A flux tower measures turbulent exchange, not a whole catchment water bill
Eddy covariance estimates turbulent exchange by relating rapid fluctuations in vertical wind to fluctuations in a transported quantity such as water vapour. For evapotranspiration, the task is to turn that measured exchange into a water flux while preserving its time interval, footprint, processing and quality limits.
The result can inform crop water use, ecosystem studies and water budgets. It does not automatically represent every field around the tower, include every storage term or supply a complete annual total without treatment of rejected and missing periods.
Think of a tower observing parcels of air passing above a landscape. The source area changes with atmospheric conditions. A stable instrument location therefore does not imply an unchanging patch of land beneath every value. This guide follows that support problem alongside energy accounting and the conversion from a flux to a water depth.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Understand the exchange
Explain covariance before interpreting the daily total
Separate latent heat, water-vapour flux and water depth
Preserve the processing chain rather than only its last column
Check spatial and physical support
Treat the footprint as part of the value
Use energy closure as a diagnostic, not an automatic correction
Show what alternative closure adjustments imply
CHAPTER 1 OF 12
1. Explain covariance before interpreting the daily total
An average wind speed is not enough to describe turbulent transport. Upward-moving air can carry more water vapour than downward-moving air, creating a net exchange even when mean vertical wind is close to zero. Eddy covariance concerns the association between those fluctuations, not simply the average humidity multiplied by average wind.
Use a fictional four-observation illustration. Vertical-wind deviations are +1, −1, +1 and −1 in arbitrary compatible units. Water-vapour deviations are +2, −2, +2 and −2. Every product is +2, so their mean product is +2. Reverse the vapour deviations and the mean product becomes −2. Both series still have zero mean deviations. Their relationship, not their separate averages, determines the covariance sign.
This is a conceptual example, not a complete flux-processing method. Real systems require high-frequency measurements, coordinate handling, time alignment, instrument corrections and quality assessment. A four-value demonstration cannot stand in for those procedures. It does, however, explain why two calm-looking averages can conceal substantial exchange.
Ask which sign convention the dataset uses. Upward water-vapour exchange, downward exchange and storage changes need consistent interpretation. Do not force every negative value to zero because the intended application expects evaporation. Investigate the conditions, measurement quality and variable definition before deciding how a value belongs in a total.
Transfer check: shift one signal in time while leaving its average unchanged. The paired products change. This reveals why synchronising wind and analyser measurements is part of the physical calculation, not clerical tidying after the flux has been obtained.
CHAPTER 2 OF 12
2. Separate latent heat, water-vapour flux and water depth
Latent heat flux expresses the energy associated with phase change per area per time. Water-vapour mass flux expresses water mass per area per time. An evapotranspiration depth over a period is a time-integrated water quantity. They are connected through latent heat of vaporisation and water density, with assumptions about temperature and sign.
For a fictional calculation, take a constant latent heat flux of 100 watts per square metre over one hour. Energy is 100 × 3,600 = 360,000 joules per square metre. Using an explicitly approximate latent heat of 2.45 million joules per kilogram gives about 0.147 kilograms of water per square metre. With liquid-water density approximated as 1,000 kilograms per cubic metre, that corresponds to about 0.147 millimetres of water.
The numerical similarity between kilograms per square metre and millimetres is a unit relationship under the density approximation. It is not permission to drop time from the calculation. One hour at that flux gives a different depth from a full day at the same flux. Real daily integration uses changing interval values, not one midday reading multiplied by twenty-four.
AmeriFlux lists latent heat turbulent flux and water-vapour turbulent flux as separate variables with explicit units. It also distinguishes storage variables. Read the metadata before treating two files labelled ET as interchangeable; one may contain observed turbulent flux, another a corrected or filled integrated product.
Transfer check: double the flux but halve the duration. The water depth is unchanged under the same conversion factor. Then change the assumed latent heat value and show how the result changes. State which approximation is adequate for the teaching task and which temperature-dependent treatment a real analysis uses.
