henhaobio
Sign in

bioRxivEcologyPreprint

The morphotype approach to classification of aerial animals in radar data

Yuval WerberiD

PDFDOIData

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 3 (13 Jul 2026) · CC BY 4.0

Splitting BirdScan MR1 radar bird classes by wingbeat frequency and radar cross section turns 4 classes into 31 morphotypes, most matching under 10 Israeli species.

Vertical looking radars track individual airborne animals at large scale but almost never resolve species, and AI classifiers so far give only coarse taxonomic classes, which limits ecological inference and uptake of radar aeroecology. This work subdivides the radar's built-in bird classes into morphotypes defined by size and wingbeat, links them to local species inventories, and opens the way to diversity measures for the airspace.

Key results

  1. ShownSplitting the radar's 4 bird classes by wingbeat interval and body size produced 31 statistically distinct morphotypes, raising classification resolution by a factor of 7.75.

    Wader 8 morphotypes (F=194, p<0.0001), Passerine 11 (F=600, p<0.0001), Bird 10 (F=285, p<0.0001), Large single bird only 2 (F=4.7, p=0.009); from 4 to 31 classes (factor 7.75) using Hula Valley radar data from August 2018 to March 2021. · Data collection and preparation; Morphotype segregation

  2. SuggestedWhen the local species inventory was mapped onto morphotypes without regard to season, a substantial share of morphotypes corresponded to one species or a small set of species.

    5 morphotypes (15%) held a single species, 9 (28%) fewer than 5 and 19 (59%) fewer than 10 (Table S2), from 332 species of 21 orders taken from the SPNI database. Baseline: the 4 MR1 classes held 65, 141, 92 and 33 species. · Hula Valley case study - Taxonomic survey; Local species inventory case study

  3. SuggestedAdding month of occurrence narrowed the candidates further: a larger fraction of morphotype/month combinations mapped to a single species or a small species set.

    From the monthly table of 3,124 species-month entries (Table S3), 69 morphotype/months (19%) were linked to a single species, 137 (38%) to fewer than 5 and 257 (71%) to fewer than 10; percentages are shares of morphotype/months, not of entries. · Hula Valley case study - Taxonomic survey; Local species inventory case study

  4. ShownWithin radar classes, birds that flapped faster were generally smaller, which supports wingbeat frequency as a size-linked axis for separating morphotypes.

    Negative WFF-RCS correlation in Wader (coefficient -0.2), Passerine (-0.63) and Bird (-0.48) classes, all p<0.0001; no correlation reported for the Large single bird class. · Morphotype segregation

  5. ClaimedThe author presents the approach as transferable to any BirdScan dataset with a species inventory, to other individual-level vertical radars, to bats and, at coarser taxonomic levels, to insects, enabling 'Aerodiversity' metrics.

    Stated in the Discussion and Perspectives sections. Only Hula Valley bird data were analyzed: bats were removed with BATScan and insects excluded, and no test across sites or radar systems is presented. · Data collection and preparation; Discussion; Perspectives and applications

Methods, data and reproducibility

Within each MR1 bird class, detections are binned into 1 or 2 Hz wing flapping frequency (WFF) intervals. The top 5% of radar cross section (RCS) values per bin are compared between neighbouring bins (ANOVA, Tukey HSD), and significant breaks define morphotypes. Local species are then mapped to morphotypes by taxonomy, size and WFF.

