1 Introduction

Scientific projects need to considerably lower their environmental impacts to contribute to a society that can flourish without surpassing planetary boundaries. In the field of ground-based astronomical observatories, the median carbon intensity per paper published reaches 24 t CO2 equivalents (CO2e), including both the construction and operation of the observatory (Knödlseder et al. 2022). In comparison, the carbon footprint per capita in 2019 amounted to 4.8 t CO2e in Chile and to 6.8 t CO2e in the European Union (EU) (World Bank n.d.).

The operational emissions of remote and off-grid telescopes primarily stem from their reliance on fossil fuel generators to meet their energy needs. This demand of mainly electrical energy is needed to power the dish motors and supply the cryogenic cooling for the instruments. In recent years, first measures have been taken to address this issue in astronomy and some observatories integrated photovoltaic arrays (PV) into their energy systems, which lowers their direct greenhouse gas (GHG) emissions during sunny hours (Rodríguez 2022). However, off-grid energy systems worldwide are overly reliant on fossil fuels for their power generation, as seen in remote settlements on Svalbard in the Arctic, in deserts in the Middle East, or in rural locations in Sub-Saharan Africa (Ringkjøb et al. 2020; Ghenai et al. 2020; Olabode et al. 2021). This reliance must change to mitigate global warming.

The Atacama Large Aperture Submillimeter Telescope (AtLAST) project, which is designing a new transformational telescope to be built on the Chajnantor plateau in the Atacama Desert in Chile, is pioneering a shift towards sustainable practices. It is the first astronomical observatory project that includes plans for an off-grid, completely renewable energy system right from its design phase (Klaassen et al. 2020; Ramasawmy et al. 2022). The results of this project will inform future efforts in other remote locations.

In our previous techno-economical analysis of AtLAST’s energy system, we conducted a comparison between a business-as-usual scenario with conventional diesel generation and various combinations of PV, hybrid energy storage systems (HESS), and diesel generators over a 25-year lifetime (Viole et al. 2023b). The AtLAST telescope, which will have a 50-m diameter primary mirror and will operate at both day and night, has an expected daily energy demand of 22 MWhe, which exhibits relative stability on an intraday basis but varies seasonally due to the addition or removal of instruments. Considering cost estimates for the projected construction year of 2030, the system with the minimal costs identified in Viole et al. (2023b) comprised a 4.8 MWp PV park, 14.2 MWh of lithium-ion batteries, a hydrogen system including a 0.34 MWe electrolyzer, a 0.39 MWe fuel cell, and a 160-kg hydrogen storage tank. Diesel generators were included as backup to meet approximately 5% of the energy demand. This system achieved a 95% reduction in direct GHG emissions, measured in CO2e, compared to relying on diesel generators only. To achieve 0% direct CO2e emissions, the PV park would need to be expanded by 42% to fuel the larger storage needs for green hydrogen production and battery charging.

1.1 Life cycle assessment (LCA) for renewable energy systems

Building the larger PV park with zero direct emissions increases the environmental impact of the production phase of renewable energy–based systems (RES). A techno-economical energy system optimization that solely takes direct GHG emissions from the operation phase into account paints a slanted picture of the CO2e emissions, giving the appearance that 100% RES would always be favorable from an environmental point of view. To counter this, we can add or include a LCA onto the energy system optimization. Previous applications of ex post LCAs on national energy system models incorporating RES have generally shown that higher shares of RES reduce climate change impacts but lead to increased mineral depletion (Hertwich et al. 2015; Berrill et al. 2016). A literature review on electricity generation LCAs found that most studies identify RES to have lower life cycle GHG emissions compared to their fossil counterparts (Barros et al. 2020). Most of the reviewed studies relied on the ecoinvent database for their life cycle inventories (LCI), that is, the underlying datasets used to receive the final environmental impact calculation. These ecoinvent LCIs have a global scope, which can distort results for specific locations, and are not regularly updated. For instance, the datasets for PV cell production in ecoinvent 3.7.1 (such as “Photovoltaic cell, single-Si wafer {RoW}| production | Cut-off, U”) are based on data that are two decades old. Tannous et al. (2019) found that environmental impacts of PV production have declined by 21–50% compared to ecoinvent 3.4 datasets from 2005, which they associate with technological advancements in electricity consumption and manufacturing processes. Therefore, it is crucial for the research field to strive for better data quality by utilizing local characteristics and more recent datasets published since 2015.

Local characteristics are particularly relevant in stand-alone or off-grid systems, such as the one discussed for AtLAST. These small- to mid-scale energy systems are influenced by the weather conditions of one specific zone, which amplifies the intermittent nature of RES generation. Additionally, they have limited or no connection to a centralized grid that could help balance the systems. LCAs on remote energy systems have been conducted, e.g., in Marsabit, Kenya; Alaska, USA; and Ginostra, Italy (Bilich et al. 2017; McCallum et al. 2021; Gandiglio et al. 2022). These studies employed different approaches, ranging from cradle-to-gate analyses to including end-of-life scenarios. They compared predefined RES, such as hydropower; PV paired with batteries; or PV, batteries, and hydrogen storage, with the previously used diesel generators in the respective areas. The findings consistently showed that the implementation of RES systems reduced the GHG impact by 89–99%. In Gandiglio et al. (2022)’s study, the renewable scenario based on PV and lithium-ion (Li-ion) batteries decreased CO2e emissions by − 88.6% over the 25 years lifetime, while mineral and metal resources increased by +80%, and the water use by +61%. McCallum et al. (2021) studied the deployment of hydropower paired with Li-ion batteries in remote Alaska, and suggested deployments of 10 or 20 years to reduce the impacts in water and metal depletion, which are linked with the metal used in the system. Bilich et al. (2017) found their PV and battery setup to significantly decrease the climate change environmental impact compared to home diesel gensets, however did not consider mineral resource depletion or water use. They noted that setting up such a PV and battery system for a rural Kenyan community could result in load-shedding issues and suggested mitigating this by installing a backup diesel generator.