CHAPTER 3 OF 12
3. Preserve the processing chain rather than only its last column
A published flux time series is the result of an analysis chain. Raw sensor signals, alignment, corrections, averaging and quality assessment lead to a reported interval flux. A later daily product may add filtering, gap filling and energy-balance adjustment. These stages should remain distinguishable.
A useful record identifies the tower and instrument configuration, sensor heights, processing software and version, averaging interval, sign convention, quality variables and any changes during the study. If a sensor is replaced, keep the change date. If processing is rerun, retain the relation between the old and revised product rather than silently overwriting the historical basis.
Consider an original comparison in which a monthly total changes from 80 to 86 millimetres after reprocessing. That difference is not automatically a climate signal. It may reflect revised lag handling, quality rules or gap filling. Reproduce the comparison with common inputs and isolate which processing stage changed before interpreting the six-millimetre difference environmentally.
Do not call a filtered interval measured without qualification when its value was generated by a model. Conversely, rejecting poor-quality flux does not mean the physical exchange was zero. A gap is a lack of accepted evidence for that interval, not a claim that the ecosystem stopped exchanging water.
Transfer check: move from a research archive to an operational dashboard. The dashboard may prefer a complete line, while the research task needs the distinction between measured and estimated intervals. Preserve both through a provenance or quality field so convenience of display does not erase the evidence chain.
CHAPTER 4 OF 12
4. Treat the footprint as part of the value
The flux footprint describes the upwind source area contributing to the measured exchange under the conditions of an interval. It can vary with wind direction, atmospheric stability, turbulence and surface structure. A tower above one field can sometimes receive a substantial contribution from another surface.
AmeriFlux documentation explicitly treats spatial and temporal representativeness as part of processing levels. Its footprint variables describe fetch distances and filtering. These are useful reminders that geographic coordinates locate the instrument, not a permanently fixed measurement polygon.
Imagine a tower intended to characterise an irrigated plot. During one wind sector, its relevant footprint is mostly irrigated crop; during another it includes a dry road and neighbouring field. Pooling all intervals as crop water use assumes away the change in source. Filtering by a defensible footprint rule can improve relevance but may leave nonrandom gaps.
A simple teaching mixture illustrates the issue: if an interval were represented by 70 percent surface A at a water flux of 4 units and 30 percent surface B at 1 unit, a linear mixture would be 3.1. This is not a substitute for a footprint model. It shows why treating the mixture as a direct measurement of surface A would understate A’s flux under the assumed example.
Transfer check: rotate the wind toward a wetland while the tower remains stationary. Ask whether an increase in flux indicates a change in the crop, the source area or both. Keep a land-cover map, wind-sector information and footprint assessment alongside the time series so those explanations can be tested.
CHAPTER 5 OF 12
5. Use energy closure as a diagnostic, not an automatic correction
A simplified surface-energy account compares available energy with turbulent sensible and latent heat exchange, while recognising ground and other storage terms appropriate to the system. When measured terms do not balance, the residual is a diagnostic problem. It does not uniquely identify an error in latent heat flux.
For a fictional interval, net radiation is 400 watts per square metre and the relevant ground term is 50. Available energy under this simplified account is 350. Measured sensible heat is 120 and latent heat is 180, summing to 300. The residual is 50 and the ratio of turbulent fluxes to available energy is about 0.857. These are descriptions of the stated account, not proof that latent heat alone should increase by 50.
Possible missing terms or measurement limitations need investigation. Storage, spatial mismatch and atmospheric transport can matter, as can instrument and processing issues. The exact account must fit the site. A recent primary study on energy-balance closure reinforces that fuller accounting and quality treatment can change the comparison; no one residual allocation is a universal truth.
Avoid dividing by near-zero available energy and then interpreting a huge ratio without context. Daytime and nighttime conditions may require different diagnostic treatment. An annual closure statistic can also hide compensating periods of different behaviour.
Transfer check: move from short vegetation to a tall forest and ask which storage terms and measurement-height considerations become important. The algebraic conservation principle survives, but the incomplete account does not. A result that looks less closed under a richer physical system is a reason to improve the account, not to choose whichever correction yields the most attractive ET total.
