Bird collision and electrocution risk

1The problem: we know they kill, we don't know which one to fix first

Distribution power lines electrocute the birds that perch on their pylons; conductors and turbine blades strike them in flight. These are the leading causes of non-natural mortality among large raptors and soaring birds in Spain.

Electrocution

Distribution pylons (1-66 kV)

The bird perches on the cross-arm and touches two phases, or a phase and earth. It depends on the pylon design, the type of insulator and whether there are taps or disconnectors. It affects vultures, eagles, storks, kites and Eurasian eagle-owl.

Collision with conductors

Any line, above all the earth wire

Birds with poor manoeuvrability or flying in flocks, at dawn and dusk: common cranes, storks, Eurasian spoonbills, grey herons, great bustards. Worse on hillsides, passes and flyways.

Collision with wind turbines

Blades on ridges and passes

Soaring birds that use the very updraughts that justify the wind farm: griffon vulture, Egyptian vulture, bearded vulture, golden eagle, kites.

What is missing

Spanish Royal Decree 1432/2008 requires dangerous lines inside protection zones to be retrofitted, and there is excellent academic work on specific sites and years. But there is no up-to-date, queryable service that tells a utility, a public authority or an NGO which specific elements to review first, in which month and for which species. Retrofitting is decided from old inventories, from mortality someone happened to come across, or segment by segment once a complaint arrives.

3Data used: all open, none behind a key

The design condition was that anyone could run the project without asking for permission. That holds; the only layer that requires a key (eBird Status & Trends) is ready but not yet loaded.

Source	What it provides	Volume in the pilots	Licence
OpenStreetMap (Overpass)	Lines with voltage and operator, pylons and poles with their tags (material, insulator, taps), wind turbines	Cantabria: 46,300 pylons, 4,535 segments, 463 turbines Strait: 2,703 pylons, 414 segments, 516 turbines	ODbL
GBIF, which integrates the eBird Observation Dataset	Georeferenced records by species, month and year since 2010. Records of all birds measure observation effort	Cantabria: 1.35 M bird records (effort), 101,000 of 18 species Strait: 165,000 records of 16 species	CC0 / CC BY / CC BY-NC depending on the dataset
Movebank Data Repository	GPS tracks published alongside the scientific papers: where tagged individuals actually fly	Strait: 296,719 positions of black kite, 64,567 of white stork Cantabria: 5 positions (no published datasets for the area)	CC0
AWS Terrain Tiles (terrarium)	Elevation at 30 m → slope and topographic position index (ridges and passes)	Both regions in full	Public domain
Literature	Sensitivity weights by species and mechanism (Bevanger 1998, Janss 2000, Lehman 2007, Marques 2014) and status in the Spanish Catalogue of Threatened Species	21 species	—

Species assessed

Strait: griffon vulture, cinereous vulture, Egyptian vulture, Spanish imperial eagle, Bonelli's eagle, golden eagle, short-toed snake eagle, booted eagle, European honey buzzard, black kite and red kite, peregrine falcon, white stork and black stork, common crane and Eurasian eagle-owl. Cantabria swaps the Mediterranean species for bearded vulture, red-billed chough, Eurasian spoonbill and grey heron. Each species carries a weight per mechanism (0-1) and a conservation weight, both adjustable in a single configuration file.

4Results so far

Two pilots with opposite profiles: the Strait, with few pylons but a migration bottleneck and a great deal of GPS data; Cantabria, with 46,000 pylons, no GPS and a mountain avifauna.

Annual risk index in Cantabria: cells coloured from yellow to dark red over a shaded relief map with the power lines in grey — **Cantabria and its surroundings** (42.75-43.55° N · 4.90-3.10° W). The highest-index pylons are in the south of the frame, in Las Merindades (Burgos), in February, with white stork, wintering red kite and griffon vulture. The riskiest line segments are 12 kV and 30 kV lines on the floor of the Liébana valley, in July, beneath the passage of booted eagles, European honey buzzards and short-toed snake eagles. Wind turbines peak in November on the eastern moorlands (red kite, griffon vulture, golden eagle). The Campoo wind farms fall in cells with an index three times the 99th percentile.

