Photovoltaics-driven power production can support human exploration on Mars

A central question surrounding possible human exploration of Mars is whether crewed missions can be supported by available technologies using in situ resources. Here, we show that photovoltaics-based power systems would be adequate and practical to sustain a crewed outpost for an extended period over a large fraction of the planet's surface. Climate data were integrated into a radiative transfer model to predict spectrally-resolved solar flux across the Martian surface. This informed detailed balance calculations for solar cell devices that identified optimal bandgap combinations for maximizing production capacity over a Martian year. We then quantified power systems, manufacturing, and agricultural demands for a six-person mission, which revealed that photovoltaics-based power generation would require<10 t of carry-along mass, outperforming alternatives over ~50% of Mars' surface.

In an effort to determine the potential of PV and PEC devices to support a crewed mission to Mars, we integrated relevant climate data from the Mars Climate Database (11) into a radiative transfer model, libRadtran (12), to predict spectrally-resolved solar flux across the Martian surface over the course of a year. The modeling overview and sample calculations for Jezero Crater are provided in Fig. 1. Sunlight incident on the surface originating from the top of the atmosphere (TOA) is mediated by orbital geometry and local atmospheric composition of gases, ice, and dust for a given location (Fig. 1A). We determined the partial pressures of constituent gases (Fig. 1B) and the concentrations and effective radii of ice (Fig.  1C) and dust (Fig. 1D) particles as a function of altitude above the surface and provided these data as inputs to a downstream radiative transfer model (diagrammed in Fig. 1E). We then calculated the spectrally-resolved solar flux (Fig. 1F). At short wavelengths (<400 nm), light transmission through the atmosphere is limited by molecular scattering (primarily by CO 2) and scattering from dust particles (13). Scattering and absorption by gas molecules is significant at wavelengths below 300 nm, but this region is not considered here because it represents a very small fraction of the available solar flux (<0.5%). Above 400 nm, most transmission loss is due to scattering from dust particles. This is markedly different from the case on Earth, where significant molecular absorption by water molecules limits the transmission of near-infrared light.
The modeling results were used to inform efficiency calculations for PV and PEC devices producing electricity and H2. Detailed balance calculations (section 4 in the Supplementary Information) (14,15) revealed ideal current-voltage characteristics for optically-thick devices consisting of 1-, 2-, and 3-junction PV and 1-and 2-junction PEC absorbers dependent on the bandgaps associated with each absorber (Fig.  2). Absorber numbers were selected to represent historical choices for PV devices on Martian rovers (16,17) and state-of-the-art PEC devices (7,18,19). For PEC devices, we assumed an electrical load consisting of the thermodynamic redox potential and a variable overvoltage term that incorporates loss mechanisms inevitable to a practical PEC device beyond radiative recombination already considered in the detailed balance (14,15).
To evaluate the potential for solar cells to supply power and commodity chemicals, we determined the maximum practical production capacity for 3-junction PV (operating at 80% of the detailed balance limit) and 2-junction PEC devices (with a 700-mV overvoltage) over the course of a Martian year (Fig. 3). Daily and seasonal variation in solar flux and temperature (Fig. 3A, B) cause substantial (~27% deviation from the yearly average) changes in production rates (Fig. 3C, D). We defined solar day (sol) 0 at a solar longitude (Ls) of 0° (vernal equinox) and assumed the solar cell operating temperature was equal to the surface temperature at all points. Dust storm season begins at sol ~372 (Ls~180°) and is primarily responsible for the drop in production capacity from a peak of ~1.7 kWh/m 2 /day at Jezero Crater to a minimum of ~1.0 kWh/m 2 /day at the height of dust storm intensity around the winter solstice (Ls ~270°, sol ~514).
