ReliefWeb (OCHA): Intelligence Source Guide

ReliefWeb is OCHA's curated archive of humanitarian reporting: situation reports, assessments, appeals, maps and analysis from thousands of organisations, indexed by disaster, country, theme and source, with a full public API. It is where a crisis is described rather than counted.

At a glance

Source	ReliefWeb (OCHA)
Category	Conflict, Crime & Human Security › Conflict & Event Databases
Homepage	https://reliefweb.int/
Machine interface	`https://api.reliefweb.int/v1/disasters`
Format	`JSON`
Access	Open — no account required
Disciplines	Open Source Intelligence, News Intelligence, Geospatial Intelligence
Mission domains	Conflict & Humanitarian

Humanitarian situation & disaster feed. — as catalogued in the platform’s own source registry.

ReliefWeb has been running since the mid-1990s and is the humanitarian sector's institutional memory. It is a curated collection rather than an automated aggregator: a team of editors reviews material from several thousand organisations – UN agencies, international and national NGOs, governments, donors, research institutes and selected media – and publishes what meets its humanitarian relevance criteria, with structured metadata attached. The content is documents. Situation reports and flash updates from the field, needs assessments, appeals and funding documents, evaluations and lessons-learned studies, analytical pieces, press releases, and a substantial body of maps and infographics. Alongside the document collection sit structured entities: a disaster registry where each event has an identifier, a type, affected countries, a status and a date; a country registry; a source registry describing the publishing organisations; and vocabularies for themes, content formats and vulnerable groups. All of this is exposed through a public API that accepts structured queries with field selection, filtering, faceting and full-text search, and returns JSON. Requests are expected to identify the calling application by name. There is no key and no charge, and the archive goes back decades, which is unusual and valuable.

Numbers tell you the scale of a crisis and documents tell you what happened, who says so, and what they were trying to achieve by saying it. That is the job ReliefWeb does. A displacement figure in a data catalogue is a value; the situation report it came from names the assessment method, the areas that could not be reached, the date the count was taken and the organisation that took it. For an analyst that context is often more important than the figure. The disaster registry gives you something else again: a canonical event spine, so that reporting scattered across hundreds of documents and dozens of organisations can be attached to a single identified event with a start date, a type and an affected-country list. That is the backbone of any crisis timeline. The source registry does the complementary job of making the reporting ecosystem legible – who publishes about a country, in what volume, with what mandate – which is how you assess whether apparent consensus reflects independent observation or a single upstream claim repeated. For OSINT and NEWSINT work on conflict and disaster, this is the corpus that supplies dated, attributable, organisationally-sourced narrative at a level of curation that general news aggregation cannot match.

Who publishes it, and why that matters

ReliefWeb is run by OCHA as a service to the humanitarian community and funded through the UN system rather than by advertising, subscription or data sales. That gives it two properties worth planning around. It is durable – it has operated continuously for decades and its archive has been maintained through multiple technology generations – and it is editorially shaped by its mandate. Editors select for humanitarian relevance and operational usefulness, and they exclude material that is primarily political advocacy or that falls outside the humanitarian frame. This is not censorship in any meaningful sense, but it is a filter with a direction: the corpus describes crises as the humanitarian system understands them, in the vocabulary the system uses, from organisations that participate in it. Local and national civil society organisations that do not publish in English or do not engage with international coordination structures are underrepresented relative to their actual role. The API is a public good with no commercial tier, and the corresponding expectation is that consumers identify themselves and behave considerately. Product changes do occur as the service modernises, and the documentation site is the authoritative statement rather than anything cached.

Provenance is the first question to ask of any dataset and the one most often skipped. Who collects it, what their incentive is, whether they publish a methodology, and whether they correct the record when they get something wrong all bear directly on how much weight a finding drawn from it can carry.

What a record actually contains

The fields you will be working with, what each one means, and whether it is something you can pivot on. Read the meanings carefully — more analysis is wrecked by misreading a field than by failing to find one, and a field that looks like an observation is often an inference.

