pith. sign in

arxiv: 2604.17859 · v1 · submitted 2026-04-20 · 🌌 astro-ph.EP

Producing High-Resolution Martian Surface Temperature Maps Using VIR-TIR Relationships

Michael A. Frazer , Eriita G. Jones , Katarina Miljkovic , Gretchen K. Benedix This is my paper

Pith reviewed 2026-05-10 03:53 UTC · model grok-4.3

classification 🌌 astro-ph.EP
keywords Marsthermal inertiaCRISMTHEMISmachine learningdownscalinghyperspectral datasurface properties
0
0 comments X

The pith

A machine learning model trained on co-registered orbital data produces thermal inertia maps of Mars at 12 meters per pixel.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper trains a regressor on paired THEMIS thermal inertia values and CRISM visible-near-infrared spectra, both resampled to 100 m per pixel resolution. The trained model is then applied to the native 12 m per pixel CRISM spectra to estimate thermal inertia at that finer scale. This produces maps that resolve surface physical properties such as particle size and cementation at the decametre scale, an order of magnitude better than standard THEMIS products. Such detail supports improved interpretation of geologic processes, resource assessment, and mission planning on Mars.

Core claim

We generate a machine learning regressor-based model to constrain relationships between THEMIS TI and CRISM VIR images at THEMIS resolution, and predict TI values from CRISM spectra with high accuracy (R2 ∼ 0.90, RMSE ∼ 23.6 TIU). We use the model to produce a downscaled TI map at a spatial resolution of 12 m/pixel, an order of magnitude finer than currently available, revealing decametre-scale features previously unresolved in THEMIS data.

What carries the argument

Machine learning regressor that learns the statistical mapping from CRISM VIR hyperspectral data to THEMIS thermal inertia at matched 100 m resolution and then applies it at native CRISM resolution.

Load-bearing premise

The statistical relationship between VIR spectra and thermal inertia learned at 100 m pixel scale continues to apply accurately when the spectra are used at their native 12 m pixel scale.

What would settle it

Independent thermal inertia measurements at roughly 10 m resolution, either from future instruments or rover traverses, that show systematic offsets from the model's predictions at matching locations.

Figures

Figures reproduced from arXiv: 2604.17859 by Eriita G. Jones, Gretchen K. Benedix, Katarina Miljkovic, Michael A. Frazer.

Figure 1
Figure 1. Figure 1: First television picture from space, taken April 1, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Global thermal inertia map from the TES instrument. Red regions are high TI (e.g. exposed rock or duricrust), blue are low TI (fine or dust-mantled areas). Credit: NASA/Arizona State University. In most cases, pixel values in TI images represent some mixture of these materials – for example, an outcrop (2000 TIU) surrounded by sand (300 TIU) and covered in thin dust (100 TIU) might return a combined TI val… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the spatial resolutions of ASTER's [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Colour image of Gale crater, 154 km wide. The red [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) CRISM HCPINDEX2 and (b) THEMIS100 TI images overlaid on a CTX background. The CRISM image shows the HCPINDEX2 band, with high (red) regions corresponding with high-Ca pyroxene abundance (basaltic sands) [32, 33]. High-TI (bright yellow) regions in the THEMIS100 TI image correspond with exposed rock, while low-TI (dark purple) correspond with sands. The white line is the traverse of the MSL Curiosity ro… view at source ↗
Figure 6
Figure 6. Figure 6: Flowchart of training and applying the model. [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: (a) The original THEMIS TI image (THEMIS1 [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: A scatter plot of the 1152 ‘test’ pixels’ observed [PITH_FULL_IMAGE:figures/full_fig_p005_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: THEMIS_R12 overlaid on a CTX background of the north-west flank of Aeolis Mons within Gale Crater. The white line is the traverse of the MSL Curiosity rover. Red crosses denote every 1000th sol. Note the high TI of the Greenheugh pediment and low TI of the Sands of Forvie. The blue and green boxes correspond to the zoomed regions in [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: MSL Curiosity's view across the Sands of Forvie, taken by the Right Navigation Camera on Sol 2990 at the location indicated in [PITH_FULL_IMAGE:figures/full_fig_p008_12.png] view at source ↗
read the original abstract

