Site Suitability & Coherence Pre-Checker

Before anyone spends a week on orbit selection, baseline picking and atmospheric correction, someone should ask the cheaper question: is this site even measurable with InSAR? A significant share of failed InSAR projects fails for reasons that were visible on day one — a rainforest canopy that decorrelates C-band signal within days, a steep quarry face that folds into layover, a wetland with no coherent surface at all. This pre-checker encodes those day-one checks into a transparent scoring model: describe the site with four parameters and a bounding box, and it returns a 0–100 suitability score, a four-dimension sub-score profile, explicit red flags where the physics is against you, and concrete recommendations for algorithm, sensor and track selection — including a ready-to-paste HyP3 job package for the ASF processing service.

The model is deliberately a heuristic. It does not ingest actual coherence data, weather reanalysis or DEM tiles; it encodes the expected behaviour of radar coherence and interferometric geometry as published in the literature, with every weight visible and every threshold explained. That transparency is the point: a screening score you can take apart is worth more in a review meeting than a black-box number you have to defend blindly.

What the four sub-scores mean

Expected coherence (weight 0.40) starts from the land-cover class — urban fabric and bare rock hold coherent scatterers for years, cropland decorrelates with every tillage pass, forest canopy is close to opaque for C-band — and applies an L-band penetration bonus for vegetation plus a penalty for high local relief. Geometry safety (0.30) compares the maximum terrain slope in your area against the sensor's incidence angle: radar cannot see slopes that face away from the beam, and slopes steeper than roughly the incidence angle fold into layover or shadow, so the sub-score decays quickly as the slope closes on that limit and triggers an explicit warning. Atmospheric quietness (0.20) reflects the tropospheric delay burden — arid stable air is nearly free, tropical humid air with strong convection over relief is the worst case InSAR encounters. Target fit (0.10) matches the processing strategy to what you are hunting: slow steady subsidence suits PS/SBAS time series, rapid landslides need per-pair DInSAR, seasonal signals want dense time series above all.

When you should use this tool

• At proposal stage, to justify (or talk a client out of) an InSAR monitoring component with a documented, reproducible rationale.
• Before committing processing effort, to choose between Sentinel-1 and L-band data and to decide whether PS, SBAS or DS-InSAR is the realistic primary strategy.
• To document due diligence — the Markdown report records exactly what was assumed and what was flagged, which is what auditors and reviewers ask for later.

When NOT to use this tool

• As a substitute for a formal feasibility study on a high-stakes project. This is a pre-check, not a study; the score is a screening indicator, not a guarantee in either direction.
• On sites where aspect distribution matters more than maximum slope — layover risk depends on the slopes facing the track, and a single max-slope figure is a crude proxy. Check both tracks' local incidence maps before final conclusions.
• For interferometric details this model does not represent: ionospheric effects, vegetation seasonality, snow epochs, or the availability and licensing of L-band archives in your region.

Reading the output & the red flags

The class labels map to ranges: 75 and above is Good, 50–74 Moderate, below 50 Poor. Red flags are the more useful output in practice: they state the physical mechanism and the standard countermeasure, e.g. dense forest with C-band triggers a volume-decorrelation warning with the L-band alternative named; water and ice trigger the fundamental no-coherent-surface warnings; steep terrain triggers the dual-track instruction. A Poor score with a fixable mechanism (wrong band, missing correction) is a very different situation from a Poor score with a physical one — read the flags, not just the number.

Common pitfalls & mistakes

• Score worship. An 88 does not mean the project will succeed; a 40 does not mean it will fail. The score ranks sites against the model's assumptions, which are stated and simplified.
• Bounding box sprawl. A box twenty times the project area averages away the very slope and land-cover heterogeneity the model needs. Draw the box around the assets.
• Ignoring aspect. Two identical max-slope sites can have opposite layover exposure depending on which way the slopes face relative to the track. The tool warns; it cannot see the aspect distribution from a box alone.
• Forgetting mission reality. L-band recommendations assume you can actually obtain ALOS data — tasking costs and archive access are real constraints the model does not price in.

Step-by-step usage

1. Define the bounding box around the assets, not around the administrative area. Type it as W, S, E, N degrees, or paste any GeoJSON geometry — a project polygon, a corridor line, a client KML — and the tool extracts the enclosing box automatically.
2. Classify the dominant land cover honestly. Mixed sites should be scored on the class that covers the measurement targets, not the class that covers the scenery; run the tool twice with both classes if the split is significant.
3. Enter the maximum terrain slope from your DEM (a quick relief figure from any topographic map is acceptable at screening level) and the climate class that best matches the site's precipitation regime.
4. Choose the sensor band you actually intend to process — not the one you wish you had — and the deformation target that matches the engineering question.
5. Run the pre-check. Read the flags first, then the class, then the recommendations; copy the HyP3 package if processing will proceed, and download the Markdown report for the project file either way — a documented negative pre-check is as valuable as a positive one.

Frequently asked questions

Why does the model not download real coherence maps or land-cover rasters?

Because the site's privacy commitment is absolute: no inputs leave your browser, so the tool cannot fetch anything either. The trade-off is transparency instead of data volume.

How sensitive is the total to the weights?

The coherence term dominates at 0.40, so land cover drives most verdicts. If your project context differs, recompute the arithmetic by hand from the sub-scores — they are all in the report.

What does the HyP3 package actually contain?

The bounding box, track plan, DEM source and atmospheric-correction suggestion in a form you can adapt into the ASF HyP3 SDK or Vertex batch interface. Always re-check the schema against your account version.

Can I use this for water-adjacent assets like ports?

Yes — with the water red flag in mind. Ports contain abundant coherent structures (quay walls, cranes, warehouses); the open water itself is unmeasurable, so constrain the box to the structures.

Why is tropical forest with L-band not automatically Good?

L-band penetrates canopy far better, but atmospheric burden, access constraints and pixel size still bite — the model reflects that with points recovered rather than points ignored.

Related tools

Once the site passes the pre-check: LOS Converter for the velocity conversion, Hazard Classifier for the spatial screening, GNSS Validation for ground truth, and the Error Budget calculator to formalise the uncertainty statement.