Field	Type	What it means	Pivot value
`id`	int	Stable numeric identifier for a report, disaster, source or other entity. This is what you key on; titles and URLs are less stable and organisational names change.	Direct entity retrieval, deduplication across collection runs, linking documents to disasters.
`title`	string	The document title as published. Frequently formulaic – the organisation's own sitrep naming convention with a number and a date – which makes it useful for series detection and poor for topic classification.	Sitrep series reconstruction for one organisation and one crisis; sequence gap detection.
`body`	string	The document text where available. This is the analytical substance and it is what full-text search runs against. Availability varies, and map and infographic entries often carry only a caption.	Named entities, place names, figures, dates and organisational references – all requiring extraction and verification.
`date.created / date.original`	timestamp	When the record was created on the service and when the source document was originally published. These differ, sometimes by days, and using the wrong one distorts every timeline you build.	Reporting lag analysis, event timeline construction, detection of retrospective publication.
`source`	array	The publishing organisation or organisations, drawn from the controlled source registry with identifiers, types and homepages. The most important metadata field in the record for assessing what a document is.	Source registry entry, organisational mandate and funding, all other reporting by the same body.
`primary_country`	string	The country the document principally concerns, with ISO codes and geographic coordinates attached. Regional and multi-country documents also carry a broader country list, and the two are not interchangeable.	Country dashboards, cross-referencing with datasets for the same location, regional aggregation.
`disaster`	array	The registered disaster event or events the document is attached to. This is the field that turns scattered reporting into an event-centred corpus and it is the most analytically valuable link in the schema.	The complete document set for one event, disaster registry metadata, the event's timeline and status.
`disaster_type`	enum	Controlled vocabulary of hazard types – flood, drought, earthquake, tropical cyclone, epidemic, conflict-related categories and others. Multiple types can apply to one event, which is common in compound crises.	Hazard-class filtering, comparison of response patterns across similar events, historical analogue identification.
`theme`	array	Controlled thematic vocabulary covering protection, health, food and nutrition, shelter, water and sanitation, coordination and others. Assigned editorially, so consistency is good by the standards of document collections.	Sectoral filtering, cluster-aligned analysis, tracking which sectors are reported on and which are silent.
`format`	enum	Content format – situation report, assessment, appeal, evaluation, news, map, infographic, analysis. This is the field that tells you what kind of claim a document is making and it should drive how you weight it.	Evidence-type stratification; separating primary field reporting from press releases and secondary analysis.
`glide`	string	The Global Identifier Number for a disaster event where assigned, a cross-system identifier used by several humanitarian information systems.	Linking the same event across independent humanitarian databases without matching on name or date.
`status`	enum	For disaster entities, whether the event is alerting, ongoing or past. This is an editorial judgement about the humanitarian response phase, not a physical statement about whether the hazard has ended.	Response phase filtering; but never cite as evidence that a situation has resolved.
`language`	array	Document language. The collection is predominantly English with meaningful French, Spanish and Arabic content, which is a real constraint on what a text-based analysis will see.	Language coverage assessment; identification of the non-English reporting your pipeline is missing.
`file`	array	Attached files, typically PDFs of the original document and images for maps. The attachment often contains data and detail that the indexed body text does not.	Full document retrieval, map and figure extraction, tabular data embedded in reports.

Coverage — and what is not in it

Coverage is deep, long and deliberately bounded. Temporally the archive spans from the mid-1990s to the present with continuous accumulation, which makes it one of very few sources in this catalogue where a genuine multi-decade comparison is possible. Geographically it covers wherever the international humanitarian system reports, which means intensive coverage of declared emergencies and protracted crises and negligible coverage of wealthy countries and of situations no agency has framed as humanitarian. Organisationally it draws on several thousand publishing sources, weighted heavily toward UN agencies and international NGOs, with national and local organisations present but underrepresented relative to their operational role. Thematically it follows the humanitarian sector's own structure, so the sectoral vocabularies map onto the coordination clusters and the gaps are the things the sector does not treat as its business – most economic reporting, most security and political analysis, most criminal justice. Update rhythm is continuous, with editors publishing throughout the day; during a major sudden-onset emergency the volume for one country can go from a handful of documents a week to dozens a day, and that volume change is itself an indicator. The disaster registry covers registered events from the same period, with conflict situations handled differently from discrete natural hazard events.

Known blind spots

Absence of evidence here is not evidence of absence. These are the conditions under which ReliefWeb (OCHA) will not show you something that is nevertheless real:

Everything here is secondary. These are documents about events written by organisations with mandates and funding interests, not observations, and the distance between the event and the record is never zero and rarely stated precisely.
Editorial selection is a filter with a direction. Material judged outside the humanitarian frame or primarily political is not included, so the corpus describes crises in the sector's own terms and excludes framings that some parties would consider essential.
Reporting lag means the first days of an emergency are documented retrospectively. The absence of reporting on a date is not evidence that nothing happened, and early figures are systematically revised upward as access improves.
Language coverage is predominantly English with partial French, Spanish and Arabic, so a text-analytic pipeline built on this corpus is analysing the internationally-facing subset of a much larger body of local reporting.
Local and national organisations are underrepresented relative to their operational role, which biases the corpus toward the perspective of international actors and away from the communities affected.
Access constraints shape what is reported, and areas controlled by parties who deny humanitarian access are described from a distance or not at all, which is exactly where conditions are usually worst.
Figures cited across many documents frequently originate from a single upstream assessment, so apparent convergence in the corpus is often repetition rather than independent corroboration.
Locations and details are sometimes deliberately generalised in published documents to protect staff and beneficiaries, so precision in a report can be lower than the organisation's actual knowledge for good reasons.
The disaster status field is an editorial judgement about the response phase rather than a statement about the hazard, and treating a past status as evidence that a crisis has ended is a straightforward misreading.

Write the blind spot into the product. A statement that something “was not observed in ReliefWeb (OCHA)” is defensible; a statement that it “did not happen” is not, and the difference is what survives cross-examination.