Thermal infrared data (TIR; 8 - 15 $\mu m$) has a wide range of applications in Earth and planetary remote sensing. On Mars, this includes deriving thermal inertia (TI), which describes surface physical characteristics (e.g. particle size, degree of cementation) and is key for understanding geologic processes, assessing in-situ resource utilisation (ISRU) environments, and assisting mission planning. However, TI data from the THEMIS instrument is limited to 100 m/pixel resolution. Hyperspectral visible and near-infrared data (VIR; 0.5 - 5 $\mu m$) compliments TIR data by providing information on surface composition and is provided by the CRISM instrument at 12 m/pixel. In this work, we generate a machine learning regressor-based model to constrain relationships between THEMIS TI and CRISM VIR images at THEMIS resolution, and predict TI values from CRISM spectra with high accuracy (R2 $\sim$ 0.90, RMSE $\sim$ 23.6 TIU). We use the model to produce a downscaled TI map at a spatial resolution of 12 m/pixel, an order of magnitude finer than currently available, revealing decametre-scale features previously unresolved in THEMIS data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript presents a machine learning regressor trained on pairs of CRISM VIR spectra (degraded to 100 m/pixel) and THEMIS thermal inertia (TI) values to learn a VIR-TIR relationship, then applies the model to native-resolution (12 m/pixel) CRISM spectra to generate a downscaled TI map. It reports R² ∼ 0.90 and RMSE ∼ 23.6 TIU on the coarse-scale data and claims this reveals previously unresolved decameter-scale surface features.

Significance. If the VIR-TIR statistical mapping is scale-invariant, the approach would deliver an order-of-magnitude gain in TI spatial resolution over existing THEMIS products, directly benefiting studies of Martian regolith properties, particle-size distributions, cementation, ISRU site evaluation, and landing-site characterization. The data-driven linkage of composition-sensitive VIR with physically diagnostic TIR is a constructive methodological step for planetary remote sensing.

major comments (1)
  1. [Abstract] Abstract: The reported R² ∼ 0.90 and RMSE ∼ 23.6 TIU are measured only after degrading CRISM spectra to THEMIS resolution; the central downscaling claim requires that this relationship remains unbiased when the same regressor is applied to full-resolution 12 m/pixel spectra. No aggregation-back test, native-resolution validation set, or assessment of sub-pixel heterogeneity effects is described, leaving the physical consistency of the 12 m map unverified.
minor comments (2)
  1. [Title] Title: The title refers to 'Surface Temperature Maps' while the work derives and maps thermal inertia (TI); a minor title adjustment or explicit statement that TI is the target quantity would improve clarity.
  2. [Abstract] Abstract: The specific regressor architecture, hyperparameter choices, feature-selection procedure, and training/validation split strategy are not stated, which limits immediate reproducibility of the quoted accuracy figures.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive review and for recognizing the potential significance of the VIR-TIR statistical mapping approach. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reported R² ∼ 0.90 and RMSE ∼ 23.6 TIU are measured only after degrading CRISM spectra to THEMIS resolution; the central downscaling claim requires that this relationship remains unbiased when the same regressor is applied to full-resolution 12 m/pixel spectra. No aggregation-back test, native-resolution validation set, or assessment of sub-pixel heterogeneity effects is described, leaving the physical consistency of the 12 m map unverified.