Annual risk index in the Strait of Gibraltar over a shaded relief map — **Strait of Gibraltar** (36.00-36.45° N · 5.95-5.30° W). The 300,000 GPS positions of black kite reinforce the sierras east of Tarifa, where the wind farms are concentrated. The riskiest distribution pylons are in that same strip (El Cuartón-Pelayo) in October and November, with cinereous vulture, red kite and Bonelli's eagle; lines and wind turbines peak in September and October, at the height of the post-breeding migration.

Model output

Clear seasonality

Cantabrian pylons peak in February (wintering red kites and arriving storks); the Liébana lines in July; the wind turbines of the Strait during the autumn passage. Retrofitting in the month before the peak is what matters to a utility.

Model output

Projects scored

The same raster scores any polygon in seconds. Applied to the notices from the public consultation observatory, the Somaloma-Las Quemadas wind farm (Campoo) reaches a maximum of 329, peaking in September.

Not yet

No validation against real mortality

The ranking is consistent with what is known about each area, but nobody has gone out to check whether there are carcasses under the top 40 pylons. That is the first action we are asking for (section 6).

5Challenges and problems encountered

Each one changed the code. We set them out because they are the same ones anyone will meet who wants to reproduce or extend the work.

GBIF is not built for downloading hundreds of thousands of recordsThe search API becomes extremely slow with deep offsets and returns 503 errors at peak times.Download in annual blocks, most recent first, with three threads, patient retries and a per-block cache so it can resume. Capped at 20,000 records per species and region.
Movebank cuts off large downloadsThe published files range from 10 to 700 MB and the server closes the connection halfway through (IncompleteRead); the direct API requires an account.Streaming download filtering by region on the fly, three retries, and files larger than 150 MB are skipped. Movebank is left out of the automated monthly process.
No published GPS dataset covers CantabriaThe data on bearded vulture, Egyptian vulture or the vultures of the Cantabrian range exist but are not held in open repositories.The GPS weight became proportional to the number of positions so that the 5 positions found would not alter anything. The real fix is an agreement with whoever holds the data.
A single record could send the index soaringA migrating black stork photographed next to a substation with 41 pylons put that cell in first place for the whole of Cantabria.Wider smoothing where there are few records, √n weighting for stacked records, a cap of 20 pylons per cell and a confidence weight per species in the aggregate.
The main Overpass server kept failingA list of mirrors with automatic fallback; the response is cached, so in production OSM is queried once a month.
The map with 41,000 pylons weighed 37 MB and froze the browserLines go in a single GeoJSON layer and only the 4,000 highest-index pylons are drawn; the rest are in the CSV and the GeoJSON. 7.8 MB.
OSM knows little about each pylonOnly 11% of Cantabria's pylons (5% in the Strait) carry any tag for material, insulator or taps. The real hazard of the design, which is what determines electrocution, is known to the utilities and is not public.A default hazard factor of 1.0 and a clear list of which attributes would change the ranking if someone provided them.
Records carry observer biasRoads, viewpoints and hotspots concentrate records even where they do not concentrate birds; and no record says at what height the bird was flying.Correction for sampling effort (records of all birds) and GPS where it exists. This is the main limitation and it motivates the next section.
There is no open mortality data in SpainWe looked for power line inventories and electrocution registers from the regional governments, NGOs and LIFE projects: aggregate figures are published, not points.What we found is documented (the Basque Government's shapefile of dangerous lines is the only open one) and the statistical calibration is ready for when the data arrive.

6What would improve accuracy, in order

The model already ranks; what is missing is to calibrate it against reality and refine the two weakest layers: the actual presence of birds in rarely visited cells and the actual condition of each pylon.

