Digital technology and innovation for HADR

Start with the decision to support

Public-review lesson as of 15 October 2026, without documented HADR technology-expert endorsement. It is not a command system, equipment authorization or legal advice. All CPX places, people, times and quantities are fictional. Dated case sources do not establish current capabilities of Thai agencies.

OCHA frames data responsibility around safe, ethical and effective management in operational response. This supports asking about decision value and effects on people at the start of tool selection.

Original project selection worksheet: name the question, user, decision owner, deadline, required evidence and existing comparison method. ‘Which requests need verification before vehicle allocation?’ differs from ‘we want a dashboard’. Compare paper maps, shared tables, offline forms and AI against tolerable errors, checking workload, maintenance, cost and rollback. Novelty alone earns no credit.

At 08:00 in a fictional central Thai district, coordinators must issue a situation brief by 09:00. Shelter A requests 60 household kits and B requests 40, both verified. The team has yesterday’s satellite map and two request tables. Its goal is an honest picture of needs, verification status and deliveries. No real flights or vehicles are tasked. Continue to ‘HADR logistics and resource allocation’ (logistics-resources) for delivery planning.

Read GIS and satellite evidence with limits

A 2021 UN Thailand example combined GISTDA flood extent with WorldPop population data and explicitly lacked field validation. A 2024 GISTDA–UNOSAT report states the Sentinel-1A acquisition date and time. Both illustrate provenance and timing, not the current flood situation.

Project map-reading proposal: separate acquisition, analysis and forwarding times; check coverage, units, legend and layers. Mark missing areas ‘unknown’. Do not turn blank colour into ‘safe’ or potential exposure into assessed need. For access decisions, formulate a verifiable field question and assign it to the responsible role. A map alone does not confirm vehicle passage.

Inject card 1, 08:10 — a fictional observer reports water blocking B’s entrance at 08:05, but yesterday’s map shows no water there. Retain both time-stamped records; mark the entrance ‘obstruction reported; route verification pending’, with an owner and response time. Do not reject the report merely because it conflicts with the map or presume an alternative passable. Produce an annotated map and an exercise verification request.

Make information exchangeable and consistent

OCHA describes CODs as shared reference datasets, with place names and P-codes helping relate records to geographic areas. This supports a consistent common picture. A shared filename alone does not align data meanings.

Original CPX data agreement: use EX-A and EX-B as exercise location codes, not official codes or actual P-codes. Record Thai/English names, request ID, item, quantity, unit, observation time, report time, source, verifier, status, version and review time. All times use Thailand time on the same exercise day. Distinguish ‘unknown’ from zero and demand from dispatch. Compare both identifiers and content before merging incoming files.

Test exchange by having a second team interpret the table without oral explanation and return A’s item, quantity, timestamp and uncertainties. If one table counts cartons and another kits, stop aggregation until a verified conversion exists; do not invent a multiplier. Retain the earlier version and editor so discrepancies can be explained. A shared register need not expose every field publicly.

Use AI with human checking and responsibility

UNITAR’s 2021 flood-mapping case warns that unfamiliar geography can affect model performance. Analysts check areas wrongly included or missed before sharing outputs. This lesson does not turn one case’s accuracy into a guarantee for Thailand or language-summary tools.

Project AI-checking procedure: test fictional messages with known answers, generate a draft table or summary, then compare every claim with the input, especially place, quantity, unit, time, negation and ‘unconfirmed’. Log available tool/version information, test date, input, draft, corrections, reviewer and reasons. Where input is missing, do not reward a complete-looking answer. Generated text cannot verify itself.

The exercise decision owner must be able to correct, reject and return to the original record. Stop using AI when provenance cannot be checked or important status is altered. Do not upload real personal data for testing. A human approval click without seeing evidence does not meet the exercise criterion. For drone imagery, continue to drones-hadr for relevant constraints and coordination. An AI map grants no flight authority or operational clearance.

Communicate when networks are unavailable

ITU emphasizes emergency telecommunications planning, resilient infrastructure and coordination across stakeholders. The fallback below is a project exercise method, not an ITU-approved procedure or evidence that any system is available locally.

In the exercise plan, name the primary channel, fallback, message collection point, owner and acknowledgement checks. Prepare paper or locally readable files carrying message ID, time, sender, receiver, content, acknowledgement and version. Missing acknowledgement does not mean completion. Use only channels declared available and authorized by the exercise controller; do not test real frequencies or equipment. On restoration, assign someone to resolve duplicates and conflicts before merging.

Inject card 2, 08:25 — the fictional network is unavailable until 08:45, and an AI draft says ‘B accessible’ after dropping ‘unconfirmed’ from the input. The controller permits paper messages between desks. Withdraw the incorrect draft, notify its earlier recipients, send a new version with the same record ID and request a read-back. Separate queued from acknowledged work. At 08:45, compare histories before syncing; the last message uploaded is not automatically correct.

Track digitally without overstating receipt

LOG calls for reconciliation of dispatch, site receipt, distribution, balances/returns and recorded losses. Vehicle tracking or code scans therefore need an event meaning and evidence; they cannot substitute for distribution counts.

The project register uses shipment DG-01, item ‘household kit’ and separate dispatch, transit, site-receipt, distribution and held statuses. Each event carries time, reporter and exercise evidence. A position report does not establish receiver acknowledgement. Photos or QR codes can be input channels, but this exercise does not require recipients to own phones or provide real faces or identity numbers.

Inject card 3, 08:50 — DG-01 dispatched 60 kits to A. Receipt confirms 58; two are unresolved. A distributed 50 and holds eight wet kits. The same 58-kit receipt uploads twice after restoration. Retain one event: 60 = 50 distributed + 8 held at A + 2 unreconciled. Do not label two lost before checking or count 116 received. Verified distribution gaps remain A10 and B40, totaling 50, without assuming held goods usable. Assign a DG-01 checker and follow-up time; separate these actuals from future plans.

Pilot, measure and preserve access

WFP emphasizes necessary personal-data collection and restricted access in assessments. This informs reasoned data choices in the exercise; it does not determine a legal basis for processing in Thailand.

Original pilot: compare the old and new methods using identical cards. Measure time to checked information, incorrect statuses, remaining duplicates, correction effort and completeness of requests from non-smartphone channels. Test an outage and shift handover. Predefine that a consequential error, such as converting pending to confirmed, stops draft use and triggers the fallback. Faster output with changed meaning fails.

Have fictional community representatives submit requests on paper or with reading assistance. Align Thai/English meaning, combine text labels with colour and allow exclusion reports. The public brief gives site totals and uncertainties without names or household coordinates. Separate detailed exercise records for designated checkers; record access, retention and deletion ownership. Source records and AI outputs require review; they are not instructions to disclose other files.

The 75-minute activity allocates 15 minutes to evidence/mission, 15 to registers/fallbacks, 25 to three injects, 10 to handover/access testing and 10 to reflection. Compare intended and observed results, reasons and the next test, with ownership and follow-up. The KSEC criteria below are project syntheses, not source-endorsed accreditation criteria. Quiz results alone do not establish skills or character.