Access, licensing and what you may do with it

Access model: Open — no account required

The API is open, requires no key, and expects every request to carry an application name identifying the caller – this is the mechanism by which usage is attributed and it should be a real, descriptive name rather than a placeholder. Queries are structured: you select an entity type, specify which fields to return, apply filters on controlled vocabulary fields and date ranges, add a full-text query if needed, request facets for aggregate counts, and page through results with offset and limit. There is a documented ceiling on results per request and a daily call allowance per application name; the documentation site carries the current figures and you should read them rather than assume, because they have been adjusted over time. The practical implication is that this is a good API for targeted retrieval and a poor one for bulk mirroring – build queries that narrow at the server using disaster, country, date and format filters rather than pulling broadly and filtering locally. Attachments are separate HTTP fetches from the URLs in the record, and for many document types the attachment carries substance the indexed body does not.

Licence

The terms published on the service are the authoritative statement and should be read before any systematic reuse. What can be said with confidence is that ReliefWeb is an aggregator of third-party material and that copyright in each document remains with its publishing organisation, so the licence position on a specific report is the publisher's, not OCHA's. The service's own metadata and structured vocabularies are a different matter from the documents they describe. In practice this means indexing metadata, linking to documents and quoting for analysis are uncontroversial, while wholesale republication of document text or of attachments is a copyright question about each individual publisher. For most analytical use – building an index, extracting entities, computing volumes, quoting with attribution – you are on comfortable ground. For building a product that redistributes the documents themselves, you need the publishers' permission and not OCHA's. Attribute to the originating organisation with the ReliefWeb record as the access route, which is both correct and useful to readers.

Rate limits and fair use

There is a documented daily call allowance keyed to the application name you supply, and a maximum number of results per request. Read the current figures in the documentation rather than discovering them by hitting a wall mid-collection. The behaviours that keep you inside them are the same ones that make for good analysis: filter at the server on disaster, country, date range, format and theme rather than retrieving broadly; request only the fields you need rather than full records with body text when you are building an index; use facet queries to get counts without retrieving documents; and cache aggressively, since published documents do not change. Set a descriptive application name that identifies your organisation, because it is the only channel through which the service can distinguish a legitimate research collector from a runaway script. If your requirement genuinely exceeds the allowance, contact the service rather than sharding across multiple application names, which is both transparent and the sort of thing that gets access withdrawn.

Licensing changes, and it changes without warning. A dataset that was free for research this year may not be free for commercial or evidential use next year. Confirm the current terms before you build a dependency on it, and record the terms you relied on alongside the data — the licence in force at the time of collection is part of the provenance.

Collecting it

How ReliefWeb (OCHA) is actually pulled, in the order you would set it up. Prefer the bulk or export interface over per-item lookups wherever one exists: it is kinder to the publisher, faster for you, and gives a reproducible snapshot rather than a series of point-in-time answers you cannot reconstruct later.

Method	Format	Cadence	Notes
Disaster registry pull	JSON	daily	Retrieve registered events with type, status, dates, affected countries and identifiers. This is the event spine everything else attaches to, and it is small enough to mirror completely.
Filtered report retrieval	JSON	hourly to daily during an active crisis	Query reports filtered by disaster, country, date and format, requesting only the fields you need. The workhorse, and the query design is where quota efficiency is won or lost.
Facet queries for volume analysis	JSON	daily	Aggregate counts by source, theme, format and date without retrieving documents. Reporting volume and its composition is an indicator in its own right and costs almost nothing to collect.
Source registry snapshot	JSON	monthly	The controlled list of publishing organisations with types and homepages. Essential for assessing whether apparent corroboration comes from independent bodies or from one institutional family.
Attachment retrieval	bulk	selective	Fetch PDFs and map images for documents that matter. Attachments frequently carry tables, methodology annexes and detail absent from the indexed text, and they are separate HTTP requests.
Incremental change collection	JSONL	continuous	Poll by creation date since your last run rather than re-querying whole windows, and append to a local store with retrieval timestamps so your corpus is reproducible.

Ingesting it into the platform

Every step below is idempotent and cursor-based: interrupt one and it resumes from where it stopped rather than duplicating rows or losing progress. Collection is recorded per source, so a feed that quietly stops publishing shows up as a stale timestamp instead of silently thinning your coverage.

Register the service with its editorial character noted — Record in sources.php that this is a curated secondary collection, so downstream analysts treat a document as a dated organisational claim rather than as an observation of an event.
Mirror the disaster and source registries first — These are small controlled vocabularies and they are the join keys for everything else. Load them through ingest.php before any document collection so that reports arrive with resolvable event and organisation references.
Preserve both dates — Store the original publication date and the record creation date separately, and derive the lag. Timelines built on the wrong date shift events by days, and the lag itself is an analytically meaningful measure of access and reporting capacity.
Attribute to the publishing organisation — Make the source registry entry the provenance on every derived record, and resolve it to organisation type and mandate through enrich.php, so that a figure carries who said it rather than where you found it.
Stratify by content format — Keep situation reports, assessments, appeals, evaluations, news and maps as distinct evidence classes rather than merging them into a document pile. A funding appeal and a field assessment are different kinds of claim and should never carry equal weight.
Extract and link entities conservatively — Pull place names, organisations, figures and dates from body text, and attach each to the document that asserted it with the assertion preserved. Never promote an extracted figure to a fact in the platform without its source document attached.
Build the event timeline — Assemble documents by disaster identifier into timeline.php so that a crisis renders as a sequence of dated claims from identified organisations, which is the form in which reporting lag and divergence become visible.
Correlate against measured data — Use correlate.php to place narrative reporting alongside conflict event data, displacement figures and remote sensing for the same place and period, and record divergence between what is reported and what is measured as a finding rather than a discrepancy to be smoothed.

