The key distinction: concentration is not flux
Aerosol concentration tells us how many particles, or how much particle mass, occupy a volume of air at a specified place and time. Aerosol flux tells us the rate at which particles move across an area. An eddy-covariance particle flux measures the correlation between turbulent vertical air motion and changing particle concentration. To translate its output into plain English, name the particle-size range, the concentration unit, the direction convention, the measurement height, the averaging interval and the surface represented by the turbulent footprint. A positive or negative sign has no meaning until upward and downward are defined.
Dry deposition velocity is an effective exchange velocity often inferred from a measured downward flux and a reference concentration. It is not the vertical speed of every particle and is not necessarily constant across particle sizes, weather, vegetation or surfaces. A net upward flux can indicate emission or other upward exchange, whereas a net downward flux is consistent with surface removal under the instrument’s conventions. Neither sign by itself establishes the mechanism or the precise source footprint.
Imagine two datasets with identical aerosol number concentrations but opposite flux signs. The atmosphere at the instrument can contain the same instantaneous number of particles in both periods while turbulence transports them differently. One period may show net deposition to a vegetated surface; another may show emission from a source or changing upwind sector. Reporting concentration alone would erase this exchange distinction. Reporting a flux alone without sampling conditions can be equally misleading.
What the eddy-covariance method actually multiplies
Eddy covariance relies on rapid, synchronized observations of vertical wind velocity w and a scalar such as particle number concentration N. Split each into a mean and fluctuation over a defined averaging interval: w = mean(w) + w′, N = mean(N) + N′. The turbulent covariance F = mean(w′N′) estimates vertical transport across an approximately horizontal plane under suitable assumptions. When upward wind is positive, a positive covariance indicates net upward transport and a negative covariance indicates net downward transport. The sign convention must be documented rather than inferred from an instrument’s colour scale.
Units provide an immediate truth test. Vertical wind w is in metres per second. Number concentration N is in particles per cubic metre. Their covariance is therefore particles per square metre per second because (m/s) × (particles/m³) = particles/(m² s). If the input number concentration is in particles per cubic centimetre, convert it before reporting SI fluxes: 1 cm³ is 10⁻⁶ m³, so 1 particle/cm³ corresponds to 10⁶ particles/m³. A million-fold mistake is possible when the word ‘cubic’ disappears during spreadsheet conversion.
The covariance formula is compact but the measurement is not simple. Turbulent motion occurs over a range of timescales. A slow instrument can miss fast eddies; tubing delays the aerosol signal; particle losses vary with size; wind distortions change turbulent statistics; and nonstationary conditions undermine the interpretation of an average. Good field analysis treats the covariance as the end of a corrected, quality-controlled signal chain, not as the effortless product of two columns in an arbitrary spreadsheet.
An elementary covariance example with four observations
For a deliberately small teaching dataset, let deviations of vertical velocity from its mean at four equally weighted instants be +0.5, −0.5, +0.5 and −0.5 m/s. Let simultaneous concentration deviations be +100, −100, +100 and −100 particles/cm³. Every product is +50 (m/s)(particles/cm³), so their average is +50 in those mixed units. After multiplying by 10⁶ to convert concentration to particles/m³, the covariance is +5.0 × 10⁷ particles/(m² s): net upward transport in the stated sign convention.
Now reverse only the concentration deviations to −100, +100, −100 and +100 particles/cm³. Every product becomes −50 in the original mixed units, yielding −5.0 × 10⁷ particles/(m² s): net downward transport. Notice that the mean concentration of the two datasets can be identical; the flux differs because the fluctuations align differently with vertical wind. This illustrates what a covariance reveals that a concentration average cannot.
Real eddy-covariance analysis uses thousands of observations and correction methods appropriate to the instruments and station, not four illustrative samples. If one channel was sampled one second later than the other, the paired fluctuations might no longer represent the same passing eddy and the covariance could be artificially reduced. The four-point exercise is useful precisely because it makes the required synchronization visible before the complexity of field processing is introduced.