Off-grid energy systems face unique challenges, as they cannot rely on a national electricity grid to balance fluctuations in power generation from intermittent RES. Optimizing off-grid energy systems with high shares of RES presents two main issues. Firstly, it can result in more frequent load shedding, which is not feasible for operating a telescope or other critical applications. Secondly, it often requires oversizing components in combination with batteries to buffer out the variability in supply, including unfavorable weather periods. As a solution, fossil fuel–based generators are commonly added as a backup alongside PV and batteries, as they offer lower levelized costs of electricity (LCOE) by avoiding the high capital costs of surplus RES capacities (Odou et al. 2020; Come Zebra et al. 2021). In a review of off-grid energy systems in developing countries, Come Zebra et al. (2021) found that energy systems combining RES and diesel generators are financially more attractive compared to 100% renewable systems.

To enhance the reliability of off-grid RES-based systems and minimize the need for excessive built capacity, another option is the integration of HESS. These systems combine multiple energy storage technologies, such as batteries for short- to medium-term energy storage, with a hydrogen system for medium- to long-term energy storage (Bocklisch 2015). Many off-grid energy system optimizations propose the usage of HESS and some real scale tests have been conducted already, such as by Endo et al. (2019) and at the Sir Samuel Griffith Centre (HyResource 2020), and are also under consideration for the telescope project AtLAST. To our knowledge, no LCA comparison between HESS to systems relying on one energy storage has been conducted thus far in either grid-connected or off-grid setups.

We identify a number of research gaps in literature: First, most studies on specific energy system case studies use global LCIs, while the quality of the LCA could be improved by using temporal- and site-specific datasets. This especially is relevant concerning the electricity mixes of the production locations of RES components, as, e.g., the carbon intensity of this production has a major influence on the life cycle environmental impact of the systems considered. Second, to our knowledge, there is no LCA on off-grid energy systems that includes LCIs as of 2020 that are in line with recent technological developments. Further, LCAs on islanded or off-grid energy systems have so far mainly been carried out comparing one or two RES options to a business-as-usual fossil scenario. A broad comparison of different renewable systems, such as those with HESS to those only using batteries, or a comparison of 100% RES scenarios to high-renewable scenarios with mostly RES and some fossil generation backup is lacking and can paint a more faceted picture of the environmental impacts.

1.2 Designing a site- and temporal-specific LCA study for a telescope in the Atacama

In this work, we contribute to close these gaps by (1) conducting the first LCA on off-grid energy systems with recent LCIs and projections of power mixes in the production countries for 2030, putting an emphasis on site-and temporal-specific inputs and (2) considering five different scenarios of implementing RES into the energy system for AtLAST, 100% RES to high-renewable systems, and batteries with HESS. This allows us to identify trade-offs between the use of different components.

Specifically, the objective of this work is to find the environmental impacts of the off-grid renewable energy systems that can supply the 22 MWhe/day and 7.7 GWhe/year energy needs of AtLAST. We conduct a comparative life cycle assessment (LCA) to evaluate three impact categories: (1) depletion of mineral resources, (2) water use, and (3) climate change impact both from direct and indirect emissions. This study examines the cradle-to-gate environmental impacts of different energy system configurations for the telescope in question, combining an ex post LCA with the findings from the highRES-AtLAST energy system optimization model, developed by Valenzuela Venegas (2022) (see methodology flowchart in Fig. 1). We adopt a specific temporal and spatial scope in this case study. Considering that the AtLAST energy system is projected to be constructed in the 2030s, we set the year 2030 as the basis for component production and implement future power mixes of production areas accordingly. We determine specific likely production areas and include the transportation distances to the location of AtLAST’s energy system to the environmental impact calculation. By utilizing LCIs from publications since 2015 and regionalizing our LCA inputs based on assumed supply chains and electricity mixes in production countries in 2030, we depart from a black-box approach. We answer the following research questions:

  1. 1.

    Which system has the lowest environmental impacts in the three impact categories considered?

  2. 2.

    What are the trade-offs between these three environmental impacts?

  3. 3.

    What is the regional distribution of these impacts?

Fig. 1
figure 1

Flowchart of methodology, inputs to the ex post LCA developed in this study in green (energy system designs, regional and temporal inventories), outputs of the LCA in blue

Full size image

2 Materials and methods

2.1 Goal and scope definition

This comparative LCA aims to evaluate the environmental impacts of different off-grid energy system configurations for powering the new telescope AtLAST on the Chajnantor plateau in Chile. This plateau is the best site to perform sub-millimeter observations from ground, and as such it hosts more than ten international sub-mm/mm observatories, including the most powerful in the world. AtLAST is planned to be located near the Atacama Large Millimeter Array (ALMA) and the Atacama Pathfinder Experiment (APEX) at approximately 5000 m above sea level (Fig. 2). At stakeholder workshops held with telescope operators and the municipality of the area (Valenzuela Venegas et al. 2023), questions regarding the life cycle greenhouse gas impacts of the off-grid energy system, especially in regard to transport of maintenance personnel and of components to the site, were raised. Further questions arose regarding the water use of the hydrogen system proposed by Viole et al. (2023b) in the Atacama Desert. This paved the way for this work.

