A fast temperature trace can inform water loss, but only through a visible chain of inference
A fine temperature sensor above a crop records repeated warming ramps followed by abrupt changes. These small fluctuations can contain information about turbulent heat exchange. They do not directly weigh the water leaving the crop. To reach evapotranspiration, a surface-renewal method interprets the temperature structures, estimates sensible heat and closes an energy account.
This guide separates those steps. It explains what a ramp means, why a calibration is not automatically portable, and how an apparently small heat-flux error can become a large relative error in residual water loss. The aim is to read a method report intelligently, not to treat an inexpensive sensor as a complete water-budget system.
All calculations below use original fictional values. The simple ramp expression is an explanatory model with supplied parameters, not a replacement for a validated research algorithm. Different surface-renewal formulations estimate or calibrate their coefficients differently. Their assumptions and validation conditions belong with the result.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1: Read the signal · Chapters 1–3
Define surface renewal without confusing it with a thermometer average
Route 2: Follow the heat estimate · Chapters 4–5
Route 3: Close the energy account · Chapters 6–8
Route 4: Audit uncertainty · Chapters 9–10
Propagate error through a residual rather than the final ratio alone
Route 5: Transfer and report · Chapters 11–12
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Read the signal · Chapters 1–3
Define surface renewal without confusing it with a thermometer average
Keep temperature amplitude, duration and sensor height separate
Understand what high-frequency sampling can and cannot recover
Follow the heat estimate · Chapters 4–5
Work through a supplied ramp-to-heat calculation
Distinguish calibration from independent validation
Close the energy account · Chapters 6–8
Close the energy budget before naming evapotranspiration
Convert latent heat to water depth with the interval visible
Handle negative residuals and incomplete closure honestly
Audit uncertainty · Chapters 9–10
Propagate error through a residual rather than the final ratio alone
Match the footprint, canopy and weather regime
Transfer and report · Chapters 11–12
Transfer the method to a different question carefully
Report the full chain from trace to water loss
CHAPTER 1 OF 12
1. Define surface renewal without confusing it with a thermometer average
An air parcel near a surface exchanges heat with it. Turbulent motions replace that parcel with another. In some conditions this exchange produces recognisable ramp-like patterns in a high-frequency temperature series: a gradual change interrupted by a more abrupt replacement. Surface-renewal analysis uses statistical descriptions of these structures to infer heat transfer.
The method is not simply the average air temperature multiplied by wind speed. The relevant information concerns fluctuations and their time scales. A daily maximum thermometer cannot recover the same structures after the fact. Nor does a fine output interval prove that the sensor and acquisition system resolved the fluctuations required by the chosen analysis.
Primary work by Spano and colleagues compared ramp-based sensible-heat estimates with eddy covariance over several crop canopies. Later USDA ARS research describes both calibrated and newer stand-alone approaches. These are related developments, not one interchangeable equation whose parameters can be copied without regard to formulation.
Picture a conveyor moving parcels past a warm counter. The temperature each parcel acquires depends on how it exchanges energy and how long it remains. That analogy helps explain renewal, but it does not prove that every jagged temperature trace is a physical ramp. Instrument noise, radiation and processing must still be distinguished from the exchange process.
CHAPTER 2 OF 12
2. Keep temperature amplitude, duration and sensor height separate
In a simplified explanatory formulation, sensible-heat flux H is proportional to air volumetric heat capacity, an effective length, ramp amplitude divided by ramp period, and a dimensionless coefficient. Write H = α ρ cₚ z Aᵀ/τ, with the symbols and sign convention explicitly supplied. Actual formulations can define effective height and time scales differently.
Here ρcₚ is in J m⁻³ K⁻¹, z in metres, ramp amplitude Aᵀ in kelvin and period τ in seconds. The resulting units are J m⁻² s⁻¹, or W m⁻². This dimensional check establishes a heat-flux unit, not the correctness of a particular coefficient or the representativeness of a sensor location.
Height is not a harmless installation detail. The observed structures and their relation to the exchanging canopy can depend on measurement position and aerodynamic regime. A coefficient used with one height definition should not be combined with another definition simply because both heights are stated in metres.
