Geoscience: What Observation Recently Changed, Misconceptions, Numbers, and Canon
Part 5 of 5 of the Geoscience reference (plugin
geoscience), covering §17–§22. Sibling skills:geo-earth-structure-tectonics-rocks-and-deep-time(§0–§4),geo-earthquakes-seismology-and-volcanism(§5–§6),geo-surface-processes-soils-and-hydrology(§7–§10),geo-oceans-cryosphere-cycles-hazards-and-observation(§11–§16). Section numbers are shared across the set; a reference written as §N →skillpoints into that sibling skill.Currency: Tectonics, stratigraphy, hydraulics and seismological theory are settled; satellite-measured continental water storage and deep learning in seismology recently changed the science. See §17 below for both.
Scope. Complements a weather-science reference (the atmosphere) and a Newtonian-mechanics reference (the physics). ⚠️ This is the solid Earth, the water, and the land surface.
⚠️ GOTCHA boxes mark genuine misconceptions and places where intuition fails badly.
The three ideas that organize the field:
- ⚠️ Deep time is the hardest thing to internalize and the most important. Processes imperceptible on human timescales — millimetres per year — build mountains and open oceans given tens of millions of years. Almost every geological misconception is a failure of timescale intuition (§3 →
geo-earth-structure-tectonics-rocks-and-deep-time).- ⚠️ The Earth runs on two engines. Internal heat (radiogenic decay plus primordial) drives tectonics, building topography; solar energy drives the water cycle and weathering, tearing it down. Everything at the surface is the interaction (§1 →
geo-earth-structure-tectonics-rocks-and-deep-time, §7 →geo-surface-processes-soils-and-hydrology).- ⚠️ Rates and residence times explain more than mechanisms do. Water in a river resides for days, in groundwater for millennia. Whether something is renewable depends entirely on the ratio of extraction rate to renewal rate — and it's why §10 →
geo-surface-processes-soils-and-hydrologyis the most consequential section here (§10 →geo-surface-processes-soils-and-hydrology, §17.1).
§17. What Observation Recently Changed — verified August 2026
17.1 ⚠️ Satellite gravimetry and continental drying
GRACE and GRACE-FO measure the mass of water by its gravitational signal — ⚠️ a genuinely new observational capability that lets us weigh continental water storage directly rather than infer it.
The findings, from peer-reviewed work:
- ⚠️ A Science Advances study reports unprecedented terrestrial water storage loss since 2002, with areas experiencing drying increasing by twice the size of California annually (~831,600 km²/yr), forming "mega-drying" regions across the Northern Hemisphere.
- ⚠️ Groundwater depletion accounts for 68% of terrestrial water storage loss over non-glaciated continental regions.
- ⚠️ Dry areas are now drying faster than wet areas are wetting — which is not what a simple "wet gets wetter" framing predicts.
- ⚠️ The continents now contribute more freshwater to sea level rise than the ice sheets, and drying regions contribute more than glaciers and ice caps. That reframes the sea level budget.
- 75% of the population lives in 101 countries that have been losing freshwater.
- A separate GRACE/GRACE-FO analysis over 21.5 years finds groundwater depletion dominating freshwater decline at continental scales, most prominently in Asia at −55 km³/yr, ⚠️ while ice mass loss remains the largest single global contributor by component — and it identifies emerging groundwater gains in some regions alongside widespread decline.
- NASA reports 21 of Earth's 37 largest aquifers have exceeded sustainability tipping points, 13 of them significantly distressed.
⚠️ GOTCHA — GRACE is powerful and it has real uncertainties, and the literature is explicit about this. ⚠️ Groundwater storage is not measured directly — it's derived by subtracting modelled soil moisture, snow, surface water and glacier contributions from total water storage, so model error propagates in. A published re-analysis found earlier GRACE-based depletion rates for the Northwest India Aquifer were likely overestimates, with constrained forward modelling giving ~14 km³/yr against a published ~18 km³/yr — and the corrected figure matched well-monitoring data. ⚠️ Where GRACE has been compared against dense well networks it generally agrees (correlations ~0.52–0.95 across major US aquifers), which is the reassuring part — but treat single-basin headline numbers with more caution than continental-scale trends.
⚠️ Why this belongs in a geoscience document rather than a news summary: it is a measurement capability change, not a policy story. We can now weigh the continents' water, and the answer differed from what models assumed.
17.2 ⚠️ Deep learning in seismology
A quieter revolution, and it changed what the observational record contains.
The problem it solved: ⚠️ STA/LTA detection had been the backbone of real-time seismic processing since the earliest digital acquisition, and manual picking by skilled analysts had become impossible to scale as channel counts grew.