Priority	Action	What it fixes	Who can help
1	Field-check the top 40 pylons in each region: photograph the design, search for carcasses under the pylon, RD 1432/2008 form.	Validates the ranking and gives the first mortality points for calibration. One weekend of volunteering per region.	Local NGO groups; environmental rangers.
1	Existing mortality records: power line inventories from the regional governments, wildlife rescue centres (cause of admission and coordinates), utility data and LIFE projects.	Replaces the literature weights with weights estimated by regression on real cases.	Public authorities and rescue centres, with the NGO as counterpart.
1	eBird Status & Trends as the main abundance layer (weekly, 3 km, modelled with effort).	Removes much of the observer bias and fills in the cells with no records. The loader is already written; what is missing is the key, which is being processed.	Cornell Lab (free research key).
2	Acoustic stations with BirdNET on wind turbines, on the pylons at the top of the ranking and in rarely visited areas. A microphone plus a single-board computer (BirdNET-Pi or BirdWeather) that identifies songs and calls continuously and publishes the detections; these can be uploaded to eBird as checklists.	Presence 24 hours a day, 365 days a year, without depending on an observer passing by: exact date and time of use of each site. It works very well for choughs, cranes, storks, eagle-owls, jackdaws and passerines; poorly for vultures and eagles, which barely vocalise. Hence it is combined with the next item.	NGOs and wind farm developers (the farms already have power and a network at each turbine).
2	Cameras and radar at the wind farms: the detection systems already fitted to many wind turbines (selective shutdown) record species, height and trajectory.	Flight height, the variable that no record provides; and ground truth for collision with blades.	Developers; the authorities can require the records to be handed over in the environmental decision.
2	Pylon design data from the utilities: type of cross-arm, insulator, jumpers, disconnectors on each pylon.	Real hazard ratings instead of the default 1.0 applied to 89% of the pylons.	Distribution companies (Viesgo/EDP, Endesa, i-DE), required to keep inventories under RD 1432/2008.
2	GPS from the local populations: bearded vultures, Egyptian vultures and the vultures of the Cantabrian range already carry transmitters from conservation programmes.	The layer that improved the Strait most does not exist in Cantabria. An agreement on aggregated use (individual-days per cell) is enough, with no need to hand over tracks.	Fundación para la Conservación del Quebrantahuesos, GREFA, regional authorities.
3	Targeted citizen science: the effort map itself flags the cells with power lines and no records; ask observers to visit them.	Fills the gaps at zero cost.	NGO members, local eBird groups.
3	Validation in the Basque Country: a new region and comparison with the Basque Government's official shapefile of dangerous line segments.	A first independent check at no cost.	Us.
3	PNOA LiDAR terrain (2-5 m) and wind variables.	Better detection of passes and ascent slopes; seasonality of thermals.	IGN (open data).

On the microphones: automatic song identification is done not by eBird but by BirdNET, from the same laboratory (Cornell Lab of Ornithology). A station costs around €100-150 in components, draws 5 W and detects at some 100-300 m depending on species and noise; on a wind turbine the microphone has to be kept away from the nacelle because of blade noise. It provides presence and phenology with hourly precision, not flight height or collisions: it complements the cameras, it does not replace them.

7What we ask for and what we offer

We ask for

An institutional counterpart to request mortality records and pylon design data from public authorities and utilities; local groups for the field check of the top pylons in the ranking; agreements on the aggregated use of GPS data; and a pilot area in which to install the first acoustic stations.

We offer

Open, documented code with no licence costs; maps and rankings updated every month; new regions in hours (all we need is the bounding box); automatic scoring of the wind farm and power line projects that go out to public consultation, ready for consultation submissions; and GeoTIFF and CSV files that open in QGIS.

Status and next steps

A working prototype (version 0.1, September 2026) with two regions, an automated monthly process and a public website. In progress: the eBird Status & Trends key. Next: field checks of the top pylons and a validation region in the Basque Country.