Keep absolute temperature and temperature difference distinct. A ramp amplitude of 0.6 K is also a difference of 0.6 °C; it is not an air temperature of 0.6 °C. The difference can be used in the explanatory expression without converting it to 273.75 K. Confusing a difference with an absolute thermodynamic temperature produces a very large and physically meaningless error.
CHAPTER 3 OF 12
3. Understand what high-frequency sampling can and cannot recover
Sampling must be fast enough for the method’s relevant fluctuations, while the sensor itself must respond adequately. Recording a slow sensor twenty times a second does not create twenty independent measurements of rapid temperature change. Sensor response, acquisition frequency and processing choices are three separate parts of the measurement chain.
Averaging can remove the structures that the analysis needs. Suppose an original fictional trace contains repeated ten-second ramps. Storing only one-minute averages may retain the mean temperature trend while losing those ramps. Interpolating the averages back to one-second values produces a smoother drawing, not restored turbulent information.
Statistical ramp extraction commonly uses more than visual identification of a few attractive examples. Structure-function approaches describe moments of temperature differences over selected lags. The details of lag choice, sign and noise treatment belong to the specific method. This guide does not replace that algorithm with hand-picked line segments.
For a supplied study, ask which raw series was retained, which frequency and averaging interval were used, how missing samples were handled and whether sensitivity tests were performed. USDA ARS work on small-eddy surface renewal explicitly examined sampling-frequency effects. A frequency that performed well in one investigation is evidence for that method and setting, not a universal instrument specification.
CHAPTER 4 OF 12
4. Work through a supplied ramp-to-heat calculation
Take a fictional interval with α = 0.50, ρcₚ = 1,200 J m⁻³ K⁻¹, z = 1.5 m, Aᵀ = 0.8 K and τ = 12 s. Under the simplified expression, H = 0.50 × 1,200 × 1.5 × 0.8/12 = 60 W m⁻². State that the coefficient and effective height were supplied for this exercise.
Now shorten the period to six seconds while holding the other supplied terms fixed. The inferred heat flux doubles to 120 W m⁻². Alternatively, retain the twelve-second period but halve the amplitude; H becomes 30 W m⁻². These are sensitivities of the stated model, not proof that a natural canopy allows each factor to vary independently.
If the fitted ramp statistic changes sign, interpret it using the method’s stated convention. Negative sensible heat can occur when heat moves from the air towards a cooler surface. A software rule that forces every H value positive would alter the energy account. Do not discard a sign merely because a daytime example used upward heat as positive.
The exercise reveals why the temperature trace is only one stage. Even with perfect arithmetic, uncertainty can enter through amplitude extraction, time-scale selection, effective geometry, coefficient estimation and volumetric heat capacity. Keeping a transparent intermediate H estimate helps identify which stage changed when a later ET result is revised.
CHAPTER 5 OF 12
5. Distinguish calibration from independent validation
A calibrated approach may estimate α by comparing uncalibrated ramp-based heat estimates with a reference sensible-heat series. The fitted relation may include an intercept, a coefficient that changes with conditions, or additional physically motivated terms. Do not reduce a published method to a single proportional factor unless that is actually the specified model.
If the same observations are used to fit and evaluate the coefficient, close agreement is partly built into the procedure. Independent validation uses data not used in fitting and asks whether the relationship survives the intended conditions. Splitting a continuous record randomly can still leave strongly related weather or crop states on both sides of the split.
Imagine a fictional training set collected over a short, uniform crop. A later evaluation concerns a taller, patchier canopy. Good training agreement does not establish that the original coefficient handles the new turbulence, height relationship or footprint. A transfer evaluation must actually include the conditions for which the new result is claimed.
Some newer surface-renewal approaches aim to avoid a separate empirical calibration against a reference flux. That does not mean they avoid assumptions or need no validation. Distinguish a stand-alone formulation from an instrument that has somehow become assumption-free. The relevant questions remain response, represented surface, physical formulation and comparison with independent evidence.
CHAPTER 6 OF 12
6. Close the energy budget before naming evapotranspiration
Define net radiation Rₙ as positive energy supplied to the surface, ground heat flux G as a positive downward loss from the chosen surface system, and storage change S as positive accumulation. In a simplified compatible account, latent-heat flux LE = Rₙ − G − S − H. Additional relevant terms must be included where that account is incomplete.
