A water-sensitive signal still needs a model of where the water is
A loop at the ground surface transmits an electromagnetic pulse and then listens for a faint magnetic-resonance response. The response is linked to hydrogen in subsurface water. This makes surface magnetic-resonance sounding unusually useful for groundwater investigation, but it does not turn the loop into a transparent window showing an aquifer’s exact boundaries.
This guide follows the evidence from excitation and a measured decay to depth-dependent water estimates. It concentrates on surface loops, pulse moments and broad subsurface sensitivity. The existing borehole NMR guide addresses a different observation geometry: a tool moving through a well. The physics connects, while the depth resolution, noise and interpretation problems differ.
The examples are original fictional calculations with simplified responses. They are not survey-design instructions, equipment settings or a groundwater-yield guarantee. The word nuclear refers to atomic nuclei involved in magnetic resonance; the method discussed here is not a radioactive-source water survey.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1: Understand the surface observation · Chapters 1–3
Begin with magnetic resonance, not electrical conductivity alone
Route 2: Interpret decay and missing signals · Chapters 4–5
Route 3: Build and test a depth model · Chapters 6–8
Route 4: Separate water from hydraulic performance · Chapters 9–10
Route 5: Audit and transfer the interpretation · Chapters 11–12
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Understand the surface observation · Chapters 1–3
Begin with magnetic resonance, not electrical conductivity alone
A surface loop has distributed sensitivity
Pulse moment changes the weighting, not a single sampling depth
Interpret decay and missing signals · Chapters 4–5
Read decay time with its magnetic context
Dead time can hide rapidly decaying water signals
Build and test a depth model · Chapters 6–8
Noise removal must preserve the signal of interest
An inversion combines a response model with constraints
Ground conductivity and magnetic properties belong in the model
Separate water from hydraulic performance · Chapters 9–10
Detected water is not automatically drainable water
Hydraulic conductivity requires an additional relationship
Audit and transfer the interpretation · Chapters 11–12
Build an interpretable water-storage estimate
Keep the observation and the interpretation reproducible
CHAPTER 1 OF 12
1. Begin with magnetic resonance, not electrical conductivity alone
Hydrogen nuclei in water respond to an applied electromagnetic field under magnetic-resonance conditions. After excitation ends, a receiver measures a weak response as the detectable magnetisation evolves. The response is related to the water present within the system’s sensitivity. It is not the same observation as an electrical-resistivity survey that infers water through bulk electrical behaviour.
This water sensitivity is valuable, yet the recorded voltage also depends on excitation, geometry, ground properties and the acquisition window. A stronger voltage does not automatically mean a proportionally larger aquifer. Before converting amplitude into water content, identify how the instrument and subsurface model connect the water distribution to the received signal.
USGS describes magnetic-resonance sounding as a surface method for indirectly estimating subsurface water and aquifer properties. The indirect qualifier matters. The observation can constrain water in a way complementary to other geophysics, while the depth distribution and hydraulic interpretation still require modelling, calibration and independent site information.
CHAPTER 2 OF 12
2. A surface loop has distributed sensitivity
The transmitted field and the receiver’s sensitivity vary throughout the ground. A measured response therefore combines contributions from a volume rather than a single point directly beneath the loop centre. Loop geometry, background magnetic field and electrical properties influence that weighting. The footprint cannot be understood from the loop outline alone.
A one-dimensional sounding usually represents a laterally simplified subsurface beneath a surface arrangement. That can be useful when layering is an adequate approximation. If a narrow channel or irregular water-bearing body crosses the sensitive region, the same simplified interpretation may assign its signal to an apparently continuous layer. The geometry is part of the model, not an observed fact.
Think of a weighted sum: one part of the subsurface may contribute strongly and another weakly even if they contain equal water volumes. Comparing that observation with a local well log requires a support argument. Agreement at the well does not by itself show that lateral variation across the surface measurement’s entire sensitive region is negligible.
CHAPTER 3 OF 12
3. Pulse moment changes the weighting, not a single sampling depth
For a simplified rectangular pulse, pulse moment is current multiplied by pulse duration. Changing it changes excitation and the resulting depth sensitivity. A sequence of pulse moments provides information used to infer a subsurface model. It does not act like lowering a probe to a succession of sharply defined depths.