Bandgap combinations that maximize production over the course of a year are 5-15% different from those that optimize efficiency at solar noon (Table S7). For both PV and PEC devices, the top junction bandgap shifted up (for H2-generating PEC devices, from 1.64 eV to 1.77 eV), while the bottom junction bandgap shifted down (from 0.95 eV to 0.83 eV). Hence, the photon absorption window for the bottom junction is broadened (by ~35% for the PEC device). This likely works to maximize productivity during the less dusty season (higher solar flux, Fig. 3B) by accounting for the relative blue-shift of surface-incident light (Fig. 1F) due to reduced scattering.
Production capacity of power and commodity chemicals must compare favorably to the demand necessary to sustain a Martian habitat and depends on the outpost location on the planet surface (Fig.  4A). Moreover, energy storage capacity is crucial for solar-powered production systems because the sun sets daily. We therefore developed a detailed process model to account for power systems demands, including habitat maintenance (for example, habitat temperature control and pressurization), fertilizer production for agriculture, methane production for ascent propellant, and bioplastics production for spare parts manufacturing ( Fig S12). We considered four different power generation scenarios: (1) nuclear power generation with the miniaturized nuclear fission Kilopower system; (2) PV power generation with battery energy storage (PV+B); (3) PV power generation with compressed H 2 energy storage produced via electrolysis (PV+E); and (4) PEC H2 generation with compressed H2 energy storage (PEC). In our calculations, we assumed a capacity factor of 75% to account for the solar flux deviation throughout the Martian year (Fig. 3) and sized energy storage systems (batteries or compressed H2) to enable 1 full day of operations from reserve power. We then calculated the carry-along mass requirements for each of the power generation systems considered.
Of the three solar-driven power generation options, only the PV+E system outcompetes the nuclear system based on carry-along mass (Fig. 4B, C; supplementary Fig. S13). For the PV+E system, the total carry-along mass increases from ~8.3 t near the equator to ~22.4 t near the South Pole (Fig. 4B), corresponding to the reduced average daily power generation of the PV array as the latitude is adjusted away from 0° (Fig. 4A). The nuclear power system is predicted to require ~9.5 t; hence, the PV+E system out-performs this option across ~50% of the planet's surface (Fig. 4B).
In addition to predicting production capacity and carry-along mass, our model provides design rules for optimal solar cell design. Optimal absorber bandgaps for the PV array are strongly dependent on the location on the surface of Mars (Fig. 4C-E). Several factors cause this variation: the total depth of the air column above a given location (i.e., the difference between the height of the atmosphere and the altitude), gradients in dust and ice concentrations and particle radii, and orbital geometry effects that cause different effective air column thicknesses for locations near the poles. Lower elevations, higher dust and/or ice concentrations, and increasing distance away from the equator (near-polar latitudes) all cause an increase in the optical depth of the air column, which enhances the fraction of light that is scattered. Because the spectrum of scattered light is slightly red-shifted with respect to direct light (Fig. 1F), optimal bandgaps decrease to capture more lower-energy photons ( Fig. 4C-E) in regions where the optical depth is higher. For example, at equivalent latitudes, the optimal bandgaps are wider for regions with higher elevations than for those with lower elevations because the fraction of light that gets scattered is lower. Regional differences in atmospheric conditions can drive countervailing effects; because the Northern Hemisphere experiences generally lower dust concentrations than the Southern Hemisphere, the lower elevation in the Northern Hemisphere does not result in (on average) narrower optimal bandgaps. Instead, the reduced dust concentration (relative to that of the Southern Hemisphere) results in a reduced optical depth, resulting in wider optimal bandgap combinations ( Fig. 4C-E).
In summary, solar cell arrays with careful attention to semiconductor choice and device construction represent a promising technology for sustaining an Earth-independent crewed habitat on Mars. Our analysis provides design rules for solar cells on the Martian surface and shows that solar cells can offer substantial reduction in carry-along mass requirements compared to alternative technology over a large fraction of the planet's surface.
Data availability: All redSun software is available through the CUBES github organization at https://github.com/cubes-space/redSun and all data is available upon request.