    Authors: We agree that the reported R² ≈ 0.90 and RMSE ≈ 23.6 TIU reflect performance only on CRISM spectra degraded to THEMIS (100 m/pixel) resolution, where paired training and validation data exist. No independent thermal inertia measurements at native CRISM resolution (12 m/pixel) are available, so direct native-resolution validation or a native-resolution validation set is not possible. The manuscript does not describe an aggregation-back test or explicit sub-pixel heterogeneity analysis. In revision we will (1) update the abstract to state explicitly that accuracy metrics apply to the degraded-scale validation and (2) add a dedicated discussion subsection on the scale-invariance assumption, the physical basis for expecting the VIR-TIR relationship to hold at finer scales, and the possible influence of sub-pixel heterogeneity. We will also implement and report an aggregation-back test (downscale then re-average to 100 m/pixel and compare with original THEMIS TI) to provide quantitative support for consistency. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation trains an ML regressor on paired THEMIS TI and CRISM VIR data at matched 100 m/pixel resolution, reports cross-validated performance (R² ∼ 0.90, RMSE ∼ 23.6 TIU) on that coarse-scale data, and then applies the fitted model to native 12 m/pixel CRISM spectra to produce the downscaled map. This is a standard supervised-learning prediction workflow; the output map is not fed back to redefine the training labels, inputs, or loss function. No equations, self-citations, uniqueness theorems, or ansatzes are present in the abstract that would collapse the claimed relationship to its own inputs by construction. The scale-invariance assumption is an untested extrapolation risk but does not constitute circularity.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that a statistical mapping learned at coarse resolution generalizes to fine resolution and that the ML regressor captures physically meaningful relationships rather than dataset-specific artifacts.

free parameters (1)
  • regressor hyperparameters and feature selection
    Tuned during training on THEMIS-CRISM overlap data; exact values and selection criteria not stated in abstract.
axioms (1)
  • domain assumption The relationship between VIR spectra and thermal inertia is scale-invariant between 100 m and 12 m
    Invoked when the model trained at THEMIS resolution is applied directly to native CRISM resolution.

pith-pipeline@v0.9.0 · 5540 in / 1290 out tokens · 47598 ms · 2026-05-10T03:53:13.623724+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    Tatem, S

    A. Tatem, S. Goetz, S. Hay, Fifty Years of Earth Observation Satellites, American Scientist 96 (2008) 390-398

    work page 2008

  2. [2]

    2025, https://wmo.int/topics/earth-observation- satellites (accessed 11/08 2025)

    WMO, Earth observation satellites. 2025, https://wmo.int/topics/earth-observation- satellites (accessed 11/08 2025)

    work page 2025

  3. [3]

    Samadzadegan, A

    F. Samadzadegan, A. Toosi, F. Dadrass Javan, A critical review on multi-sensor and multi- platform remote sensing data fusion approaches: current status and prospects, International Journal of Remote Sensing 46 (2025) 1327- 1402

    work page 2025

  4. [4]

    E. Neinavaz, et al., Thermal infrared remote sensing of vegetation: Current status and perspectives, International Journal of Applied Earth Observation and Geoinformation 102 (2021) 102415

    work page 2021

  5. [5]

    Green, M

    D. Green, M. Schodlok, Characterisation of carbonate minerals from hyperspectral TIR scanning using features at 14 000 and 11 300 nm, Australian Journal of Earth Sciences 63 (2016) 951-957

    work page 2016

  6. [6]

    McMillin, Estimation of sea surface temperatures from two infrared window measurements with different absorption, Journal of Geophysical Research 80 (1975) 5113-5117

    L.M. McMillin, Estimation of sea surface temperatures from two infrared window measurements with different absorption, Journal of Geophysical Research 80 (1975) 5113-5117

    work page 1975

  7. [7]

    Peng, et al., A review of spatial downscaling of satellite remotely sensed soil moisture, Reviews of Geophysics 55 (2017) 341-366

    J. Peng, et al., A review of spatial downscaling of satellite remotely sensed soil moisture, Reviews of Geophysics 55 (2017) 341-366

    work page 2017

  8. [8]

    Pablos, et al., Multi-temporal evaluation of soil moisture and land surface temperature dynamics using in situ and satellite observations, Remote Sensing 8 (2016) 587

    M. Pablos, et al., Multi-temporal evaluation of soil moisture and land surface temperature dynamics using in situ and satellite observations, Remote Sensing 8 (2016) 587

    work page 2016

  9. [9]

    Li, et al., Satellite Remote Sensing of Global Land Surface Temperature: Definition, Methods, Products, and Applications, Reviews of Geophysics 61 (2023)

    Z.-L. Li, et al., Satellite Remote Sensing of Global Land Surface Temperature: Definition, Methods, Products, and Applications, Reviews of Geophysics 61 (2023)

    work page 2023

  10. [10]

