C-DRONE GUIDE · 2 SEPTEMBER 2026

Measuring river discharge by drone: image velocimetry (LSPIV), method, accuracy, price

A conventional gauging requires putting something in the water: a current meter on a wading rod, an acoustic profiler towed from bank to bank, sometimes a hydrographer standing in the flow. During a flood — precisely when the measurement is most valuable — that operation becomes slow, hazardous, sometimes impossible. Image velocimetry proposes the opposite: film the water surface from a hovering drone, measure how visible tracers move between two frames, derive a surface velocity field, then a discharge by combining it with the geometry of the bed. The method is old in the laboratory, mature in rivers, and supported by free software. It is not magic for all that: it rests on two assumptions that must be openly discussed — the presence of usable surface tracers, and the step from surface velocity to depth-averaged velocity. Here is the actual method, its validity conditions, the accuracy that can honestly be claimed, and what a campaign costs in 2026.

Published on 2 September 2026, reviewed on 2 September 2026 — regulations in force as of September 2026.

Image velocimetry: what a drone actually measures above a river

The technique is called LSPIV, for Large Scale Particle Image Velocimetry. The principle fits in one sentence: if a pattern visible on the water surface (foam, bubbles, a leaf, a ripple, scum, debris) moves by so many pixels between two frames separated by a known time interval, and if you know how many real centimetres a pixel is worth, then you know its velocity. Repeated over a grid of points covering the whole visible water surface, this yields a two-dimensional surface velocity field.

Two families of algorithms coexist. LSPIV proper works by correlating patterns: it splits the image into interrogation windows and looks, in the next frame, for the position where the pattern most resembles itself. LSPTV (Large Scale Particle Tracking Velocimetry), and PTV approaches more generally, instead track individual objects identified frame after frame. LSPIV is more robust when the surface is densely textured; LSPTV behaves better when tracers are sparse and well contrasted. Other variants exist and are used in both research and operations: KLT-IV, OTV, SSIV.

On the software side, two tools dominate European practice and are available without a commercial licence. Fudaa-LSPIV is free and open-source software, co-developed and distributed by EDF and INRAE together with the company DeltaCAD since 2010, available in French and English on Windows and Linux, and supported by annual user workshops since 2017. RIVeR (Rectification of Image Velocity Results), published by Patalano, García and Rodríguez in Computers & Geosciences in 2017 and today distributed as open source under the AGPL-3.0 licence, offers an equivalent chain for LSPIV and LSPTV. A serious deliverable states in its report which software and which version produced the result: that is the first condition for a third party to be able to reproduce it.

One crucial point, often misunderstood at purchase: the drone does not measure a discharge, it measures surface velocities. Discharge is a later computation, which needs two further ingredients — the wetted cross-section and the velocity coefficient — covered below. An offer that presents LSPIV as a direct discharge measurement is glossing over most of the uncertainty.

What a drone gauging actually involves, step by step

1. Choosing the reach. You look for a reach as uniform as possible over a few tens of metres: flow roughly parallel to the banks, no major eddy, no strong reverse current, a visible surface without continuous vegetation cover. A reach just upstream or downstream of an existing gauging station lets the result be tied to a stage reading taken at the same moment, which is the prerequisite for feeding a rating curve.

2. Placing and surveying ground control points. This is the step inexperienced providers rush, and it is the one everything else depends on. You need at least four points identifiable in the image, spread around the filmed area, whose coordinates are known: targets laid on the banks, structure corners, or characteristic points surveyed by GNSS. These points drive orthorectification, that is, the transformation from image coordinates to ground coordinates. Without them the scale is wrong, and so is the discharge. An RTK or PPK survey is the fastest way to acquire them at useful accuracy: see our guide on RTK/PPK drones and centimetre accuracy. If the site is to be gauged repeatedly, those points are permanently marked once and for all.

3. The flight. The drone holds a hover, nadir (camera pointing straight down) or near-nadir, typically a few tens of metres above the water — published work on medium-sized rivers commonly operates between 20 and 30 m. Two requirements pull against each other: climb high enough to cover the full width, stay low enough that each pixel covers little ground and tracers are resolved. The nadir view is preferred because it limits perspective distortion and rectification error; an oblique view remains possible but degrades accuracy on the far side of the river.

