DECEMBER RESEARCH DEMONSTRATION

From rainfall to evidence

Compare sources. Inspect coverage. Keep uncertainty visible.

COMMON WINDOW · PHTSame interval and geographic support for both sources
Research evidence, not a loss estimate. Thresholds are illustrative; observed, missing and scenario inputs stay separate.
IMERG EarlyProvincial mean rainfall
GSMaP NRTProvincial mean rainfall
Source difference
Observed flood support

SPATIAL COMPARISON

Two products, one shared scale

DURATION COMPARISON

Rainfall across windows

Different durations can select different periods. These are exploratory pooled maxima, not independent events.

EVIDENCE READINESS

What this event can support

All products are retrospective. Availability immediately after the historical event has not been established.

ILLUSTRATIVE RULES

Inspect source-sensitive decisions

Comparison uses ≥. OR may be exceeded despite an unknown flood branch. AND may be not exceeded despite an unknown branch. Both retain a review requirement.

EVIDENCE OR SCENARIO

Keep assumptions explicit

Stage is recorded as context only. No historical stage is inferred and no unvalidated damage multiplier is applied. This package does not yet test stage-conditioned triggers.

Two sources, transparent results

No combination authorizes payment. The flood denominator is observed geographic area, not insured rice. Threshold changes are sensitivity scenarios, not evidence of improved protection.

LOCAL PREVIEW ONLY

Bring an insured boundary into the review

Upload longitude/latitude GeoJSON polygons. Files remain in this browser session and are not uploaded to the hosting server. Preview does not establish historical insurance, rice presence, valid topology or rainfall/flood overlap.

No parcels loaded. No actual farms assessed.

Next qualification requires event-year insurance and crop evidence, sowing/transplanting date semantics and spatial overlay review. Imported attributes are unverified. No person names are required.

EVIDENCE REVIEW

Ask about the current event

The offline guide uses explicit rules. It is not a language model or an RH3 performance result.

Questions are processed in your browser by the rule-based guide. Nothing is sent to an AI service. No shared database or server-side parcel storage is enabled.

TRACEABLE RESPONSE

Answer and supporting evidence

Select a question to begin.

The assistant cannot change measurements, thresholds or trigger calculations. AI explanations still require review; software safeguards are not a guarantee of factual accuracy.

ADVISER DISCUSSION

Adviser review · version 30 October 2026

The contribution under development is a flood-centred WIBI research application that reveals how rainfall sources, observation gaps and exposure definitions affect the evidence and illustrative decisions. Current results demonstrate source sensitivity, not validated crop-loss prediction.

Five minute demonstration

  1. Compare Egay's 24-hour provincial rainfall and paired maps.
  2. Use a 150 mm rainfall threshold to expose source disagreement.
  3. Switch to flood only and inspect why partial coverage is unassessable.
  4. Load a synthetic flood scenario; contrast OR and AND rules.
  5. Inspect parcel gaps and ask the assistant whether either source is accurate.
  6. Export the review record, including assumptions and source identifiers.

Download Word adviser brief · Read PDF brief

Remaining scientific decisions

  • Approve the amendment from depth/loss calibration to the feasible evidence and sensitivity scope.
  • Secure paired rice masks and event-year parcel/crop records or revise RH1 explicitly.
  • Prespecify untouched evaluation episodes for RH2; three inspected cases do not create confirmatory power.
  • Freeze a model and matched comparison design for RH3; do not score development fixtures as held-out evidence.

Evidence register

This is a fixed research snapshot; new R runs are not imported automatically. The package preserves original source paths and SHA-256 fingerprints. Research sources were not modified.