Introduction and Overview
A central question surrounding possible settlement of Mars is whether human life can be supported by available technologies using in situ resources. Here we present a detailed analysis showing that photovoltaic and photoelectrochemical devices would be adequate and practical to sustain a crewed outpost for an extended period over a large fraction of the planet's surface. Climate data were integrated with a radiative transfer model to predict spectrally-resolved solar flux across the Martian surface, which informed detailed balance calculations for solar cell devices supporting power systems, agriculture, and manufacturing. Optimal design and the corresponding production capacity over a Martian year revealed the size and mass of a solar cell array required to support a six-person mission, which represents less than 10% of the anticipated payload.
The following SI describes the redSun software created as an integration of available software and custom code written in Python 3.6 with UNIX and Fortran backends. It can be found at https://github.com/cubes-space/redSun.

Mars Climate Database
Downstream radiative transfer calculations require a number of input streams describing the Martian environment. We make use of the Mars Climate Database (MCD) 1 developed by Le Laboratoire de Meteorologie Dynamique (LMD) in Paris, queried via the mcd-python package, to model most climate and environmental constraints, including photon flux and power spectra over time and location. The software engineering processes for building and using MCD somewhat efficiently are illustrated in Figure 1, along with input parameter profiles and sample output plots.

Initial Geotemporalspatial Grid
We began by first initializing the geotemporalspatial grid from which all downstream radiative transfer and PV/PEC calculations would be based. The grid was composed as a .netCDF file with dimensions of 19 points of 10 • latitude × 37 points of 10 • longitude × 25 points of 15 • areocentric longitude × 13 points of 2 (Martian) hours. Additionally, we included the dimension of altitude above the Martian datum in 20 points ranging from 0 to 120 km. The dimensions for the initial grid are shown in Table 1

Atmospheric Variables
Through a combination of custom code in redSun and modifications to the Python-based extension of MCD, we then looped through Lat, Lon, Hr, and Ls dimensions to initialize the data variables in Table 2.  Table 2. Initial atmospheric grid variables sourced from MCD.

Planetary Variables
While most of the required environmental variables could be sourced from MCD, additional efforts were made to add data on the planetary albedo and zMOL as shown in Figure 2 and in Table 3.

Variable Units Dimensions Dimension Number
Albedo None lat,lon 2 zMOL None lat,lon 2 Table 3. Initial planetary grid variables sourced from MCD.

Solar Variables
In addition to atmospheric and planetary variables, our initial environmental data for downstream radioactive transfer required that we calculate the solar flux at the top of the atmosphere (TOA). Downstream radiative transfer calculations required as input the spectral flux in W/(m 2 ·nm) whereas MCD only provided an integrated solar flux in W/(m 2 ). For a given Lat, Lon, Hr, and Ls, we were able to calculate the spectral flux F 0 via 2 where r is the Sun-Mars distance along its orbit, d is the mean Sun-Mars distance of 1.52 AU, µ is the cosine of the solar zenith angle z, e is the Martian eccentricity (e = 0.0934), L s is the aerocentric longitude, L s,p is the aerocentric longitude of perihelion (250 • ), θ is the latitude, is the Martian obliquity (25.2 • ), P is the duration of the Martian solar day (88775 s), t is any time measured from local noon, and F 1.52 is the flux at the average Sun-Mars distance 3 .
While the separation of the aerocentric longitude and hourly time dimensions was helpful in indexing our grid, these two dimensions are related. For any aerocentric longitude index, there are 13 time points, and as these times correspond to movement of Mars around the sun, so does the aerocentric longitude. Therefore, when computing the TOA flux F 0 , we updated L s to correspond to the change in time t using the build in functions Ls2Sol and Sol2Ls from the MCD package. These functions relate L s and t through Kepler's Problem via where D s is the sol number, t peri is the time at perihelion, N s is the number of sols in a Martian year, ν is the true anomaly, E is the eccentric anomaly, M is the mean anomaly, and N s is the number of sols in a Martian year.
The data variables shown in Figure 3 were then added to the grid for downstream use as shown in Table 4.  be ∼156 W/m 2 . Our calculated equatorial annual-mean TOA flux was found to be 159.43 W/m 2 which differs by ∼1.5% from the theoretical value. We extended this calculation across all latitudes as shown in Figure 4 to confirm our methods.

