Gimpo Habitat Profile: Does the Incheon AVIX-AI Model Generalise?
A fitness study testing whether AVIX-AI BirdThreat trained at Incheon Technopark transfers to Gimpo Airport's Han River basin habitat — what matches, what diverges.
By Park Moojin · Topic: Gimpo Airport Habitat Profile: Generalising the Incheon ModelAVIX-AI BirdThreat trained at Incheon Technopark partially generalises to Gimpo Airport — shared EAAF stopover species and tidal flat foraging logic transfer well, but Gimpo's Han River riparian corridor introduces resident species assemblages and seasonal flooding patterns that require a dedicated habitat layer before full deployment.
Gimpo Habitat Profile: Does the Incheon AVIX-AI Model Generalise?
Abstract
The AVIX-AI BirdThreat system validated at Incheon Technopark in April 2026 represents the most rigorously documented machine-vision bird-threat pipeline operating in Korean low-altitude airspace today. Nineteen of nineteen inference calls returned HTTP 200 across the 4-stage habitat treatment pipeline, and animal-class entities now publish natively into Anduril Lattice — a provenance chain that satisfies both KAS audit requirements and dual-use operator standards. The question this article addresses is not whether the system works. It is whether the trained model, built on Incheon Airport's reclaimed-tidal-flat ecology, transfers with sufficient fidelity to Gimpo Airport's Han River riparian corridor to justify deployment ahead of the 2027 commercial K-UAM window without full retraining.
The answer is conditional. Shared EAAF flyway species — the migratory shorebirds and waterfowl that use both sites as stopover nodes — provide a statistically robust transfer anchor. Outside migration windows, Gimpo's resident riparian guild introduces species assemblages and foraging-altitude distributions that diverge materially from the Incheon training corpus. A targeted habitat-layer delta, not full retraining, is the appropriate intervention. This article defines where the generalisation holds, where it breaks, and what vertiport operators and K-UAM working-group members need to commission before a Gimpo site enters the 2027 operational licensing pipeline.
1. Operational Anchor — Gimpo Airport and the Han River Basin
The Site
Gimpo International Airport (IATA: GMP) sits at the western edge of Seoul's metropolitan fabric, bounded to the north and east by the Han River and its tidal-influenced lower reach before the river widens toward Incheon Bay. The airport handles approximately 15–17 million domestic and short-haul international passengers annually under Korea Airports Corporation (KAC) management. Its runway system runs parallel to the Han River's south bank, placing the approach and departure corridors directly above a managed floodplain and riparian reed-bed system — a structurally distinct ground envelope from Incheon's open reclaimed-tidal landscape. For vertiport planning purposes, two candidate zones are under early-stage review by MOLIT's K-UAM working group: a northern rooftop terminal expansion apron and a riverside ground-level pad adjacent to the Magok district corridor.
Environmental Read
The Han River basin supplies a predictable set of ecological variables. The riparian corridor hosts year-round wetland vegetation — reed beds (Phragmites australis), willow margins, and managed grassland levees — that provide nesting, roosting, and foraging substrate for resident heron, egret, and cormorant guilds throughout all twelve calendar months. Seasonal flooding pulse between June and August temporarily expands shallow-water foraging area, drawing additional wading species. The EAAF northbound migration window (March–May) and southbound window (August–October) overlay this resident baseline, adding shorebird and waterfowl species shared with the Incheon coast. The combined result is a strike-risk profile with two distinct regimes: a migration-period spike driven by EAAF species behaviorally identical to those at Incheon, and a non-migration baseline driven by resident riparian species with altitude and flight-path distributions the Incheon model has not been trained to characterise.
Differential Factor
The critical structural difference between Gimpo and Incheon is vertical habitat layering. Incheon Technopark's ground habitat is predominantly open tidal flat and reclaimed grassland — a low-canopy environment where bird flight occurs in open airspace that aligns cleanly with a vertiport's low-altitude threat envelope. Gimpo's riparian corridor introduces reed-canopy concealment up to four to six metres, willow canopy to twelve metres along levee margins, and river-surface foraging behaviour where birds approach vertiport airspace from below expected detection altitude in conventional coastal models. This means that detection-confidence thresholds calibrated against Incheon's open-ground prior will systematically underweight the probability of low-angle canopy-emergence events — the exact strike geometry most relevant to eVTOL rotorwash ingestion scenarios.
Modern Bridge
For vertiport operators reviewing 2027 slot applications at Gimpo, this habitat-layer divergence is not an argument against deploying AVIX-AI BirdThreat. It is an argument for deploying it with a correctly scoped habitat delta survey commissioned 12–18 months before commercial operations commence. The 4-stage pipeline is designed to ingest updated habitat classification inputs; the Incheon-validated commit (fbcb327, 2026-04-20) provides the inference architecture. The decision point is whether operators treat the Incheon validation as a turnkey transfer or as a validated baseline requiring a documented domain-shift assessment — a distinction that MOLIT's wildlife-hazard working group is expected to formalise in the 2027 operational licensing criteria.