The temperature-ramp analysis estimates H, not LE directly in this particular residual approach. Net radiation, ground heat and storage require their own observations or models. A project with a low-cost temperature sensor can still depend strongly on the quality of these other components. The uncertainty of ET cannot be inferred from the thermometer specification alone.
For a fictional half-hour, let Rₙ = 420, G = 50, S = 20 and H = 60 W m⁻². The residual LE is 290 W m⁻². If storage is omitted, LE becomes 310 W m⁻². That difference would be attributed incorrectly to water exchange rather than to unaccounted energy accumulation.
Define the physical boundary consistently. Heat stored in canopy air, biomass or the soil layer above heat-flux plates may matter, depending on the system and interval. Including the same storage twice is also wrong. An explicit boundary drawing is more useful than adding every available correction without checking whether it is already represented in another term.
CHAPTER 7 OF 12
7. Convert latent heat to water depth with the interval visible
Latent-heat flux is energy per area per time. Divide by latent heat of vaporisation λ to obtain water mass flux, then use water density to express equivalent depth. For a teaching value λ = 2.45 MJ kg⁻¹, a sustained 290 W m⁻² over 1,800 seconds gives 522,000 J m⁻²/2,450,000 J kg⁻¹ = about 0.213 kg m⁻², or 0.213 mm.
Multiplying that half-hour depth by 48 would imply a constant 24-hour flux. It is not an observed daily total. To estimate a day, integrate the accepted interval values and account for missing periods explicitly. Night conditions and heat storage can differ substantially from those during the illustrated sunlit half-hour.
Latent heat varies with temperature. The fixed value used here makes the arithmetic readable and is labelled an approximation. A real processing system should state its temperature convention and units. Mixing a λ value in kJ kg⁻¹ with energy in joules produces a thousandfold error even if the formula is otherwise correct.
A depth estimate is not yet a volume over a farm. Multiply by the represented area only after establishing what surface the flux describes. A sensor’s location above a plot does not automatically make every turbulent interval representative of that plot. Rate, duration, spatial support and area conversion remain distinct pieces of the calculation.
CHAPTER 8 OF 12
8. Handle negative residuals and incomplete closure honestly
A negative LE result can have several interpretations. Under appropriate conditions it may correspond to condensation or another downward latent exchange. It may also reflect inconsistent signs, omitted terms or measurement errors. Automatically replacing negative values with zero chooses an interpretation without investigating the evidence.
Take a fictional interval with Rₙ = 25, G = −10, S = 5 and H = 45 W m⁻². The stated account gives LE = −15 W m⁻². A negative G means heat moves upward into the chosen surface system. Treating that G as positive would change the result again. Read signs as directions, not as undesirable values to remove.
Energy closure does not uniquely identify every error. If the residual is implausible, changing H until the answer looks familiar does not prove that H was the faulty component. Net radiation, storage, ground heat and incompatible footprints can contribute. Investigate independent evidence and keep alternative explanations visible.
A physical plausibility check is therefore a diagnostic, not permission to tune the account towards a preferred ET. Preserve raw component values, processing versions and flagged intervals. When an estimate is replaced by a model, label that substitution so a later analyst can distinguish inferred continuity from an uninterrupted observation record.
CHAPTER 9 OF 12
9. Propagate error through a residual rather than the final ratio alone
With independent standard uncertainties, the variance of LE = Rₙ − G − S − H is the sum of the four component variances. Suppose uncertainties are 15, 8, 10 and 12 W m⁻². Their combined standard uncertainty is √(225 + 64 + 100 + 144), approximately 23.1 W m⁻². The independence assumption needs justification.
For LE = 290 W m⁻², that uncertainty is about 8%. For a different interval with LE = 30 W m⁻² and the same component uncertainty, it is about 77%. A similar absolute energy error produces very different relative confidence depending on the residual’s size. Percentage precision cannot be transferred from a large daytime flux to a small nighttime flux.
Correlations can change the residual uncertainty. Shared calibration, common forcing or common processing may make components dependent. Include covariance with the appropriate signs rather than always adding variances. The connected water-budget closure guide develops the same general lesson for fluxes and storage measured in water units.
A coefficient fitted to a reference heat flux also carries fitting uncertainty and possible conditional bias. Treating the coefficient as exact while quoting only ramp-fit scatter understates the chain’s uncertainty. Separate random variation from a systematic transfer error: repeating an unchanged biased setup many times does not average away that bias.
