C‑DRONE
Agricultural drone flying over a farm field

C-DRONE GUIDE · 1 SEPTEMBER 2026

Detecting Downy and Powdery Mildew in Vineyards with a Multispectral Drone: What the Imagery Really Shows

Vines cover roughly 3% of France's farmed area yet account for close to 20% of national pesticide consumption; their treatment frequency index sits around 13, against 3.8 for arable crops (French Senate report no. 42, 2012-2013). Most of that gap comes down to two pests: downy mildew and powdery mildew. Hence the value, for an estate or a growers' group, of knowing where pressure is rising, not just when. Multispectral drones have been sold for a few years now as the answer to that question. They do add something real — just not what the sales brochures imply. A multispectral sensor does not detect a pathogen: it measures reflectance, meaning a drop in photosynthetic activity and a change in canopy structure. It is a prioritisation tool, not a diagnostic one. Here is exactly what it shows, what it confuses, when to fly it, what it changes in a reasoned protection plan, and what it costs in 2026.

Published on 1 September 2026, reviewed on 1 September 2026 — regulations in force as of September 2026.

What multispectral imagery actually detects: a plant losing vigour, not a fungus

A multispectral camera carried by a drone measures the light reflected by the canopy in a handful of narrow spectral bands — typically green, red, the red edge (the transition zone between red and infrared, around 720 nm) and near-infrared. From those bands you compose vegetation indices: NDVI, NDRE, GNDVI and a few others. Those indices are arithmetic ratios, nothing more. They describe the state of a canopy: chlorophyll content, internal tissue structure, canopy density, quantity of green biomass per pixel.

The consequence is direct and too rarely stated: the sensor sees neither Plasmopara viticola nor Erysiphe necator. It sees a leaf photosynthesising less, a leaf area shrinking, a shoot whose growth has stopped. Disease appears in the imagery only through the physiological effect it produces on the plant, and only once that effect rises above the block's agronomic background noise. Between infection and aerial legibility of the symptom lies a delay that nothing shortens.

What that delay covers has been measured precisely. A study by F. Portela, J. J. Sousa, C. Araújo-Paredes, E. Peres, R. Morais and L. Pádua, published in 2025 in Agronomy, followed a block in Portugal's Vinho Verde region where 84 vines were instrumented, half receiving the usual fungicide protection and half deliberately left untreated for the whole season; seven multispectral flights were flown at successive phenological stages (see the study on Google Scholar). The authors describe a progressive, consistent signature on infected vines: increased reflectance in the visible and red-edge bands, together with a progressive decline in near-infrared reflectance, with infection also restricting vegetative growth measurably in morphometric parameters. In other words: the signal exists, it is reproducible, and it builds over time rather than appearing all at once.

That distinction — anomaly detected versus disease diagnosed — is the dividing line between an honest service and a sales pitch. It applies to all plant imagery monitoring, in fact; we set it out the same way in our guide to drone-based tree health and fall-risk assessment, where the drone documents without ever replacing expert judgement. On the fundamental differences between a multispectral sensor and a plain RGB camera, our comparison guide multispectral or RGB camera on an agricultural drone details what each one measures; and for those wondering whether hyperspectral would change the picture, our guide to the hyperspectral drone camera explains what that level of spectral detail adds — and at what price.

Two diseases, two biologies, two levels of aerial detectability

They are always mentioned together, but downy and powdery mildew share neither nature, nor behaviour, nor visibility from the air. First point, often overlooked: downy mildew is not a fungus. Plasmopara viticola is an oomycete, an organism long filed among the "lower fungi" but which molecular analysis now places with the Chromista, alongside algae and diatoms; it reached France around 1878, imported from North America (Ephytia-INRAE factsheet). Powdery mildew, by contrast, is a true fungus, the ascomycete Erysiphe necator. That difference is no taxonomist's flourish: it explains why the two diseases respond neither to the same weather conditions nor to the same families of active substances.

Downy mildew needs free water. Its primary infections start in spring, when overwintering oospores germinate: the empirical "rule of three tens" — 10 cm shoots, 10 mm of rain, 10 °C average temperature — remains a handy field marker, nowadays cross-checked against simulation models. The classic symptoms (oil spot on the upper leaf surface, white felting underneath) translate, at block scale, into patches that appear first where water stands and air moves poorly: low ground, dips in the terrain, badly drained headlands, shaded margins. That is exactly the geometry an index map is good at revealing: a spatially organised anomaly, not diffuse noise.