In particular, larger pulse moment should not be translated into a universal depth using one fixed multiplier. The relationship depends on the loop, field and subsurface response. Sensitivity regions overlap, and the excitation response can be nonlinear. A graph against pulse moment is therefore not already a water-content log against depth.
For interpretation, retain the pulse sequence and the sensitivity calculation used in inversion. Ask which depth intervals the available data distinguish, rather than how many depth pixels appear on the finished plot. More plotted cells do not establish more independent subsurface information. The scientifically meaningful resolution belongs to the observation and model together.
CHAPTER 4 OF 12
4. Read decay time with its magnetic context
A simple response envelope can be written E(t) = E₀exp(−t/T₂*). Here E₀ is an extrapolated initial amplitude and T₂* is an effective decay time for the stated observation. Real data may combine several components and need a more detailed representation. One fitted exponential is a model choice, not proof of one uniform pore population.
Surface free-induction decay can be influenced by magnetic-field inhomogeneity as well as relaxation associated with the material and fluid. T₂* should not be treated as interchangeable with a borehole spin-echo T₂ value. A short effective decay can therefore have several explanations, and its direct conversion to a unique pore size is not justified.
Knight and colleagues’ field comparison connected surface and borehole observations by accounting for magnetic effects and acquisition limits. The important transfer lesson is to compare compatible quantities rather than demand identical decay numbers from different methods. A physically justified difference can be more informative than an artificial agreement obtained by relabelling parameters.
CHAPTER 5 OF 12
5. Dead time can hide rapidly decaying water signals
After transmission, there is an interval before usable reception begins. Signals that decay strongly during this interval may be difficult to recover. Relaxation during the pulse can also matter. These effects mean that a weak observed response does not necessarily imply little total water, especially where some water contributes very short-lived signals.
In the teaching exponential, a 100-unit initial response with T₂* = 20 ms has only about 13.5 units left at 40 ms. A second 100-unit response with T₂* = 100 ms retains about 67.0 units. Their initial amplitudes match, while their first usable measurements can differ greatly. These values are not a specification for any instrument.
Extrapolation back to E₀ depends on the decay model and signal quality. It does not recreate information with unlimited certainty. When short components are unobserved, fitting a smooth long component may underestimate the total detectable-at-zero-time response or miss water classes. Report the acquisition window and what kinds of signals it may exclude.
CHAPTER 6 OF 12
6. Noise removal must preserve the signal of interest
Surface magnetic-resonance responses are weak and can be affected by environmental electromagnetic noise. Repetition, reference information and processing can improve an estimate when their assumptions hold. But a cleaner-looking curve is not automatically a more faithful water signal. Filtering can also alter early-time behaviour or remove part of the response.
For independent random noise with stable variance, averaging N comparable repetitions reduces the standard error approximately as one over sqrt(N). Coherent interference does not necessarily cancel that way. A persistent unwanted oscillation can survive repeated averaging and appear increasingly convincing if the analyst assumes every retained feature belongs to the subsurface.
Keep raw or adequately preserved diagnostic records, rejected observations and processing versions. Compare plausible processing choices against synthetic or independent checks. The question is not simply which method gives the smoothest decay, but whether it preserves the amplitude, phase and timing information needed for the water and depth interpretation.
CHAPTER 7 OF 12
7. An inversion combines a response model with constraints
A conceptual forward model adds weighted contributions from water in different parts of the subsurface. The inversion seeks water and decay parameters whose predicted responses agree with the measurements, subject to chosen constraints. Several models can fit noisy or incomplete data similarly well. A best-fitting section is not automatically a unique reconstruction.
Smoothness or other regularisation stabilises the problem but can spread a sharp layer over a wider interval. Strong prior constraints can make the model look precise because alternatives were disallowed. Those may be reasonable choices, but the report should identify them and test whether the central interpretation survives plausible alternatives.
The practical check is a family of compatible models, resolution or sensitivity information, and comparison with independent geology and water observations. Distinguish a stable integrated water estimate from an unstable layer boundary. A dataset may constrain total weighted response reasonably well while leaving its vertical distribution much less certain.
