 1). For short time scales, these differences can be estimated as a decrease of 1% in storage output frequency for each 18 GW of installed renewable capacity.
Overall, our findings suggest that renewable capacity, in particular offshore wind with a projected capacity factor of about 60%, has a greater impact on storage size and usage than any other variable (insulation, PV share, or interconnections), and it is therefore an effective measure to reduce storage need.
2.3. The role of system storage on long-term time scales
We evaluated storage capacity and use patterns by simulating the operation of the three core architectures with 35 years (1980–2015) of historical meteorology data. This approach allowed us to analyse relatively rare meteorology, such as periods of low wind and temperature (as in 2010), and to describe the seasonal and the inter-annual variability in storage use. We found that the most frequent output from heat and hydrogen storage is around 20 GW and can balance supply in a wide range of energy deficits, up to 160 GW in winter periods, with an almost linearly decreasing frequency (Fig. 3A). Heat storage output shows a less skewed distribution than the other two storage types, with a higher usage than the hydrogen storage at values below 10 GW, and a lower peak at 20 GW. Electricity storage shows an additional peak for very small outputs, as it is also used to meet short-term small fluctuations of electricity-specific (e.g. appliances, lighting) end-use demand. The distribution of electricity storage output has a shorter range than the other two storage types because the electricity storage size is set to a smaller capacity, due to its high unit cost. An example of the actual temporal distribution of the storage output is given by the DH heat output along the 35 years, aggregated into a single year time (Fig. 3B); storage his usually full, as shown by the weekly median of zero, and is characterised by frequent hourly variations – up to 100 GWth, as indicated by the interquartile range – which are uniformly spread along the year, with a total range between 40 GWth in summer to 160 GWth in winter. The storage output values recorded in summer are partially due to mild wind speeds (compared to winter) and to nightly heat demand, resulting from our assumptions on human activity patterns and on the thermostat cut-off temperature set to 20 oC for space heating demand. As the simulated architectures were manually created, the size of heat and hydrogen storages are not optimised to balance supply exactly; instead, they were slightly oversized to make sure that no biomass or natural gas is consumed during periods of peak heat demand.
These storage output patterns suggest that, regardless of the heating architecture, all storage types would play the same double role: as seasonal back-up with a wide range of output capacities, and as a short-term flexible buffer to balance hourly and small daily supply-demand mismatches, in which the electricity storage would cover the smaller end of the output capacity range. Regarding long-term energy needs, the storage outputs should be in the order of 100 GW to cope with rare very high demand, as indicated by the wide range of the modelled storage output.
2.4. Annual energy consumption and peak demand
Together with storage needs, our 2009–2011 simulations also provided valuable insights on end-use energy demand (Supplementary Figs. 2 and 3). During 2010, heat annual consumption and peak demand are, respectively, 540 TWhth and 186 GWth in the core DH architecture; these values are unchanged across all scenarios, except in the building insulation and climate change scenarios. In the DH CChLo, consumption is 478 TWhth, with a peak of 174 GWth, while in the DH CChHi these quantities are 422 TWhth and 162 GWth, respectively, suggesting that for each + 1 oC in annual average ambient temperature, annual heat consumption decreases by 31 TWhth (1% of the consumption in the core DH) while peak demand is reduced by 6 GWth (3% of the peak demand in the core DH). As an additional note, air conditioning demand increases by 100% in DH CChLo and by 200% in DH CChHi, that is, by 50% per oC. For electric vehicles, the change in annual energy consumption due to climate change is quite small, around 1% per oC, as less cabin heating is nearly balanced by increased cabin cooling in summer, as are the impacts on EV battery and heat pump efficiencies. These results suggest that climate change will have a bigger impact on cooling than on heating demand; this modelling needs refining, but it illustrates the importance of an integrated system modelling approach to evaluate the cumulative impact of different demands on the system.
As the remaining non-heat demands are the same in all scenarios, electricity consumption – inclusive of electricity delivered to consumers, DH heat pumps, electrolysers, and transmission losses – is similar across almost all scenarios (Fig. 4), except for the hydrogen boiler and Climate change scenarios (Supplementary Table 1). In the core H2 architecture, the total electricity consumption is about double that of the architectures using heat pumps (which have an average COP of 300–450%), because the overall efficiency of the hydrogen supply chain is only about 70% (assuming 80% electrolysis efficiency and 85% boiler efficiency). Therefore, the H2 architecture uses 4 to 6 times more electricity per heat output than the core HP and DH architectures. In the DH CCh scenarios, electricity consumption is reduced by about 2.5% per degree increase in annual mean ambient temperature, since the heat demand is less than the DH scenario.