CHAPTER 6 OF 12
6. Show what alternative closure adjustments imply
When an application uses an energy-balance adjustment, name it, preserve the unadjusted flux and quantify the consequence. Alternative allocations of a residual can produce different water totals. They are modelling choices with assumptions, not additional direct observations.
Return to available energy 350, sensible heat 120 and latent heat 180. One illustrative proportional adjustment multiplies both turbulent terms by 350/300, giving sensible heat 140 and latent heat 210. Another extreme assigns all 50 residual units to latent heat, giving 230. A third assigns it all to sensible heat, leaving latent heat at 180. All three close this simplified numerical account, yet they imply different evaporation.
The proportional example preserves the measured sensible-to-latent ratio. The other examples do not. None is endorsed here as a universal procedure. Their purpose is to demonstrate that closure alone cannot select the correct water flux. Evidence about the site, instruments, missing terms and intended analysis is required.
Using the one-hour conversion from chapter 2, latent heat values of 180, 210 and 230 watts per square metre imply approximately 0.264, 0.309 and 0.338 millimetres. A budget sensitive to that difference should not report one adjusted total without disclosing the treatment.
Transfer check: apply the adjustment during a period with different residual structure or a changed footprint. Explain why the old proportionality assumption might fail. Keep a sensitivity comparison of adjusted and unadjusted totals, and avoid evaluating a remote-sensing product against a tower correction chosen because it improves agreement with that same product.
CHAPTER 7 OF 12
7. Describe gaps by their conditions, not only their percentage
Missing flux intervals can arise from power loss, maintenance, rain effects, analyser problems or failed quality conditions. Their hydrological importance depends on when they occur. A ten-percent gap concentrated in hot dry afternoons differs from ten percent distributed across many ordinary nights.
Use an original two-period day. Twelve hours have latent heat flux 50 watts per square metre and twelve have 200. The mean is 125 if both are observed. If every high-flux period is missing, the observed mean is 50. Multiplying it by the full day underestimates the stated daily exchange by sixty percent. This is not an argument that all gaps are high flux; it is a demonstration that completeness and representativeness are separate.
Plot missing and rejected intervals against radiation, weather, season and footprint conditions. Report both the fraction of intervals and their potential contribution to integrated exchange. A file with ninety-five percent valid records can still be weak for one consequential period.
AmeriFlux guidance encourages preservation of non-gap-filled fluxes alongside appropriately identified filled products. That distinction allows users to evaluate how much of a seasonal or annual total depends on reconstruction. Do not erase the original gaps after the filled series is produced.
Transfer check: a storm knocks out the tower for two days. Nearby fair-weather observations may not describe interception evaporation or changing canopy wetness during those days. The gap method should be assessed against similar withheld conditions where evidence exists, and its limits should be explicit when no such comparison is possible.
CHAPTER 8 OF 12
8. Test gap filling on withheld blocks
Gap filling estimates what might have occurred during missing intervals using an empirical or process-based relationship. A useful test hides accepted observations, predicts them using only the information the method would have during a real gap, and compares the predictions with the hidden record.
Randomly withholding isolated intervals can be easier than reconstructing a three-day instrument failure. Neighbouring observations often carry strong information about an isolated gap. Use withheld blocks that resemble the duration and conditions of actual missing periods. Keep model selection and final evaluation separate.
For a fictional validation, method A has small average error on scattered half-hours but underestimates a hot three-day period by 15 millimetres. Method B is slightly worse on isolated intervals but loses only 4 millimetres in that block. The better choice depends on the actual gap structure and water-budget consequence, not one pooled score.
Include uncertainty from reconstruction in the final integrated product. It is not enough to report a residual standard deviation for interval predictions and then assume all interval errors cancel independently. Persistent model bias across a long missing block can accumulate. Weather-driven errors can also be correlated.
Transfer check: train the filling relationship before a harvest, then apply it after canopy removal. The surface process has changed. A relationship based on the previous canopy may reconstruct a plausible-looking but physically inappropriate total. Identify disturbance dates and decide when the method needs a new regime or a narrower claim.
