Predicting Subsurface Structure From Surface Observations

Context

Mountain watersheds route water, solutes, and carbon through a critical zone whose function depends as much on what lies beneath the surface — fractured bedrock, weathered saprolite, variable soil mantles — as on what grows above it. Remote sensing platforms now image surface topography, vegetation, and snow at fine resolution across entire basins, while geophysical methods can probe the subsurface only at discrete locations and significant cost. Whether surface signals reliably encode subsurface structure determines if watershed-scale hydrologic and biogeochemical models can be parameterized from airborne data alone, or whether ground-based campaigns will remain indispensable.

Frontier

The unresolved question is the strength and transferability of surface-subsurface covariance across lithologies, climates, and vegetation regimes. Co-evolution of topography, soils, vegetation, and weathering profiles suggests that surface patterns should carry information about depth to bedrock, fracture density, and weathering intensity — yet how generalizable those relationships are, and where they break down, remains poorly mapped. Advancing the boundary requires integrating remote sensing, near-surface geophysics, and pedology under a shared inferential framework, and confronting machine-learning predictors with independent geophysical ground-truth at sites that span the relevant geologic and ecological gradients. The deeper integration challenge is connecting empirical surface-subsurface mapping to process-based understanding of how landscapes evolve, so that learned relationships can be extrapolated with mechanistic confidence rather than treated as black-box correlations valid only within their training domain.

Key questions

• Can machine-learning models trained on lidar and multispectral imagery predict soil thickness and depth-to-bedrock with useful accuracy across watersheds they were not trained on?
• Which surface covariates (topographic curvature, vegetation structure, snow persistence) carry the most information about subsurface architecture, and does this ranking shift with lithology?
• How does prediction skill degrade as one moves from a training watershed to a new lithology, climate regime, or disturbance history?
• At what density and configuration of geophysical ground-truth do hybrid surface-plus-subsurface models outperform either approach alone?
• Can fracture density and bedrock weathering depth — which control deep water storage — be inferred indirectly through their imprint on vegetation phenology and topography?
• Do co-evolution principles linking landscape form and subsurface structure provide enough physical constraint to enable transferable predictions in data-sparse regions?

Barriers

The principal blockers are data gaps (paired remote sensing and subsurface datasets are rare and clustered in a few well-instrumented catchments), scale mismatch (airborne pixels versus point-scale boreholes versus 2D geophysical transects), method gaps (machine-learning generalization across geologic settings has not been systematically tested), and coordination gaps (geophysics, remote sensing, and pedology operate in separate communities with different conventions). Translation gaps also limit uptake: hydrologic and biogeochemical modelers often lack the subsurface parameterizations these methods could deliver.

Research opportunities

A coordinated paired-sites campaign could assemble matched lidar, hyperspectral, and airborne electromagnetic data with co-located electrical resistivity tomography transects and borehole logs across watersheds spanning contrasting bedrock types, aspects, and vegetation communities. Such a dataset would enable rigorous cross-site testing of machine-learning transferability and explicit quantification of where surface-only inference fails. Complementary opportunities include developing hybrid models that combine data-driven prediction with mechanistic constraints from landscape evolution and critical-zone theory, building open benchmarks for subsurface-from-surface prediction analogous to those in computer vision, and embedding lightweight geophysical sensors in long-term monitoring networks so that ground-truth accumulates passively over time. Coupled simulation platforms that ingest predicted subsurface fields into watershed hydrologic models would close the loop, showing which prediction errors actually matter for streamflow, groundwater storage, and biogeochemical fluxes — and which can be tolerated.

Pushing the frontier

Concrete, fundable actions categorized by kind of work and effort tier (near-term = single lab; ambitious = focused multi-year program; major = multi-institutional; consortium = agency-program scale).

Data gaps surfaced in source statements

Descriptions of needed data (not existing datasets), drawn directly from the atomic statements feeding this frontier.

• geophysical subsurface surveys at multiple sites
• paired remote sensing and borehole datasets across lithologies
• soil thickness transects on contrasting aspects

Impacts

Reliable subsurface prediction from surface observations would directly support watershed management in the upper Colorado River headwaters, where Bureau of Reclamation operations at Aspinall and downstream water deliveries depend on understanding mountain water storage. Forest Service and BLM Resource Management Plan revisions in the Gunnison Basin would benefit from spatially explicit estimates of soil and weathered-bedrock water holding capacity, which inform drought vulnerability and vegetation treatment decisions. CWCB instream flow filings and headwater conservation prioritization could draw on basin-wide subsurface mapping rather than extrapolating from a handful of instrumented sites. More broadly, hydrologic modelers, biogeochemists, and ecologists working anywhere in the mountain west would gain a parameterization pathway currently bottlenecked by the cost of ground-based geophysics.

Linked entities

Sources

Every claim in the synthesis above derives from the source atomic statements below, grouped by their research neighborhood of origin. Click a neighborhood to follow its primer and full citation chain.

Framing notes: Single-statement cluster, but the question is sharply defined and supports a concrete experimental and synthesis program; tractability rated medium because methods exist but transferability has not been systematically tested.