Powdery mildew works differently. It needs no rain: mild overcast weather, moderate humidity and a dense canopy are enough, and it thrives in dry years when downy mildew stays quiet. Its costliest damage happens on the bunch — greyed, split berries, then an entry point for rot — that is, on an organ the drone barely sees from directly above, hidden by foliage. This has to be said plainly: powdery mildew is markedly less detectable from the air than downy mildew, and any contractor promising powdery mildew mapping at the same reliability as downy mildew mapping is overpromising. What the imagery picks up, in the powdery mildew case, is a weakening of the canopy in the worst-affected sectors — later and more diffuse.

A third player often muddies the picture: black rot, whose pressure has been rising in several French vineyards and whose leaf symptoms are easily mistaken for downy mildew by an untrained eye — and, all the more so, by a vegetation index. Here again, the map locates; identification happens in the row.

What the research says, and what it does not establish

The literature here is real and of good quality. The most-cited French work is that of M. Kerkech, A. Hafiane and R. Canals (PRISME laboratory, Orléans-Bourges), published in 2020 in Computers and Electronics in Agriculture. The authors set out a complete downy mildew detection chain from drone imagery: optimised registration between visible and infrared images — a harder problem than it looks, the two sensors sharing neither optics nor shutter instant — then segmentation by a convolutional neural network classifying every pixel into four categories: ground, shadow, healthy vegetation, symptomatic vegetation. Reported performance reaches over 92% detection at vine level and 87% at leaf level (see the study on Google Scholar). The gap between those two figures is instructive: identifying which vine is affected is markedly more reliable than delineating which leaf is.

On the proximal side — that is, close to the canopy rather than from the sky — L. Ghiani, S. Serra, A. Sassu, A. Deidda, A. Deidda and F. Gambella published in 2025 in Smart Agricultural Technology a joint automated detection of downy and powdery mildew symptoms, built on a field image set annotated by experts over two years of collection, with a YOLO-type model reaching a mean average precision (mAP) of 0.730 (see the study on Google Scholar). The authors stress a rarely highlighted point: performance depends heavily on the diversity of the dataset and on how it is split between training and validation — a model trained on a single vintage, a single variety and a single block posts flattering scores that collapse elsewhere.

What this work cautiously supports: a well-timed multispectral flight, processed with the right indices and properly trained models, lets you rank a block and spot organised outbreaks before they become obvious at row scale. What it does not establish: that a map suffices to name the disease, that laboratory performance transfers unchanged to a heterogeneous commercial holding, or that a model trained in Vinho Verde or Sardinia works without recalibration on a Burgundy hillside. The Portela design in particular rests on half the vines being deliberately left untreated — a legitimate experimental setup, but one with nothing in common with the residual pressure of a protected vineyard.

The limits: everything that looks like an outbreak without being one

This is the section most sales brochures leave out, and yet it decides whether a campaign is useful or merely expensive. A vegetation index map does not show "downy mildew": it shows an area where the canopy is less vigorous than the block average. The possible causes of such a decline are numerous, and several are more common than disease itself:

Two practical safeguards limit these confusions. The first is the baseline flight early in the season, which fixes the block's "normal" vigour map: anything departing from it afterwards becomes a signal, whereas a standalone map is indistinguishable from a soil map. The second is ground truthing: some twenty geolocated sample points, scored by eye by the technician on the day of the flight or the day after, let you check what the image is claiming and adjust thresholds. Without those two precautions, a multispectral campaign produces attractive maps that decide nothing. On the general principle of vigour mapping, our guide to drones in viticulture: mapping vine vigour lays the groundwork.

The flight calendar: phenological stages, rainfall, weather windows

The right calendar is not a fixed one. It combines three clocks: vine phenology, the season's rainfall, and the weather constraints specific to flying.

Phenology first. The baseline flight goes in once the canopy is established and before any significant pressure, around the 8 to 12 unfolded leaves stage. Then comes the critical period, from the end of flowering to bunch closure: this is when berries are most receptive and when an uncontrolled outbreak is paid for in yield. One or two passes here offer the best information-to-cost ratio. After veraison, susceptibility drops and a health-monitoring flight loses its point — the multispectral camera then shifts to another use, mapping grape ripeness for selective harvesting.

