The morphotype approach to classification of aerial animals in radar data

This paper introduces a computational "morphotype" approach that analyzes movement and morphology in vertical radar data to significantly increase taxonomic resolution, thereby enabling the study of "Aerodiversity" and bridging the gap between large-scale radar observations and specific species identification.

The Gist

The Big Problem: The "Blind" Sky Watchers

Imagine you are standing in a dark room with a powerful flashlight (the Radar). You can see thousands of tiny, glowing specks flying around you. You know exactly how fast they are moving, how big they are, and how they are flapping their wings.

But here's the catch: You can't see their faces.

For decades, scientists using radar to study birds and insects in the sky have been like that person in the dark room. They can tell, "Oh, that's a big bird flapping slowly," or "That's a tiny insect buzzing fast." But they couldn't say, "That's a Sparrow" or "That's a Hawk."

This is a huge problem. In science, if you can't name the species, you can't really understand the ecosystem. It's like trying to study a forest by only counting "trees" without knowing if they are oaks, pines, or maples. You miss the whole story.

The Solution: The "Morphotype" Detective

This paper introduces a clever new method called the "Morphotype Approach." Think of it as a detective game where the radar data is the clue, and the goal is to narrow down the suspect list.

Instead of trying to guess the exact species immediately (which is impossible with radar alone), the author groups the flying animals into "Morphotypes."

The Analogy: The Airport Security Line
Imagine an airport security line.

• Old Way: The security guard just yells, "Next!" and lets everyone through. They know people are there, but they don't know who they are.
• New Way (Morphotype): The guard looks at the person's height, how fast they walk, and the size of their bag. They group them: "Tall people walking slowly with big bags," "Short people running with small bags."

Even if the guard doesn't know the person's name, they can say, "That group is almost certainly only business travelers," or "That group is almost certainly only families with toddlers."

The author did this with birds. By looking at wingbeat speed (how fast they flap) and size (how big their radar "shadow" is), he split the generic "Bird" category into 31 specific groups (Morphotypes).

How It Works (The Magic Trick)

The author used a radar station in Israel (the Hula Valley) that has been watching birds for years. Here is the step-by-step magic:

1. The Radar Data: The radar measures two things for every bird:
   • Wingbeat Frequency: How many times per second the bird flaps its wings. (Big birds flap slowly; tiny birds flap like crazy).
   • Radar Cross Section (RCS): How much of the radar signal bounces back. This tells us the size of the bird's body.
2. The Sorting: The author took the data and said, "Okay, let's look at all the 'Passerine' (small songbird) detections. If I group them by how fast they flap, do the sizes change?"
3. The Discovery: Yes! He found that within the "Passerine" group, there were actually 11 distinct sub-groups.
   • Group A: Flaps 10 times a second, small body.
   • Group B: Flaps 15 times a second, tiny body.
   • Group C: Flaps 8 times a second, slightly larger body.
4. The Match: He then took a list of every bird species known to live in that area (332 species) and matched them to these groups based on their known size and wingbeat speed.

The Result: From "Maybe" to "Probably"

Before this method, the radar data was like a blurry photo where you could only see a blob.

• Old Resolution: "There are 4 types of birds here."
• New Resolution: "There are 31 types of birds here."

This is an 8-fold increase in detail!

The Best Part:
The author found that most of these new groups were very specific.

• Some groups contained only 1 species.
• Most groups contained fewer than 10 species.

The Analogy:
Imagine you are looking at a crowd of people.

• Before: You say, "I see a crowd of people."
• After: You say, "I see a group of 3 red-hatted mailmen, a group of 2 blue-hatted chefs, and a group of 10 people wearing green suits."

You still don't know their names, but you know exactly what kind of people they are. In many cases, the radar data now points to just one or two possible bird species.

Why Does This Matter? (The "Aerodiversity" Concept)

The author coins a new term: "Aerodiversity."

Just as we talk about "biodiversity" (how many different species are in a forest) to measure the health of nature, we can now talk about "Aerodiversity" (how many different flying morphotypes are in the sky).

• Conservation: If we know that a specific "Morphotype" (which likely equals one specific endangered bird) is disappearing from the radar data, we can act fast to save it.
• Science: We can finally ask real questions like, "Are small birds migrating earlier this year because of climate change?" instead of just guessing.
• Communication: It helps scientists talk to the public and policymakers. Instead of saying "We detected biological targets," they can say, "We detected a high volume of Swallows and Swifts."

The Bottom Line

This paper is a bridge. It connects the high-tech, blurry world of radar with the specific, named world of bird species.

It doesn't require new, expensive cameras or AI supercomputers. It just requires a smart way of looking at the data we already have. By sorting birds into "Morphotypes" based on how they move and how big they are, we can finally stop looking at the sky as a blur of dots and start seeing it as a vibrant, diverse community of living creatures.

In short: We went from seeing a "flying blob" to seeing "a flock of 50 specific birds," and that changes everything for how we protect our skies.

