A cool patch is a useful clue—not a complete groundwater measurement
A thermal image shows a cool streak entering a warm stream. It is tempting to draw a groundwater arrow immediately. But the camera records radiation from visible surfaces. Shade, a tributary, reflected surroundings, shallow water and groundwater can produce patterns that need separating before the image becomes a hydrological explanation.
This guide follows the chain from surface radiation to a mapped anomaly and then to a testable water-source hypothesis. It complements the existing stream heat-budget and subsurface heat-tracing guides. Its particular job is to explain the image: what its pixels represent, what they miss, and why a persuasive colour pattern is not yet a discharge estimate.
All numerical examples are fictional. The simplified radiation and mixing calculations expose assumptions rather than reproduce a camera’s retrieval software. They are not instructions for aircraft operation, stream access or environmental approval. You can practise the complete reasoning using supplied images and data without entering the water.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1: Read what the camera sees · Chapters 1–3
Route 2: Test the thermal pattern · Chapters 4–5
Colour is a display choice, not a universal thermal threshold
Route 3: Move from anomaly to mechanism · Chapters 6–8
Route 4: Check uncertainty and scale · Chapters 9–10
Route 5: Practise attribution and transfer · Chapters 11–12
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Read what the camera sees · Chapters 1–3
Begin with a visible surface, not a transparent river
Separate radiance from the temperature assigned to it
A pixel can contain several different surfaces
Test the thermal pattern · Chapters 4–5
Colour is a display choice, not a universal thermal threshold
Change the season before trusting the sign
Move from anomaly to mechanism · Chapters 6–8
Mixing can hide an inflow or spread its signature
Distinguish groundwater from other thermal sources
Use independent measurements to test different links
Check uncertainty and scale · Chapters 9–10
Account for geometry, obstruction and missing coverage
Quantify sensitivity before reporting a flux
Practise attribution and transfer · Chapters 11–12
Use a repeatable anomaly record rather than a gallery
Transfer the reasoning to an unfamiliar stream
CHAPTER 1 OF 12
1. Begin with a visible surface, not a transparent river
Thermal-infrared imaging responds to radiation arriving from the surfaces within the camera’s field of view. Over open water, that is a very shallow surface layer rather than a direct view through the water column to the bed. A groundwater plume must affect the observed surface sufficiently to leave a detectable thermal signature.
This distinction explains both the method’s usefulness and its blind spots. A surface-reaching inflow can create a spatial pattern quickly recognised in an image. A deeper inflow that mixes before reaching the surface may leave little contrast. An image without an anomaly therefore cannot establish that no subsurface exchange occurs beneath that reach.
USGS thermal-imaging guidance describes the use of surface contrasts to locate water exchange and guide further investigation. Keep the order clear: image the surface, identify a pattern, develop a mechanism, then seek confirming evidence. The camera’s broad view can improve where you look without directly answering every question about the water moving below it.
CHAPTER 2 OF 12
2. Separate radiance from the temperature assigned to it
A camera detects radiation in its operating spectral range. Temperature is inferred using assumptions and instrument calibration. Surface emission, reflected radiation from the surroundings and atmospheric transmission can contribute to the received signal. The displayed temperature is not simply a thermometer reading taken from every pixel’s water volume.
For an opaque surface, a useful conceptual expression combines emitted radiation weighted by emissivity and reflected environmental radiation weighted by one minus emissivity. This omits several operational details deliberately. It shows why changing the apparent surroundings or viewing conditions can affect the result even when the physical water temperature is unchanged.
For an ideal broad-band blackbody illustration, emission scales with absolute temperature to the fourth power. Celsius values cannot be inserted directly into that relationship. A real thermal camera operates over a band and applies its own response model, so the broad-band example is not a calibration equation. Use it only to understand that the radiation–temperature conversion is nonlinear.
CHAPTER 3 OF 12
3. A pixel can contain several different surfaces
A pixel represents a finite angular footprint, and its ground footprint depends on viewing geometry and distance. Near a bank, the received signal may combine water, wet sediment, vegetation and exposed ground. The temperature assigned to that mixed signal need not equal the water temperature or the area-weighted arithmetic mean of the component temperatures.