The models: PhaseNet (⚠️ Zhu & Beroza 2019 — a U-Net that reformulates phase picking as image segmentation), EQTransformer (⚠️ Mousavi et al. 2020 — CNN + LSTM + self-attention doing joint detection and picking, and notably compact at ~379k parameters), GPD, and association methods like GaMMA.
⚠️ The consequence for the science, which is the important part: these models detect earthquakes missed by standard methods, and the resulting catalogues have ⚠️ contributed to uncovering fault-structure complexity and earthquake swarm dynamics, giving new insight into aseismic crustal processes. One 2026 study applying a DL workflow to a single day containing a major mainshock obtained 5,315 earthquakes — reduced to 3,839 after location-quality filtering — against 1,086 in the manually reviewed catalogue. ⚠️ The record got several times denser, and denser catalogues change what questions you can ask.
⚠️ DAS is the amplifier. PhaseNet-DAS applies this to fibre-optic distributed acoustic sensing, ⚠️ turning existing telecom cable into an ultra-dense seismic array. The scale is different in kind: applied to ~9,839 catalogued earthquakes near one array, it produced ~36 million P-picks and ~53 million S-picks. Submarine and ocean-bottom variants (DeepSubDAS, PickBlue, OBSTransformer) extend it to marine environments — ⚠️ which matters because seismometers are scarce at sea and most plate boundaries are underwater.
⚠️ The honest caveats, and the literature is candid: models trained on regional surface stations at 100 Hz ⚠️ transfer poorly to high-frequency borehole data (2000 Hz) and to DAS without retraining; there is documented prediction inconsistency and parameter dependence in neural pickers, with active work on mitigation; and ⚠️ catalogue performance is variable enough that a 2026 paper is titled, in effect, "which is better: deep learning or manual picking?" — it is not a settled rout.
§18. Misconceptions
| Misconception | Correction |
|---|---|
| Plates float on molten rock | ⚠️ The asthenosphere is solid and ductile. S-waves prove it (§1.1 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| Convection currents drag plates | ⚠️ Slab pull dominates; plates are the top of the system (§1.2 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| Continental crust subducts | ⚠️ Too buoyant — it collides and thickens (§1.2 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| ¹⁴C dates rocks and dinosaurs | ⚠️ Organic only, ~50 ka limit (§3 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| You date a sedimentary rock directly | ⚠️ The grains predate the deposit. Bracket it (§3 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| The rock record is continuous | ⚠️ It's mostly gaps (§3 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| Richter is the modern magnitude scale | ⚠️ M_w is; Richter saturates above ~7 (§5.3 → geo-earthquakes-seismology-and-volcanism) |
| Magnitude describes shaking at a place | ⚠️ That's intensity. Magnitude is the source (§5.3 → geo-earthquakes-seismology-and-volcanism) |
| Earthquakes are predictable short-term | ⚠️ They are not. Early warning ≠ prediction (§5.4 → geo-earthquakes-seismology-and-volcanism) |
| Lava is the main volcanic killer | ⚠️ Pyroclastic flows and lahars are (§6 → geo-earthquakes-seismology-and-volcanism) |
| All volcanoes erupt similarly | ⚠️ Viscosity and gas escape decide everything (§6 → geo-earthquakes-seismology-and-volcanism) |
| Fine sediment erodes most easily | ⚠️ Cohesion — see the Hjulström curve (§7 → geo-surface-processes-soils-and-hydrology) |
| Porosity and permeability are the same | ⚠️ Clay: high porosity, no permeability (§10 → geo-surface-processes-soils-and-hydrology) |
| Groundwater flows downhill | ⚠️ It follows hydraulic head, and can flow up (§10 → geo-surface-processes-soils-and-hydrology) |
| Aquifers refill if you stop pumping | ⚠️ Compaction is often permanent (§10 → geo-surface-processes-soils-and-hydrology, §17.1) |
| "Fossil water" is renewable | ⚠️ Millennial residence times. It's mining (§10 → geo-surface-processes-soils-and-hydrology) |