    M. Cetin, et al., Determination of land surface temperature and urban heat island effects with remote sensing capabilities: the case of Kayseri, Türkiye, Natural Hazards 120 (2024) 5509-5536

    work page 2024

  11. [11]

    Armstrong, T.N

    J.C. Armstrong, T.N. Titus, H.H. Kieffer, Evidence for subsurface water ice in Korolev crater, Mars, Icarus 174 (2005) 360-372

    work page 2005

  12. [12]

    Mellon, et al., High-Resolution Thermal Inertia Mapping from the Mars Global Surveyor Thermal Emission Spectrometer, Icarus 148 (2000) 437-455

    M.T. Mellon, et al., High-Resolution Thermal Inertia Mapping from the Mars Global Surveyor Thermal Emission Spectrometer, Icarus 148 (2000) 437-455

    work page 2000

  13. [13]

    Fergason, P.R

    R.L. Fergason, P.R. Christensen, H.H. Kieffer, High-resolution thermal inertia derived from the Thermal Emission Imaging System (THEMIS): Thermal model and applications, Journal of Geophysical Research: Planets 111 (2006)

    work page 2006

  14. [14]

    Beddingfield, J.E

    C.B. Beddingfield, J.E. Moersch, H.Y. McSween, Investigating crater rim thermal inertia variations on Mars: A case study in Tisia Valles, Icarus 314 (2018) 345-363

    work page 2018

  15. [15]

    Presley, P.R

    M.A. Presley, P.R. Christensen, Thermal conductivity measurements of particulate materials 2. Results, Journal of Geophysical Research: Planets 102 (1997) 6551-6566

    work page 1997

  16. [17]

    Hook, et al., The MODIS/ASTER airborne simulator (MASTER) — a new instrument for earth science studies, Remote Sensing of Environment 76 (2001) 93-102

    S.J. Hook, et al., The MODIS/ASTER airborne simulator (MASTER) — a new instrument for earth science studies, Remote Sensing of Environment 76 (2001) 93-102

    work page 2001

  17. [18]

    Ramsey, Mapping the City Landscape from Space: The Advanced Spaceborne Thermal Emission and Reflectance Radiometer (ASTER) Urban Environmental Monitoring Program, in G

    M.S. Ramsey, Mapping the City Landscape from Space: The Advanced Spaceborne Thermal Emission and Reflectance Radiometer (ASTER) Urban Environmental Monitoring Program, in G. Heiken, Fakudiny, R., Sutter, J., (Eds), Earth Science in the City: A Reader, American Geophysical Union: Washington, 2003, pp. 337- 361

    work page 2003

  18. [19]

    Murchie, et al., Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) on Mars Reconnaissance Orbiter (MRO), Journal of Geophysical Research: Planets 112 (2007)

    S. Murchie, et al., Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) on Mars Reconnaissance Orbiter (MRO), Journal of Geophysical Research: Planets 112 (2007)

    work page 2007

  19. [20]

    Christensen, et al., The Thermal Emission Imaging System (THEMIS) for the Mars 2001 Odyssey Mission, Space Science Reviews 110 (2004) 85-130

    P.R. Christensen, et al., The Thermal Emission Imaging System (THEMIS) for the Mars 2001 Odyssey Mission, Space Science Reviews 110 (2004) 85-130

    work page 2001

  20. [21]

    W. Kustas, et al., Estimating subpixel surface temperatures and energy fluxes from the vegetation index-radiometric temperature relationship, Remote Sensing of Environment 85 (2003) 429-440

    work page 2003

  21. [22]

    Anderson, et al., A Multiscale Remote Sensing Model for Disaggregating Regional Fluxes to Micrometeorological Scales, Journal of Hydrometeorology 5 (2004) 343-363

    M.C. Anderson, et al., A Multiscale Remote Sensing Model for Disaggregating Regional Fluxes to Micrometeorological Scales, Journal of Hydrometeorology 5 (2004) 343-363

    work page 2004

  22. [23]