Masking non-water surfaces is therefore a physical step, not merely image tidying. A cool vegetated bank can create an apparent stream-edge anomaly if included as water. A narrow groundwater seep can also be diluted within a larger pixel. Increasing spatial resolution may help distinguish components, but it does not automatically solve emissivity, reflection or attribution problems.
Keep the visible-light context and spatial alignment with the thermal observations where available. A small registration shift can move a bank feature into the water mask. Compare stable landmarks and the actual water boundary, rather than assuming that two images align perfectly because their maps use the same coordinate system.
CHAPTER 4 OF 12
4. Colour is a display choice, not a universal thermal threshold
Thermal palettes translate values into colours. If each image automatically stretches its own minimum and maximum, the same blue can represent different temperatures on different dates. A high-contrast display can make a small physical difference look dramatic. Read the numerical scale before comparing the visual strength of two anomalies.
Define an anomaly relative to a suitable reference, such as nearby stream water observed under comparable conditions. A two-degree contrast means something different from a pixel merely appearing dark in a palette. Retain both the absolute temperature information and the reference definition so another reader can reproduce what was classified as unusual.
A screening threshold should also reflect uncertainty and the purpose of the survey. A weak contrast may be worth investigating without supporting a confident source label. Conversely, a strong contrast can still be caused by an outfall or tributary. Detection and attribution are separate decisions; increasing confidence in one does not automatically establish the other.
CHAPTER 5 OF 12
5. Change the season before trusting the sign
Groundwater can be cooler than a stream during one period and warmer during another. A discharge feature may therefore reverse from a cool anomaly to a warm anomaly as the stream’s thermal background changes. The expected sign belongs to the season, weather and hydrogeological setting, not to the word groundwater alone.
Time of day also matters. Solar loading, shade, wind and heat exchange alter surface temperatures. A feature that emerges only during strong daytime heating may have a different explanation from one that persists through contrasting conditions. Repeat observations can test those alternatives, provided the water boundary and viewing context remain comparable.
The multiscale temperature study linked below discusses why surface expression depends on contrast and mixing. Do not translate that into a universal best survey hour or season. The informative window depends on the expected mechanism and site. State the condition that should strengthen or weaken the hypothesised signature, then see whether the observations follow it.
CHAPTER 6 OF 12
6. Mixing can hide an inflow or spread its signature
When an inflow and stream water mix, their temperatures tend toward a mixture determined by their heat and water contributions, with additional exchanges if present. A small inflow with a large initial temperature contrast may leave a visible streak. A larger inflow at nearly the same temperature as the stream may be difficult to see thermally.
For a simplified, fully mixed, steady junction with equal water properties and negligible external heat exchange, QsTs + QgTg = (Qs + Qg)Tm. Here the temperatures must be representative bulk-water values and the flows must share a consistent time basis. A single surface pixel does not automatically supply Tm for that equation.
If stream water is 20°C, groundwater 12°C and the measured bulk mixture 18°C, then Qg/Qs = (20 − 18)/(18 − 12) = one third. That result is conditional on the stated model and end members. Without independent bulk temperatures, mixing completeness and heat-budget evidence, applying it to a colour patch would overstate the image’s information.
CHAPTER 7 OF 12
7. Distinguish groundwater from other thermal sources
A tributary, a drainage outlet, treated effluent, shade or shallow bank storage can all complicate a temperature pattern. A plume’s position and persistence help develop hypotheses, but neither appearance nor a temperature sign uniquely identifies its source. Trace the physical connection and compare independent water or hydraulic information where appropriate.
The USGS Meduxnekeag study used mapped thermal anomalies to direct ground observations. Those checks distinguished groundwater seeps from wastewater outfalls and drainage features. The transferable lesson is the classification process, not a claim that every anomaly in another catchment belongs to the same set of sources or has the same importance.
Separate source identity from source quality. A confirmed groundwater contribution is not automatically clean water, and an unconfirmed anomaly is not proof of pollution. Temperature alone cannot certify chemical composition or ecological suitability. If quality matters, the sampling question and the relevant analyses must be designed for that claim.