CHAPTER 8 OF 12
8. Ground conductivity and magnetic properties belong in the model
Electrical conductivity affects electromagnetic propagation and can alter the amplitude and phase relationships used in a surface sounding. Magnetic properties can influence the local field and effective decay. A water-sensitive measurement is therefore not a measurement independent of every other ground property. Those properties enter through the observation physics.
Complementary electrical or electromagnetic observations can help constrain a response model where appropriate. That does not mean copying a resistivity section into a water-content image. Each method has its own support, nonuniqueness and uncertainty. Joint interpretation is useful when the shared parameters and independent information are explicitly defined.
Do not treat all reduced amplitudes as dry zones. A change in conductivity, geometry, noise or short-decay sensitivity may explain part of the difference. The model should test material alternatives, and uncertainty in those inputs should be carried into the final water estimate rather than removed by assuming one convenient background.
CHAPTER 9 OF 12
9. Detected water is not automatically drainable water
A water-content estimate describes water per bulk volume under the method’s detection and modelling assumptions. Specific yield describes the water released from an unconfined material under an appropriate drainage change. Total porosity, detectable water content and specific yield are different quantities, even though all may be expressed as fractions.
For example, a layer can contain substantial water in pores that do not readily drain during the period of interest. Conversely, acquisition limits may miss rapidly decaying contributions. A reported 0.20 m³/m³ surface-NMR water estimate is not automatically a promise that 20% of the formation volume can be pumped out.
To estimate recoverable water, connect the geophysical result with hydrogeology, boundaries, drawdown, recharge and the relevant operational time scale. The surface method may improve the conceptual model without independently establishing a sustainable abstraction rate. A resource claim needs more than multiplying an interpreted water fraction by a large mapped volume.
CHAPTER 10 OF 12
10. Hydraulic conductivity requires an additional relationship
Relaxation information can be linked empirically with hydraulic properties under suitable conditions, but hydraulic conductivity is not directly recorded by the receiving loop. Relationships depend on pore structure, fluid and material properties, and calibration. A long decay should not simply be labelled high yield without checking the applicable model.
An illustrative power-law model K = CθT² exposes sensitivity: doubling the chosen time parameter multiplies predicted K by four if all other terms are held fixed. This is a hypothetical model structure, not a universal NMR conductivity equation. It shows why differences between T₂ and T₂* or uncertain decay components can matter greatly to an inferred hydraulic parameter.
Compare hydraulic predictions with independently supported pumping or other hydraulic evidence over comparable materials and scales. Keep calibration and validation separate. A relation fitted to a few local values may be useful within that setting, yet transfer to different lithology or magnetic conditions can fail even when the numerical calculation is performed correctly.
CHAPTER 11 OF 12
11. Build an interpretable water-storage estimate
If a justified model assigns volumetric water content θ to a layer of thickness b, its water-equivalent depth is θb. That is an integration of interpreted water content over thickness, not a direct measurement of recoverable storage. Both θ and b may be uncertain or correlated because different layer models can reproduce similar responses.
Suppose a 10 m layer at θ = 0.15 and a 5 m layer at θ = 0.30 each contain 1.5 m of water-equivalent storage under a uniform-layer interpretation. Their integrated amounts agree, but their layer geometries do not. The sounding must have enough depth information to distinguish them before one geometry is presented as established.
For area-wide volume, add a further spatial-representativeness argument. A surface sounding at one location is not automatically the mean for a basin. Multiple soundings, geology and appropriate interpolation can constrain the extension, with uncertainty retained. Reporting more volume digits does not compensate for unsampled lateral variation.
CHAPTER 12 OF 12
12. Keep the observation and the interpretation reproducible
A usable interpretation preserves geometry, field context, pulse sequence, recording window, noise information, processing choices, forward-model assumptions and inversion constraints. Report which water classes may be weakly detected and which depth intervals remain poorly resolved. This lets another reader distinguish observation limits from hydrological absence.
The transfer test changes the context deliberately: a more conductive overlying layer, a shorter decay, different noise or lateral heterogeneity. Predict how the data and inference would change. A robust explanation identifies which conclusions survive and which need new evidence, rather than carrying an attractive colour section unchanged to another site.
Finish with a bounded resource statement. Surface magnetic resonance can provide valuable water-sensitive evidence and complement boreholes or other geophysics. It does not uniquely establish every layer boundary, total pore water, hydraulic conductivity and sustainable well yield at once. Clear boundaries make the evidence more useful, not less.