From net flux to an effective deposition velocity
With upward defined as positive, a downward net particle flux F is negative. A commonly used effective deposition velocity is v_d = −F/N_ref where N_ref is a stated positive reference particle number concentration. Suppose F = −4.0 × 10⁷ particles/(m² s) and N_ref = 8.0 × 10⁹ particles/m³. Then v_d = 0.005 m/s = 5 mm/s. These are hypothetical numbers for unit-checking. The value is not a universal deposition speed: it summarizes observed exchange relative to a particular concentration, height, size range and time interval.
If F is positive under the same sign convention, the same algebra gives a negative value of −F/N_ref. Rather than call that a negative physical deposition speed, it is often clearer to label a signed exchange velocity or classify the period as net emission. The definition used in a particular study must be read before comparing published velocity plots. Some authors define downward as positive, which reverses the sign language. A formula copied without its coordinate convention can turn emission into deposition in a report.
When atmospheric chemistry, particle nucleation or source emission occurs between measurement height and surface, net observed flux may not equal a pure sink caused by deposition. The effective quantity can be very informative, but it reflects the integrated exchange process sampled by the measurement system. Reporting a ‘deposition velocity’ under bidirectional conditions should therefore identify whether the analysis isolated downward-flux periods, modelled separate emission and deposition terms, or retained a signed exchange metric.
The footprint is a moving source-and-sink landscape
A flux instrument does not measure a single square metre directly below the sensor. Turbulence carries air from an upwind area whose location and weighting depend on wind direction, measurement height, atmospheric stability, roughness and other conditions. This footprint may encompass grass, soil, roads, buildings, water or several surface types. A time-series change can reflect a shift of the footprint rather than a change in the physical exchange properties of one stable surface.
Suppose a tower measures a net downward number flux when wind arrives over forest but a net upward flux when the wind swings toward a nearby industrial road. It would be misleading to conclude that the forest suddenly began emitting particles. The observed sectors and their surface composition changed. An appropriate analysis assigns wind-sector and footprint information to each accepted averaging interval, rejects obstructed or unsuitable sectors under declared criteria, and avoids treating the whole experiment as one homogeneous source.
Footprint models carry their own assumptions and cannot precisely trace each particle’s actual trajectory. Their job is to estimate how different upwind distances contribute to the measured turbulent exchange. Where surfaces are mixed, uncertainty in footprint allocation becomes part of the interpretation. A map may clarify the result, but only if the source area is treated as a weighted region rather than a decorative pin directly beneath the inlet.
Time response, inlet residence time and lag alignment
Particle instruments differ in how rapidly they can resolve concentration changes. A condensation particle counter may detect a number concentration with a finite response and internal transport delay; an optical particle counter may provide size bins with different effective sampling characteristics. Aerosol sampling tubing adds residence time, wall losses and high-frequency smoothing. Turbulent fluctuations can be lost even when the instrument reports an accurate average number concentration. This creates a distinct challenge for flux that does not arise from ordinary time-mean concentration analysis.
Align the vertical wind and aerosol signals before calculating covariance. An empirical covariance-lag search can help identify transport delay, but unconstrained maximization can produce a spurious peak from noise. A physical estimate based on tubing length, flow and instrument response should inform the allowable lag range. Preserve the lag choice, the search window and diagnostic plots. A sudden change in optimal lag after a pump change may reflect a hardware event rather than a change in atmospheric transport.
High-frequency attenuation needs appropriate correction or explicit qualification. Underestimated covariance magnitude can make both upward and downward fluxes appear too small. A correction factor designed for one sensor response or particle-size bin should not be pasted onto another by habit. At the same time, large empirical corrections can amplify noise; report the basis and sensitivity of any applied correction. A credible flux estimate must have a credible instrument response.
Rotation, averaging and statistical stationarity
Flux methods usually transform measured wind components into a coordinate system in which vertical velocity has an interpretable mean relative to the local flow. Choices such as double rotation or planar-fit approaches should follow the site methodology and remain consistent across analyses. A tilt error can introduce contamination from a much larger horizontal wind component into apparent vertical motion. Mechanical orientation and coordinate processing are part of the measurement, not optional plotting settings.