CHAPTER 8 OF 12
8. Use independent measurements to test different links
A contact temperature profile can test whether the surface pattern extends through the water column. Hydraulic-head information can constrain exchange direction under an appropriate local interpretation. Seepage measurements can examine a small area of bed flux. Solute or isotope evidence can help compare source hypotheses. These observations do not all have the same spatial support.
Choose the confirming measurement for the unresolved explanation. If the question is whether a cool pixel is actually vegetation, better surface classification may be the first useful check. If the question is water-source identity, repeating the palette may add little. If the question is total inflow, representative discharge or a justified mass-and-heat model becomes necessary.
Agreement is strongest when methods constrain different parts of the chain and their uncertainties are understood. Two temperature-based methods may share thermal assumptions. A point seepage result does not validate an entire image corridor automatically. Preserve each observation’s location, time, support and conversion before comparing conclusions.
CHAPTER 9 OF 12
9. Account for geometry, obstruction and missing coverage
Vegetation, overhanging banks and viewing angle can obscure water. An apparently continuous map may contain stretches that were never observed clearly. Retain those gaps instead of allowing interpolation to imply evidence everywhere. A masked or hidden location is different from a valid water pixel with no detected anomaly.
Changing water level alters the visible bank, water depth and surface pattern. It can also change the exchange process itself. Before comparing dates, identify which differences belong to observation geometry and which might be physical. The same geographical coordinates do not guarantee the same surface or water volume was observed.
Define the spatial claim to match the usable coverage. A survey may support a catalogue of visible candidate discharge locations rather than a complete count of all inflows. Its screening value can remain high even when completeness is limited. State what could systematically escape detection so absence on the map is not mistaken for absence in the river.
CHAPTER 10 OF 12
10. Quantify sensitivity before reporting a flux
In a mixing calculation, an end-member temperature close to the mixture can make the inferred flow ratio unstable. Small temperature errors can then produce large relative changes in the inferred contribution. More decimal places on the displayed image do not fix a poorly separated set of end members.
Using the preceding 20°C stream and 12°C inflow, a mixture of 18.2°C gives Qg/Qs ≈ 0.290, while 17.8°C gives about 0.379. These are sensitivity scenarios, not confidence bounds. They show why the temperature definition and uncertainty should travel with any quantitative interpretation.
Heat exchange after entry can produce another mismatch. If the plume warms before the chosen downstream observation, the simple two-source equation may underestimate its cool-water contribution. If another cool tributary also enters, it may overattribute cooling to groundwater. A flux claim needs a control volume and all material heat and water pathways, not just plausible algebra.
CHAPTER 11 OF 12
11. Use a repeatable anomaly record rather than a gallery
A useful record identifies observation time, image calibration and processing version, spatial reference, water mask, display range, anomaly rule and confirming evidence. Keep the original values and the classification separate. If a later interpretation changes a candidate’s source label, preserve why it changed rather than silently replacing the earlier explanation.
For a repeat survey, compare temperature differences on a common numerical basis and track changes in water level and coverage. A moving anomaly may reflect a moving water boundary or mixing zone, not migration of a spring. A missing anomaly may reflect loss of contrast rather than cessation of discharge. Each comparison should identify which explanation the evidence actually separates.
A final map can usefully distinguish observed thermal feature, independently supported inflow and quantitatively estimated discharge. Those categories communicate progress through the evidence chain. Combining them under one groundwater symbol hides important differences in confidence and measurement support. Clarity comes from keeping the categories meaningful, not from filling every blank.
CHAPTER 12 OF 12
12. Transfer the reasoning to an unfamiliar stream
Begin the new case with the camera’s observation boundary: visible surfaces, pixel support and usable coverage. Then identify the expected thermal contrast and alternative sources. Only after those steps should you ask whether a mixing or exchange model is appropriate. This sequence works even when the new site looks unlike the training example.
Change one consequential condition. Reverse the thermal contrast, deepen the stream, add bank vegetation or move the inflow below a well-mixed surface. Predict which parts of the image should change and which claims become weaker. A good transfer answer explains the mechanism; it does not simply apply the previous colour threshold to a new picture.
