Lunar surface PV, storage and mobility#
orbitalpv.lunar connects lunar terrain, solar geometry, photovoltaic output,
panel temperature, battery dispatch and rover movement. It reuses the existing
SPICE engine, calibrated PVWatts reference, thermal balance and battery model.
The six gallery examples run offline from source-checked inputs.
Inputs and evidence#
Connector |
Inputs |
Evidence boundary |
|---|---|---|
|
PGDA polar 80 m DEM; 5 m DEMs and height-error maps for Sites 01, 04, 11. |
Laser-altimetry-derived, interpolated terrain. Pixel spacing is not the effective measurement resolution. Missing heights are retained. |
|
PDS4 seasonal polar cumulative temperature tables and XML labels. |
Binned bolometric brightness measurements from 2009–2020; the supported family covers L_s 0–60 degrees and 24 subsolar-longitude bins. Not a dated future temperature forecast or measured cell temperature. |
|
DE421 planetary SPK, lunar orientation PCK, frame kernel and leap seconds. |
Computed Sun/Earth vectors in MOON_ME_DE421, consistent with the terrain. The downloader never changes the global SPICE kernel pool. |
|
NASA SVS 4930 metadata, reference images and explicitly selected frames. |
Qualitative rendered references. Pixel brightness is not irradiance and these images are not used as numerical validation targets. |
All four use anonymous HTTPS, bounded downloads, retrieval times and SHA256 receipts. Cached files are checked against the requested product and checksum; corrupt or mismatched caches raise errors. Provider errors do not produce synthetic fallback data. No additional account connector is needed.
The data preparation script retains original XML and verbatim Diviner subset
rows. It records original and derived hashes, extraction rules and geometry
settings in examples/lunar/_data/provenance.json. The checked-in bundle is
about 5 MB; refreshing its originals downloads roughly 550 MB once.
Methods#
Terrain and sunlight#
Positions use planetocentric latitude, east-positive longitude and elevations above the 1,737,400 m lunar sphere. Polar stereographic axes use a south-pole origin and longitude-zero central meridian. The raster reader rejects Earth CRSs, alternate scale parameters and unsupported calibration offsets.
terrain_horizon transforms grid centres to Cartesian lunar coordinates and
then to the observer’s true east/north/up basis. It finds the maximum elevation
angle in each azimuth sector, including lunar curvature and tower height.
Three terrain layers cover nearby 5 m detail, a 40 km-wide 80 m window and a
broad 1280 m sampled grid. Their observer elevation is held consistent; coarse
inner samples are excluded where finer data is available.
This is a sampled sector-envelope model. It can overestimate shading within a sector and miss subpixel peaks; it does not interpolate a continuous surface along every ray. Finite map extent, DEM interpolation and layer disagreements remain limitations. Missing sectors raise errors. Published height-error maps can be downloaded, but their correlated uncertainty is not propagated into confidence intervals in this version.
lunar_ephemeris requires explicitly loaded kernels and propagates coverage
errors. It uses simultaneous geometric vectors, without light-time corrections.
lunar_solar_geometry adds topocentric azimuth/elevation, inverse-square solar
flux and spherical-Earth eclipse visibility. Earth’s atmosphere is omitted.
solar_disc_visibility integrates a uniform solar disc above a locally
straight horizon. Limb darkening and variation across the disc are omitted.
Terrain and Earth visibility fractions are multiplied; simultaneous overlap
is approximate. Solar incidence uses the disc centre.
PV and thermal balance#
lunar_pv_power supports fixed orientation, azimuth tracking at a fixed tilt,
and ideal two-axis Sun tracking. The explicit AM0 reference feeds the existing
PVWatts equation. There is no atmospheric diffuse component. Ground reflection,
mutual array shading, tracker electricity and dust deposition are not implicit;
optical retention is a caller-supplied scenario.
The yield/storage examples assume 28% reference efficiency, -0.2%/K relative temperature coefficient, one square metre and constant 25 C cells. These are illustrative hardware parameters, not measurements for a particular cell.
lunar_panel_thermal separately solves the existing transient radiative and
conductive energy balance. The ground view factor is averaged over all emitting
faces and the sky contribution is reduced accordingly. For Diviner bolometric
brightness temperatures, use ground_emissivity=1: radiance already encodes
emissivity. Physical ground temperatures instead require their emissivity.
Incoming ground IR is thermal forcing, never photovoltaic irradiance.