Mie Scattering Calculations
The presence of dust and cloud particles in the Martian atmosphere affect the propagation of sunlight. The size of such dust and cloud particles falls within the Mie scattering range.
The libRadtran package was used for Mie scattering calculations of the scattering phase matrices and corresponding Legendre polynomials 6 . Input files for both dust and ice were constructed (Listing 1) and fed to the MIEV0 tool.

15/34
where r 0 is the logarithmic mode of the distribution, calculated from r eff . Through a series of trial-and-error attempts, we specified additional parameters for clouds such as the number of phase matrix elements set at 1, the number of Legendre terms to be computed set at 6000, the maximum number of scattering angles set to 2000. The resulting output from MIEV0 was a .netCDF file of ∼100 MB.
For dust, an effective radius r eff grid was set between 0.00310352 and 10.1 µm in steps of 1.0 µm and with a lognormal distribution with standard deviation σ = 1.3616. Again, through a series of trial-and-error attempts, we specified additional parameters for dust such as the number of phase matrix elements set at 1, the number of Legendre terms to be computed set at 2500, the maximum number of scattering angles set to 2000. The dust calculations provided more computationally expensive than those for clouds due to the smaller r eff grid size. The resulting output from MIEV0 was a .netCDF file of ∼10 MB.
The output .netCDF files include the dimensions and variables in Table 5 and a sample of the output variables are shown in Figure 6.

uvspec
The uvspec program was designed to calculate the radiation field of the atmosphere for Earth. Modifications were carried out such that uvspec could be leveraged for similar calculations of the Martian radiative transfer. Input to the model are the constituents of the atmosphere including various molecules, aerosols and clouds. The absorption and scattering properties of these constituents were calculated via the MIEV0 tool. Boundary conditions are the solar spectrum at the top of the atmosphere and the reflecting surface at the bottom 8 . The uvspec program was called for each point in the geotemporalspatial grid and provided with a custom, programmatically generated input file -an example of which is shown in Listing 2.

Listing 2. Sample input file for uvspec calculation
Due to the peculiar way uvspec must be called, input for atmosphere, solar flux, dust conditions, and cloud conditions are required in the form of text-based .DAT files. Because multiple uvspec calls were carried out in parallel, a random string was generated ("2WKSII17KG" in the case of Listing 2) and used to identify specific .DAT files. For each point of the grid, an input .INP file was created along with correspond .DAT files for atmosphere, solar flux, dust conditions, and cloud conditions. The atmosphere file contained the altitude above sea level in km, pressure in hPa, temperature in K, air density in cm −3 , ozone density in cm −3 , Oxygen density in cm −3 , water vapor density in cm −3 , CO 2 density in cm −3 , and NO 2 density in cm −3 . The dust and cloud aerosol files contained altitude above sea level in km, dust/cloud content in kg/kg, and effective radius in µm. The solar flux file contains the wavelength in nm and the spectral flux for that wavelength in mW/(m 2 nm). Data for each of these files was sourced from the MCD data organized in the Stupidgrid.nc file and converted to the appropriate units using functions in the redSun codebase.
The wavelength range was set from 300.5 to 4000 nm. This range was selected to match available data for solar flux and significance to downstream photovoltaic calculations. Wavelengths outside these bounds were found to have negligible impact on bandgap calculations or to require substantial computational efforts, and were thus ignored. The mixing ratios for atmospheric CH 4 , N 2 O, and greenhouse gases (GHG) F11, F12, and F22 were set to 0.0 to reflect the change from Earth to Mars conditions. The altitude for the location was also programmatically added to the input file to specify the exact position of the surface in relationship to the Martian datum. The filenames from the Mie scattering calculations for dust and cloud aerosols were passed as well. The radius of the planet was changed to the Martian value of 3389.5 km. The albedo of the grid-point was also provided programmatically.
We selected the DIScrete ORdinate Radiative Transfer solvers (pseudospherical disort) radiative transfer solver for our calculations using 6 streams. The discrete ordinate method was first developed in 1960 with significant updates in 1988 and 2000 and offer 1D calculations of radiance, irradiance, and actinic flux. We opted for pseudo-spherical methods to offset any spherical effects associated with using the smaller Martian geometry. In pseudo-spherical calculations, the monochromatic radiative transfer equation in 1D can be formulated as