CHAPTER 9 OF 12
9. Integrate accepted and estimated intervals transparently
To produce a daily or seasonal depth, convert each interval flux using the stated physical factors and multiply by its duration. Sum over the defined period. Keep accepted observations, adjusted observations and filled intervals identifiable so the total can be decomposed.
Suppose a fictional month has 60 millimetres from accepted measured intervals, 12 from model-filled gaps and a closure adjustment adding 8. The published adjusted total is 80. It is more informative to report those components than to call all 80 measured. The component breakdown does not make the estimates bad; it makes their role inspectable.
Check timestamps, daylight-saving conventions where relevant, leap days, duplicate rows and irregular intervals. A thirty-minute average with an end-of-interval timestamp should not accidentally be aligned as a beginning value against rainfall. Missing rows and explicit missing values both need detection.
Do not sum watts per square metre directly as if the result were water depth. Energy flux needs time integration and conversion. Nor should interval water rates be summed without duration when intervals differ. A units test with one constant-flux hour is a valuable independent check of a much larger processing script.
Transfer check: combine products from two processing versions halfway through a year. First reproduce an overlap comparison and determine whether the conversion, correction or quality basis changed. A complete calendar does not guarantee a consistent annual measurement basis.
CHAPTER 10 OF 12
10. Compare with a water balance at a compatible scale
A water balance can provide an independent perspective on evapotranspiration, but only with compatible boundaries, periods and storage accounting. Catchment precipitation minus streamflow is not automatically ET if storage change, groundwater exchange or water transfers are important.
For a fictional bounded catchment year, precipitation is 1,000 millimetres, streamflow is 400 and storage increases by 100, with other boundary flows assumed absent. The implied ET is 1,000 − 400 − 100 = 500. Ignoring storage would give 600. The arithmetic demonstrates the role of the missing term; it does not supply a real catchment estimate.
A tower value of 450 millimetres need not contradict the catchment value of 500. The tower footprint may represent one land cover, while the catchment includes several. The tower record may also have incomplete conditions or a different correction basis. Compare support before declaring one method wrong.
For a land-cover aggregation, identify representative areas and weights independently. Do not choose weights merely to force balance. If the tower is used to tune the catchment model, that tuned model is not fully independent validation of the tower-derived estimate.
Transfer check: add irrigation imported from outside the catchment. The water-balance ET can exceed what precipitation minus streamflow alone suggests. The new input belongs explicitly in the account. This is a practical reminder that conservation is powerful only when the system boundary includes the relevant pathways.
CHAPTER 11 OF 12
11. Avoid turning ET into an unsupported irrigation instruction
Measured or estimated ET is one part of irrigation reasoning. A field decision also depends on rainfall, soil storage, rooting, application efficiency, crop stage, water quality and operational constraints. A tower above a mixed footprint cannot alone prescribe a uniform watering depth for every field nearby.
Consider a fictional field with 25 millimetres of ET over a week and 10 millimetres of effective input retained from rain. Under a simplified target of restoring the starting storage with no other changes, net replacement is 15 millimetres. If an assumed application efficiency is 0.75, a gross depth of 20 millimetres would supply that net amount. Every step is conditional; real scheduling needs a valid definition of effective rainfall and efficiency.
If the root zone already stores enough usable water, immediate replacement may not be necessary. If drainage or salinity management matters, the simplified balance omits important requirements. These limits should be stated rather than converting the worked calculation into an operational recommendation.
Compare ET with its uncertainty and spatial support before evaluating water savings. A lower flux after a management change may reflect weather, crop development, reduced canopy or stress rather than a successful efficiency improvement. The outcome of interest might include yield, crop health or environmental effects as well as water use.
Transfer check: move the same weekly ET total to a shallow-rooted crop on a freely draining soil. The timing and storage constraints change. The depth calculation remains useful, but the schedule cannot be copied from a deeper-rooted system.