The highest peak of delivered electricity to all consumers of 156 GWe occurs in the core HP scenario, in which the peak input to consumer heat pumps is 49 GW, converted to 131 GWth of heat at an average COP of 2.7 (Supplementary Fig. 4). On the supply side, the peak renewable and nuclear generation is 270 GWe and sufficient network capacity is needed to utilise some of the surpluses. In the H2 scenario, the delivered peak (136 GWe) is lower than the HP case, but the overall peak on the UK transmission network is the highest, 325 GWe, mainly because of the electricity input to the electrolysers, which reaches 250 GWe. Around 40–50% of electricity generation (~ 500 TWh) is spilled annually across all scenarios, when supply is greater than demand and the surplus cannot be stored or exported because there is no spare capacity. Spillage occurs on average about 30% of annual hours and is distributed quite uniformly across each year of the simulations. Among the core architectures, spillage is 50% in the H2 and HP scenarios, and 46% in DH; in the derived DH scenarios, spillage is the lowest in DH GenLo (41%), and reaches the maximum of 50% in DH GenHi and DH CChHi (Supplementary Fig. 5). Altogether, our findings highlight that the hydrogen architecture would require a significant increase in renewable generation and network transmission capacities with respect to the other scenarios, due to its considerable electricity demand.
2.5. Whole system and heating costs
The component capacities drive the dominant fixed capital and O&M costs, with variable O&M small, and no fuel costs in the scenarios. We annuitized capital costs, and aggregated annual system costs by consumer, intermediate and primary energy categories (Fig. 5). In terms of consumer capital cost, the HP scenario is the most expensive (23 G£/a, ~ 57% higher than DH and ~ 30% higher than H2); similarly, electricity storage and network costs are higher in the HP (19.6 G£/a) than in the DH (+ 42%) and H2 (+ 16%) architectures. Storage and network account for 22.8 G£/a in DH, and 20.9 G£/a in H2. In total, intermediate technologies amount to 29.3 G£/a in HP, 27.8 G£/a in DH, and 37.9 G£/a in H2. With regards to renewables, the H2 architecture is the costliest (46 G£/a), because of the greater need for primary energy. This results in the highest total system cost (102.6 G£/a), which is ~ 26% more expensive than HP and ~ 40% higher with respect to DH (Supplementary Table 2). Specifically, both capital and O&M system costs are higher in the H2 architecture than in the other two core scenarios (Supplementary Table 3).
Although total heat-related costs for consumer and intermediate heating technologies are similar across the core architectures (around 33 G£/a), in HP these are mostly allocated to consumers (23G£/a), in DH to district heating schemes (23G£/a), and in H2 equally distributed between consumer and intermediate conversion (Fig. 6). However, H2 incurs a 270% higher cost for electricity supply for heat (52G£/a) than the other core scenarios (14 G£/a), because of its higher electricity consumption (58%, as opposed to 24% in DH and HP) necessary for hydrogen production. Consequently, the total heat cost in H2 reaches 83 G£/a, which corresponds to 81% of the total system costs, which is more than the 47 G£/a of DH and HP (representing about 60% of the total system cost) (Supplementary Table 4). These costs correspond to an average cost for heating of 17.8 p/kWh in H2, and around 10.2 p/kWh in DH and HP; and for electricity of 7.3 p/kWh in H2, 7.9 p/kWh in DH, and the highest cost of 9.4 p/kWh in HP.
Overall, we found that ZEESAs costs are predominantly due to capital costs with O&M costs accounting for 34% of total costs, which are the highest in the H2 architecture due to high electricity demand; system costs are similar for the DH and HP scenarios, although in HP they mostly bear on the consumer side.
In this study, we used a methodology that combines a meteorology-driven whole system modelling approach with hourly simulations to understand the requirements and operation of future ZEESAs for the United Kingdom. We focussed on scenarios for heating, one of the most challenging energy consuming sectors to decarbonise. Before describing the advantages of our study, there are a couple of caveats that must be considered. The architectures were designed manually, though informed by other optimisation work. To reduce computational time, we opted for a simplified but functional representation of the international European network through a star topology, to simulate the electricity flows among countries. The algorithm controlling hourly energy flows within a node and between nodes aims to maximise the use of zero emission supplies for a given system design. This algorithm is one of the most complex parts of the model.