The practical endpoint is a bounded conclusion: a thermal anomaly has been observed under stated conditions; particular independent evidence supports a water-source interpretation; discharge remains estimated or unresolved. That conclusion helps someone choose a better next observation without turning remote sensing into a promise that the instrument cannot fulfil.
Independent worked case and evidence checks
Consider a fictional summer image with stream-surface pixels near 22°C and a streak near 17°C. Start by recording the five-degree apparent contrast and checking the numerical scale. Do not label the streak groundwater yet. Its location could reflect an inflow, shade, a mixed bank pixel or another surface effect. The initial finding is a thermal pattern.
The display automatically stretches a second image between 10 and 14°C. A blue patch in that image is 10°C, while the first image’s blue streak was 17°C. The shared colour does not indicate a shared physical temperature. For an honest comparison, retain the actual values, reference temperatures and display ranges.
A visible-light check shows that part of the first streak overlaps overhanging vegetation. Excluding those pixels reduces the candidate water area from 30 to 18 m². This does not prove the remaining area is groundwater. It removes one identified classification error and changes the support of the claim.
Suppose a warm bank occupies 40% of a boundary pixel and water occupies 60%. In an ideal equal-emissivity radiation mixture, the observed radiance combines both surfaces. Treating the assigned pixel temperature as pure water would be inappropriate. The correct first move is to resolve or exclude the mixed surface, not subtract an arbitrary bank temperature.
For a separate ideal blackbody exercise, compare 293.15 K and 303.15 K. The broad-band emitted-radiation ratio is (303.15/293.15)⁴, about 1.144. A ten-degree Celsius increase here corresponds to roughly a 14.4% radiation increase, not a temperature ratio computed from 30/20. This is not a real camera-band retrieval.
Now imagine independent bulk temperatures of 20°C upstream, 12°C in the inflow and 18°C after complete mixing, with negligible external heat exchange. The inferred Qg/Qs is 2/6, or one third. If Qs is independently measured as 0.60 m³/s, the conditional inflow estimate is 0.20 m³/s.
Check the result by reconstructing the mixture: (0.60 × 20 + 0.20 × 12)/0.80 = 18°C. This confirms arithmetic within the assumed model. It does not demonstrate that a real channel was fully mixed or that its surface-image temperature represented the bulk mixture.
Change only the mixture temperature to 18.2°C. The ratio becomes 1.8/6.2, approximately 0.2903, and the corresponding inflow is about 0.174 m³/s. At 17.8°C, the ratio is 2.2/5.8, approximately 0.3793, giving about 0.228 m³/s. This sensitivity matters before attaching precision to the flux.
Hold the 18°C mixture but change the inflow end member from 12 to 16°C. The ratio becomes (20 − 18)/(18 − 16) = 1. The same mixture now requires an inflow equal to Qs under the model. End-member information, not the image pattern alone, determines that difference.
If the inflow and stream were both 20°C, the two-source thermal contrast would vanish regardless of their flow ratio in this idealisation. A thermally invisible contribution could still carry substantial water. Absence of contrast is therefore not a general upper bound on groundwater discharge.
For a winter transfer, let stream water be 4°C and the inflow 10°C. Equal 0.30 m³/s flows would mix to 7°C in the simple model. The inflow now produces a warm rather than a cool feature. A rule retaining only cool anomalies would miss the mechanism it was meant to detect.
Next place a 12°C inflow beneath a deeper, well-mixed 20°C stream surface. Without specifying transport and heat exchange, do not assume the camera must display 12°C at the entry point. It observes the resulting surface expression, which can be diluted, displaced or absent.
A separate cool tributary enters between the upstream and downstream bulk-temperature stations. The original two-source equation is no longer a complete heat-and-water balance. Its numerical fit cannot identify how much cooling came from each source. Add the independently supported pathway or narrow the claim.
Suppose an anomaly persists at the same mapped position on two dates, but one survey had clear water visibility and the other had canopy obstruction. A stable location does not establish stable observed area. Compare usable masks and coverage before interpreting an apparent change in plume size as a change in inflow.
For reach coverage, imagine 800 m of a 1,000 m stream corridor is visible. Ten candidate anomalies occur in the visible part. Reporting ten observed candidates is supported. Reporting a complete reach total of ten is not, and multiplying by 1,000/800 assumes the hidden part has comparable occurrence and detectability.