The thermal example uses nearby measured spatial temperature percentiles as three constant boundary scenarios. It does not reconstruct time histories from spatial samples. These boundaries drive July 2026 modelled sunlight. Efficiency’s linear temperature approximation remains a screening model.
Storage, tower design and rover routing#
size_lunar_storage enumerates array areas and nominal battery capacities,
reusing energy-conserving dispatch with charge/discharge power limits, losses
and SOC bounds. A feasible candidate must serve the load and, by default, end
with at least its initial SOC. Initial stored energy cannot silently compensate
for an undersized energy supply. Capacity and efficiency are constant; there is
no chemistry-specific cold-temperature or ageing model.
The example uses a continuous 1 kW payload plus 103 W heating. Heater demand
comes from an explicitly assumed equivalent thermal conductance (not lunar
convection): max(UA*(Tset-Tsink)-internal_heat, 0). The separate panel model
does not model a lander’s enclosure temperature.
The tower analysis compares the lowest feasible candidate mass at each tested height, using 3 kg/m² panels, 150 Wh/kg storage and an assumed tower mass of 20 kg + 5 kg/metre. This excludes many balance-of-system components and is not a structural design or continuous global optimum.
energy_aware_route searches a time-expanded four-neighbour grid, including
waiting. It respects slope, speed, charge/discharge power and energy reserve,
with rolling resistance, ascent, housekeeping and heaters. Generation and heat
loads are averaged over edge endpoints and assumed constant over each step.
The highest-energy state at each cell/time dominates lower-energy states. The
objective is earliest feasible arrival, then highest remaining energy. Terrain
below grid resolution, wheel slip, navigation and communication constraints
are outside scope. All rover parameters in the example are scenarios.
Run and reproduce#
From a source checkout of this change:
python -m pip install -e ".[examples,ground,orbit]"
python analysis/lunar_surface_pv/run.py --output outputs/lunar_surface_pv
python examples/lunar/plot_lunar_terrain_yield.py
python examples/lunar/plot_lunar_tower_orientation.py
python examples/lunar/plot_lunar_storage_survival.py
python examples/lunar/plot_lunar_thermal_diviner.py
python examples/lunar/plot_lunar_rover_route.py
python examples/lunar/plot_lunar_data_sources.py
No network is needed for these commands after installation. Figures appear
interactively; use MPLBACKEND=Agg for headless execution. The analysis runner
writes numerical CSV/JSON outputs. Optional original-data regeneration:
python analysis/lunar_surface_pv/fetch_inputs.py --cache /path/to/lunar-cache
Default analysis inputs include a full 2026 year at two-hour boundaries and July 1–31 at half-hour boundaries. Energy/storage uses left-endpoint constant power; the thermal solver uses right-endpoint constant forcing and 300 s substeps. Darkness durations are censored by the time window. One year does not cover the approximately 18.6-year lunar precession cycle. The runner exports a July time/azimuth-resolution sensitivity comparison; it is not an uncertainty distribution or proof of convergence at all sites/seasons.
For the supplied Site01 July case, changing sector width from 4 to 1 degrees at a half-hour time step changes modelled 30-day yield from about 222 to 231 kWh/m² and longest darkness from 49.5 to 29.5 hours. This substantial sensitivity is why the coarse screening outputs must not be treated as precise storage requirements. The 2-degree default is documented rather than presented as a converged high-resolution ray-tracing solution.
Tests cover geometric axes, spherical horizons, monotonic height response, finite-disc limits, Earth occultation, thermal equilibrium and energy closure, storage terminal SOC, rover reserve/detours/obstacles, PDS labels and cache integrity. These establish implementation checks, not validation against measured lunar PV output.
References#
NASA PGDA large-area LOLA products and 5 m site maps and uncertainties.
Gläser et al. (2014), Illumination conditions at the lunar south pole using high resolution Digital Terrain Models from LOLA, doi:10.1016/j.icarus.2014.08.013.
Ross et al. (2023), Preliminary quantification of the available solar power near the lunar South Pole, doi:10.1016/j.actaastro.2023.06.040.
Schorghofer and Williams (2020), Mapping of Ice Storage Processes on the Moon with Time-dependent Temperatures, doi:10.3847/PSJ/abb6ff.
Dhere and Santiago (1993), Computation of photovoltaic parameters under lunar temperature variation, NASA NTRS 19940006939.
The cited studies motivate these experiments; their numerical results are not claimed to have been independently reproduced here.