Registered sources and their last-collected state are listed in sources.php, and the scheduled chain that keeps them current is in automation.php.

How it is wrong, and how to tell

Every dataset is wrong in characteristic ways. Knowing which ways is the difference between using a source and being used by one, and it is the part of source evaluation most often skipped because it is the part that takes work.

Judged as what it is – a curated collection of organisational reporting – the quality is high and the editorial layer is doing real work. Documents are attributed to identified organisations, metadata is applied consistently by editors rather than by contributors, the controlled vocabularies are maintained, and the disaster registry provides a reliable event spine. That is considerably better than any general news aggregation and better than most sectoral archives. What the editorial layer does not and cannot do is verify the substance of what organisations report. A situation report's figures are the reporting organisation's figures, produced under access constraints it may or may not describe, with methodology that may or may not be documented, in a context where the organisation is often also seeking funding for the response it is describing. That is not a criticism of the sector – the incentive is structural and widely acknowledged within it – but it means the analyst's quality assessment has to operate at the level of the individual publisher and document. The reliable signals are the same ones that work elsewhere: does the document state its method, does it state what it could not reach, does it distinguish assessed from estimated figures, and does the organisation have a track record of revising its own numbers. Documents that do all four are strong evidence. Press releases from any organisation are not.

Characteristic false positives

A figure repeated across twenty documents is usually one assessment cited nineteen times. The corpus manufactures apparent consensus efficiently, and tracing every number to its originating document is the only defence.
Publication date is mistaken for event date. Reporting lags events by days and sometimes weeks, and a timeline built on record creation dates will place a crisis later than it happened and compress its early phase.
Press releases and appeals are read as field reporting. Both describe conditions, both are published by credible organisations, and both are written to secure funding or visibility; the content format field exists to distinguish them and is routinely ignored.
Early figures are treated as final. Initial estimates in sudden-onset emergencies are systematically revised, usually upward as access improves, so citing a first-week number without noting it is preliminary misrepresents the evidence.
Absence of reporting is read as absence of events. In access-denied areas the reporting stops precisely when the situation deteriorates, so a quiet period in the corpus for a conflict zone frequently marks the worst phase.
The disaster status field is read as a statement about the hazard. Past means the humanitarian response phase has concluded in the editors' judgement, not that the drought ended or the conflict stopped.
Deliberate generalisation of locations to protect staff and beneficiaries is read as imprecision in the underlying knowledge, leading analysts to discount reporting that is vague for entirely sound reasons.
English-language dominance is mistaken for the complete picture, so a text-analytic finding about what is or is not being reported may be a finding about the language of international humanitarian communication.

None of these make the source unusable. They make it a source that requires corroboration before an assertion built on it goes into a product, which is true of every source and admitted by few.

Ageing

Documents do not age – a dated situation report remains an accurate record of what an organisation said on a date, and that is its evidential value. What ages is the situation it describes, and the rate varies enormously by document type. Flash updates and situation reports describe conditions that change within days and are superseded by the next issue in the series, so citing sitrep number four when number nineteen exists is a straightforward error that the title numbering makes easy to avoid. Needs assessments have a useful life measured in months. Evaluations and lessons-learned studies age slowly and remain valuable for years because they are about how a response worked rather than about current conditions. The disaster registry ages through status transitions that reflect editorial judgement rather than physical change. The subtler ageing problem is in your own derived analysis: an assessment built on the reporting available at the time will not incorporate the retrospective documentation that arrives weeks later, and revisiting a conclusion once the fuller record exists is a discipline rather than an option. A stale record here looks like a current-sounding figure quoted from a document whose series has moved on twelve issues, and the fix is to collect the series rather than the document.

What this source feeds

A source is only worth what it lets you conclude. These are the disciplines that collect through it, the mission domains it serves and the data points it yields — every one is a tag, so you can follow any thread from here into the rest of the library.

Collected by these intelligence disciplines

Serves these mission domains

Conflict & Humanitarian

Yields these data points

How each sector uses ReliefWeb (OCHA)

The same dataset is worked very differently depending on who you are, what authority you hold, and what you are ultimately producing. A military analyst is supporting a commander’s decision; a journalist is meeting a publication standard; an NGO caseworker is protecting a person. The records are shared — the constraints, thresholds and outputs are not.