| A 100-year flood happens once a century | ⚠️ 1% annual probability. ~26% chance in 30 years (§9 → geo-surface-processes-soils-and-hydrology) |
| Levees eliminate flood risk | ⚠️ They transfer it and encourage exposure (§9 → geo-surface-processes-soils-and-hydrology) |
| Melting sea ice raises sea level | ⚠️ It's floating. Land ice is the one that matters (§12 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
| Soil is renewable on human timescales | ⚠️ 0.01–0.1 mm/yr (§8 → geo-surface-processes-soils-and-hydrology) |
| Rare earths are geologically rare | ⚠️ Processing and supply concentration are the constraint (§14 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
| Reserves = how much exists | ⚠️ Reserves are economic; resources are geological (§14 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
| Silicate weathering will fix CO₂ | ⚠️ It's the thermostat, and it's ~10⁵–10⁶ years too slow (§13 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
| Disaster losses rise because hazards rise | ⚠️ Exposure and vulnerability dominate (§15 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
| A geophysical inversion gives the answer | ⚠️ Non-unique. Needs constraints and uncertainty (§16 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
§19. Numbers
EARTH
Radius 6371 km · Age 4.54 Ga · Crust 0–35 km (oceanic ~7 km)
Lithosphere ~100 km · Mantle to 2890 km · Core to 6371 km
⚠️ Deepest borehole ~12 km · Plate motion 10–100 mm/yr
⚠️ Oldest ocean floor ~200 Ma · Oldest continental crust ~4 Ga
TIME
4.54 Ga formation · 2.4 Ga Great Oxidation · 541 Ma Cambrian
252 Ma Permian-Triassic (~90% marine species) · 66 Ma K-Pg · 11.7 ka Holocene
¹⁴C half-life 5730 yr, ⚠️ useful to ~50 ka
EARTHQUAKES
⚠️ +1 magnitude = ~32× energy · +2 = ~1000×
P ~6 km/s crust, S ~3.5 km/s · ⚠️ brittle-ductile transition ~10–15 km
M₀ = μAD
WATER
Ocean 96.5% of Earth's water · ice caps ~1.7% · ⚠️ groundwater ~1.7%
Rivers and lakes ~0.01% · Ocean salinity ~35 psu
Thermohaline overturning ~1000 yr
⚠️ Groundwater residence: days to millions of years
Darcy: Q = −KA(dh/dl)
RATES
⚠️ Soil formation 0.01–0.1 mm/yr
Continental drying expansion ~831,600 km²/yr (§17.1)
⚠️ Groundwater = 68% of non-glaciated continental TWS loss (§17.1)
Asia groundwater trend ~−55 km³/yr (§17.1)
§20. Books
| Author | Work | Why |
|---|---|---|
| Marshak | Earth: Portrait of a Planet | ⚠️ The best broad introduction |
| Press, Siever et al. | Understanding Earth | Classic survey |
| Stein & Wysession | An Introduction to Seismology, Earthquakes, and Earth Structure | ⚠️ §5 → geo-earthquakes-seismology-and-volcanism and §16 → geo-oceans-cryosphere-cycles-hazards-and-observation, definitively |
| Fetter | Applied Hydrogeology | ⚠️ §10 → geo-surface-processes-soils-and-hydrology, the standard |
| Freeze & Cherry | Groundwater | ⚠️ The classic, and now free online |
| Anderson, Woessner & Hunt | Applied Groundwater Modeling | §10 → geo-surface-processes-soils-and-hydrology quantitatively |
| Dingman | Physical Hydrology | §9 → geo-surface-processes-soils-and-hydrology |
| Bierman & Montgomery | Key Concepts in Geomorphology | §7 → geo-surface-processes-soils-and-hydrology |
| Montgomery | Dirt: The Erosion of Civilizations | ⚠️ §8 → geo-surface-processes-soils-and-hydrology's stakes, superbly written |
| McPhee | Annals of the Former World | ⚠️ The best writing about geology, full stop. Read it for deep time (§3 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| Talley et al. | Descriptive Physical Oceanography | §11 → geo-oceans-cryosphere-cycles-hazards-and-observation |
| Cuffey & Paterson | The Physics of Glaciers | §12 → geo-oceans-cryosphere-cycles-hazards-and-observation |
Practical: USGS publications and data (⚠️ enormous, free, authoritative), national geological survey maps, IRIS/EarthScope for seismic data, GRACE/ GRACE-FO at JPL (§17.1), Copernicus/Sentinel and Landsat archives, ObsPy and SeisBench for seismological work in Python, MODFLOW for groundwater modelling, and QGIS.