Independent worked case and evidence checks
Consider a fictional single-component response with E₀ = 100 units and T₂* = 50 ms. Use E(t) = 100exp(−t/50), with t also in milliseconds. At 10 ms the response is about 81.87 units, at 50 ms about 36.79 and at 100 ms about 13.53. Matching time units is essential before interpreting the curve.
If the first usable observation occurs at 40 ms, the response is about 44.93 units. Calling that value the initial amplitude would underestimate the specified E₀ by more than half. Extrapolating correctly requires a supported decay model and sufficient useful observations; the missing interval cannot simply be ignored.
Change T₂* to 20 ms while retaining E₀ = 100. At 40 ms, only about 13.53 units remain. Change it instead to 100 ms, and about 67.03 remain. Equal initial responses can therefore look very different in the same acquisition window. This is a changed-detectability scenario, not evidence that the short-decay case contains less initial water response.
Now combine two fictional components: 40 units with a 10 ms decay and 60 units with a 100 ms decay. At 40 ms their contributions are about 0.733 and 40.219 units, totalling 40.952. The observed late response is dominated by the slow component, even though the fast component supplied 40% of the initial amplitude.
A late-time single-exponential fit could describe much of that slow tail while missing the early component. A visually good tail fit is therefore not a complete test of the initial water response. Compare the observation window with plausible short components before equating extrapolated amplitude with all water in the formation.
For an ideal rectangular pulse, let current be 2 A and duration 0.02 s. Pulse moment is 0.04 A·s. Doubling current doubles that moment; doubling duration does too. This arithmetic does not imply identical excitation histories or identical relaxation effects during the pulse, and it does not supply a universal depth in metres.
Use an intentionally simple two-zone sensitivity model to illustrate inversion. Let one measurement equal 0.8θ₁ + 0.2θ₂ and another equal 0.3θ₁ + 0.7θ₂. If θ₁ = 0.10 and θ₂ = 0.30, the two responses are 0.14 and 0.24. These dimensionless weighted responses are not a real surface-NMR forward kernel.
If only the first response were available, θ₁ = 0.15 and θ₂ = 0.10 would also produce 0.14. One weighted observation cannot uniquely determine two water fractions. The second independent weighting helps distinguish these particular models, illustrating why multiple informative pulse responses matter more than the number of colours in a final section.
Even with two responses, nearly identical weightings would provide weak separation in noisy data. If the second weighting were 0.79 and 0.21 rather than 0.3 and 0.7, its information would largely repeat the first. Repeated data are valuable for precision, but new independent sensitivity is what helps distinguish the zones.
For a noise example, suppose one comparable observation has independent random-noise standard deviation of 12 units. Averaging 36 independent observations would reduce that component to 2 units under the ideal model. An additional coherent 3-unit interference would not automatically shrink to 0.5 merely because the same number of repeats was used.
If processing removes an unwanted oscillation but also suppresses the first usable part of the water response, a lower residual-noise figure is not the whole quality assessment. Check the recovered amplitude and decay against a known synthetic response or an appropriate independent comparison. Signal preservation and noise reduction are related but separate requirements.
For depth integration, take a uniform interpreted layer 8 m thick at θ = 0.20. Its water-equivalent amount is 1.6 m. Over a representative 10,000 m² area that would equal 16,000 m³. The word representative is a substantive assumption, and this amount is not automatically drainable or sustainably extractable.
If an independent unconfined-aquifer interpretation assigns specific yield 0.08, a 1 m water-table decline over that area would correspond to 800 m³ released under the simplified relation. That differs from total water held in the 8 m layer. Water content and the response to a head change answer different questions.
Test layer ambiguity with 8 m at θ = 0.20 versus 4 m at θ = 0.40. Both integrate to 1.6 m of water. A model may distinguish them only if the pulse responses constrain the distribution sufficiently. Equal integrated storage is not proof that either vertical arrangement is correct.
For the hypothetical K = CθT² sensitivity, hold C and θ fixed and increase T from 40 to 60 in consistent time units. Predicted K increases by (60/40)² = 2.25. If the two T values arose from incompatible decay definitions, that numerical contrast would not be a defensible hydraulic comparison.