Technical Summary

1. Problem Statement

Radar aeroecology utilizes Vertical Looking Radars (VLRs) to monitor large-scale biological activity in the airspace, providing unique datasets on migration, foraging, and behavior. However, a critical limitation exists: taxonomic blindness.

• Current radar methodologies can detect and track individual targets (from insects to large birds) but cannot directly identify specific species.
• Existing automated classifiers (e.g., Random Forest) provide only coarse categories (e.g., "Passerine," "Wader," "Insect"), which limits the ecological inference and the ability to correlate radar data with specific conservation needs or species-level biodiversity metrics.
• This lack of taxonomic resolution hinders the integration of radar data with traditional ecological studies and conservation planning.

2. Methodology

The author proposes a computational method to bridge the taxonomy gap by creating "aerial morphotypes"—operational units defined by morphology and movement rather than strict taxonomy.

Data Source & Preprocessing:

• Instrument: BirdScan MR1 (X-band VLR, 9.4 GHz) located at the Hula Valley Research Station, Israel (a major migration stopover).
• Dataset: Bird detections recorded between August 2018 and March 2021.
• Filtering:
  • Excluded rain periods, ground clutter (<50m AGL), and high-altitude targets (>1,000m AGL) to ensure detection of the full regional vertebrate range.
  • Retained only detections with >50% classification probability for bird classes ("Passerine," "Wader," "Large single bird," "Bird").
  • Excluded insects and bats using specific classifiers.

Morphotype Segregation Process:
The core innovation relies on the biological principle that wing flapping frequency (WFF) and body size are correlated with taxonomic affiliation.

1. Parameter Extraction: The radar automatically calculates:
   • Wing Flapping Frequency (WFF): Derived from the dominant frequency component of the raw echo signal (Fast Fourier Transformation).
   • Radar Cross Section (RCS): A measure of reflected radiation, serving as a proxy for body surface area (specifically the bottom surface area of birds).
2. Accurate RCS Subset Selection: Since RCS is often biased downward due to flight angle, the author selected the top 5% of RCS values within 1 Hz WFF intervals. This subset represents the most accurate size estimations for organisms of that specific flapping frequency.
3. Statistical Segregation:
   • The dataset was divided into 1 Hz WFF intervals.
   • An Analysis of Variance (ANOVA) followed by Tukey's HSD post-hoc test was applied to compare the median RCS of adjacent WFF intervals.
   • Significant differences in size between adjacent WFF intervals indicated a boundary between distinct morphotypes.
4. Taxonomic Mapping:
   • A local species inventory (332 species from the SPNI database) was compiled with estimated WFF and body dimensions.
   • Each species was assigned to a radar class and then mapped to the newly defined morphotypes based on their WFF and size.

3. Key Contributions

• The Morphotype Concept: Introduces a non-taxonomic, morphology-based classification system that acts as a proxy for species groups, allowing for finer resolution than standard radar classes.
• Computational Pipeline: Demonstrates a reproducible method to split coarse radar classes into subgroups using WFF and RCS statistical breaks.
• Taxonomic Bridging: Provides a framework to link radar data directly to local species inventories, moving from "unknown birds" to "likely species pools."
• Aerodiversity Framework: Proposes the concept of "Aerodiversity" (analogous to biodiversity), enabling the calculation of richness and diversity metrics for aerial habitats.

4. Results

• Resolution Increase: The method increased the classification resolution by nearly 8-fold, expanding the 4 original radar classes into 31 distinct morphotypes:
  • Passerine: 11 morphotypes
  • Wader: 8 morphotypes
  • Large single bird: 2 morphotypes
  • Bird: 10 morphotypes
• Species Correlation:
  • Annual View: 15% of morphotypes corresponded to a single species; 28% corresponded to fewer than 5 species; 59% corresponded to fewer than 10 species.
  • Monthly View: Resolution improved further due to seasonal turnover. 19% of morphotype-months were unique to a single species, and 71% were associated with fewer than 10 species.
• Biological Validation: The results confirmed the expected inverse relationship between WFF and body size (slower flappers are generally larger), validating the morphotype segregation.

5. Significance and Implications

• Ecological Inference: By reducing the "species pool" for a given radar detection to a small number of candidates (often 1–10), researchers can make realistic inferences about specific species' behaviors, migration timing, and abundance.
• Conservation Applications: The approach allows for the monitoring of specific threatened species or groups without requiring species-level identification algorithms, which are currently unreliable for radar.
• Standardization: It enables the application of standard ecological metrics (richness, diversity indices) to aerial habitats, facilitating comparisons with terrestrial and aquatic ecosystems.
• Future Outlook: While AI may eventually improve direct species identification, the morphotype approach offers an immediate, statistically robust solution to the taxonomy gap. It encourages radar aeroecologists to integrate local species data, thereby increasing the scientific impact, dissemination, and acceptance of radar-derived ecological findings.

In conclusion, Werber's work transforms radar data from a "black box" of biological activity into a structured, taxonomically informed dataset, paving the way for a new era of "Aerodiversity" research.