§21. Quick Reference
21.1 Picker
| Question | Approach |
|---|---|
| How old is this rock? | ⚠️ U-Pb zircon; bracket sediments with ash or intrusions (§3 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| How old is this organic material (<50 ka)? | ¹⁴C (§3 → geo-earth-structure-tectonics-rocks-and-deep-time) |
| Where was the earthquake? | ⚠️ S−P times from ≥3 stations (§5.2 → geo-earthquakes-seismology-and-volcanism) |
| How big was it? | M_w from seismic moment (§5.3 → geo-earthquakes-seismology-and-volcanism) |
| How much will it shake here? | ⚠️ Site conditions — soft sediment amplifies (§5.3 → geo-earthquakes-seismology-and-volcanism) |
| Will this slope fail? | ⚠️ Factor of safety; pore pressure after rain (§7 → geo-surface-processes-soils-and-hydrology) |
| How much water will this well yield? | Darcy, aquifer tests, cone of depression (§10 → geo-surface-processes-soils-and-hydrology) |
| Is this water source sustainable? | ⚠️ Compare extraction to recharge, and check residence time (§10 → geo-surface-processes-soils-and-hydrology) |
| Is the ground subsiding? | ⚠️ InSAR (§16 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
| Is the aquifer losing storage regionally? | ⚠️ GRACE, with §17.1's caveats (§17.1) |
| Are there earthquakes we're missing? | ⚠️ DL phase picking; DAS if fibre is available (§17.2) |
| What's the flood risk here? | ⚠️ Return period ≠ schedule; check non-stationarity (§9 → geo-surface-processes-soils-and-hydrology) |
| Map faults under forest | LiDAR bare-earth (§16 → geo-oceans-cryosphere-cycles-hazards-and-observation) |
21.2 Sanity checks
- What timescale is this process on, and does my intuition match it? (§3 →
geo-earth-structure-tectonics-rocks-and-deep-time) - Am I confusing rate with total, or magnitude with intensity? (§5.3 →
geo-earthquakes-seismology-and-volcanism) - Is this a probability or a schedule? (§9 →
geo-surface-processes-soils-and-hydrology) - Is this resource renewable at the rate I'm using it? (§10 →
geo-surface-processes-soils-and-hydrology, §14 →geo-oceans-cryosphere-cycles-hazards-and-observation) - Is this measurement direct, or derived by subtracting models? (§17.1)
- Is this inversion unique, and where's the uncertainty? (§16 →
geo-oceans-cryosphere-cycles-hazards-and-observation) - Is the trend hazard, or exposure and vulnerability? (§15 →
geo-oceans-cryosphere-cycles-hazards-and-observation)
§22. Method
§1–§16 → geo-earth-structure-tectonics-rocks-and-deep-time, geo-earthquakes-seismology-and-volcanism, geo-surface-processes-soils-and-hydrology, geo-oceans-cryosphere-cycles-hazards-and-observation and §18–§19 rest on settled science — plate tectonics (confirmed 1960s),
stratigraphic principles (Steno, Hutton, Lyell), radiometric dating, Darcy's law (1856),
elastic rebound (Reid, 1910), and seismological wave theory — sourced from the references
in §20, chiefly Marshak, Stein & Wysession, Fetter, Freeze & Cherry, and
Bierman & Montgomery. ⚠️ None of that needed verification.
Scoped to complement: the atmosphere sits in a weather-science reference, and the
mechanics in a Newtonian-mechanics reference. ⚠️ §9 → geo-surface-processes-soils-and-hydrology's floods and §12 → geo-oceans-cryosphere-cycles-hazards-and-observation's cryosphere touch
both deliberately.
Two searches were run in August 2026, and ⚠️ both were about measurement capability rather than events — which is why §17 is framed as "what observation changed" rather than as news. The science moved because we can now measure things we couldn't.
Confidence. High in §1–§16 → geo-earth-structure-tectonics-rocks-and-deep-time, geo-earthquakes-seismology-and-volcanism, geo-surface-processes-soils-and-hydrology, geo-oceans-cryosphere-cycles-hazards-and-observation. High in §17's factual content, which came from
peer-reviewed primary literature — Science Advances and EGUsphere/Copernicus
for §17.1, NASA/JPL mission pages, and Nature Communications, Geophysical Journal
International, Geophysical Journal International (2026), and Scientific Reports for
§17.2.
⚠️ The §17.1 caveat is the one I'd want carried forward, and I've given it a gotcha box rather than a footnote. GRACE does not measure groundwater — it measures total mass change, and groundwater is what's left after subtracting modelled soil moisture, snow, surface water and glaciers. ⚠️ Model error propagates directly into the headline number. The published re-analysis finding earlier Northwest India depletion rates were likely overestimated is exactly the kind of correction that gets far less attention than the original alarming figure — and the validation against ~23,000 monitoring wells across US aquifers (correlations 0.52–0.95) is the reason to trust continental-scale trends more than single-basin headlines.
⚠️ §17.2 I have deliberately not overstated. Deep learning genuinely transformed seismic catalogue density — several-fold more events in the case I cited — and the limitations are documented in the same literature: poor transfer to high-frequency borehole and DAS data without retraining, prediction inconsistency requiring mitigation, and an active 2026 paper still asking whether deep learning or manual picking is better. A denser catalogue is not automatically a better one, and the field knows it.