Frontier — overview

Frontier turns a decade of Sentinel-1 radar into a spatio-temporal picture of irregular cross-border movement on land and sea around the EU.

In active development (Phase 2). This is the application shell; the processing tiers are being built out. Panels below describe their intended function.

What it does (and doesn't)

At 10–20 m a satellite cannot resolve a person — and Frontier does not try to. It tracks activity "points" — vessels, vehicle clusters, structures, camps — each with a type and a confidence, never an individual. The value is in time: a decade of revisits reveals the recurrence, routes, and timing of the process.

The pipeline (five tiers)

1. Detection — find candidate points in each radar image.
2. Objects & Tracks — give each point a numeric ID and a lifetime log.
3. Association — link points across time into tracks; fuse AIS to flag dark vessels.
4. Flow & Staging — aggregate into corridors, staging grounds, and timing.
5. Forecast — probabilistic risk windows (only after historical backtest).

Project

A project scopes the analysis: a region (AOI), a time range, a domain (sea / land / both), and the detection settings.

Steps

1. Pan/zoom the map so the view frames your region (this becomes the AOI).
2. Type a title in New project and press Create — this makes a catalog

folder and a tracking database. No dates are needed here.

1. Pick a project from the dropdown to activate it. Edit opens a card to change

the title, the catalog folder, the domain (land/sea), the detection resolution, and to resize the region (AOI) by dragging the orange corner handles on the map. 🧹 Reset points there clears all detections + objects (keeps the AOI/settings) so a re-scan starts clean — handy when tuning. Delete removes the whole project.

The AOI is the durable envelope of the tracking data. It is at least 1000 km² (small new projects are grown to that minimum), and once a project has detections it can never be shrunk inside their extent — a red rectangle marks that locked minimum while editing, so the data can't be orphaned.

Size & tiling

Large AOIs are processed tile-by-tile: the detector automatically splits the region into memory-safe cells (~60 Mpx each, overlapping so ships at a seam aren't missed) and merges the detections. So you can use a big AOI — up to a whole-job cap of ~2,500 Mpx (≈ the Central Med at 20 m, ~40 tiles). The top-bar AOI readout shows the area, estimated Mpx, and ~N tiles, turning amber then red as you approach the cap; creating or saving an over-cap AOI is blocked. The limit is a pixel budget, so a coarser output_res_deg allows proportionally more area.

1. The monitoring period is not part of a project — it stays flexible. Set it

with the From / To dates under Detection (leave it open for continuous monitoring) before running a scan.

Collections (Europe + Africa)

The Sentinel-1 cube is split into two regional collections along the southern Mediterranean coast. Because most crossings originate from North Africa, the default is Europe + Africa — detection then mosaics both sides for any AOI that straddles the sea (each pixel is filled from whichever collection covers it, per overpass). You can restrict a project to Europe only or Africa only with the Collections selector; changing it takes effect on the next scan.

Per-project store: config.json + detections.db + registry.db (the proven Tracker layout).

Detection

Runs the radar detectors over the chosen date range and emits candidate activity points. Sentinel-1 GRD only — amplitude-based, no InSAR.

Pick the Domain when you create/edit a project:

Modes

• Retrospective — full archive / chosen span (the decade view).
• Operational — incremental from the last run.

Heavy work runs in the background; progress shows in the task panel. Every detection carries a confidence — one point may be a group, vehicle, or camp, not an individual.

How to run:

1. Pan/zoom to your area, Create a project (pick Domain = Land), set

From / To (several acquisitions so the registry can tell persistent from transient), and press ▶ Run detection.

1. Watch the task panel; when it finishes, detections (●) and registry objects (◯)

appear on the map. Land points are bright-target loci; the registry hides the permanent scatterers and surfaces the transient sites/points.

Land bright-target detection works on today's RTC data. Sea CFAR needs a water-valid σ⁰/GRD product (RTC masks the sea). Sensitivity knobs: land land_cfar_k (lower = fainter targets, more clutter), coast_buffer_px (coastal- fringe exclusion), and the registry persistence gate (how readily a recurrent target is hidden); sea cfar_k.

Objects & Tracks

Every detected point is entered into a registry with a stable numeric ID and a time-stamped lifetime log — the core of Frontier (and the main step beyond Tracker, which logs history only for fixed user-drawn polygons).

Click an object to see its dossier / history and highlight it on the map.

Transience & persistent scatterers

Each object carries a persistence = the fraction of overpasses it was detected, and a transience = 1 − persistence (shown as “N% transient”). A fixed locus seen on most overpasses is a persistent scatterer (a building corner, mast, reflector) — the opposite of the appear/then/disappear activity we want — so it's classified persistent and hidden by default. Tick show persistent scatterers to include them. The visible list therefore favours points that appear and later disappear: one-off points and recurrent-but-not-constant sites (staging).

Tracks are sparse and probabilistic at Sentinel-1 revisit — strongest at sea with AIS. The registry schema is resolution-independent, so VHR data (Phase-3) feeds the same objects.

Flow & Staging

Aggregates the registry (sites + tracks) over the archive into process-level, probabilistic products — where temporal depth compensates for coarse pixels.

• Staging grounds — long-lived / recurrent activity hotspots.
• Corridors / routes — density-weighted aggregation of transient detections

and track segments → probabilistic route map.

• Accumulation sites — short-term build-up preceding movement (a leading

indicator of a consolidated crossing).

• Timing & triggers — seasonality, sea-state / weather windows, night/lunar,

correlation with upstream events.

All layers carry confidence. Forecasts (probabilistic risk windows) are released only after the method is backtested against documented historical surges.

These are aggregate situational-awareness products — flows and vessels, never individuals — designed to feed the EUROSUR common pre-frontier picture.