A covariance calculated across a calm half hour and a period containing a sharp wind shift may combine incompatible flow regimes. Stationarity checks ask whether turbulent statistics behave sufficiently consistently for the selected averaging interval to have meaning. Shortening the interval can improve local consistency but may exclude important low-frequency transport; lengthening it can capture slow structures while mixing different meteorological states. There is no magical duration that is always ideal for every site and sensor.
Because concentration and wind fluctuate, uncertainty arises from finite sampling of turbulent events. Repeated intervals and flux quality flags matter. A small positive flux with uncertainty spanning zero is not strong evidence of emission. Similarly, the most negative single half-hour measurement may reflect a nonrepresentative event or instrument disturbance. Aggregate results should show accepted periods, signs, dispersion and any corrections, rather than elevating one extreme datum into a definitive surface property.
Particle size changes the physics of exchange
Particle deposition responds to size-dependent processes including Brownian diffusion, interception, inertial impaction, gravitational settling and interactions with turbulent boundary layers. Different mechanisms dominate in different size ranges and surfaces. An instrument measuring particles between 10 and 30 nm cannot automatically represent the flux of larger coarse particles. A number flux may be dominated by many ultrafine particles even when the mass flux is dominated by fewer larger particles. State whether the reported flux is number, mass, or size-resolved.
Suppose one size class contains 10⁹ particles/m³ with very little mass per particle while another contains 10⁶ particles/m³ whose individual mass is a million times larger. Their number concentrations differ by a factor of a thousand, yet the second class could dominate mass despite its lower count. Actual mass conversion additionally depends on particle shape, density and diameter definition. Reporting a number deposition velocity as though it were a mass-specific chemical removal rate is not legitimate without a justified conversion model.
Airborne nanoparticles may be lost to an inlet by diffusion, while large particles may be lost through inertial or gravitational effects depending on tubing geometry and flow. Particle charge and humidity can also matter. These instrument losses can bias number concentrations and their high-frequency fluctuations differently. Before concluding that the surface preferentially deposited one size bin, check whether the sampling train itself removed that bin.
Net exchange is not gross emission or gross deposition
A measured turbulent covariance can capture a net result of simultaneous upward and downward contributions. Suppose a surface emits 8 × 10⁷ particles/(m² s) while physical removal represents 5 × 10⁷ particles/(m² s), under an illustrative accounting convention. The net exchange is +3 × 10⁷ particles/(m² s). An observer measuring only the net flux cannot uniquely recover those two components from the single number. Many combinations of emission and deposition could produce the same net result.
This matters over water, roads, vegetation and snow where emissions and removal may coexist, vary by particle size, or depend on wind and surface processes. The literature sometimes estimates one component with a separately parameterized deposition model and subtracts it from observed net flux to infer emissions. Such results depend on the additional model. A report must label which values are direct net flux estimates and which are calculated gross flux components.
If a field paper presents a downward ‘deposition velocity’ only during selected intervals, it may intentionally exclude net emission periods. Comparing that number with a signed median over all periods would be an apples-to-oranges comparison. The correct translation describes filtering rules, positive/negative convention and why a net effect should not be misrepresented as a single physical pathway.
Field data quality must be visible
A practical audit asks whether vertical wind and aerosol channels were synchronized, whether sample flow was stable, whether temperature and humidity influenced the instrument, whether the turbulent footprint was acceptable and whether nonstationary intervals were excluded. Quality criteria should also examine spikes, missing-data fractions, wind sector, instrument diagnostics, frequency response and estimated random uncertainty. Not every spike is bad: genuine plumes exist. The analyst’s task is to distinguish instrument artifacts from environmental events without designing exclusions only after viewing a desired outcome.
A dust pulse could increase concentration sharply without representing a continuous turbulent flux that can be reliably estimated over a full half-hour. A maintenance operation near the sampling inlet could introduce a highly local artifact. A rain event may interfere with an optical counter in a different way from a condensation counter. Site notes, photographs, maintenance logs and redundant measurements can resolve ambiguities that a statistical filter alone cannot.