  • BirdScan MR1 vertical radar bird detections, Hula Valley Research Center, Israel, Aug 2018 to Mar 2021; short-pulse mode, rain-free, 50 to 1,000 m AGL, classifier probability >50% · birds (multiple species)
  • Subset of accurately sized detections: top 5% of RCS values per WFF interval, used for the morphotype statistics · birds (multiple species) · n = 5% of all detections
  • SPNI-based inventory of bird species in Mediterranean Israel with genus, class, mass, body length and monthly prevalence (Table S1) · birds · n = 332 species of 21 orders
  • Monthly table of species occurrence mapped to morphotype/month labels (Table S3) · birds · n = 3,124 species-month entries
  • Species wingbeat frequencies from published literature, with Rayner (1995) allometric estimates where none were published · birds
  • Rainfall measurements from a nearby meteorological station, matched to detections to exclude rainy periods
Methods & tools
BirdScan MR1 X-band vertical looking radar (Swiss BirdRadar Solution AG, software version 1.6), MR1 random forest target classifier, BATScan classifier (Werber et al. 2023b) for bat removal, On-radar fast Fourier transform for WFF estimation, Top-5% RCS size-accuracy filter (Schmid et al. 2019), ANOVA (aov, base R), TukeyHSD (base R), R Studio version 4.5.0, Rayner (1995) allometric WFF-mass equation, SPNI bird database (birds.org.il)
Reagents & environment
R Studio version 4.5.0 using base R aov and TukeyHSD; BirdScan MR1 radar software version 1.6
Reproducibility gaps
The Discussion mentions published code but gives no repository link. No deposition or accession is stated for the radar detection data. Species traits and morphotype assignments are only in supplementary Tables S1 to S3.
Limitations (5)
  • RCS matches true body surface only in level flight directly above the radar and is biased downward at the beam edge or in pitched or banked flight, so sizing depends on the top 5% subset and on the radar's WFF being computed correctly. · stated by the authors
  • Detections during rain (unreliable radar output) and below 50 m AGL (ground clutter) were discarded, and those above 1,000 m AGL were dropped so that the full range of local bird sizes stayed detectable. · stated by the authors
  • Wingbeat information is not available from every radar system, and the great diversity of insects would require higher taxonomic levels and body-shape parameters; neither bats nor insects were segregated in this study. · stated by the authors
  • Species were assigned to morphotypes from literature or allometric WFF and body length, not from ground-truthed radar detections (e.g., cameras or acoustics), so the 1 to 10 species correspondence is projected rather than validated, at a single site. · noted in this summary
  • There are internal inconsistencies: the per-class species counts (65, 141, 92, 33) do not add up to the stated 332; the 8 passerine and 12 Large single bird unrepresented species are described as <1% of species; and total radar detection counts are never given. · noted in this summary

Read this if you work with vertical looking or BirdScan radar data and want a simple statistical way to split coarse bird classes into size and wingbeat morphotypes for diversity analyses.

Citations
1

Abstract

Radar aeroecology makes ecological inferences about aerial wildlife using radars. While generating unique, large-scale datasets describing airborne biological activity, radars are largely incapable of relating detections to specific species and are thus taxonomically limited. I describe an analytical method to increase taxonomic resolution in vertical looking radar data by dividing detected organisms into morphology- and movement-based "aerial morphotypes". The Birdscan MR1 radar, widely used in aerial wildlife research and monitoring, is deployed in dozens of locations across Europe, the Near East and the United States. Using the MR1s target classifier, wing flapping frequency calculations and target size estimations, I demonstrate a nearly 8-fold increase in bird classification resolution using radar data from the Hula Valley Research station, Israel. I show that most newly defined morphotypes correspond to small numbers of local species (1-10). Thus, the approach provides a realistic opportunity to bridge the taxonomy gap in radar aeroecology. By using the approach, radar aeroecologists can estimate "Aerodiversity", analogous to biodiversity, a fundamental, widely used measure in ecology and conservation. By explicitly addressing taxonomy in radar aeroecology, practitioners will increase the impact and dissemination of their work and contribute to a deeper, more complete understanding of the aerial habitat.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISShi '25Nussbaumer '21Lochmann '26Ostojić '26Sheldon '26Nussbaumer '19
72% semantic matchEcology
Xu Shi, Jacob R. Drucker, Jason W. Chapman+2 more·Ornithological Applications·Feb 27, 2025· 5 citations