4. The video. A continuous sequence is recorded, typically on the order of a minute, at stable frame rate and resolution, with exposure and focus locked. The duration must cover enough turbulent fluctuation for the time-averaged velocity field to be stable. The exact time and the stage read on the staff gauge are noted.

5. Stabilisation and rectification. Even in a hover, a drone drifts and oscillates. Before any computation the sequence is stabilised by registering each frame onto fixed landmarks outside the water (rocks, structures, trees). Each frame is then orthorectified from the control points. A significant share of the final uncertainty is decided at this stage.

6. Velocity field, filtering, cross-section, discharge. The software computes displacements over a grid, averages them in time, filters out spurious vectors (shadowed areas, glare, motionless floating vegetation), then integrates velocities along a cross-section whose geometry must be known by other means. Discharge comes out of that integration, corrected by the velocity coefficient.

From surface velocity to discharge: the velocity coefficient and the cross-section

The drone sees the surface, and only the surface. Yet water does not flow at the same speed from bed to surface: friction on the bed slows the lower layers. To move from surface velocity to depth-averaged velocity, a velocity coefficient is applied, written alpha, defined as the ratio of depth-averaged velocity to surface velocity. Discharge is then written, panel by panel: discharge = alpha × surface velocity × wetted area.

The commonly adopted default value is 0.85. It follows from assuming a logarithmic vertical velocity profile, or a 1/6 power law, over a bed of ordinary roughness. But that 0.85 is a fallback convention, not a physical constant. A review by Biggs, Smart, Doyle, Eickelberg, Aberle, Randall and Detert published in 2023 in Water (Surface Velocity to Depth-Averaged Velocity — A Review of Methods to Estimate Alpha and Remaining Challenges) surveys the methods for estimating alpha and recalls that, under conditions commonly met during current-meter or acoustic profiler gaugings, alpha typically varies between 0.7 and 0.9: lower on small rough-bedded rivers, higher in smooth artificial channels where depth greatly exceeds roughness height.

The commercial consequence is direct. Taking 0.85 by default where the site is really worth 0.75 means overestimating discharge by roughly 13 % — a gap that will not survive scrutiny by a regulator when an annual abstracted volume or a compensation flow is at stake. A competent provider:

• states explicitly the alpha value used and its justification;
• offers, where the stakes justify it, to calibrate it on site against a reference gauging with an acoustic profiler or current meter carried out the same day;
• reports the sensitivity of discharge to that choice, that is, how much the result moves between alpha = 0.80 and alpha = 0.90.

In other words: the first campaign on a new site almost always benefits from including a simultaneous reference gauging. It is not a superfluous expense, it is what turns an indicative measurement into a defensible one, and what makes later — purely aerial — passes usable.

The second ingredient is the geometry of the bed, and this is the structural limit of the method: a surface velocity alone yields no discharge. It must be multiplied by an area, so the bed below the waterline at the chosen section must be known, along with the stage at the time of the flight. Three ways to get there, depending on the site:

• Section surveyed at low flow. When the bed is dry or nearly so in late summer, the section is surveyed by conventional aerial photogrammetry, exactly like a topographic survey — see the drone surveying and photogrammetry hub. It is then reused for every later gauging, as long as the morphology has not shifted.
• Bathymetry. On a watercourse that never dries out, measurement has to happen underwater: green bathymetric laser in clear shallow water, or single-beam echo sounder on a boat or surface water drone in turbid water. Success conditions, achievable depths and pricing are detailed in our guide on drone bathymetry of lakes, reservoirs and waterways.
• Known engineered section. On a concrete canal, a calibrated weir or a tailrace structure, the geometry comes from the as-built drawings. This is the most favourable configuration — and it is why LSPIV deploys easily downstream of a hydropower scheme or on an irrigation canal.

Two precautions deserve to be written into the contract. The section must be re-checked after any morphogenic event: a flood shifts bars and makes a section surveyed last year lie. And bank morphology evolves between campaigns, which affects both the section and the control network; on reaches monitored over time, gauging benefits from being tied to a riverbank and watercourse monitoring round, which uses the same flight and the same georeferencing. Finally, if the section is not known and cannot be, one honest way out remains: deliver the surface velocity field alone, as input or validation data for a hydraulic model. That is a legitimate deliverable for an engineering firm, provided it is not invoiced as a gauging.