Fig. 2
figure 2

Map of the Chajnantor area with a potential site of AtLAST connected to the planned energy system site via a power line (in black), surrounded by the telescopes ALMA and APEX, and the city of San Pedro de Atacama; elevation between the two sites is shown in the bottom graph

Full size image

This LCA study targets the decision-makers for the new telescope AtLAST, next to energy system modelers, astronomers, and off-grid system operators interested in RES implementation. We follow the ISO14040:2006 and ISO14044:2006 standards (Technical Committee ISO/TC 207 2006; Technical Committee ISO/TC 207/SC 5 2006) and employ a comparative, attributional cradle-to-gate LCA. The aim is to identify the key activities within the life cycle of the off-grid energy systems and their components that contribute the most to environmental impacts (Guinée et al. 2018). Given the scope of the attributional assessment, we do not expect the decisions made for the telescope’s energy system to have significant far-reaching consequences that would require consequential modeling. This may differ when conducting an LCA on a national or international energy system. The functional unit is defined as delivering annual power of 7.7 GWhe, based on the estimated demand curve of the telescope, over a lifetime of 25 years. The start of operation is defined as 2030. Techno-economic optimization details regarding the energy system, including the telescope’s demand, component choices, and timeline, can be found in Viole et al. (2023b). Figure 3 shows the system boundary, which includes the production of components, their transport to the energy system site, and the operation and maintenance of the system. Self-consumption of, e.g., the hydrogen system components which need to power their balance of plant components was included in the efficiencies of components applied in the model highRES-AtLAST (Valenzuela Venegas 2022), and will not be explicitly mentioned in this work. While the system boundary includes a DC/AC transformer, other power electronics that would be needed in a finalized system were cut-off in this LCA due to their negligible impact.

Fig. 3
figure 3

Flowchart of the system boundaries (black line), foreground system with energy system components

Full size image

The end-of-life (EoL) treatment of the power supply system is cut off from the system boundary. In this, we consistently adopt the cut-off model based on environmental significance, for example, the ecoinvent “cut-off” database was used through the entire analysis (see also ISO14044:2006, Art 4.2.3.3 (Technical Committee ISO/TC 207/SC 5 2006)). This entails that the EoL management of system components, which is likely to be recycled in 2050, is cut off from the product systems studied; the impact of recycling process is therefore allocated to the recycled materials and products. Presently, research on recycling and EoL treatments for RES components, batteries and hydrogen systems is insufficient. Mahmoudi et al. (2019)’s review on solar PV systems from 2019 highlighted a lack of LCA research on PV waste management. A recent acceleration in PV recycling research effort was noted by Akram Cheema et al. (2024), as the first generation of PV modules reached their EoL. In energy storage, Pellow et al. (2020) found a lack of addressed EoL options for stationary Li-ion batteries in their reviewed literature, and Valente et al. (2017) observed that most LCAs on hydrogen energy systems employed cradle- or gate-to-gate system boundaries. If the infrastructure of the energy system in this case study will be recycled, which is likely to occur in the far future (2050), we see three options to deal with this in a LCA: (1) cut-off, (2) allocate the impact of the recycling process, or (3) giving credit for recycled material (e.g., in a consequential LCA). To maintain the consistency of our analysis, we cut off the recycling of the materials, so that it is allocated to the life cycle of the second lives of those materials.

Our life cycle impact assessment (LCIA) applies the Environmental Footprint (EF) 3.0 method, implemented in the SimaPro software (v9). From the EF’s 16 environmental impact categories and their respective impact assessment models, we selected the three most relevant for the analysis of a remote energy system such as AtLAST, namely, climate change, mineral resource depletion, and water use (see Table 1 for units and impact assessment models). The first two categories were chosen as current energy policies put much emphasis on climate change and the depletion of mineral resources. The environmental impact of water use is considered due to the location of the proposed energy system in a hyperarid desert. The median humidity at the APEX telescope, close to the potential AtLAST site, was 25% in 2022 (Atacama Pathfinder Experiment (APEX) 2023). The EF 3.0 method includes the AWARE (Available WAter REmaining) regionalized water scarcity footprint (Boulay et al. 2018). AWARE distributes regional weighting factors relating to the local water scarcity, e.g., Chile has a weighting factor of 81.37, Norway 0.76, and factors in China vary between 0.39 in Hunan and 87.07 in Hubei. That means that 1 m3 of water in Norway accounts for 0.76m3 water eq of deprived water in the LCIA, while 1 m3 of Chilean water results in 81.37 m3 water eq.

Table 1 Selected impact categories and models
Full size table

2.2 Product systems and sizes: Energy system optimization

We consider four different energy system components: (1) a diesel generator, (2) a ground-mounted mono-crystalline silicon (mc-Si) PV park, (3) lithium iron phosphate (LFP) batteries, and (4) a hydrogen storage system consisting of an alkaline electrolyzer (AE), a 350 bar compressed gas (CG) hydrogen storage, and a proton-exchange membrane fuel cell (PEMFC). Different system compositions of these components can provide the energy required by the telescope according to the six scenarios described in the following. The component selection follows the technological feasibility on and nearby the Chajnantor plateau, as discussed in Viole et al. (2023b)’s techno-economic optimization. Here, we make some simplifications by (a) only including on-site hydrogen production and (b) focusing on systems located at a valley site of 2500-m altitude, at 23°02′S 68°06′W, supplying the telescope on the plateau via a 43-km-long subterranean power line (see Fig. 1). We exclude the option of an energy system at the telescope’s site at 5000 m for operational reasons, such as the easier access for maintenance, less probability for snowfall, and more available building space.

The sizing of system components follows the GAMS-based techno-economic optimization highRES-AtLAST (Valenzuela Venegas 2022). Key features in the linear optimization are:

  • Six scenarios allowing different system components to be built by the optimization model:

    1. (a)

      Reference: Diesel generators

    2. (b)

      PD: PV and diesel generators

    3. (c)

      PDB: PV, diesel generators, and batteries

    4. (d)

      PB: PV and batteries

    5. (e)

      PDBH: PV, diesel generators, batteries, and hydrogen technology

    6. (f)

      PBH: PV, batteries, and hydrogen technology

  • The annual demand of the telescope, divided in 1-h segments, is met without load shedding.

  • To model hourly PV capacity factors, we apply the weather year 2020 from the ERA5 Reanalysis (Hersbach et al. 2020).

  • We use 2030 as the year for costs and electricity demand.

  • The optimization equation minimizes the annualized total costs of the system composed of investment and dispatch costs.