2. Problem Definition — The Domain-Shift Gap in Riparian Vertiport Planning
The K-UAM Roadmap 2030 targets 200+ vertiports across the Korean metropolitan network, with a significant proportion sited along river corridors and urban waterfront zones that intersect the EAAF flyway pinch point. Gimpo represents the archetype of this category: high passenger-demand location, existing aerodrome infrastructure, and a riparian habitat envelope that does not match the coastal or urban-rooftop profiles that most initial K-UAM bird-strike research has addressed.
ICAO Doc 9332 requires that aerodrome wildlife hazard management programmes include a formal habitat assessment that identifies attractants within 13 km of the aerodrome reference point. Gimpo's 13 km radius encompasses the lower Han River system, the Gyyang Stream confluence, and multiple managed agricultural impoundments that serve as documented waterfowl congregation areas during the non-migration baseline period. A model trained exclusively on coastal open-ground data — however rigorously validated — cannot satisfy the Doc 9332 habitat assessment requirement for this radius without supplemental riparian species inventory data.
The quantitative gap is material. Field ornithological surveys at comparable Han River riparian corridors (Hanam, Goyang, and Gimpo-si wetland management zones) document between 34 and 47 resident bird species active within the strike-risk altitude band (0–120 m AGL) during the non-migration period. The Incheon training corpus was built against a coastal assemblage of approximately 22–28 species in the same altitude band during validated survey periods. The species overlap between the two assemblages in non-migration months is estimated at 40–55% — sufficient to provide partial generalisation but insufficient to meet the KAS Part 25 site-level airworthiness documentation standard without supplemental validation. For vertiport operators and their insurers, that gap represents a quantifiable compliance exposure heading into the 2027 commercial window.
3. UAM KoreaTech Solution — Habitat-Layer Delta Protocol
AVIX-AI BirdThreat's 4-stage habitat treatment pipeline is architecturally modular. Stage 1 (habitat classification input) and Stage 2 (species-probability prior construction) are the two stages where Gimpo-specific riparian data must be inserted. Stages 3 (real-time inference) and 4 (Lattice entity publication) are domain-agnostic once the priors are correctly specified — the Incheon validation at commit fbcb327 demonstrates the inference and publication architecture is production-stable.
The proposed Gimpo habitat-layer delta protocol involves three steps. First, a 60-day riparian survey covering the Han River corridor within the 13 km Doc 9332 radius, structured to capture resident species altitude distributions and canopy-emergence flight-path geometry during the non-migration baseline period. Second, a prior update injection into the Stage 2 probability construction, merging the new riparian guild codes with the existing EAAF shared-species priors from the Incheon corpus — preserving the migration-window inference quality while extending coverage to the resident assemblage. Third, a re-validation run against the 19/19 HTTP 200 standard at a Gimpo hardware deployment, with animal-class entities publishing into Anduril Lattice under the same audit chain established at Incheon.
The Acoustic Vibration Mat dimension is also site-specific at Gimpo. The rooftop terminal expansion apron candidate site sits above occupied office space with Korean municipal noise ordinance obligations. The mat's 90% absorption performance at 8–40 Hz, verified by accelerometer audit at installation, directly addresses the low-frequency vibration transmission risk that conventional acoustic matting fails to manage — and generates the documented KAS Part 25 installation record that rooftop vertiport permit applications require. Both products address discrete compliance gaps; neither is redundant to the other.
4. Strategic Context — Why Gimpo, Why 2027
Gimpo occupies a structurally advantaged position in the K-UAM Roadmap 2030 corridor network. Its proximity to the Magok innovation district, the Yeouido business hub, and the Seoul metropolitan rail interchange makes it a high-frequency-demand vertiport candidate under MOLIT's demand modelling. The Kakao Mobility federation — which provides the routing and booking backbone for the UAM Korea Travel app's transactional layer — has existing ground-transport density at both Gimpo Airport and the Magok catchment, creating a multimodal interlock that makes Gimpo more commercially viable per vertiport unit than many greenfield riverside sites.
The 2027 commercial window is specifically pressured by two regulatory calendars running in parallel. KAC's aerodrome development schedule for Gimpo's northern apron expansion has a permit decision gate in Q4 2026 — meaning wildlife hazard management documentation must be submission-ready by Q3 2026 to avoid slipping to the next review cycle. Simultaneously, MOLIT's K-UAM operational licensing criteria are expected to formalise wildlife-hazard audit obligations for vertiport tenants in H1 2027, creating a compliance cliff for operators who have not already commissioned site-specific assessments.
The EAAF flyway's permanence is also a strategic constant. Flyway-corridor habitat obligations under the Ramsar Convention and EAAFP partnership agreements do not expire with regulatory cycles. Any vertiport operating within the Yellow Sea key biodiversity area must maintain ongoing wildlife-hazard documentation — not a one-time permit deliverable but a continuous audit obligation. The Lattice entity-publication architecture embedded in AVIX-AI BirdThreat is the only current Korean low-altitude-airspace system designed to generate that continuous audit record natively.