CHAPTER 12 OF 12
12. Build a reproducible flux-to-water explanation
Prepare an audit package containing the exact source product, variable definitions, processing version, interval timing, accepted-quality rule, footprint treatment, energy-accounting basis, closure adjustment, gap method and integration equation. Include both original and final series, with the components of the total visible.
A practical review can start with three calculations from this guide: reproduce 0.147 millimetres from one hour at 100 watts per square metre; reproduce the 50-watt energy residual; and compare the alternative latent-heat adjustments. These show that conversion, accounting and correction are distinct tasks.
Then hide a block of good observations resembling a real failure and evaluate the reconstruction. Finally move the hypothetical wind sector to another land cover. An explanation that survives these changes should state which conclusions remain supported and which need new evidence.
For students, the transferable lesson is not a list of tower components. It is the connection between a physical exchange, an observed relationship, a changing source area and a time-integrated quantity. Once those are clear, a seasonal number can be questioned constructively rather than accepted or rejected because it looks sophisticated.
The finished conclusion should say whether the reported ET is turbulent-flux-derived, storage-treated, closure-adjusted and gap-filled, over what period and support. Avoid calling it exact or purely measured when the evidence includes modelling. Clear qualification makes the number more useful, not less scientific.
For an additional worked comparison, imagine two annual products reporting 500 millimetres. Product A contains 450 millimetres from accepted intervals and 50 from gap filling. Product B contains 300 from accepted intervals, 150 from gap filling and 50 from a closure adjustment. The totals match, but the evidential compositions differ. It would be wrong to infer equal reliability from the matching totals alone. Ask whether the missing periods are representative, whether the adjustment is justified and whether independent comparisons support both products.
Next suppose product A samples mostly crop-dominated footprints, while B includes a wetland sector. The annual equality now conceals both methodological and spatial differences. A fair comparison needs either a matched support or an explicit explanation of why the differing footprints belong to the question. Deleting inconvenient sectors until totals agree is not validation; it changes the represented system.
Finally set a fictional management target of detecting a twenty-millimetre annual change. If alternative defensible processing choices move the estimate by thirty millimetres, the proposed comparison requires more careful uncertainty analysis. That does not prove a real change is absent. It means the current processing sensitivity is material to the target. Report the change estimate, the competing processing bases and the evidence needed to distinguish them rather than announcing success from one preferred column.
Worked comparison
| Quantity or step | Fictional value | Meaning |
|---|---|---|
| Net radiation minus ground term | 350 W/m² | Simplified available energy |
| Sensible plus latent heat | 300 W/m² | Measured turbulent terms in the example |
| Residual | 50 W/m² | Unresolved difference, not automatically latent heat |
| Proportional adjusted latent heat | 210 W/m² | Assumption-dependent correction |
| Unadjusted latent heat | 180 W/m² | Preserved comparison basis |
Questions and practical checks
Is a tower value a catchment average?
Not automatically. The interval footprint and land-cover representation must justify the spatial comparison.
Does an energy residual prove ET is too low?
No. The residual can reflect several missing terms, measurement limitations and scale differences. Its allocation requires evidence and assumptions.
Can I multiply noon flux by a day?
Only under the explicitly unrealistic assumption that the flux remains constant. Real totals integrate changing interval values.
Are filled intervals measurements?
They are estimates informed by a method and other evidence. Preserve them separately from accepted observed intervals.
What should a student verify?
Reproduce the unit conversion and energy account, then explain how changed wind direction and a long gap alter the claim.
Continue with the connected subjects
Evaporation depth, changing area and reservoir water balance
Drought indices and evaporative demand
Paired catchments and attribution of change
Bowen-ratio energy balance and latent heat
Eddy-covariance CO₂ flux — compare sign conventions and gap filling
Scintillometer sensible-heat measurements and path weighting
Soil evaporation curves and cumulative water loss
Sources and scope
AmeriFlux variable definitions — flux, storage, units and footprint fields
AmeriFlux processing levels — spatial support, filtering and gap filling
AmeriFlux data FAQ — retain non-filled and identified gap-filled variables
Primary research — energy-balance accounting and closure in eddy-covariance measurements
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.