18/34
where B[T (r)] is the Planck function, β is an extinction coefficient, µ 0 is the solar zenith angle, φ 0 is the azimuth angle, p is the phase function, and the single scattering albedo ω(r) is Additionally, f ch is the Chapman function 9, 10 for describing the extinction path in a spherical atmosphere and is formulated as where R is the planet radius and r 0 is the distance above the atmosphere.
The output of each uvspec call was a text-like file that was indexed with a matching random string identifier. Each file consisted of the direct, global, diffuse downward, diffuse upward, net and sum irradiance in mW/(m 2 nm) for each nm in the input flux file. The output file was then read back with additional functions from redSun for use in downstream calculations.

Photovoltaic Power and Photoelectrochemical Commodity Calculations
We use the detailed balance model to calculate the energy efficiency of one-, two-, and three-bandgap photovoltaic solar cells and one-and two-bandgap photoelectrochemical devices. This model has been used to calculate the limiting efficiency of ideal photovoltaic and photoelectrochemical devices for single and multiple bandgap architectures previously [11][12][13] .
The current density (J)-voltage (V) dependence J (V, E g ) for a single bandgap is given by where J G is the photogeneration current, J R is the recombination current due to radiative recombination, and E g is the bandgap of the absorber material. The generation current J G is calculated according to where q is the electronic charge, Γ(E) is the photon flux at a given photon energy E, and E max is maximum photon energy in the solar spectrum. We used a minimum wavelength of 300 nm in our calculations, corresponding to a maximum photon energy of ∼4.14 eV because photons above 4 eV contribute negligibly to the photon flux 11 . The recombination current density J R is calculated according to where c is the speed of light in vacuum, h is Planck's constant, k is Boltzmann's constant, and T is the temperature of the device (we assume the local surface temperature in these calculations).
The photovoltaic energy efficiency η P V at a given operating voltage is written as where F is the calculated total power flux at the Martian surface. The operating voltage can then be selected to maximize the efficiency for a given bandgap. In technoeconomic calculations (see below), we assume the device efficiency is 80% of the calculated detailed balance limit to account for absorber material and device inefficiencies (i.e., nonradiative recombination losses not captured by the detailed balance limit).
The photoelectrochemical device energy efficiency η PEC is given by

19/34
where E 0 is the minimum thermodynamic potential required to drive the electrochemical reaction (1.23 V for H 2 generation from water splitting). In practical devices, the operating voltage of the photoelectrochemical device will be larger than E 0 to account for anode and cathode overpotentials and resistive potential drop in the electrolyte and electrodes. Hence, for these devices the operating voltage is where V o is the overpotential associated with the above-mentioned losses. In all technoeconomic calculations (see below) we assume the overvoltage is 700 mV, corresponding to a practical minimum that also accounts for absorber material inefficiencies (i.e., nonradiative recombination losses not captured by the detailed balance limit) 12 .
For two-and three-bandgap tandem devices, we assume the absorber layers are connected optically and electronically in series. Generation and recombination currents are calculated as described above, with the modification that E max is substituted with E g,n−1 for absorber n (counted sequentially starting with the top absorber) to reflect the assumption that each absorber layer is optically thick (i.e, absorbs all the above-bandgap light incident on its surface). In tandem devices, the total current density must be equal in each absorber layer, while the total operating voltage is given by the sum of the voltages developed across each cell. For example, for a three-absorber photovoltaic device For tandem devices, the efficiency is calculated analogously to the single-junction devices but as a function of each absorber bandgap.