Validity conditions: when the method works, and when it does not

Image velocimetry is not universal. Six conditions determine whether a flight will be usable, and an honest provider declines the job or flags the risk when they are not met.

1. Visible surface tracers. This is the primary condition. In flood, the river carries foam, scum, leaves and driftwood: the surface is richly textured and the method is at its best. At low flow, on smooth debris-free water, there is nothing to track — the computation returns noise. Two possible responses: artificial seeding (releasing biodegradable floating tracers from upstream, a technique used in research), or abandoning the method. This is the useful paradox of LSPIV: it performs best precisely when the situation is dangerous for a human operator.

2. Little or no wind. Sustained wind does two distinct kinds of damage: it destabilises the drone, degrading rectification, and above all it drags the surface film independently of the flow, directly biasing the measured velocity. Published work explicitly recommends avoiding windy and rainy conditions. A cross gust on a slow river can distort the result by tens of per cent.

3. A near-nadir viewing angle. The further the line of sight departs from vertical, the larger the rectification error becomes on the far part of the image, and the more a small altitude or attitude error translates into a large ground-distance error.

4. Careful georeferencing. At least four control points, well distributed, surveyed to an accuracy consistent with the accuracy sought on discharge, and unambiguously visible in the video.

5. A genuinely stationary hover, or software stabilisation. Any residual drift of the drone reads as water velocity unless corrected by registration onto fixed landmarks outside the water.

6. Light without specular reflection. Low or overhead sun reflecting off the water creates a blown-out zone where no correlation is possible; overcast sky is often the best condition. Shadows cast by bankside trees, canopy hiding part of the width, and motionless floating vegetation (which produces false zero vectors) are the three other classic traps.

To this are added the ordinary airspace constraints: flying over a watercourse crossing a built-up area, proximity to an aerodrome, wildlife protection zones, waterway manager consent. These fall under the usual regulatory scoping of a mission and are checked before travelling, not on site.

Expected accuracy: what the literature honestly allows you to claim

No serious provider should quote a single universal uncertainty figure for LSPIV, because none exists: performance depends on the site, on tracer density, on georeferencing quality, on the choice of alpha and — the most often ignored factor — on the operator processing the video. Three published results usefully frame expectations.

On the gap to a reference gauging. A technical note by Anette Eltner, Hannes Sardemann and Jens Grundmann published in 2020 in Hydrology and Earth System Sciences (Flow velocity and discharge measurement in rivers using terrestrial and unmanned-aerial-vehicle imagery) compares discharges obtained from drone imagery with acoustic profiler reference measurements on two German rivers, with flights at 20 and 30 m. On the river with a regular section, the maximum error recorded is 5 %. On the river with an irregular section and incomplete tracer coverage, discrepancies rise to 7 to 31 %. That gap between the two sites captures what is at stake better than any average: the method is good where the site suits it, mediocre where it does not.

On the operator effect. This is the most useful result for a buyer, and the most counter-intuitive. Guillaume Bodart, Jérôme Le Coz, Alexandre Hauet and Magali Jodeau published in 2024 in Water Resources Research an analysis (Quantifying and Reducing the Operator Effect in LSPIV Discharge Measurements) based on an intercomparison in which fifteen to twenty-odd participants processed the same videos with the same Fudaa-LSPIV software. The result: the interquartile range of relative discharge error reached 17 % and the median bias −9 %, purely because of processing choices (time interval between frames, grid of points, filters). With assistance tools and automated filters, that spread falls to about 2 % and the bias to about 1 %. In other words: for an identical video, skill and processing method weigh as much as the flying hardware. That is a provider selection criterion, not a detail.

What this means concretely for the delivered report. A usable deliverable must contain: the raw video and the stabilised sequence; the coordinates and accuracy of the control points; the software and its version; the time interval, interrogation window size and filters applied; the cross-section used and its survey date; the stage reading and the time; the alpha value and its justification; the mean velocity field and the discharge; and a reasoned uncertainty estimate rather than an isolated number. It is worth knowing that full uncertainty quantification for video-based measurements remains an open research topic in hydrometry: a deliverable that displays an error bar without saying how it was built should be questioned.