The resulting capacity and generation are reported in Table 2 and Fig. 4. The scenarios PDBH and PDB have the lowest LCOEs, while scenarios PB and PBH have the lowest direct GHG emissions. These direct emissions consist only of the diesel burned in the generators on site, applying an emission factor of 0.86t CO2e/MWh (Gaete-Morales et al. 2018). PB and PBH scenarios require five times the amount of capacity installed compared to the reference scenario and generate more than double the electricity. The system curtails generation from PV when the solar generation exceeds the demand and the storage options.

Table 2 Overview of compared scenarios, sizing optimized with highRES-AtLAST (Valenzuela Venegas 2022)
Full size table
Fig. 4
figure 4

a Annual generation (left axis), and levelized costs of electricity (LCOE, right axis); b capacities installed of the six compared systems setups optimized with highRES-AtLAST to supply 7.7 GWhe/year to the telescope

Full size image

2.3 Compilation of the life cycle inventories (LCI)

In this LCA, we include ground-mounted, mc-Si PV panels, LFP batteries, AE, CG hydrogen storage, PEMFC, and diesel generators, as the system components, connected to the telescope via subterranean AC power lines. The component selections in this LCA are driven by economic considerations while also taking into account their mineral resource impacts. We chose Li-ion over lead acid batteries due to their longer cycle life and reduced need for replacements. Within Li-ion battery chemistries, we opt for LFP batteries based on the findings of Yudhistira et al. (2022), which showed lower impacts on climate change and mineral resource depletion compared to other battery types like NMC (Nickel Manganese Cobalt) and lead acid ones. LFP batteries’ cathodes consist of ~ 8% lithium, with the remaining composition mostly made up from iron and phosphorus (International Energy Agency 2022a). In the hydrogen system, we consider the more economical AE over PEM electrolyzers. For hydrogen storage, we apply CG due to its cost-effectiveness. Other storage options include liquid hydrogen, which is more expensive, underground storage (e.g., in salt caverns) better suitable for larger storage volumes, and storage in metal hydrides, not widely available yet (Egeland-Eriksen et al. 2021).

2.3.1 Production of system components

For the production LCIs, we use inventories published after 2015 wherever possible, adjusted to the temporal and regional scope of the work. In PV production, we refer to Frischknecht et al. (2020)’s LCIs, modifying the background processes to ecoinvent 3.7.1 datasets. This includes the production of PV cells, made from silicon wafers based on Czochralski single silicon crystals. PV modules are formed by connecting multiple PV cells. Regarding LFP batteries, we relied on Porzio and Scown (2021)’s LCI, which included a LiFePO4-based cathode and a graphite-based anode. Additionally, we employed Quan et al. (2022)’s inventories for the production of LiFePO4 from lithium carbonate and iron phosphate.

In the hydrogen system, LCIs are based on Koj et al. (2017) for the AE with a Zirkon membrane, on Boureima et al. (2011) and Wulf et al. (2018) for the steel-based CG hydrogen storage tank, and on Notter et al. (2015) and Weber et al. (2018) for the PEMFC with a Nafion membrane. The diesel-electric generating set and background datasets are taken from the ecoinvent database (v. 3.7.1). Detailed activity levels are available as an open-source dataset in Viole et al. (2023a).

Setting the temporal and regional scope

For this case study, we choose specific production locations based on production assumptions for 2030. These locations impact the water use, power mixes, and shipping distances. For PV panels, the majority of shipments to Latin America in recent years originated from China, specifically Jiangsu and Xinjiang regions (International Energy Agency 2022b). It is not possible to determine the exact ratio between the two supplying regions. Therefore, we allocate these regions in a 50:50 proportion as our sourcing locations for PV panels. In batteries, China currently accounts for three-quarters of the global Li-ion production and is expected to maintain a 70% production capacity share through 2030 (International Energy Agency 2022a). Cathode production for LFP batteries in China is distributed across various provinces, including Hunan, Yunnan, and Hubei. We include these three regions as equal sourcing locations for the batteries in this work. For hydrogen systems, manufacturers today are in the process of scaling up their capacities. Europe houses some notable MW-scale AE producers, such as NEL Hydrogen in Norway, Sunfire in Germany, and John Cockerill in France, while China with AE producers like Sinopec leads the global list of AE electrolyzer production in 2023 (International Energy Agency 2023; John Cockerill; Nel Hydrogen; Sunfire). PEMFC production of stationary fuel cells applicable for AtLAST’s energy system is conducted by Plug Power in New York State, USA, and by PowerCell in Sweden (Plug Power 2022; PowerCell 2022). In our LCA, we implement Herøya, Norway, as the production location for AE and Slingerland, NY, USA, for PEMFC. Local CG storage production is available in Chile. We accounted for 43 km of 24 kV underground power lines to connect the power plant at 2500-m altitude with the telescope at 5000 m, and adapted Arvesen et al. (2015)’s LCI for 22-kV subterranean distribution lines to a production facility in Chile.

We include region-specific electricity mixes for 2030 in the LCIs. For the Chinese production sites, we use a global Chinese power mix following energy system optimizations from Zhang et al. (2022). We could not use regionally varying power mixes for the different Chinese production locations due to the continuous increasing of the Chinese targets of renewable shares. Regional power mix calculations like Shen et al. (2019) fall far short of the most conservative assumptions in renewable expansion by Zhang et al. (2022); hence, we include the latter country-wide national power mixes, which assume a carbon-neutral power generation share of 54% in 2030, including RES and nuclear power. For the Chilean and US-American power mixes in 2030, we follow the national outlines of RES integration for 2030 based on Ministerio de Energía, Gobierno de Chile (2020), and, for Norway, we adapt DNV (2022)’s 2030 forecast.

To understand local circumstances relevant for AtLAST, the authors visited APEX, a 12-m single-dish telescope on Chajnantor, in 2022. Learnings like common practices in PV cleaning and staff commuting are fed into the assumptions for the energy systems’ LCA, which we shall discuss in the next section.