SinglePoint Calculation
The calculation of a single gridpoint's spectral flux (via libRadtran) and the corresponding photovoltaic and photoelectrochemical production quantities ran for ∼5 minutes. Given the grid of 228475 geotemporalspatial points composed of 19 points of 10 • latitude × 37 points of 10 • longitude × 25 points of 15 • areocentric longitude × 13 points of 2 (Martian) hours, a serial calculation would require 2.17 years. Wanting to avoid that lengthy calculation, we opted for an "embarrassingly parallel" computing method shown in Figure 7. Since our computations require some initial or final communication (generally in the distribution and collection of data, then we call it nearly embarrassingly parallel. In parallel computing, an embarrassingly parallel workload or problem is one where little or no effort is needed to separate the problem into a number of parallel tasks. This is often the case where there is little or no dependency or need for communication between those parallel tasks, or for results between them. In the ideal case, all the sub-problems or tasks are defined before the computations begin and all the sub-solutions are stored in independent memory locations (variables, array elements). Thus, the computation of the sub-solutions is completely independent 1 .
Files were not constructed for grid-points that did not receive sunlight, and so the result was the storage of ∼150k .netCDF files, each with a size of ∼4-5 MB.

Stitching
The ∼150k singlepoint .netCDF files were initially stitched across time dimensions of hours and areocentric longitude to produce ∼700 time series .netCDF files, each for a different pair of latitudes and longitudes using the tcsh scripts provided in Listing 3 and 4.

Production Mapping
The resultant timeseries .netCDF files were then used for constructing the final maps of PV and PEC production. For each time series .netCDF file, we began by calculating PV power P and PEC production rateṁ via where Γ is the solar flux in W/m 2 sourced from the MCD data in StupidGrid.nc, is the electrochemical equivalency factor, η is the calculated PV/PEC efficiency, Z is the molar mass, n is the number of moles of electrons required to make one mole of the product, F is the Faraday constant, and V is the voltage. The c term corresponds to the chemical of interest in the set of H 2 , NH 3 , and AA. The values used to produce the for each chemical is given in Table 6.
We calculated the optimal sol-averaged 3-junction PV P opt and 2-junction PECṁ copt across all bandgap combinations given  Table 6. Electrochemical equivalency factor parameters.
the form The computational instance of calculations for 2J H 2 production is provided in Listing 6. Listing 6. Function for calculating the optimal H 2 production rate The results from the calculation of the optimal sol-averaged 3-junction PV P opt and 2-junction PECṁ copt and their corresponding bandgap combination were again saved as .netCDF files with dimensions of latitude and longitude.
The resulting PV power and PEC production for H 2 is provided in Figure 8-10 with the corresponding Bandgaps distributions over the Martian grid. The distribution of bandgaps are provided in Figure 11.  Table 7. Comparison of optimal bandgaps for different optimization strategies

Missing Location Values
We were able to complete the calculations for ∼97% of the 228475 geospatial points across the Martian grid. We found that ∼6000 of these points could not be completed due to a number of issues our method of using libRadtran for Mars-based calculations. Upon inspection, we found that the missing values were generally concentrated in areas with very low elevation below the Martian datum. Further inspection confirmed that the issues in resolving the radiative transfer were caused by errors in interpolation by the solver for the gas concentrations below the datum. However, these ∼2% of missing values do not prevent us from offering a meaningful analysis.