CHAPTER 10 OF 12
10. Match the footprint, canopy and weather regime
The air reaching a sensor can have exchanged heat with an upwind area rather than the ground directly beneath it. That represented area can change with wind, atmospheric stability and measurement height. A fixed sensor position therefore does not imply a fixed spatial footprint for every interval.
Suppose the fictional intended field is irrigated, but some wind directions bring air from a dry adjacent surface. Combining heat exchange influenced by the dry area with net radiation and soil heat measured in the wet field creates a boundary mismatch. The resulting residual can look arithmetically tidy while lacking a coherent physical support.
Canopy growth changes geometry and exchange conditions. Record crop height and relevant sensor-height definitions rather than assuming a coefficient remains appropriate throughout a season. Pruning, harvest or a change from uniform cover to isolated rows may require a new evaluation even when the sensor stays attached to the same mast.
Stratify comparisons by conditions that matter. An overall daily agreement can hide opposing errors during calm nights and windy afternoons. Useful diagnostics include accepted wind sectors, canopy states, stability classes where used by the method, and missing-data distribution. Report the operating range actually supported by the evaluation, not only the range of dates on the file.
CHAPTER 11 OF 12
11. Transfer the method to a different question carefully
Change the exercise from estimating crop ET to separating soil evaporation from plant transpiration. The residual method still estimates an aggregate latent exchange over its represented surface. It does not by itself identify the two components. An independent partitioning method or additional information is needed; dividing by leaf area does not create that missing identification.
Next change from a uniform crop to open water. The energy and aerodynamic context changes, including potentially important water heat storage. A soil ground-heat correction cannot simply stand in for the lake’s thermal storage. The temperature-ramp concept may be studied in a new setting, but its previous calibration is not evidence of validity there.
Finally shorten the averaging interval drastically. Higher-frequency outputs may look more responsive, while reducing the number of independent exchange structures available for a stable statistical estimate. Fine temporal resolution and low uncertainty do not automatically improve together. Use the validation and sampling requirements of the chosen formulation.
The transferable skill is identifying what remains invariant: conservation, units and the difference between observed temperature and inferred water loss. The coefficient, effective height, quality filters and represented surface are conditional. Ask what evidence would need to change before each new use, rather than asking whether the method name alone authorises the transfer.
CHAPTER 12 OF 12
12. Report the full chain from trace to water loss
A useful report identifies the sensor and response, acquisition frequency, ramp-statistic method, averaging interval, height definition, coefficient or stand-alone formulation, and sensible-heat sign convention. It then gives the radiation, ground-heat and storage account used to reach LE. Only after that should it present the water-depth conversion.
Separate observed, fitted and modelled fields. The temperature trace is observed through an instrument. Ramp statistics are processed. H is inferred through the surface-renewal formulation. LE is a residual in the approach discussed here. ET depth additionally depends on λ and interval integration. These distinctions help readers understand what an independent comparison can actually test.
For an illustrative statement, write that the supplied formulation produced H = 60 W m⁻², the compatible energy account produced LE = 290 W m⁻², and the half-hour water equivalent was 0.213 mm using λ = 2.45 MJ kg⁻¹. Add uncertainty, quality status and the represented surface. Do not label the final depth a direct humidity measurement.
The method’s value lies in a defensible connection between fast temperature structure and water exchange, not in hiding the connection. A clear chain makes low-cost monitoring more useful because it identifies where further evidence would improve a decision. It also protects a learner from mistaking an impressive high-frequency graph for a complete and independently validated water account.
Independent worked case and evidence checks
Use a further blind-check exercise to separate two possible failures. In supplied interval A, an independent reference supports the ramp-derived H, but the radiation record has a documented offset. In interval B, the radiation and storage terms are supported, while ramp extraction fails a response test. Both intervals might produce the same erroneous final ET depth, yet they require different repairs. Looking only at the residual would conceal the distinction. Diagnose the chain at its intermediate stages before adjusting its last number.
The same logic applies to a very good daily total. Two large half-hour errors with opposite signs can cancel during integration. Check interval behaviour as well as the daily sum, especially if the intended use is detecting water stress or scheduling a time-sensitive comparison. Conversely, a small interval mismatch may have limited practical importance for a coarse seasonal account. The required validation follows the decision being made, not the most visually impressive graph available.