2.3.2 Transport of components to site, operation, and maintenance

We estimate transportation routes between the production places and the energy system’s site based on common shipping routes and streets, where goods are mainly shipped to the Antofagasta port and then lorried to the site (see Table 6 in the “Appendix”). We apply ecoinvent 3.7.1 datasets to integrate transportation by train, lorry, and containership. The operation and maintenance of the energy system for AtLAST include the fuels burned by the system’s components during operation, the travels of the maintenance personnel to and from the site, the PV cleaning, and component replacements.

Maintenance workers

To estimate travels of the maintenance crew, we conducted qualitative interviews with observatory operators from APEX during 2022. Scaling up the operations of this 12 m to our prospective 50-m-dish telescope, we foresee the need for three workers on duty for maintenance of the system, with two in day-shift, and one in night-shift. In the operation of telescopes in the Atacama, personnel typically fly in for 8-day shifts, after which they fly home and are off-duty for 6 days. In the LCI for the maintenance workers, we include one return-trip from Santiago de Chile to the energy system site per week per worker. We account a third of the energy system workers’ shift and commuting emissions towards the energy system, as these engineers hold other duties, e.g., technical handling of astronomical instrumentation.

Replacements

Further in maintenance, we include the necessary replacements over the system’s lifetime, as given in Table 3. For the PV degradation, we choose a higher value than the commonly used 0.5–0.8%/year, as we are operating in arid desert conditions, which leads to soiling. Additionally, high ultraviolet (UV) irradiation is present in the Atacama, typically ~ 40% above the annual UV-B doses in Northern Africa (Cordero et al. 2018). High UV exposure can lead to encapsulant discoloration, which lowers the output power. Field studies in deserts report degradation rates ranging from 0.74 ± 0.29%/year for poly-crystalline silicon (pc-Si) panels under maritime desert conditions (Hassan Daher et al. 2022), 0.15–2.83%/year for pc-Si in the Atacama (Vásquez et al. 2021), 1.89%/year for mc-Si panels in the Sahara (Bouraiou et al. 2015), to 1.25–4.22%/year in pc-Si panels operated in a hot semi-arid climate (Bansal et al. 2022).

Table 3 Lifetime and needed replacement per 25 years of system components
Full size table

We implement 1.56%/year of degradation as the average over these studies’ median values. This requires adding 8% of total PV capacity every 5 years to offset the degraded modules. In the lifetime for the batteries, AE, and PEMFC, we attribute 290 cycles/year for the batteries, 3400 operation hours/year for AE stacks, and 4000 h/year for PEMFC and follow the US DOE’s estimated component lifetimes (Viswanathan et al. 2022).

Photovoltaic cleaning

In operating PV parks in the Atacama Desert, Vásquez et al. (2021) found soiling as the most frequent problem, since layers of dust on the panels reduce their power output. To reduce soiling, common cleaning options include natural cleaning by wind and rain, manual cleaning with water, and cleaning robots. We follow the declarations of the operators of the PV installations of the Energy Cooperative of San Pedro the Atacama (CESPA) to assume that PV Arrays are cleaned twice a year, and include biannual manual cleaning with mild detergent and solar cleaning brushes in our maintenance LCI, assuming the need for 0.5 L deionized water per m2 of panel cleaned (Kazem et al. 2020).

Water usage of the hydrogen system

Proposing hydrogen systems in the Atacama Desert, including hydrogen production, led to questions related to water use in stakeholder workshops run by the AtLAST project (Valenzuela Venegas et al. 2023). San Pedro de Atacama has access to brackish water with ~ 2.3% salt content. In calculations for the desalination and deionization of the water for the hydrogen system and the PV cleaning, we account for water losses of 27% in the reverse osmosis desalination and 20% in the deionization (GHD 2022; Chululo - Revista Informativa de la Comuna de San Pedro de Atacama 2022). For the proposed hydrogen system, we use 9 L of deionized water to produce 1 kg hydrogen (H2). Further, we assumed leakages in the hydrogen system to calculate the overall water need. In their literature review on hydrogen leakage, Fan et al. (2022) found almost no data on actual hydrogen leakage rates beyond required hydrogen safety management. We include losses in the AE of 1L water/kg H2 produced, 5% H2 lost during compression and storage (a value consistent with Fan et al. and between the losses found ranging from 2 to 10% at a hydrogen fueling station in California (Genovese et al. 2020)), and 13% H2 lost in the PEMFC, mainly in the purging process (assumed based on proposed purging ratio of 7:1 in prior PEMFC simulation for AtLAST (Øien 2023)). Note that these hydrogen losses do not equal the efficiency losses along the hydrogen value chain (see Valenzuela Venegas (2022) for efficiencies assumed in the hydrogen system components).

The water needs for scenarios PDBH and PBH are calculated with 205 and 323 m3/year of deionized water, respectively, or 352 and 553 m3 of brackish water extracted at the local sources, respectively (see “Supplementary information”). These numbers do not justify the need of an own desalination unit in our case; hence, the usage of a local desalination plant with 1.9 kWh/m3 water desalinated is included in the LCI (Antonyan 2019; Cetinkaya and Bilgili 2019), where the energy stems from a local PV and battery park.

3 Results and discussion

We compare the environmental impacts of the six energy system scenarios over a 25-year lifetime (see absolute values in Table 4). To visualize the results, Fig. 5 shows relative values to the maximum value in the impact category. The climate change impacts are set relatively to the CO2e emissions of the reference scenario, which holds most emissions over the system’s lifetime. Scenario PB shows the lowest GHG impact across the life cycle, by solely supplying AtLAST with energy from PV and batteries. At the same time, this scenario has the greatest impact in mineral resource depletion. The highest impact in water use is seen in scenario PBH, which has the second-lowest climate change impact. There are trade-offs between the three impact categories, as each scenario has its own strengths and weaknesses in terms of environmental impact.

