What should a terrain risk report include?

Data provenance and dates, CRS and scale, an explicit statement of whether it maps susceptibility, hazard, or risk, the method and weights, a validation result, an uncertainty and limitations section, confidence notes per finding, and clear review boundaries.

How do you communicate uncertainty in a risk report?

State the input data accuracy and age, the model's validation performance (e.g. AUC), the scale at which findings are valid, and where confidence is low. Avoid single hard lines where the data does not justify them; use graded classes and explicit caveats.

How do you separate observed hazards from modelled susceptibility?

Keep them as distinct layers with distinct legends. An observed, dated landslide or flood extent is evidence; a modelled susceptibility class is a prediction. Never merge them into one symbol or the reader cannot tell fact from forecast.

When should a terrain risk screen be escalated to specialist review?

Whenever the output informs safety, design, permitting, insurance, or public communication, or when a finding sits near a decision threshold. A GIS screen identifies candidates; engineering and regulatory decisions require qualified professional assessment.

Terrain Risk Report Checklist

Why the report, not the map, is the deliverable

A terrain risk map without a report is a coloured raster that an outside reader cannot weigh, audit, or act on safely. The report is where the map becomes defensible: it states what was mapped, from what data, by what method, how well it was validated, and how far it can be trusted. This checklist sets out every section a terrain risk report should contain, with the specific content each must capture. It applies whether the underlying product is a flood screen, a landslide susceptibility surface, or a multi-hazard constraints overlay, and it is built around one principle: a reader should never have to guess what a finding means or how confident to be in it.

1. Scope and the susceptibility / hazard / risk statement

Open by naming exactly what the report delivers. The most common and most damaging ambiguity is conflating susceptibility (spatial propensity), hazard (propensity plus magnitude and probability), and risk (hazard combined with exposure and vulnerability). State which one this is in plain language on the first page, because each licenses very different conclusions. A susceptibility screen cannot tell a client what a flood will cost; saying so up front prevents misuse downstream.

2. Data provenance

List every input with the attributes that determine its reliability:

Source and product name (e.g. Copernicus GLO-30, FEMA FIRM panel number, national geology sheet).
Date / vintage — hazard data ages; a decade-old land-cover layer may misstate current exposure.
Resolution and source scale — a 1:50,000 layer cannot adjudicate a parcel boundary.
CRS as an EPSG code and, for elevation, the vertical datum.
Licence / authority — and whether the source is the governing regulatory product or a modelled approximation.

A short provenance table near the front lets a reviewer judge the whole report's foundations at a glance.

3. Method and parameters

Describe how findings were produced so they could be reproduced:

DEM derivatives and their settings (slope algorithm and units, cell size).
For composite products, the conditioning factors, their reclassification breakpoints, and the weights with their justification (and the AHP consistency ratio if used).
Software and key commands or model files (gdaldem, QGIS Raster calculator, GRASS modules).

Hidden weights and undocumented breakpoints are the single biggest reason a risk report cannot be defended when challenged.

4. Observed events versus models — kept separate

Maintain a strict line between evidence and prediction:

Observed, dated events — mapped landslides, recorded flood extents, an erosion inventory — are evidence and should carry their date and source.
Modelled susceptibility or hazard classes are predictions and must be symbolised and labelled distinctly.

Merging the two into one symbol destroys the reader's ability to separate what happened from what might. Where an inventory exists, present it both as a layer and as the basis for validation.

5. Validation

State how the product was tested. For a susceptibility model, report the validation against a held-out event inventory — typically a success-rate / ROC curve and the AUC (around 0.5 is chance; approaching and above 0.8 is useful). For a flood layer, note whether it reconciles with recorded high-water marks. If the product is not validated, say that explicitly and treat its findings as provisional.

6. Uncertainty and limitations

Be specific rather than ritual:

Which inputs are weakest (age, resolution, coverage gaps)?
At what scale are findings valid, and where do they break down?
Where is confidence low, and why (sparse data, model extrapolation, DSM-vs-DTM ambiguity)?

Prefer graded classes and explicit caveats over crisp lines the data cannot support. A hard boundary drawn from coarse data implies a precision that does not exist.

7. Confidence notes per finding

Attach a short confidence statement to each significant finding — high / moderate / low with a one-line reason. "High-susceptibility zone on the north slope (high confidence: steep slope confirmed in lidar and two inventoried failures)" tells a decision-maker far more than a colour alone.

8. Review boundaries and escalation

State plainly what the screen can and cannot decide, and when it must be escalated. A GIS terrain screen identifies candidate issues; it does not replace geotechnical, hydraulic, regulatory, or environmental assessment by a qualified professional. Specify the triggers for escalation: any finding that informs safety, engineering design, permitting, insurance, or public communication, and any finding sitting close to a decision threshold.

9. Reproducibility appendix

Include the input checksums, the exact commands or model, and the full parameter set, so a third party can regenerate the analysis from the same sources. A report that cannot be reproduced cannot be independently verified.

Worked example: applying the checklist to a flood screen

Scope: state it is a regulatory-flood hazard screen, not a damage/risk estimate.
Provenance: FEMA FIRM panel, effective date, plus GLO-30 (EPSG:32617, EGM2008) for terrain context.
Method: floodway and Special Flood Hazard Area extracted directly; no modelling, so weights are not applicable — say so.
Evidence vs model: FIRM zones are regulatory observations; any added DEM-derived low-lying flag is labelled as a separate modelled hint.
Validation: reconcile against any recorded flood events; note agreement or gaps.
Uncertainty: FIRM effective date and panel scale; areas of recent development not yet remapped.
Confidence: high inside mapped zones, lower for the DEM-derived hints.
Review boundary: elevation certificates and engineering required for any build decision.

Common pitfalls and why they happen

"Risk map" on a susceptibility product — overstates conclusions and invites misuse.
No provenance table — readers cannot judge whether inputs are fit for purpose.
Evidence and model fused — the reader cannot separate fact from forecast.
No validation, no AUC — the model's weights are unaccountable.
Crisp boundaries from coarse data — false precision that a court or regulator will expose.
No escalation triggers — a screen gets used as a final assessment by default.

QA before issue

The susceptibility/hazard/risk wording is consistent throughout, including the legend.
Every layer's source, date, scale, and CRS appear in the provenance table.
Validation result is reported, or its absence stated.
Each key finding carries a confidence note.
Escalation triggers and review boundaries are explicit.
The reproducibility appendix lets the analysis be rerun.

Bathyl perspective

We treat the report as the product and the map as one figure within it. Every finding traces to a dated source, carries a confidence note, and names the boundary at which a specialist must take over. That is what lets a client act on a terrain risk screen quickly and still defend the decision later.