One of the main advantages of ESTIMO is to model both demand and supply using long-term historical weather data, thus providing valuable insights on the long-term dynamics of renewable-based systems that many investment modelling approaches (e.g. the TIMES family of models) cannot yield accurately. We demonstrate the importance of using long time simulations for capturing meteorology and climate variability and quantifying impacts on heat demands and storage; our results indicate that there is an inverse linear relationship between increasing ambient temperature and energy consumption for heating.
Furthermore, our integrated system approach allows to evaluate the implications of components on the whole system and, therefore, to estimate the technology requirements accurately. In particular, we show that future systems will need to store energy in the order of tens of TWh to cope with demand driven by extreme weather events; storage has a dual role to meet small demands frequently and high peaks more rarely. Moreover, we highlight the non-linear inverse relationship between renewable capacities and storage size, with the balance between these depending on the relative costs. We analysed 35 years of weather data, but there may be more extreme meteorology in the future and to provide resilience against this, options such demand reduction or generators using stored fossil fuels or scarce biomass could be deployed. However, if fossil fuels, even with carbon capture and storage, were used regularly then emissions would have to be balanced with environmental carbon capture.
We have calculated the costs of all components, but there are uncertainties particularly relating to network costs: for hydrogen there no large-scale networks and consumer adoption. A key insight of our study of heat decarbonisation within a zero-emission energy system is that heat pump technology represents the most cost-effective solution, thanks to its efficiency and therefore low electricity consumption. An additional advantage is they can be reversible and used for cooling. District heating has lower heat costs than consumer heat pumps, because of its scale economies, and wider role in system management. In contrast, a heating with electrolytic hydrogen boilers leads to higher total system costs and to heat related costs than architectures dominated by consumer or district heat pumps due to its greater requirements for renewable capacity. This means a faster rate of renewable installation is required, with implications for the difficulty of reaching the net-zero target mandated by the UK Government (Climate Change Act 2008, Order 2019). Technology improvements can be expected in the future in terms of cost and performance. Some future costs are very uncertain, particularly for batteries, wind and solar generators: the past decade’s reduction of 50–80% in the costs of these, with further reductions forecast. The capacity factors of new, larger offshore wind turbines, currently estimated to be around 60%, have a large impact on storage need. Our results show that increasing UK interconnections to Europe would reduce storage need.
This is rapidly changing the optimum least cost balance between technologies. The consequent reduction in heat pump heat costs reduce the cost-effectiveness of deep building retrofit. Consumer heat pumps and hydrogen are inflexible in that there is only one last conversion pathway to heat. District heating, in contrast, has a range of possible heat production components – heat pumps, storage, CHP, boilers, solar and geothermal – which mix can be changed hour by hour and is evolvable as the mix of component capacities can be changed across the years without physical impact on consumers. Moreover, district heating can facilitate competition between heat providers and thereby reduce costs for consumers.
To conclude, our findings indicate that a zero-emission energy system for the UK based on renewables would need efficient technologies, such as heat pumps, as well as system stores, to resiliently manage emissions during rare meteorology periods. Reducing emissions from heat is a high priority since aviation and some industrial processes are more difficult to decarbonise. The speed of implementation is important to reduce emissions as rapidly as possible and thereby the cumulative emission to 2050. We have not reported here the required rates of installation of the various components to achieve net zero by 2050, nor we have modelled the required social capacity (labour force, financing) to meet installation rates. These data would be needed to detail explicit policies to reach targets, such as mandating replacing gas boilers with heat pumps or renewable capacity auctions (Supplementary notes 2). Future developments of ESTIMO will prioritise the addition of: (1) district cooling, to explore its costs and synergies with heating; (2) sustainable fuel supply for aviation, to calculate the sector emissions and primary energy demand; (3) carbon capture and storage, to provide a carbon feedstock for synthetic fuels and to balance residual emissions; (4) an optimisation algorithm. These improvements will enable us to explore a larger set of scenarios and address additional questions, for example assessing the consequences of undersized storages on biofuels consumption or evaluating the trade space of alternative architectures.