Table 4 Life cycle environmental impact associated with 7.7 GWh/year of power delivered to the telescope over a lifetime of 25 years
Full size table
Fig. 5
figure 5

Relative environmental impacts of six compared energy system scenarios to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years, set in relation to maximum value within respective impact category

Full size image

3.1 Life cycle GHG emissions

The scenarios’ GHG emissions can be broken down into the contribution of production, transportation, operation, and maintenance steps (Fig. 6). In the reference scenario, most life cycle GHG emissions are made up by direct emissions during the operation phase, where diesel generates all the power needed by the telescope. Adding PV panels to this system, scenario PD, lowers these direct emissions by 41%. The production phase now includes slightly more emissions, mostly due to the manufacturing of PV panels. Over the considered life cycle, the PD scenario lowers the GHG emissions by 39% compared to the reference scenario.

Fig. 6
figure 6

Climate change impact of the six energy system scenarios to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years, in 103 t CO2e

Full size image

3.1.1 Running diesel as backup—Scenarios PDB and PDBH

Scenarios PDB and PDBH, which use mostly RES and run diesel generators as backup, reduce life cycle CO2e by 87%, compared to the reference scenario. The detailed distribution of relative GHG emissions in scenario PDB is shown in Fig. 7. The diesel burned to supply 7% of the annual demand in this scenario makes up 50% of this system’s emissions, while the production of the RES, PV panels, and batteries, supplying the other 93% of the telescope’s demand, account for another 47% of the total emissions. The scenarios PDB and PDBH reach life cycle carbon intensities of 0.115 and 0.113 kg CO2e/kWh, respectively (Table 4). Over the 25-year life cycle, the GHG emissions for replacements, which make up 15% of the life cycle GHG emissions, are about half the size of the initial installation’s GHG emissions. Both the transport of the components from production sites to the Atacama location and staff commuting of the maintenance personnel have a rather miniscule influence.

Fig. 7
figure 7

Relative contributions to the life cycle climate change impact on the PDB (PV, diesel and batteries) scenario to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years vs the total of 100% climate change impact on the right side

Full size image

3.1.2 Systems with 100% RES—Scenarios PB and PBH

Scenario PB, one of the two 100% RES scenarios, has the overall lowest life cycle GHG impact, consisting of 0.077 kg CO2e/kWh supplied. Compared to PDB, we see an increase in emissions from PV and battery production by 40% and 51%, respectively (Fig. 6). This is interlinked with the higher capacities built in this scenario, as the PV and battery system must supply the telescope throughout the given weather year, that is also during multiple days of cloudiness, without having a diesel generator as backup. However, the upstream emissions from these larger capacities do not exceed the evaded CO2e emissions from diesel burned. Hence, this scenario’s life cycle GHG emissions are the overall lowest in this study.

When we add a hydrogen system into the mix, that is scenario PBH, slightly higher GHG emissions are present, as the production of a second energy storage technology has to be included. At 0.080 kg CO2e/kWh, PBH is the scenario with the second-lowest life cycle GHG emissions out of the six scenarios compared.

3.2 Water use

In the renewable energy systems’ life cycle water use, a high impactor is the PV panel production, accounting for 1.3 (PD) to 3.5 (PBH) million m3 water eq (see Fig. 8). This relates to 44–57% of the scenarios’ total water use. Moreover, the battery production makes for 1.8 (PDBH) to 3.7 (PB) million m3 water eq, a share of 31–49% of the scenarios’ total water use. In the upstream processes, especially the Czochralski process in the PV production and the phosphoric acid production for the LFP cathode material push up the water use. The water use on site in the Atacama during operation and maintenance, depicted with an “o” hatch in Fig. 8, mainly comprises water for the hydrogen system. This accounts for 0.4–0.6 million m3 water eq, that is, 16,500–26,000 m3 water eq/year. Compared to the water used in PV cleaning, the hydrogen systems have about 15 times the impact. The water use for the hydrogen system’s operation had a share of 7–8% in the systems where hydrogen was included.

Fig. 8
figure 8

Water use of six energy system scenarios to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years, in 106 m3 water eq

Full size image

3.3 Mineral resource depletion

In mineral resource depletion, we find a positive correlation between the size of RES capacities built and the associated environmental impact (see Fig. 9). The relative mineral resource flow of scenario PDBH is depicted in Fig. 10. Most of the mineral resource impact stemmed from the mining of copper and silver. Copper is needed in the production of batteries and PV panels, as it is used in the current collector of the battery’s anode and the wires and cell interconnections of solar panels. PV additionally requires silver as part of the metallization paste applied on the wafers for electric contacts. Hydrogen systems contribute to the mineral resource depletion with the use of platinum in PEMFCs. These used resources are weighed to kg Sb eq in the EF 3.0 method with factors of 1.37⋅10−3 (copper), 2.22 (platinum), and 1.18 (silver). Lithium, with a weighting factor of 1.15⋅10−5, accounts for only up to 0.013% of the mineral resource depletion in the compared systems. The share of battery and PV production far outweighs any other influence on the mineral resource depletion.

Fig. 9
figure 9

Mineral resource depletion of six energy system scenarios to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years, in kg Sb eq

Full size image
Fig. 10
figure 10

Relative contributions to the mineral resource depletion of the PDBH (PV, diesel, batteries, hydrogen system) scenario to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years vs the total of 100% mineral resource depletion on the right side

Full size image

According to the EU’s foresight study on critical raw materials, the materials used in AtLAST’s energy systems are categorized from low to moderate supply risks. Natural graphite, which is not considered in EF 3.0’s mineral depletion assessment, is ranked with a moderate supply risk. Lithium, titanium, and silicon metal have low supply risks, while nickel and copper have very low risks. Although materials like lithium are not currently considered scarce, possibilities of future scarcities emerge as demand continues to grow. The EU has projected a significant increase in lithium demand, estimating a multiplication factor between 8 and 18 from the year 2020 to 2030 (European Commission. Directorate General for Internal Market, Industry, Entrepreneurship and SMEs 2020). Compared to this general overview of supply risks, the resource and mineral depletion impact assessment of this work showed rather low numbers. It is however important to note that emerging scarcities in graphite and lithium could potentially alter these results for batteries in the mid-term future.