An evidence-backed report can therefore describe both accuracy at the intended aggregation scale and known limits at finer scales. It need not promise performance that the comparison did not test.
A fictional project supplies α = 0.60, ρcₚ = 1,200 J m⁻³ K⁻¹, effective z = 2.0 m, ramp amplitude 0.50 K and period 10 s. Under the stated simple formulation, H = 72 W m⁻². The project also supplies Rₙ = 350, G = 35 and S = 18 W m⁻² for the same half-hour and represented surface.
Calculate LE = 350 − 35 − 18 − 72 = 225 W m⁻². Over 1,800 seconds this is 405,000 J m⁻². With λ = 2.45 MJ kg⁻¹, the water equivalent is approximately 0.1653 mm. The estimate concerns that interval, not an entire day.
Changed context one: use an untested coefficient of 0.90 while leaving the trace unchanged. H becomes 108 W m⁻² and LE becomes 189 W m⁻², giving 0.1389 mm. The altered water result came entirely from the heat calibration; there was no new direct observation of water vapour.
Changed context two: retain the original α but discover that 18 W m⁻² was a storage release, not accumulation. S is −18, so LE becomes 261 W m⁻² and the interval equivalent is 0.1918 mm. Correcting a direction changes the inference without changing any temperature-ramp statistic.
Evidence question: can the same half-hour establish how much ET was soil evaporation? No. The aggregate residual does not separately identify evaporation and transpiration. A chamber or microlysimeter observation could provide additional information, but its surface, interval and disturbance would have to be reconciled with the larger flux footprint.
Finally suppose Rₙ has standard uncertainty 12, G 7, S 9 and H 15 W m⁻², treated independently for this exercise. The combined uncertainty is √(144 + 49 + 81 + 225) = 22.3 W m⁻², or about 0.0164 mm over the half-hour. These figures exclude unbounded footprint and transfer bias; naming that exclusion is part of the answer.
Worked comparison
| Stage | Quantity | Independent question |
|---|---|---|
| Acquisition | High-frequency air temperature | Did the sensor resolve the required structures? |
| Processing | Ramp amplitude and time scale | Was the extraction algorithm and lag choice tested? |
| Heat inference | Sensible-heat flux H | Is the coefficient or stand-alone formulation appropriate? |
| Energy residual | Latent-heat flux LE | Do all terms describe the same boundary and interval? |
| Water conversion | ET depth | Were latent heat, units and integration handled correctly? |
Questions and practical checks
Does surface renewal directly measure transpiration?
Not in the residual approach explained here. Temperature structures inform sensible heat; a compatible energy balance then gives latent heat and an aggregate water-loss estimate.
Is a high sampling rate enough?
No. Sensor response, processing and spatial support also matter. Oversampling a slow sensor does not restore missing fluctuations.
Does calibration remove the need for validation?
No. Check performance on independent data representing the intended canopy, weather and measurement geometry.
Must a negative latent-heat residual be deleted?
No. Investigate sign conventions, conditions, storage and measurement errors before deciding whether it represents downward exchange or an invalid interval.
Can the same coefficient be used after harvest?
Only with evidence that the formulation remains appropriate. Canopy and aerodynamic changes can alter the relationship being used.
Continue with the connected subjects
Check the turbulent and energy accounts
Eddy-covariance ET: footprints, energy closure and missing intervals
Residual fluxes, correlated errors and identifiability
Separate local components from total ET
Microlysimeters: soil evaporation and the isolated-core boundary
Stem heat-balance sap flow and whole-plant water accounts
Lake heat storage: why the energy boundary changes over water
Compare a connected measurement problem
Soil vapour adsorption: distinguish net mass gains from the source of non-rainfall water. Follow the changed boundary and test the relevant storage, mixing or attribution assumptions rather than transferring a correction by name alone.
Sources and scope
Spano and colleagues: surface-renewal structure functions and sensible-heat comparison
USDA ARS: stand-alone surface renewal compared with weighing lysimetry and eddy covariance
USDA ARS: small-eddy surface renewal, free convection and sampling frequency
Castellví: surface-renewal and flux-variance approaches over olive trees
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.