5. Forward Outlook
The 12-to-24-month roadmap for Gimpo vertiport wildlife-hazard compliance aligns to three decision gates. By Q3 2026, the riparian habitat-layer delta survey should be complete and submitted alongside the KAC northern apron permit application. By Q1 2027, the AVIX-AI BirdThreat prior update and re-validation run should produce a Gimpo-specific 19/19 HTTP 200 record, satisfying MOLIT's anticipated operational licensing wildlife criteria. By Q3 2027, the Acoustic Vibration Mat installation on the rooftop apron should carry its accelerometer-audited KAS Part 25 documentation into the final commercial operations certificate package.
The broader implication is methodological. Gimpo is the first Han River riparian site to be evaluated against the Incheon-validated AVIX-AI standard — but it will not be the last. The K-UAM Roadmap 2030's 200+ vertiport target includes Yeouido, Hanam, and Goyang corridor sites that share Gimpo's riparian habitat profile. The domain-shift assessment protocol developed here becomes the template for every river-corridor vertiport entering the 2027–2030 licensing pipeline, making the Gimpo fitness study a methodological asset with network-wide value rather than a site-specific exercise.
Conclusion
Gimpo Airport's Han River basin habitat is neither a disqualifier for **AVIX-
Frequently Asked Questions
Why does the Incheon AVIX-AI model not transfer directly to Gimpo without modification?
Incheon International Airport sits on reclaimed Yeongjong tidal flats dominated by migratory shorebirds and waterfowl moving along the East Asian–Australasian Flyway (EAAF). Gimpo Airport, 18 km east, is embedded in the Han River riparian corridor — a structurally different habitat featuring reed-bed margins, inland wetlands, and a resident bird community (e.g., herons, egrets, cormorants) supplementing the same migratory pulse. The species-detection priors and seasonal confidence windows built from Incheon's coastal open-ground profile underweight reed-canopy foragers and river-surface feeders that represent Gimpo's dominant strike risk vectors. A habitat-layer delta file covering riparian guild codes and resident species phenology must be appended before the 4-stage AVIX-AI BirdThreat pipeline can achieve the same 19/19 HTTP 200 validation standard demonstrated at Incheon Technopark. This is not a model failure; it is an expected domain-shift artefact requiring controlled transfer learning rather than wholesale retraining.
What EAAF species do Incheon and Gimpo share, and how does that commonality anchor the generalisation case?
Both sites sit within the EAAF's Yellow Sea key biodiversity pinch point, where species such as Far Eastern Curlew (Numenius madagascariensis), Black-faced Spoonbill (Platalea minor), and multiple Charadrius plover species make stopovers during the northbound (March–May) and southbound (August–October) migration windows. Because the detection logic for these EAAF-listed species is already encoded in the Incheon-trained model — including their altitude-band preferences during approach and their response to acoustic deterrents — this shared phenology provides a statistically robust transfer anchor. The critical divergence point is outside the migration windows, where Gimpo's year-round riparian resident community drives a baseline strike-risk floor that Incheon's coastal open-ground baseline does not represent. Operators planning 2027 commercial vertiport slots at Gimpo can use the Incheon model's EAAF inference layer directly but must commission a Han River riparian resident survey to complete the habitat profile.
What regulatory obligations govern bird-strike risk assessment at a K-UAM vertiport site like Gimpo Airport?
Korean Aviation Safety (KAS) Part 25 establishes airworthiness standards for transport-category aircraft, including wildlife-strike resilience specifications that vertiport operators must satisfy at the site level. ICAO Doc 9332 (Wildlife Control and Reduction) provides the international framework for aerodrome wildlife hazard management, requiring habitat assessments, species inventories, and documented deterrence programmes. At Gimpo, the Korea Airports Corporation (KAC) is the aerodrome operator accountable for wildlife hazard management under Civil Aviation Act Article 68. For K-UAM vertiport slots embedded within or adjacent to licensed aerodromes, MOLIT's K-UAM Roadmap 2030 working group has flagged that UAM operators share wildlife-hazard obligation proportionally with the host aerodrome operator. This creates a compliance gap for vertiport tenants who assume KAC coverage absorbs their obligation — a gap that a site-specific AVIX-AI BirdThreat deployment, publishing animal-class entities into Anduril Lattice for audit, directly closes.
References
- ICAO Doc 9332 — Manual on the ICAO Bird Strike Information System (IBIS)(2012)
- East Asian–Australasian Flyway Partnership — Key Biodiversity Areas(2024)
- Korean Aviation Safety Regulations (KAS) — Ministry of Land, Infrastructure and Transport(2023)
- Korea Airports Corporation — Gimpo International Airport Operations Overview(2024)
- MOLIT K-UAM Roadmap 2030 (국토교통부 한국형 도심항공교통 로드맵)(2023)
- Anduril Industries — Lattice Platform Overview(2024)
- EAAFP — Far Eastern Curlew Species Profile(2023)