Report how many intervals were attempted, how many passed each major quality gate and the time periods excluded. When an uncertainty interval includes both emission and deposition directions, say the flux was not resolved from zero rather than selecting a sign. Transparency about lost data is not evidence of poor work; it enables readers to judge what the accepted observations can support.
Worked example: use concentration and flux in matching units
An invented half-hour dataset has a measured number concentration Nref = 4,000 particles/cm³ and a corrected turbulent number flux F = −2.0 × 10⁷ particles/(m² s), upward defined positive. First convert concentration: 4,000 × 10⁶ = 4.0 × 10⁹ particles/m³. Then calculate v_d = −F/Nref = 2.0 × 10⁷ / (4.0 × 10⁹) = 0.005 m/s, or 5 mm/s. This quantity is the effective downward exchange velocity for the size interval and footprint represented by that period. It does not tell us whether each individual particle physically fell at 5 mm/s.
Now imagine the next period yields Nref = 8,000 particles/cm³ and F = −2.0 × 10⁷ particles/(m² s). The magnitude of flux is unchanged while the reference concentration doubles, so the inferred effective velocity falls to 0.0025 m/s. This does not prove that the surface physically changed. It highlights that exchange velocity is a ratio involving an environmental concentration and an estimated net flux, both of which have uncertainties and changing atmospheric contexts.
For a third period, Nref = 4,000 particles/cm³ and F = +2.0 × 10⁷ particles/(m² s). The flux is now upward. If one mechanically calculates −F/Nref the signed velocity is −0.005 m/s, but the communicative label should be net upward exchange, not a negative gravitational settling speed. A table that retains the signed flux and an explicitly defined exchange velocity is less ambiguous than one that displays a column called deposition without sign conventions.
Worked example: residence time and spectral loss
Suppose a number-concentration instrument draws through a 6 m inlet line at a mean air speed of 3 m/s. The idealized bulk transit time is approximately 2 s, before instrument response and dispersion are considered. Pairing the concentration at recorder timestamp 12:00:00 with wind at precisely the same timestamp may miss the matching eddy if the two channels were not hardware-synchronized after transit. A physical transit estimate narrows the search for the correct lag, but an actual response test or constrained covariance analysis remains necessary.
Consider a simplified high-frequency correction in which the measured covariance retains only 80% of the magnitude expected under a validated response model. A measured F = −8 × 10⁶ particles/(m² s) would correspond to an estimated corrected flux F ≈ −1 × 10⁷ particles/(m² s). That correction has its own uncertainty. Applying it to an unrelated instrument or particle-size range without a response basis would be invalid. Reporting the raw and corrected values reveals how much of a conclusion depends on modelled recovery.
The correction must not be used to hide an unreliable trace. When raw covariance is indistinguishable from noise, dividing by a small response fraction can magnify meaningless fluctuations. Inspect lag-covariance curves, spectra and timing stability before accepting a correction. A statement that a correction increased flux magnitude by 25% is informative, but the reader also needs to know whether the correction was independently calibrated and whether that 25% is small or large relative to total uncertainty.
Uncertainty and the difference between sign and evidence
Every flux estimate has random and systematic uncertainty. Random uncertainty arises because the chosen averaging interval includes only a finite sample of turbulent structures. Systematic biases may arise from calibration drift, uncorrected high-frequency response, inlet losses, footprint mismatch, wind tilt and nonstationarity. A covariance of +1 × 10⁷ particles/(m² s) with an uncertainty range spanning −1 × 10⁷ to +3 × 10⁷ does not decisively establish upward exchange. Report the interval, not a categorical source statement.
Uncertainty in a ratio can become large when the denominator is small or noisy. For v_d = −F/Nref, the numerator and denominator may also have correlated errors if particle detection varies during an event. Simply adding their percentage uncertainties ignores correlation. A first-order propagation can sometimes be used under stated assumptions, but sensitivity analysis over plausible flux and reference concentration ranges is often clearer for readers. If the sign of F itself is unresolved, a positive deposition velocity derived from its central estimate is not strong evidence of deposition.