Now hold observed amplitudes fixed but change the assumed conductivity of an overlying layer. The forward sensitivity can change, and the inversion may assign different water fractions or depths. The observation did not change; the model interpretation did. Independent electrical information can test this alternative rather than forcing the original section to remain unchanged.
A borehole next to the loop intersects a thin permeable bed. That point observation can help test geology, but the surface sounding samples a broader weighted region. If the bed pinches out laterally, the borehole value may not represent the full sensitive volume. Neither method should be declared wrong solely because their reported water profiles differ.
Finally, an interpreted low-water zone coincides with an interval where short decays would be poorly observed. Preserve both possibilities: genuinely less water and water outside the effective detection window. Additional suitable acquisition or independent measurements may separate them. A blank part of the section does not identify its own cause.
For a detection-threshold illustration, let the usable amplitude threshold be 5 units under a declared fictional noise rule. A response starting at 20 units with a 10 ms decay retains only 0.366 units at 40 ms and would not pass that rule. A 10-unit response with a 100 ms decay retains 6.703 units and would pass. The weaker initial response can therefore be more detectable in the late window.
Do not turn that threshold into a real minimum water fraction without the appropriate sensitivity model. Amplitude units, loop response, geometry and noise all belong to the conversion. A method can report a data-detection threshold more directly than a universal groundwater-detection limit. Keep those two statements separate when explaining an apparently empty part of a section.
Finally, compare a sounding acquired before and after a change in electrical interference. A lower usable signal after the change may reflect a poorer observation window rather than lost water. Preserve quality and processing information for both dates before treating the difference as a storage trend. Repeating an indirect method creates a temporal comparison only when its relevant observation conditions are understood.
Your independent answer should recover the decay values, the 40.952-unit two-component response, the two-zone ambiguity and the 1.6 m integrated water amount. Then separate pulse-moment arithmetic, water-sensitive observation, depth inversion and hydraulic inference. That separation is the transferable skill, not memorising a loop-to-aquifer conversion.
Worked comparison
| Quantity or feature | Useful interpretation | Limit to retain |
|---|---|---|
| Received response | Water-sensitive magnetic observation | Geometry and noise still matter |
| Pulse moment | Changed excitation weighting | Not a unique depth coordinate |
| Effective decay T₂* | Observed signal persistence | Not interchangeable with T₂ |
| Small late-time signal | Little signal in that window | Not necessarily little total water |
| Interpreted θ × thickness | Water-equivalent layer amount | Not sustainable well yield |
Questions and practical checks
Does this method use radioactive sources?
The surface magnetic-resonance method described here excites and detects magnetic responses of nuclei. The term nuclear does not mean that a radioactive source is used for this measurement.
Is it the same as borehole NMR?
The magnetic-resonance connection is real, but observation geometry, sensitive volume, acquisition windows and decay definitions can differ. Compare compatible parameters rather than copying a borehole interpretation.
Does larger pulse moment mean one known deeper point?
No. It changes distributed sensitivity under the survey model. The response is not a point sample indexed directly by depth.
Can an undetected response prove dry ground?
Not alone. Noise, short decay, conductivity and sensitivity may limit detection. State the alternatives and the evidence needed to separate them.
Can NMR alone certify a productive well?
No. Water-sensitive evidence can improve an aquifer model, but hydraulic performance and sustainable abstraction require additional site-specific hydrogeological assessment.
Continue with the connected subjects
Compare magnetic-resonance observations at another scale
Borehole NMR: T₂ logs, detected water and hydraulic-conductivity interpretation
Connect water-sensitive evidence to hydrological storage
Hydrogeological engineering: groundwater pathways and independent evidence
Repeat microgravity: mass change, reference stability and water-storage attribution
Groundwater recharge estimation and the distinction between storage and input
Sources and scope
USGS: surface magnetic-resonance sounding and its measured parameters
US EPA: surface nuclear magnetic resonance, depth sensitivity and limitations
Knight and colleagues (2012): field comparison of surface and borehole NMR
Primary study: surface-NMR modelling, complex data, static dephasing and inversion
The numerical cases in this guide are original fictional teaching examples, not measurements from a real site. This is an educational explanation, not a field safety instruction, engineering approval or an official certification. For a real project, use qualified practitioners, current product documentation and site-specific evidence. Sources were checked on 30 October 2026.