🎖 Military and defence

For civil-military coordination, humanitarian assistance operations and stabilisation planning, this is the open record of what humanitarian actors are doing, where they are present, what they assess the situation to be, and where they cannot go. That last item is operationally significant: access constraints described in reporting map the areas where movement, permission or security prevents humanitarian operations, which is a real geography. The corpus also supports historical analogue analysis – how comparable disasters and displacement crises unfolded and what the response required. The constraint that must be stated plainly is that this material is published by organisations whose safety and access depend on being perceived as neutral and independent, and use that could associate them with military activity endangers them. Consume the public record; do not approach the organisations through it.

🕵 National intelligence

The distinctive analytical products here are narrative context, reporting lag and source structure. Narrative context tells you what happened around a measured figure – method, access, caveats – which is what makes the figure interpretable. Reporting lag is an indicator in its own right: a widening gap between events and their documentation signals access deterioration, and a sudden change in reporting volume for a location is frequently the earliest open signal of a shift on the ground. Source structure lets you distinguish genuine multi-source convergence from a single claim propagating, which is a discipline that matters at least as much in humanitarian reporting as in any other open source. For NEWSINT and OSINT work on conflict and disaster this is a higher-quality corpus than general news aggregation because attribution and dating are systematic. Note that no assertion here originates from a model – the platform's only language-model function writes prose about records that already exist.

👮 Law enforcement

The law enforcement uses are contextual and strategic rather than case-level. Documents describing displacement routes, camp populations, protection concerns and border conditions establish the environment in which transnational trafficking and smuggling investigations sit, and evaluations frequently describe the vulnerabilities that criminal exploitation targets. Nothing here identifies individuals and nothing should be used to attempt to. For any matter touching trafficking, child protection or exploitation the correct route is the established referral mechanism – the national hotline, the relevant protection agency, the international reporting channel – rather than analysis of published documents, and material that would help locate a vulnerable person has no place in an investigative product regardless of its source.

🔍 Private investigation and corporate security

Corporate security, supply chain and due diligence work uses this to understand operating environments in crisis-affected countries with sourced, dated material rather than commercial summaries. Situation reports describe infrastructure damage, market functioning, access routes and population movement in a level of practical detail that risk products compress away. Evaluations and assessments identify which organisations are actually operating where, which matters for anyone assessing local partners. The limits are the sector's limits: the corpus is about humanitarian need and response, not about crime, corruption or commercial risk, and treating a humanitarian assessment as a security assessment misreads what it is measuring.

📰 Journalism and OSINT media

This is a working reference for crisis reporting and one of the few places where a journalist can find dated, attributable primary organisational documents rather than press summaries of them. Its practical strengths are the disaster registry for establishing what an event is called and when it started, the source registry for identifying who publishes about a country, and the archive depth for historical comparison. The reporting practices that matter are to trace every figure to the document that first asserted it rather than to the most recent repetition, to cite the publishing organisation rather than the platform, to state the reference period, and to be explicit when a figure is preliminary. Reporting volume itself is a story when it changes sharply, and the facet queries make that measurable in seconds.

🌍 NGO, humanitarian and human rights

For humanitarian and human rights organisations this is both a source and a channel, and the useful advice concerns both. As a source, it is the fastest route to what other organisations have assessed in a location, which is how duplication is avoided and how a new assessment is designed to add rather than repeat. The evaluation and lessons-learned collection is genuinely underused and contains the accumulated institutional knowledge of decades of response. As a channel, publishing here is how documentation enters the shared record and becomes citable by others, and the metadata you supply determines whether anyone finds it. For advocacy work, the archive is where you demonstrate what was known and when, which is frequently the decisive question in accountability.

🎓 University and research

The corpus supports disaster studies, humanitarian and development research, conflict analysis, organisational sociology and computational text analysis, and its multi-decade span with consistent structured metadata is genuinely rare. Three methodological cautions belong in any paper. The corpus is editorially selected against humanitarian relevance criteria, so it is a sample of humanitarian discourse rather than of events, and the selection mechanism must be modelled. Language distribution is skewed toward English, so text-analytic findings about salience may reflect the language of international communication rather than of local reporting. And documents are organisational products with funding and mandate incentives, which makes them excellent objects of study in their own right and problematic as unexamined evidence about the world. Archive the exact records and attachments used, with identifiers and retrieval dates.

Playbook: working ReliefWeb (OCHA) end to end

A repeatable sequence from first pull to finished product. Each phase states what you are trying to establish, not merely what to click — the objective is a defensible chain of reasoning, not a completed checklist.

Phase 1 — Anchor on the event, not the search term

Start from the disaster registry and identify the registered event and its identifier before searching text. Working from a keyword produces a pile of documents from several events and several years; working from an event identifier produces the actual corpus for the thing you are studying, with dates and affected countries already established.

Phase 2 — Map the reporting ecosystem before reading anything

Run facet queries by source, format and date for the country and period. This tells you in one request who is publishing, how much, in what document types and when the volume changed – and it frames every subsequent judgement about whether a claim is widely supported or narrowly sourced.

Phase 3 — Stratify by content format immediately

Separate field situation reports and assessments from appeals, press releases, news and secondary analysis. These carry very different evidential weight and the format field makes the separation free. An analysis that treats them as one corpus will overweight whichever category is most numerous, which is usually the least evidential one.