Maps, rankings and downloads Repository and full method Public consultation observatory

How it is computed

Everything happens on a grid of 0.005° cells (about 450 × 550 m). For each species s, month m and cell, three indices are computed:

electrocution[s,m] = P[s,m] · w_electro[s] · equivalent_pylons            (pylons ≤ 66 kV only)
line_collision[s,m] = P[s,m] · w_col_lin[s] · km_of_line · (1 + terrain)
turbine_collision[s,m] = P[s,m] · w_col_aer[s] · n_wind_turbines · (1 + ridge)

Download the infrastructure from OpenStreetMapLines classified by voltage: up to 66 kV is distribution (electrocution risk, the scope of RD 1432/2008); above that, collision only. Pylons inherit the voltage of their line. Where no pylons are mapped, 6 per km of distribution line are assumed.
Hazard rating of each pylonA factor of 0.5-2 from the OSM tags: wood 0.5; pin insulator 1.5; strain insulator 1.3; taps, sectioning points and line terminations 1.3; horizontal cross-arms 1.1; portal frames 1.3. With no tags, 1.0. Capped at 20 pylons per cell (more than that is a substation, not a power line).
Monthly presence P by speciesDensity of records of the species divided by the density of records of all birds in the same month (this corrects for the fact that people observe more near roads and viewpoints). Adaptive Gaussian smoothing: 1.5 km with 300 records or more in the month, wider with fewer. Records repeated at the same point, month and year count as √n. Normalised to 0-1 per species.
GPS tracking where it existsIndividual-days per cell and month from the Movebank tracks, blended into P with a weight that grows with the number of positions (up to 50% from 2,000 positions). Five fixes from a passing bird do not move the map; 300,000 do.
TerrainSlope (up to doubling the risk of collision with lines above 25°) and topographic position index at 1.5 km (ridges and passes, where soaring birds cross low and where wind farms are built).
Aggregation and confidenceEach species adds to the total index with a weight of status × confidence, where confidence is min(1, records/500) or grows with the GPS positions: a species with 120 records in the region (common crane in Cantabria) keeps its own layer but does not dominate the ranking.
Scaling and rankingEach index is rescaled so that the region's 99th percentile equals 100, without clipping the maxima. Each pylon takes the electrocution index of its cell multiplied by its hazard rating; each segment, the mean along its route; each wind turbine, that of its cell. The peak month and the three species contributing most are stored.
Publish and repeatInteractive map with layers by month, mechanism and species; report; CSV and GeoJSON of the ranking; 12-band GeoTIFF for QGIS. An automated process recomputes everything on the 2nd of each month with the new records.

Parameters

Grid cell	0.005°
Maximum distribution voltage	66 kV
Pylons assumed where none are mapped	6 per km
Pylon cap per cell	20
Record smoothing (σ)	1,500 m with ≥300 records in the month
GBIF records	from 2010, up to 20,000 per species
Minimum records to include a species	30
Full confidence	500 records or 2,000 GPS fixes
Maximum GPS weight (Movebank)	50%
eBird Status & Trends weight	70%
Slope at which line collision doubles	25°
Topographic position radius	1,500 m
Scale	99th percentile = 100

Validation

Pending. The ranking has not yet been checked against real mortality: there is no open dataset of electrocuted or collided birds in Spain. The first planned check is against the Basque Government's shapefile of hazardous power lines (region euskadi); once mortality points are available, the species weights will be fitted with a logistic regression.

Limits

The index is relative: 100 is the 99th percentile of the cells with infrastructure in the region, and values above 100 are the top 1% and keep their order for the ranking. It is not a probability of mortality: it is meant for prioritising, not for certifying.
OSM coverage is incomplete, especially for pylons and voltages.
The occurrence data carry observer bias (roads, viewpoints, hotspots) even after the effort correction, and carry no flight height; only some GPS datasets do.

Power-line risk for birds