The French framework: hydrometry quality charter, HydroPortail, ISO 748

Drone gauging is not a separate regulatory object: it sits within the general framework of French hydrometry, which must be understood to deliver reusable data. Three references structure practice, current as of September 2026.

The hydrometry quality charter — good practice guide. Published by the French ministry responsible for the environment, in its second edition in 2017 (the first dated from 1998), written by Christian Perret, Stéphanie Poligot-Pitsch and Rachel Puechberty, it is the methodological reference for the field. It covers gauging methods — velocity-area exploration, tracer dilution, volumetric methods and non-intrusive technologies including video analysis and radar —, the construction of rating curves relating stage to discharge, and the management and communication of uncertainty at every stage of data production. A gauging delivered to a public client benefits from explicitly adopting its vocabulary and reporting structure.

HydroPortail. Since 25 January 2022, HydroPortail (hydro.eaufrance.fr) has replaced Banque HYDRO as the reference site for accessing public stage and discharge data for French watercourses. It is administered by SCHAPI, the central service for hydrometeorology and flood forecasting support, with responsibility for data quality remaining with station operators. If a gauging is intended to consolidate an official rating curve, the format and coding must be agreed with the station operator — regional environment directorate, flood forecasting service, river basin authority, concession holder — before the flight, not after.

ISO 748:2021. The international standard "Hydrometry — Measurement of liquid flow in open channels — Velocity area methods using point velocity measurements", fifth edition, cancels and replaces ISO 748:2007. It describes the velocity-area method that remains the backbone of the calculation: measuring velocities on verticals distributed across the section, then integrating. It also covers surface velocity measurement systems, floats included — that is, the family of methods of which LSPIV is the instrumented version.

Two professional use cases draw particular benefit from this framework. Industrial sites subject to a prefectural discharge order, which must justify an effluent flow or a receiving watercourse flow: traceability of the calculation then matters as much as the value. And hydropower operators, for checking compensation flows downstream of an intake, where aerial gauging avoids working in a deeply incised bed — a job adjacent to dam and levee inspection by drone, which often shares the same site visits. For local authorities, discharge data also feeds hazard studies: see our guide on flood-prone area and flood risk plan mapping by drone.

Observed 2026 pricing for drone gauging

Pricing is built on three independent items: acquisition (travel and flight, short), processing (the heaviest share of time), and knowledge of the cross-section (done once, then reused). A site already instrumented and already surveyed therefore costs markedly less on the next pass than on the first campaign. Ranges observed in France in 2026, excluding VAT:

Three remarks for reading these figures. First, the flight is the short part: a usable sequence is filmed in a few minutes, whereas stabilisation, rectification, computation, filtering and writing take up most of the billed time. A quote priced in proportion to flight time betrays a poor grasp of the method. Second, pooling is the real saving: a single site visit can carry the gauging, the bank survey round and a structure inspection. Third, the relevant comparison is not with a current-meter gauging in calm weather, but with what it takes to gauge a flood: a crew, safety measures, access, a window of a few hours. To place these amounts among the site's other services, see our guide on how much a drone service costs.

Getting a reach gauged

To price an intervention, the useful information fits in a few lines: where the reach is and whether a gauging station stands nearby, the order of magnitude of width and depth, the target discharge range (low flow, mean flow, flood), whether a cross-section has already been surveyed, and the final use of the data — rating curve, compensation flow check, justification to a regulator, hydraulic model validation. Also state whether the site is reachable on foot from a bank, which determines how control points can be placed.

Describe your requirement on the request a quote page: the reply will state whether image velocimetry is relevant on your site, which conditions must be met on the day of the flight, and what else is needed to obtain a discharge rather than surface velocities alone.

Request a free quote

Also worth reading

• FEDER: Funding a Drone Project with EU Regional Funds (Local Authorities, SMEs)
• Protected species on facades: surveying nests by drone before renovation or demolition, price
• Drone water sampling: equipment, laboratory acceptance and pricing