Reporting a long time-series average requires care about sign cancellation. Suppose one accepted period shows +5 × 10⁷ and another −5 × 10⁷ particles/(m² s). Their arithmetic mean is zero, but the surface did not necessarily have no exchange. It experienced substantial opposing net fluxes. Show positive and negative fractions, distribution and environmental context as well as the net mean. The distinction between weak flux throughout and large cancelling fluxes can matter for mechanistic interpretation.
Site example: road, trees and a changing wind direction
Consider a tower near a tree stand and a busy road. During winds from the tree stand, the footprint model suggests that most weighted exchange originates above vegetation. The observed corrected net flux is downward in a certain ultrafine size range. During a wind shift toward the road, short upward bursts coincide with traffic movement. A causal story is plausible but not yet proven: vehicle-related emissions may contribute, while turbulence, other upwind sources, and instrumental response must be considered.
To evaluate the road hypothesis, compare wind sectors with traffic records, size-resolved concentration, background conditions and flux quality. An upward burst when traffic is absent weakens a simple traffic-only explanation. A flux signal that vanishes after time-lag correction or when inlet disturbances are removed may have been partly artifactual. A robust report identifies the strength of corroborating evidence rather than declaring a source from a covariance sign alone.
The forest comparison is equally vulnerable to simplification. Leaf surfaces can capture particles, but vegetation may also emit volatile compounds that contribute to secondary particle formation under appropriate conditions. A measured flux at tower height does not separate every process without additional analysis. Distinguish surface deposition, above-canopy nucleation and transport from neighbouring sectors. The lesson is to use the footprint and supporting measurements to test a mechanism, not to decorate a preselected conclusion.
Site example: coastal aerosols and apparent upward exchange
Over water, waves, wind and bursting bubbles can emit sea-spray aerosol, while particles can also deposit into the surface. A flux instrument records a turbulent net effect within its target size range. During high winds an upward particle number flux may be consistent with sea-spray generation, but source attribution is strengthened by concurrent wind, wave, size-distribution and composition evidence. A model estimating dry deposition can be combined with measured net flux to infer emission, but the two should not be described as independently measured components.
Suppose measured net upward exchange is +3 × 10⁷ particles/(m² s) and a separately modelled gross downward deposition is 1 × 10⁷ particles/(m² s). Under a consistent sign convention, estimated upward emission is 4 × 10⁷ particles/(m² s), because emission minus deposition equals net flux. If the deposition model is uncertain by 50%, that uncertainty propagates to the inferred emission. A paper that prints only the 4 × 10⁷ result obscures the important modelling dependency.
In coastal datasets, the upwind fetch may change as wind rotates from open water toward shore. Aerosol size distributions can also change with humidity. The estimated flux of dry-equivalent particle number may differ from an optical counter’s apparent wet-particle classification. Surface type, inlet elevation and aerosol drying conditions must be stated before comparing coastal campaigns. The same arithmetic cannot rescue measurements that characterize physically different populations.
Site example: snow, deposition and bidirectional exchange
Snow-covered landscapes can receive atmospheric particles but may also display intermittent upward number fluxes under particular meteorological and chemical conditions. An apparent exchange velocity can vary strongly across size ranges and stability regimes. Field studies have reported both directions in some environments, which is why forcing all negative or positive periods into a single deposition-only interpretation loses information. Study the accepted footprints and whether local particle formation or blowing material might matter.
Imagine a measurement reports a median signed exchange velocity of −0.2 mm/s under a convention where positive means downward. A reader cannot interpret the sign until that convention is confirmed; another study might define upward positive. Translate it as the authors defined it, and state whether the median includes periods of emission. If someone converts −0.2 mm/s to −0.2 m/s, the result is a thousand-fold error. Always write 1 mm/s = 0.001 m/s before comparing studies.