Phase 4 — Build the series, not the document

Most operational reporting is serial – numbered situation reports issued by one organisation for one crisis. Collect the whole series and read the trajectory, because the changes between issues are where new information actually is, and quoting one issue from the middle of a series is how outdated figures get published.

Phase 5 — Trace every figure to its origin

For each number that will appear in your product, find the first document that asserted it and identify the assessment behind it. Repetition across the corpus is not corroboration and is very often a single upstream source propagating. This step alone prevents the majority of published errors in humanitarian analysis.

Phase 6 — Measure the reporting lag deliberately

Compute the interval between original publication date and record creation, and between events described and the documents describing them. A widening lag is a signal about access, capacity or security, and it frequently precedes explicit reporting that conditions have deteriorated.

Phase 7 — Read the silence as data

Identify the districts, populations and sectors that the reporting for an event does not cover, and establish why. In conflict, absence usually means access denial rather than absence of need, and the geography of what is not reported is often the most important finding available from this corpus.

Phase 8 — Cross-read organisations with different mandates

Compare how a UN agency, an international NGO, a national organisation and a donor describe the same situation. Divergence in emphasis is informative about interests, and divergence in fact is a question that has to be resolved rather than averaged. Consensus among organisations that share a funding stream is weaker evidence than it appears.

Phase 9 — Pull the attachments for anything load-bearing

The indexed body text is frequently a summary. Methodology annexes, data tables, assessment coverage maps and caveat sections live in the attached PDF, and a conclusion drawn from the summary without opening the attachment is a conclusion drawn from a press-facing abstract.

Phase 10 — Correlate narrative against measurement

Place the document timeline alongside conflict event data, displacement figures, market prices and remote sensing for the same place and period. Where narrative and measurement agree, the finding is strong; where they diverge, that divergence is a finding about reporting, access or incentive that deserves its own paragraph.

Phase 11 — Handle sensitive content protectively

Protection reporting, trafficking documentation and material on vulnerable populations require victim-centred handling. Do not aggregate details that could identify individuals or locate people at risk, respect the deliberate generalisation publishers apply, and route anything requiring action to the appropriate protection agency or national referral mechanism rather than into an analytical product.

Phase 12 — Record provenance to the publisher and freeze what you used

Cite the publishing organisation, the document identifier, the original publication date and the retrieval date, and archive the records and attachments your analysis rests on. Attribution to the platform rather than the publisher obscures the only thing a reader needs to assess your evidence.

The platform ships this as a step-checked workflow in playbooks.php, so progress is recorded against a case rather than held in someone’s head.

What to pair it with

No single source carries a finding. These are the datasets that corroborate, extend or contradict this one — and a source that contradicts is worth more than one that agrees, because it is the only thing that will tell you when you are wrong.

Source	Relationship	What it adds
Humanitarian Data Exchange	extends	The quantitative counterpart from the same parent organisation. Where ReliefWeb describes a crisis, HDX supplies the boundaries, baselines and structured measurements underneath it.
ACLED	corroborates	Independently coded conflict event data with published methodology, providing measurement against which narrative reporting of violence can be tested rather than merely echoed.
UCDP	corroborates	Academic conflict data with conservative inclusion thresholds and a long series, useful as a second independent measurement with different coding assumptions.
GDACS	prerequisite	Automated hazard alerting and impact estimation that fires within hours of a sudden-onset event, filling the window before humanitarian reporting exists at all.
Copernicus Emergency Management Service	corroborates	Satellite-derived damage and extent mapping, which measures physical impact independently of what any organisation reports and covers areas humanitarian access does not reach.
IOM Displacement Tracking Matrix	extends	Systematic displacement measurement with documented survey methods, supplying the figures that situation reports cite and the methodology those reports usually omit.
IPC	extends	The food security classification system that converts assessment data into a comparable severity scale, and the reference for interpreting food security claims in reporting.
GLIDE	extends	Cross-system disaster event identifiers, allowing an event in this archive to be matched to records in independent humanitarian and disaster databases without name matching.
UN OCHA	prerequisite	The parent organisation, its coordination architecture and the humanitarian principles that determine what is in this collection and how it may appropriately be used.

Legal, ethical and operational constraints

The dominant constraint is copyright, and it sits with each publishing organisation rather than with OCHA. Indexing metadata, linking, quoting with attribution and analysing text are unproblematic in most jurisdictions; wholesale republication of documents or attachments is a permission question for each publisher. Data protection engages where reporting names individuals – beneficiaries, staff, victims of violations – and in most regimes storing and processing that material requires a lawful basis, a purpose limitation and a retention position, none of which the public availability of the document supplies. The ethical layer is heavier than the legal one. This corpus documents people in situations of extreme vulnerability, and material about protection cases, trafficking, sexual violence and child protection carries handling obligations that go well beyond copyright: it should be minimised, not disseminated, and never aggregated in ways that could identify or locate individuals. Where documents deliberately generalise locations to protect staff and beneficiaries, do not attempt to resolve that generalisation. And as with any humanitarian source, uses that would associate neutral organisations with military, intelligence or enforcement activity carry real consequences for people in the field and should be settled as a matter of policy rather than left to the analyst.