Rainfall next. Intermediate passes are triggered by events: after a rainfall episode that met infection conditions, you must let the incubation period run — variable with temperature, from a few days to around ten — before the symptom becomes marked enough to register on a vegetation index. Flying the morning after the storm achieves nothing; flying three weeks later means arriving once the outbreak is visible from the track. Scheduling is therefore a two-way conversation with the adviser or technician following the risk models.

Flight constraints last, and they are not details. A usable multispectral campaign requires stable illumination: a slot around solar noon to limit shadows cast between rows — critical in trellised vines, where the neighbouring row's shadow occupies a large share of the pixel — a homogeneous sky (bright clear or uniform cover, never a parade of cumulus), moderate wind, and a reading of the radiometric calibration panel before and after the flight to make passes comparable. Without that rigour, two maps taken a fortnight apart cannot be compared, and the whole multi-temporal argument collapses. Over large areas — a group of several hundred hectares, a cooperative winery — the choice of platform becomes structural: our guide to fixed-wing versus multirotor over a large area details the threshold beyond which a multirotor no longer holds up.

What it changes in a protection plan: scouting rounds, variable rate, traceability

The drone's main contribution is not to trigger a spray — protection against downy and powdery mildew stays preventive, and a map revealing an outbreak reveals a protection strategy already caught out. The contribution lies elsewhere, and it is very concrete for any operation managing area.

Prioritising scouting rounds. On an 80-hectare estate spread across a dozen blocks, a technician cannot walk everything every week at the same intensity. A dated anomaly map turns a uniform scouting round into a ranked itinerary: you go first to the five sectors that have declined, with a GPS point in your pocket. The gain is measured in hours of skilled labour and, above all, in how early the finding comes, in a trade where forty-eight hours change a season's outcome. It is the same reasoning we apply to any large-area survey, regulatory ones included.

Matching dose to actual canopy. Canopy-adjusted dose approaches reason on the leaf area to be protected, not on cadastral hectares. A vigour map supplies exactly the input variable that reasoning lacks at within-block scale: a young planting, a thin sector and a vigorous row do not call for the same spray volume. Acting on it assumes a sprayer capable of variable rate, which not every estate has; but even without automatic modulation, splitting a block into two or three dose zones is within reach of any operation.

Managing the copper budget. For organic or converting estates, the constraint is numerical: implementing regulation (EU) 2018/1981 of 13 December 2018, which renewed the approval of copper compounds for seven years, caps use at 4 kg/ha/year on average, i.e. 28 kg/ha over seven years, with the option to smooth from one year to the next. That smoothing is a multi-year management decision: knowing which blocks genuinely consume the copper budget, and which get by on less, has direct value. An archived annual map contributes more than memory does.

Documenting. Dated, georeferenced, archived maps build a block history that stands up to scrutiny: useful for an environmental certification audit, for an insurance file after a difficult vintage, for the due diligence of an estate acquisition, and simply to compare two strategies from one year to the next. On hillside blocks where machinery cannot pass, the drone can also spray: that is a separate regulatory framework, detailed in our guide to drone spraying in steep-slope vineyards, and in the one on the emergency derogation after a flood. For animal pests, drone-based biocontrol also exists — see our guide to mating disruption with pheromone dispensers.

What the drone does not replace: the plant health bulletin, the scouting round, the laboratory

Three systems remain primary, and a drone service can only honestly be sold as a complement to them.

The plant health bulletin (Bulletin de santé du végétal, BSV). Produced under France's biological surveillance of the territory scheme, it rests on a network of observers — chambers of agriculture, cooperatives, merchants, volunteer estates — applying harmonised observation protocols whose weekly reports feed a regional risk analysis combining phenological stages, damage thresholds and weather conditions. BSVs are distributed free of charge by the regional agriculture directorates and the regional chambers of agriculture, under the national Ecophyto plan. A regional bulletin tells you the pressure across the district; a drone map tells you where it is expressing itself on your land. Neither substitutes for the other: the bulletin informs the calendar, the map informs the geography.

The scouting round. Turning a leaf over to look for downy mildew's white felting underneath, opening a bunch, recognising a still-fresh oil spot: no airborne sensor does this, because those observations happen under the leaf, inside the canopy, on lesions a few millimetres across. That is precisely why the Ghiani study cited above works on proximal imagery: at symptom scale, you have to be in contact.