Snow-surface observations may also have weak turbulent signals during stable conditions and occasional bursts under stronger winds. A mean over all intervals can be sensitive to which intervals passed quality control. Readers should see a description of meteorological regime and sample count, not merely the largest statistically significant value. A measurement programme is strongest when it treats conflicting conditions as informative parts of the physical system rather than inconvenient noise.
What cannot be inferred from one net particle flux
A number flux alone cannot determine mass flux without information about particle sizes, density and morphology. It cannot identify the chemical species in a mixture. It cannot reveal a unique source location without footprint and contextual evidence. It cannot separate two simultaneous counteracting gross processes. It also cannot guarantee that a measured covariance at instrument height equals the flux entering an individual leaf surface, because vertical divergence, particle formation and storage can matter between the surface and the sampling plane.
Suppose the instrument detects 5 × 10⁷ particles/(m² s) of upward exchange in particles between 10 and 100 nm. It would be incorrect to label that 5 × 10⁷ micrograms/(m² s). A particle count is not a mass, and even selecting a typical diameter does not give precise mass without density, shape and distribution. A scientifically careful conversion could instead introduce a measured size distribution, an explicit particle volume model and uncertainty due to particle density or refractive index when relevant.
Similarly, a forest footprint showing net downward number flux does not by itself establish that the forest improved public-health outcomes by a known amount. Human exposure depends on concentrations in occupied places, time, particle composition and many other variables. Flux measurements provide powerful evidence about atmosphere–surface exchange, but their public interpretation should remain within the process they actually represent.
A reproducible field-data workflow
Start from synchronized high-frequency wind and aerosol concentration channels. Record sample rate, instrument clocks, inlet dimensions, flow, measurement height, particle-size detection, data gaps and maintenance history. Validate physical units. Remove documented instrument failures under predeclared criteria, not simply because individual flux values look inconvenient. Apply a justified coordinate treatment, examine time lag and response corrections, and compute turbulent covariance over selected averaging intervals. Keep raw and corrected results separately accessible.
Next inspect wind direction, atmospheric stability and footprint suitability for each interval. Flag contamination by nearby obstructions or changing surface fetch. Evaluate whether turbulent statistics are sufficiently stationary, inspect spectral and ogive diagnostics as suitable, and quantify uncertainty. In size-resolved work, keep the definitions and detection characteristics of each size bin. A consistent workflow makes it possible to compare atmospheric changes without mixing instrument changes with environmental variation.
Finally publish the essential bookkeeping: measurement period, data quantity, exclusion counts, correction methodology, sign convention, units and summaries of accepted positive and negative fluxes. If the data do not resolve a sign, use the words ‘not distinguishable from zero’ rather than an unsupported direction. A visually polished figure is valuable only when the reader can trace a plotted number to an observation, a processing choice and a defensible interpretation.
Independent transfer problems with checked answers
Problem A. Vertical velocity deviations are +0.2, −0.2, +0.2, −0.2 m/s, and synchronous number-concentration deviations are −50, +50, −50, +50 particles/cm³. Find the covariance. Answer: each product is −10 in mixed units, so F = −10 × 10⁶ = −1.0 × 10⁷ particles/(m² s). With upward-positive w this indicates net downward turbulent transport over the four-point illustrative sequence. The mean concentration is not needed for the covariance itself.
Problem B. The reference number concentration is 2,000 particles/cm³ and the net downward flux is −1 × 10⁷ particles/(m² s). Find v_d. Answer: 2,000/cm³ is 2 × 10⁹/m³, so v_d = 1 × 10⁷/(2 × 10⁹) = 0.005 m/s, or 5 mm/s. Problem C. Why is 0.005 m/s not the speed of every particle? Answer: the ratio is an effective bulk exchange metric; actual transport depends on particle size, turbulence and other processes.
Problem D. During an hour, F1 = +6 × 10⁷ and F2 = −6 × 10⁷ particles/(m² s) for two equal-length quality-controlled periods. What is the time-mean net flux? Answer: zero. Does this mean there was no exchange? No: substantial opposite fluxes cancelled in the mean. Problem E. A sensor samples at 10 Hz for 30 minutes with no gaps. How many time points? Answer: 10 × 1800 = 18,000 per stream. If 900 are missing, 17,100 remain, but counting remaining points alone does not establish a valid flux.