Operational security

Every API request carries the application name you supply, which is a self-declared identity, along with your network address and the exact query – country, disaster, date range and search terms. A pattern of queries concentrated on one country or one event is a clear statement of interest, and in humanitarian contexts the parties with an interest in knowing who is examining reporting about their territory are not hypothetical. The mitigations are routine: collect broadly and continuously rather than querying narrowly when something happens, mirror the archive segments you care about so that repeated analysis is local, and keep the application name honest but not more revealing than it needs to be – it should identify a real organisation without describing an operation. The larger exposure is in what you publish. Analysis that names specific facilities, organisations or locations can endanger field staff and the people they serve, and identifying which organisation supplied a piece of reporting is in some contexts sufficient to put it at risk. Separate the provenance you record internally from the provenance you publish.

Two rules that hold regardless of jurisdiction. Collection that is lawful is not automatically proportionate, and a dataset assembled for one purpose does not carry consent for another. Where the records concern identifiable people, the question is not only whether you may hold the data but whether holding it serves the purpose you are accountable for.

Is it earning its place?

Sources accumulate. Feeds get added during an incident and are never reviewed again, and a decade later the pipeline is carrying dead weight that nobody dares remove. These are the measures that show whether ReliefWeb (OCHA) is contributing anything, and they are worth baselining now so the answer is available later.

Reporting volume per country and event over time, from facet queries, tracked as an indicator in its own right because sharp changes usually precede explicit reporting that conditions have shifted.
Median interval between original publication date and record creation, monitored per country, as a measurable proxy for access and reporting capacity.
Proportion of figures in your products traced to an originating document rather than to a repetition, which is the single clearest measure of analytical discipline against this corpus.
Distribution of the documents you actually use across content formats, to check that the analysis is resting on field reporting and assessments rather than on press releases and appeals.
Source diversity behind each significant finding, counted as genuinely independent organisations rather than as document count, to detect manufactured consensus.
Share of load-bearing conclusions for which the attachment was retrieved and read, rather than only the indexed summary text.
Number of findings where narrative reporting diverged materially from independently measured data, which is where this source earns its place rather than merely confirming what other sources already showed.

Beware of volume. Indicator counts rise easily and say almost nothing. Unique contribution — findings this source produced that no other source in your stack would have — is the measure that matters, and it is usually far lower than anyone expects.

Tradecraft notes

The distinctions that separate a competent analyst from a fast one:

The event identifier is the unit of work, not the search term. Anchoring on the disaster registry produces a defined corpus with dates and countries attached; anchoring on keywords produces an unbounded pile that quietly mixes events and years.
Content format is an evidence class. A field assessment, a funding appeal and a press release make different kinds of claim, and an analysis that pools them will be dominated by the category that publishes most, which is never the most evidential one.
Repetition is not corroboration. Humanitarian reporting propagates figures efficiently, and tracing each number to its first assertion is the step that separates analysis from aggregation.
Read serial reporting as a series. Sitreps are numbered for a reason, the trajectory carries the information, and quoting a single issue out of a run is how superseded figures reach publication.
Reporting lag is a measurement, not a nuisance. Compute it, monitor it, and treat a widening gap as a signal about access rather than as a data quality complaint.
Silence is geography. The districts that vanish from reporting during an escalation are usually the ones where access was lost, and mapping the absence is frequently more informative than analysing the presence.
Distinguish organisational mandate from independent observation. Agencies describing crises they are also appealing to fund are not dishonest, but the incentive is structural and it belongs in your weighting rather than in a footnote.
Open the attachment. Methodology, coverage maps, caveats and tables are in the PDF and not in the indexed body, and conclusions drawn from indexed text alone are conclusions drawn from an abstract.
Handle protection material as protection material. Minimise, do not aggregate, do not attempt to resolve deliberate generalisation, and route anything actionable to a protection agency or national referral mechanism rather than into a report.

Questions analysts actually ask

Do I need an API key?

No key and no charge, but every request must carry an application name identifying the caller, and that should be a real descriptive name rather than a placeholder. Usage allowances are keyed to it, and it is the only channel through which the service can distinguish a legitimate collector from a runaway script.

How far back does the archive go?

To the mid-1990s, continuously. That depth is unusual and is one of the source's strongest properties, because it supports genuine historical comparison of how similar crises unfolded and how responses were structured, rather than only current-situation monitoring.

Is this a substitute for news monitoring?

No, and it is better than news for some jobs and worse for others. It is curated, attributed and dated with consistent metadata, which makes it far more tractable analytically. It is also editorially bounded to humanitarian relevance, lagged relative to events, and skewed toward international organisations, so it will not surface breaking developments or non-humanitarian framings.

What is the disaster registry and why does it matter?

It is a controlled list of registered events with identifiers, types, statuses, dates and affected countries. It matters because it turns scattered reporting into an event-centred corpus: instead of searching text and hoping, you retrieve the documents attached to a specific event. It is small enough to mirror completely and should be the first thing you load.

Why do numbers differ between reports on the same event?