The laboratory. Where the identity of an agent is in doubt — downy mildew, black rot, yellows, trunk disease — only a plant clinic analysis settles it. This is especially true of regulated organisms, where confirmation triggers legal obligations. A flight report is never worth a laboratory result, and no serious drone contractor will present a map as a diagnosis.

Put differently: the drone shifts the constraint, it does not remove it. It turns an impossible question ("where do I look across 200 hectares?") into a tractable one ("let's go check these seven sectors"). That is a real gain, measurable in technician hours and in how early decisions are made — and that is already a great deal in a trade where the window for action is counted in days.

Method and 2026 prices (excl. VAT)

A campaign runs in four stages. Scoping: listing blocks, varieties, planting dates and airspace constraints (proximity to an aerodrome, restricted zones), and setting the calendar with the technical adviser. Baseline flight: multispectral acquisition at the 8-12 leaves stage, with a radiometric calibration panel and a saved flight plan so it can be replayed identically. Monitoring passes: triggered by rainfall events and phenological stage, same settings, same solar hour. Deliverables: orthophoto, index maps (NDVI, NDRE, GNDVI), a map of deviation from the baseline flight, an anomaly-zone layer exportable as shapefile or GeoJSON for the tractor console or the estate's GIS, and a reading note ranking the sectors to check on the ground — never naming a disease the imagery cannot name.

Orders of magnitude observed in France in 2026 (excl. VAT):

The business case rests not on the price of the flight but on what it helps avoid. In a vineyard where the treatment frequency index sits around 13, a half-dose saved on the least exposed blocks, an intervention brought forward by forty-eight hours on an incipient outbreak, or an hour of technician time better spent each week weigh more than the imagery budget. Conversely, on a uniform few-hectare holding the grower walks twice a week in person, the contribution is marginal and we say so: the drone's value grows with area and with the distance between blocks. We work across every French wine region — from Bordeaux to the Languedoc around Béziers, taking in Champagne and Burgundy; the regional specifics of agricultural missions are set out in our guide to drone services in Nouvelle-Aquitaine. To price a campaign on your holding, request a quote stating the area, the number of blocks, the varieties and the number of passes planned.

Frequently asked questions

Can a drone tell downy mildew from powdery mildew?

Not on its own. A multispectral camera measures light reflected in a few broad bands (green, red, red-edge, near-infrared): it translates a physiological state — a leaf photosynthesising less, a canopy degrading — not the identity of a pathogen. Two different diseases can produce similar signatures, and a nutrient deficiency or water stress can produce a comparable one. What the map delivers is a spatial ranking: the sectors of the block where something is wrong, sorted by intensity. Identification itself happens in the row, by eye, and where doubtful, in the laboratory. In practice, agronomic context helps a great deal: after a heavy rainfall episode, an anomaly appearing as a patch at the bottom of a block points strongly to downy mildew; the same anomaly in a hot dry year, on sheltered rows, points rather to powdery mildew. But that is a field inference, not a direct reading of the image.

When in the season should the flights be scheduled?

The first flight is a baseline flight: it goes in once the canopy is established but before any significant pressure, typically at the 8 to 12 unfolded leaves stage. Without that starting image, a vigour index map cannot be read — there is no way to tell a naturally less vigorous sector (thin soil, young planting, different rootstock) from a sector that has just declined. Subsequent passes are triggered by events rather than a fixed calendar: after a rainfall episode that met infection conditions, around flowering and fruit set — when the bunch is most susceptible — then at bunch closure. Beyond veraison, berry susceptibility falls sharply and a health-monitoring flight loses ground to ripeness monitoring. Three to five passes over a season cover the essentials; beyond that, cost rises faster than information.

Does a drone campaign genuinely cut pesticide use?

It can help, but the mechanism deserves honesty. Protection against downy and powdery mildew remains overwhelmingly preventive: you spray ahead of infection, on the strength of a forecast risk, not after seeing symptoms. A map that reveals an outbreak therefore reveals, by definition, a protection strategy already caught out. The gain lies elsewhere than in triggering the spray: spreading scouting effort better across a large holding, matching dose to actual canopy volume rather than to cadastral hectares, isolating at-risk blocks instead of reasoning across the whole estate, and documenting block history to adjust the following season's strategy. No serious contractor promises a guaranteed percentage cut in treatment frequency from flying a drone alone: that figure depends on the vintage, the variety, the sprayer and the grower's decisions far more than on the sensor.

Request a free quote

Put it into practice

Also worth reading