Problem F. A measured magnitude is 8 × 10⁶ with a justified 80% frequency-response recovery. What corrected magnitude does the simple model imply? Answer: 8 × 10⁶/0.8 = 1 × 10⁷ in the same flux units. The sign stays the same. Problem G. A 4 m tube has mean transit speed 2 m/s. What is the nominal transit delay? Answer: 2 s, to be checked against additional sensor and dispersion response. Do not maximize covariance over unlimited delays without physical constraints.
Problem H. A surface emits 9 × 10⁷ particles/(m² s) and deposits 7 × 10⁷ in the same unit convention. Net flux is +2 × 10⁷ upward. If only +2 × 10⁷ were measured, can the two gross components be uniquely recovered? No. Problem I. A flux is −2 × 10⁷ with uncertainty ±3 × 10⁷ in the same units. Is deposition decisively established? No: the uncertainty spans zero and upward values. Problem J. Wind shifts from forest to road during the averaging interval. What must be checked? Footprint heterogeneity, stationarity, possible source change and whether to split or exclude that period.
Frequently asked questions
Is aerosol deposition velocity the same as wind speed? No. Wind is bulk air motion; deposition velocity is a flux-to-concentration ratio summarizing net particle transport toward a surface under a chosen convention. Can a deposition velocity exceed a simple gravitational terminal speed? Effective transport includes turbulence and collection mechanisms, so treating it as a settling speed is inappropriate. Is a negative sign automatically bad data? No. It may indicate direction, or merely the authors’ convention. Read definitions before interpreting.
Why does concentration remain high during deposition? Because sources, transport and mixing can replenish the air as a surface removes some particles. Why can particle flux be positive at night? The sign depends on actual exchange processes and local conditions, not on a simple day/night rule. Why do different particle-size bins have different velocities? Diffusion, interception, inertial effects, settling and sampling biases have size dependencies.
Can a single tower measure a whole city? No. Its footprint represents an upwind weighted area under particular conditions. Can number flux be converted to PM2.5 mass flux directly? Not without size, density and classification assumptions or independent mass-resolved measurements. Does a robust covariance prove an exact source? It demonstrates a turbulent exchange signal; attribution requires spatial, chemical and meteorological corroboration. What is the essential reporting habit? Always show sign convention, particle population, units, height, averaging and QC.
The practical reading skill: put the denominator back
Specialist plots become easier to read when the denominator is made visible. A concentration is per volume of air. A flux is per area and time. A deposition velocity is flux divided by a reference concentration. A footprint is not an exact boundary; it is a weighted contribution region. Once these are separated, the apparent contradiction between rising concentration and downward flux often disappears. The two statements describe different physical dimensions of the atmosphere.
This same discipline supports understanding mobility particle-size distributions and cascade impactor deposition on internal stages, while remembering that sampler collection is not the same as atmospheric dry deposition. The mathematical principle of checking each operation is useful here: convert units before forming ratios, identify assumptions and verify the direction of every reported effect.
A complete interpretation might read: ‘For accepted intervals in the defined wind sector, number flux in this particle-size interval was predominantly downward under the upward-positive convention. The flux-to-concentration ratio gives a stated effective deposition velocity, with uncertainty and instrumentation corrections. The result describes net exchange over the modelled footprint, not a unique gross sink or chemical composition.’ That statement is longer than a headline, but it preserves precisely what makes the science useful.
Primary research and methodological references
Schmidt and Klemm (2008): size-resolved turbulent particle fluxes using eddy covariance and an electrical low-pressure impactor. Baltic Sea aerosol eddy-covariance flux and sea-spray emission comparison (2024). High-Arctic particle flux and gradient relationships over changing surfaces (2025). ALPACA snow-surface particle dry-deposition and signed exchange velocities (2025). Exercises in this explainer are hypothetical arithmetic examples, not observations from these studies.