Because they were produced at different times, by different organisations, using different methods, covering different areas that were accessible to different degrees. Early figures are preliminary and usually revised upward. The correct response is to trace each figure to its assessment and report the range with methods and dates, not to average them.

Does a disaster status of past mean the crisis is over?

No. The status is an editorial judgement about the humanitarian response phase, not a physical statement about the hazard or a social statement about need. Droughts, displacement and protracted conflict outlast the response cycle routinely, and citing a past status as evidence of resolution is a misreading of the field.

Can I republish the documents?

Copyright stays with the publishing organisation, so republication is a permission question for each publisher rather than for OCHA. Indexing, linking, quoting with attribution and analysing are on comfortable ground in most jurisdictions. Attribute to the originating organisation and use the record as the access route.

How do I handle reporting on trafficking or protection cases?

Victim-centred and referral-first. Do not aggregate details that could identify or locate individuals, respect the deliberate generalisation publishers apply for protection reasons, and route anything requiring action to the appropriate protection agency or the relevant national referral mechanism. Analytical curiosity is not a lawful basis and is not a justification.

What is the fastest way to understand a new crisis?

Retrieve the disaster record, then run facet queries by source, format and date to see who is reporting and how the volume has moved, then read the two or three most recent situation reports from organisations with a field presence, opening the attachments. That sequence takes under an hour and gives you the event, the reporting ecosystem and the current assessment with their caveats.

Standards, formats and interoperability

What this source speaks natively, and what it has to be translated into before a partner can consume it. Work that arrives in a recognised format is easier to defend, easier to hand over and easier to automate against:

GLIDE numbers provide cross-system disaster event identifiers, linking events here to records in independent humanitarian and disaster information systems.
The humanitarian cluster system structures the sectoral vocabularies used for themes, so thematic filtering aligns with how the response is actually coordinated.
ISO 3166 country codes underpin the country and primary country fields, making geographic joins to other datasets straightforward.
The Common Operational Datasets and P-codes maintained across the humanitarian system are the geographic reference against which locations mentioned in reporting should be resolved.
IPC and Cadre Harmonisé phase classifications are the standard vocabulary for food security severity appearing throughout the reporting.
The humanitarian principles of humanity, neutrality, impartiality and independence are the framework under which this material is produced and the reference for judging appropriate use.
OCHA data responsibility guidance governs the handling of sensitive information about affected populations and applies to your derived products as much as to the originals.
The platform exports derived events, entities and locations in STIX 2.1, MISP, CSV, JSON and JSONL, so humanitarian reporting travels into a case alongside conflict, geospatial and technical observables.

References

Primary documentation and authoritative references for this source. Publishers revise and retire material, so treat the retrieval date as part of the citation and re-check before relying on any of it in a formal product.

ReliefWeb — UN OCHA. The service itself. Browse a country or disaster page first to understand how the collection is structured before writing any query code.
ReliefWeb API documentation — UN OCHA. The authoritative reference for entity types, fields, filters, facets, result ceilings and usage allowances. Read the current text rather than any secondhand summary.
UN Office for the Coordination of Humanitarian Affairs — UN OCHA. The parent organisation and the coordination architecture that determines what is in this collection and in what vocabulary it is described.
Humanitarian Data Exchange — UN OCHA Centre for Humanitarian Data. The quantitative counterpart. Where a report cites a figure, this is frequently where the underlying dataset and its methodology live.
GLIDE — Asian Disaster Reduction Center. The cross-system disaster identifier scheme, and the mechanism for matching an event here to records in independent disaster databases.
ACLED — Armed Conflict Location and Event Data Project. Independently coded conflict events with published methodology, the standard measured cross-check against narrative reporting of violence.
GDACS — European Commission and UN. Automated hazard alerting and impact estimation covering the hours before any humanitarian reporting on a sudden-onset event exists.
Copernicus Emergency Management Service — European Union. Satellite-derived damage and extent mapping that measures physical impact independently of organisational reporting and reaches areas access does not.
IOM Displacement Tracking Matrix — International Organization for Migration. The displacement measurement programme behind many of the movement figures cited in reporting, with the methodology those reports usually omit.
Integrated Food Security Phase Classification — IPC Global Partners. The classification framework behind food security statements in humanitarian reporting, and the reference for interpreting a phase claim correctly.
INHOPE — INHOPE. The international network of hotlines for reporting child sexual abuse material, and the correct referral route if protection-related material is encountered in the course of research.

Link integrity: every reference above was verified with a live request when this page was generated. Where a publisher had moved or withdrawn a document, the link was repointed at a preserved copy in the Internet Archive and marked as archived. Anything with no reachable copy anywhere had its link removed rather than left to rot — the source is still credited, it simply cannot be linked.

Put it into practice

The Quantus Intel threat intelligence platform operationalises this source: it mirrors the disaster and source registries as the event spine, preserves publication and ingest dates separately so reporting lag becomes measurable, stratifies documents by content format so appeals never carry the weight of assessments, and traces every extracted figure back to the organisation that first asserted it.. Browse the full source catalogue, or follow any tag above into the rest of the library.