    Agam, et al., A vegetation index based technique for spatial sharpening of thermal imagery, Remote Sensing of Environment 107 (2007) 545-558

    N. Agam, et al., A vegetation index based technique for spatial sharpening of thermal imagery, Remote Sensing of Environment 107 (2007) 545-558

    work page 2007

  23. [24]

    Agam, et al., Utility of thermal sharpening over Texas high plains irrigated agricultural fields, Journal of Geophysical Research: Atmospheres 112 (2007)

    N. Agam, et al., Utility of thermal sharpening over Texas high plains irrigated agricultural fields, Journal of Geophysical Research: Atmospheres 112 (2007)

    work page 2007

  24. [25]

    Li, et al

    W. Li, et al. Selection of Predictor Variables in Downscaling Land Surface Temperature using Random Forest Algorithm, 39th IGARSS - IEEE International Geoscience and Remote Sensing Symposium, 2019, Yokohama, Japan, 28 July - 2 August

    work page 2019

  25. [26]

    Dominguez, et al., High-resolution urban thermal sharpener (HUTS), Remote Sensing of Environment 115 (2011) 1772-1780

    A. Dominguez, et al., High-resolution urban thermal sharpener (HUTS), Remote Sensing of Environment 115 (2011) 1772-1780

    work page 2011

  26. [27]

    G. Yang, et al., A Novel Method to Estimate Subpixel Temperature by Fusing Solar- Reflective and Thermal-Infrared Remote- Sensing Data With an Artificial Neural Network, IEEE Transactions on Geoscience and Remote Sensing 48 (2010) 2170-2178

    work page 2010

  27. [28]

    S. Bonafoni, Downscaling of Landsat and MODIS Land Surface Temperature Over the Heterogeneous Urban Area of Milan, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 9 (2016) 2019-2027

    work page 2016

  28. [29]

    Gao, W.P

    F. Gao, W.P. Kustas, M.C. Anderson, A Data Mining Approach for Sharpening Thermal Satellite Imagery over Land, Remote Sensing 4 (2012) 3287-3319

    work page 2012

  29. [30]

    Yang, et al., Spatial Downscaling of Lunar Surface Temperature Based on Geographically Weighted Regression, IEEE Geoscience and Remote Sensing Letters PP (2023) 1-1

    X. Yang, et al., Spatial Downscaling of Lunar Surface Temperature Based on Geographically Weighted Regression, IEEE Geoscience and Remote Sensing Letters PP (2023) 1-1

    work page 2023

  30. [31]

    L. He, et al., Surface Kinetic Temperatures and Nontronite Single Scattering Albedo Spectra From Mars Reconnaissance Orbiter CRISM Hyperspectral Imaging Data Over Glen Torridon, Gale Crater, Mars, Journal of Geophysical Research: Planets 127 (2022)

    work page 2022

  31. [32]

    J.R. Christian, et al., CRISM-Based High Spatial Resolution Thermal Inertia Mapping Along Curiosity's Traverses in Gale Crater, Journal of Geophysical Research: Planets 127 (2022) e2021JE007076

    work page 2022

  32. [33]

    C.E. Viviano, et al., Revised CRISM spectral parameters and summary products based on the currently detected mineral diversity on Mars, Journal of Geophysical Research: Planets 119 (2014) 1403-1431

    work page 2014

  33. [34]

    Malin, et al., Context Camera Investigation on board the Mars Reconnaissance Orbiter, Journal of Geophysical Research: Planets 112 (2007)

    M.C. Malin, et al., Context Camera Investigation on board the Mars Reconnaissance Orbiter, Journal of Geophysical Research: Planets 112 (2007)

    work page 2007

  34. [35]

    Dickson, et al., The Global Context Camera (CTX) Mosaic of Mars: A Product of Information-Preserving Image Data Processing, Earth and Space Science 11 (2024)

    J.L. Dickson, et al., The Global Context Camera (CTX) Mosaic of Mars: A Product of Information-Preserving Image Data Processing, Earth and Space Science 11 (2024). 288 https://doi.org/10.52202/083076-0035

    work page doi:10.52202/083076-0035 2024