3.4 Trade-offs in combining energy system components

Trade-offs between the assessed environmental impacts are unavoidable. The reference scenario with the overarching highest CO2e emissions had the smallest impact in water use and mineral depletion. Scenarios that significantly decrease the carbon intensity of the power supplied to the telescope entailed the usage of energy storage systems, either batteries or a HESS with batteries and a hydrogen storage system. Scenarios with solely batteries resulted in slightly higher mineral depletion, those with HESS had higher water use. The biggest impactor in mineral resource use and water use was the PV park size. Its initial capacities ranged between 2.5 and 6.5 MWp, depending on scenario setup, where the larger capacities were present in 100% RES scenarios.

3.5 Global distribution of environmental impacts

While this work compares the life cycle impacts of an off-grid energy system to operate in Chile, the sourcing locations of minerals and the impacts during the production and transportation phase are distributed globally. Figure 11 shows the global distribution of climate change and water use impact, and Fig. 12 maps out the shares of mineral resource depletion. Note that the global water use in the reference scenario resulted in a negative value, as in the background data some water is extracted on country-level, but released back on a global/rest-of-the-world level.

Fig. 11
figure 11

Global distribution of a climate change and b water use of the six energy system scenarios to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years

Full size image
Fig. 12
figure 12

Map of global distribution of mineral resource depletion of the six energy system scenarios to supply 7.7 GWhe/year to the telescope over a lifetime of 25 years, bars from left to right: Reference scenario, PD, PDB, PB, PDBH, PBH

Full size image

Comparing the global distribution of environmental impacts, we can see a shift in impacts depending on the share of RES. The highest climate change impact in Chile takes place in the reference scenario, stemming from the diesel burned in the generators. The less diesel we use in the energy systems, the lower the Chilean climate change impact. Apart from this, GHG emissions are mainly accounted as rest-of-the-world and global (both depicted as global in Fig. 11), that is, without a specific country affiliation. In water use, the reference case mainly accounts the upstream of the diesel production, e.g., in the Middle East. When using systems with RES, the water use in energy system component production globally and in China rises. We can clearly see the impact of the hydrogen system with its distributed water use in Chile in scenarios PDBH and PBH, while scenarios with PV and batteries solely have the water use for the PV cleaning to account for in Chile. The mineral resource depletion of the compared energy system scenarios follows the underlying share of sourcing locations for minerals in ecoinvent. Big shares in this environmental impact category are held by Chile, which serves as a source for copper and silver, China for copper, and South Africa and Russia for platinum. Copper, which had the bear share in this impact category in this study, is further sourced from Russia, Australia, and North America, amongst others.

3.6 Sensitivity analysis

In the comparative LCA conducted, we found two important contributors to the climate change impacts: (1) the assumed 2030 power mixes in the Chinese production areas of PV panels and LFP batteries and (2) the battery replacements, the impact of which could be sensitive to the replacement rates.

To test the sensitivity of our results given different power mixes in 2030 China, we change them towards a slower and a faster transition towards carbon-neutral power supply (includes both RES and nuclear generation). The lower boundary is set with a share of 35.5% in carbon-neutral generation, applying the unaltered Chinese power mix from 2021 (China electricity council 2023). Our reference case in the study (called reference transition in Table 5) had applied a share of 54% (Zhang et al. 2022). The upper boundary is set with a power mix where GHG emissions in China peak in 2025, with a carbon-neutral generation share of 64.9% (Zhang and Chen 2022). We find that by applying these power mixes, the life cycle GHG impact of our energy systems deviates by −2.8%/+6.0% (see Table 5). Changes in the power mix affect scenario PB the most, as it has the highest battery capacity built.

Table 5 Sensitivity analysis of life cycle GHG emissions depending on different Chinese power mixes in 2030, associated with 7.7 GWh/year of power delivered to the telescope over a lifetime of 25 years
Full size table

Continuing with testing the sensitivity of the battery-related assumptions in this work, we change the battery replacement rates in the LCA. These replacements were a main contributing factor in all three environmental impact categories. We had set the original replacement rates of the LFP batteries at every 2640 cycles and 2.7 replacements over 25 years (see assumptions in Table 3, Visnawanathan et al. (2022)’s assumption for 2030). In studying their sensitivity, we change these rates to the lower boundary at every 15 years and 1.7 replacements over 25 years (Cole et al. (2021)’s not-cycle specific lifetime), and the upper boundary at every 2400 h and 3.3 replacements over 25 years (Visnawanathan et al. (2022)’s assumption for 2021).

Figure 13 shows the relative changes in environmental impacts given the boundary replacement rates. We see changes in the climate change impacts by up to −13%/+5%, with scenario PB again affected the most, as it contains most battery capacity. Different replacement rates in batteries have the strongest effect on the mineral resource depletion, with relative changes up to −30%/+10%. The water use deviates up to −19%/+7%.

Fig. 13
figure 13

Sensitivity analysis of battery replacement rates, relative changes in climate change impact, water use, and mineral resource depletion with either 1.7 (lower boundary), 2.7 (baseline), and 3.3 replacements (upper boundary) of the batteries over the energy system’s lifetime of 25 years

Full size image

3.7 Comparing the carbon footprints of the energy system with the construction of the 50-m-dish telescope

The climate change impacts of space activities and astronomy have come to the foreground in recent years, for one with recent touristic space missions of the very rich, but also with debates around the carbon footprint of astronomers, whose need for telescopes results in median carbon intensities of 22 (space observatories) and 24tCO2e (ground-based observatories) per scientific paper (Stevens et al. 2020; Knödlseder et al. 2022; Ryan et al. 2022). To put our energy system LCA for the AtLAST telescope into perspective, we compare its life cycle climate change impact to the rough estimation of the CO2e impact of building the telescope itself. Knödlseder found monetary emission factor for ground-based observatories at 240t CO2e /M€, where their need for concrete and steel were the biggest contributor (2022). First cost estimates for AtLAST account for a construction budget of 250–300 M€, resulting in roughly 60,000–70,000t CO2e, provided the technological production of steel and concrete remains the same. The 14,700 to 107,600 t of lifetime CO2e emissions calculated for the renewable power scenarios in this work compare with 56 to 490% of the projected telescope’s construction GHG impact.