Independent checking identifies six candidates as seeps, two as drains and leaves two unresolved. Preserve those classes. The unresolved features are not confirmed seeps, and the drains should not remain in a groundwater count merely because they were initially selected by temperature.
A later survey finds no contrast at one confirmed seep while contact observations still show water movement. The observation is consistent with reduced thermal separation or changed mixing, among other possibilities. The image alone does not establish that the seep stopped and restarted between visits.
Compare apparent cooling with heat transport in another bounded example. If 0.50 m³/s of water cools by 1°C, using an illustrative volumetric heat capacity of 4.18 million J/(m³·K) gives a heat-rate difference of about 2.09 MW. That is an energy quantity, not a groundwater volume. Radiation, evaporation, mixing or several processes could contribute. The calculation cannot assign the cause by itself.
For a detection-only exercise, suppose repeat observations of a stable reference give a 0.15°C standard deviation. A hypothetical screening rule at three times that value would flag contrasts above 0.45°C. This is a declared rule for the exercise, not a universal accuracy specification or proof of source identity. Changing the noise context requires reassessing the rule rather than carrying the numerical threshold to another camera.
An anomaly with 0.60°C contrast passes that fictional screen, but a 0.40°C feature does not. The latter could still be physically real; the rule simply does not classify it as detected under the chosen criterion. A screen trades missed features against false detections. Preserve the numerical contrasts and the decision rule so later evidence can revisit the classification.
Finally, ask whether the management question concerns locating a source, estimating its flux or assessing a thermal refuge for organisms. The same image supports those tasks differently. A mapped cool surface patch can help choose sampling locations, yet habitat interpretation needs persistence, depth and ecological context. Choosing the question before the threshold prevents a useful reconnaissance image from being presented as a complete ecological or water-budget assessment.
Retain the original image values after drawing the final map. A later reviewer may need to change the water mask, compare another reference temperature or test an alternative source classification. A coloured screenshot without its numerical and spatial context would make those checks unnecessarily difficult.
Your independent solution should recover the one-third mixing ratio, the 0.20 m³/s conditional inflow and its temperature sensitivity. Then explain why surface visibility, source classification, bulk-temperature support and complete mixing are independent requirements rather than details the algebra can verify.
Worked comparison
| Observation | What it supports first | What still needs evidence |
|---|---|---|
| Cool surface streak | Thermal contrast | Groundwater source identity |
| No visible contrast | No detected surface anomaly | Absence of subsurface exchange |
| Mixed bank pixel | Combined radiance | Pure water temperature |
| Confirmed seep location | Local water-source interpretation | Whole-reach discharge |
| Matched bulk temperatures and flow | Conditional mixing calculation | Heat-budget and support assumptions |
Questions and practical checks
Can a thermal camera see the streambed through water?
Not in the direct sense assumed by that question. Thermal-infrared water observations primarily concern the surface expression, not a transparent view through the water column.
Is groundwater always the cold part?
No. The relative temperature can reverse seasonally or vary by setting. Start with expected end-member temperatures rather than a fixed colour rule.
Does a larger plume prove a larger flow?
Not on its own. Mixing, depth, contrast, visibility and display or masking changes can alter the apparent area.
Can an anomaly prove pollution?
No. Source identity and water quality require their own evidence; temperature alone does not certify either contamination or safety.
What is a useful next check?
Choose the unresolved link: surface classification, contact temperature, source chemistry, hydraulic context or independently supported flow. Repeating the same display is not always the most informative check.
Continue with the connected subjects
Connect surface images to exchange and heat
Groundwater–surface-water exchange, subsurface heat tracing and residence time
Stream heat budgets, shade, advection and groundwater attribution
Compare independent spatial and flow evidence
Seepage meters and the limits of local-to-reach scaling
River image velocimetry and the separate conversion from pixels to discharge
Sources and scope
USGS: thermal-imaging cameras for groundwater–surface-water exchange
Primary multiscale temperature study: surface expression and complementary exchange observations
Iwasaki and colleagues (2023): real-time thermal video and groundwater-discharge mapping
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.