Primary Power and Energy Demands
We consider four different power production and energy storage scenarios for comparison ( Fig. 12): (1) Nuclear power generation with the Kilopower system; (2) Photovoltaic power generation with battery energy storage; (3) Photovoltaic power generation with compressed H 2 energy storage, and (4) Photoelectrochemical H 2 generation with compressed H 2 energy storage.
In all cases, power and/or energy demand is driven by continuous power required for habitat operations, including lighting, heating/cooling, pressurization, power draw for ISRU processes, and power draw for rover travel, and by materials demand for ISRU manufacturing. We assume that ammonia, methane, and plastics are produced using H 2 as the starting material (along with N 2 and CO 2 sourced from the atmosphere), which we use to calculate power demands based on water electrolysis to produce H 2 . We note that methane could be diverted for bioprocess production (dashed lines in Fig. 12), although we don't explicitly consider this scenario here since it would not change the relative mass requirements of the four systems we consider.
To compare the carry-along mass necessary for each system, we include the mass of elements unique to or uniquely sized for a given energy supply scenario. For example, we consider the mass of photovoltaic cells because the area of cells necessary to power the habitat and ISRU manufacturing will be different depending on the strategy for energy storage. However, we don't include the mass of the Sabatier reactor for methane production, since this mass will be equivalent regardless of the upstream processes producing H 2 and collecting CO 2 from the atmosphere. In this way, we can determine the mass contributions only of the uniquely necessary components for each energy supply scenario. The carry along masses are provided in Figure 13.

Nuclear Power
Power derived from the Kilopower nuclear reactor system is fed directly to habitat power systems and to an electrolyzer producing H 2 for ISRU manufacturing. Hence, the power draw is given by: where P K is the total power draw for Kilopower nuclear reactor system, P Hab is the power draw for the habitat, α E is the energy demand per unit of H 2 produced for the electrolyzer,Ṅ is the ammonia demand rate,Ṁ is the methane demand rate,Ḃ is the bioplastic demand rate, and α i is the conversion factor between, e.g., the ammonia demand rate and the H 2 demand rate for the Haber-Bosch process. We also define Λ =Ṅ α HB +Ṁ α S +Ḃα HB .
The carry-along mass requirements for this scenario is given by where p K is the specific power of the Kilopower reactor (6.25 W/kg) and p E is the specific productivity of the electrolyzer (kg H 2 /h/kg).

Photovoltaic power with battery energy storage (PV+B)
Power generated by photovoltaic cells can be transferred either directly to power-drawing systems (habitat systems, water electrolysis) or diverted to battery stacks for storage to enable continuous operation either at night or during low-sunlight days (due to high dust conditions). We define the fraction of power supplied directly to power systems as χ, which, for photovoltaic systems, can be thought of as the fraction of the day that solar cells produce equal or more power than what is consumed by power-drawing systems. Unless otherwise stated, we assume in our calculations χ = 1/3. Hence, the total power draw for the PV+B system is given by: where P PV+B is the total power draw for the PV+B system and η B is the energy efficiency of the battery storage system. More compactly,

28/34
The carry-along mass required for the PV+B scenario is given by where p PV is the specific power of photovoltaic cells, t store is the desired back-up power availability time, and e B is the specific energy of the battery stack (units of energy per mass).  . Carry-along mass for different power generation scenarios. Carry-along mass across the Martian surface for PV+B, PV+E, and PEC power generation systems. PV+B and PEC systems cannot reach parity with nuclear power generation in terms of carry along mass (no locations at which the projected mass of the PV+B or PEC systems is less than the projected mass of the nuclear system).

Photovoltaic power with H2 energy storage
In this scenario, power generated by photovoltaic cells can either be directly fed to habitat systems or to an electrolyzer, which produces H 2 for consumption in ISRU manufacturing and for consumption by fuel cells the supply power to the habitat and other demands when direct power cannot (e.g., at night). Here, the total power demand for the system is given by 30/34