4 Possibilities to expand this LCA

This work presents a cradle-to-gate comparative LCA on various off-grid energy systems in Chile using components from selected production regions. As such, it is subject to various limitations and opens up opportunities for future work. Our scope was specific to certain locations and timeframes, incorporating LCIs of future power mixes in production and transportation routes. If we selected other locations than China (for PV, batteries), the USA (for PEMFC), Norway (for AE), and Chile (for other components), the environmental impacts of the respective energy systems for AtLAST would differ. Faster or slower transitions towards RES in the production areas’ power mixes can further lead to varying results. By applying a sensitivity analysis, we could address some of these uncertainties and show that the climate change impact varies by up to −2.7/+6.4% with lower and upper boundary carbon-neutral generation shares in the 2030 Chinese power mix. When using upper and lower boundaries in battery replacement assumptions, the GHG emissions change up to −13%/+5%.

Regarding future work, the life cycle GHG emissions from this LCA shall be integrated into the techno-economical analysis for a multi-objective optimization considering both CO2e emissions and LCOEs. This allows for finding the Pareto front between the lowest LCOE and lowest GHG emissions to supply the telescope (Konak et al. 2006). We can further conduct a similar study on the currently installed capacities supplying energy to the other observatories on Chajnantor, to help understand if an energy transition of all these telescopes would make sense environmentally, also regarding the expected lifetime remaining of the respective observatories. Options such as substituting the fossil diesel in the PDB or PDBH scenario with sustainable biodiesel could also be looked upon there.

To further advance the LCA community in the field of energy systems, it would be beneficial to reduce our reliance on black-box inputs such as ecoinvent datasets from 2009. Instead, we should strive to incorporate temporal and regional realities into our LCIs. This approach would significantly enhance the accuracy and relevance of our studies, leading to more reliable and insightful findings. This work showcased that applying more recent LCIs compared to the ones available in ecoinvent can be done, though the literature search and implementation in SimaPro was a tedious process. To ease the path for upcoming studies on energy system LCAs, we publish the LCIs applied in this work as detailed activity level datasets in Viole et al. (2023a).

5 Conclusions

With this comparative LCA, we demonstrate that the production of RES components, especially PV panels and batteries, next to the burning of diesel in generators has the biggest impact on the CO2e emissions over the life cycle of off-grid energy systems supplying a telescope in the Atacama Desert. Other operations, maintenance, and the transport of components to the site, in comparison, have minor impacts. Regarding our three research questions, we find the following.

5.1 The system with the lowest environmental impact in the three impact categories considered

We show that depending on the choice of components for the off-grid system, trade-offs between our three environmental impacts must be made, which are climate change, water use, and mineral resource depletion. The four scenarios with high shares in RES have emissions of 0.077–0.115 kg CO2e/kWh supplied, compared to GHG emissions of 0.559 kg CO2e/kWh when combining PV with diesel, and 0.917 kg CO2e/kWh employing only diesel generation in the reference scenario. This reference scenario has the lowest impact in both mineral resource depletion and water use, as these two impact categories are mainly impacted by RES components.

5.2 Trade-offs between the environmental impacts

Due to the desertic location of the energy system for the telescope AtLAST, this case study paid special attention to the water use. We find that over the life cycle, the water use of the production of PV panels and batteries in China has a much higher impact compared to the water use for the hydrogen production or the PV cleaning in the Atacama. The hydrogen system at most made up for 8% of the water use impact. We find systems with hydrogen storage to have a slightly higher water use than their counterparts with only batteries.

The mineral resource use is highly impacted by the required capacities of both PV panels and batteries. The highest impact in this category stems from copper, zinc, silver, and titanium. Lithium, often cited as a scarce commodity for the expansion in battery production, has a minuscule share in this impact category, due to its low characterization factor in the applied ADP method. This result is in line with the EU projections, which foresee a low supply risk for lithium. We find that systems with a HESS using hydrogen and batteries have a lower impact in this category than the systems only applying batteries as energy storage.

In trade-offs between the impact categories, we find that the systems with the lowest GHG emissions, the 100% RES ones, have the highest impact in water use and mineral resource depletion. The high-renewable scenarios have slightly higher GHG emissions; however, their water and mineral resource use lies around 27% below that of the 100% RES scenarios. This is due to the smaller PV and storage capacities that they require.

5.3 Regional distribution of the environmental impacts

In assessing the regional distribution of the impacts, we find them localized in Chile in the reference scenario with solely diesel generators and spread out much over the globe when applying RES, which largely are produced in China from mineral resources sourced all over the globe. The water use impact is located at the production facilities as well as on site in Chile for the PV cleaning and hydrogen system. The sensitivity analysis of this work found that sensitivities regarding the Chinese power mix do not play a major role in the LCA, contrary to what was originally anticipated. We see this as a relevant result that others can learn from.

This work shows that lowering the environmental impact like for example GHG emissions in off-grid remote energy systems is a challenging task. Replacing the business-as-usual diesel generators with 100% RES lowers the GHG intensity per kWh supplied, while trade-offs in mineral depletion and water use have to be taken. We show that a variety of energy systems with high shares of RES could be deployed to power the planned telescope, and that systems with some fossil generation next to high shares of renewables can evade oversizing of components, thus saving on resources and have a higher security of supply. Given the high importance that telescope operators put on power quality, the authors recommend hybrid energy systems combining renewables, energy storage systems and conventional power generators for use in the mid-term future.