# Official river observations, rainfall records and flood documents for the Nepal cascade

Research and automated asset assessment completed on 8 September 2026. This phase adds **10 candidate resource families and 106 assets**: 79 station-page/query routes and 27 document links. All 106 assets have source-linked metadata assessments. The catalogue source at the end of this phase contains 27 reviewed resources with 270 assets and 96 candidate resources with 6,042 assets. Individual records retain their actual retrieval and assessment dates.

The four new executions bring the search register to **44 runs: 35 completed, eight partial and one unavailable**. Completion describes a declared search operation, not the completeness of the event evidence or scientific validation. No independent human eligibility decisions or manuscript inclusions were created, and no overall search-completeness date is assigned.

## What was searched

| Run | Reader-facing search title | Outcome | Native records screened | Curated families | Assets |
|---|---|---|---:|---:|---:|
| RUN-20260908-041 | River-gauge observations along the Bhote Koshi–Trishuli–Narayani corridor, 25–31 August 2026 — Nepal DHM | Partial | 332 | 1 | 13 |
| RUN-20260908-042 | Official flood reports, bulletin listings, station catalogues and data-access guidance for the Nepal cascade — Nepal DHM | Partial | 209 | 4 | 15 |
| RUN-20260908-043 | Rainfall-station observations across central Nepal, including the cascade corridor, 25–31 August 2026 — Nepal DHM | Completed | 371 | 1 | 66 |
| RUN-20260908-044 | Official event reports and daily flood bulletins for the Bhote Koshi–Trishuli–Narayani cascade, August–September 2026 — Nepal CHWRR | Completed | 252 | 4 | 12 |

Run 041 screened 331 national river-map entries and one additional locally known station page. Run 042 screened 55 download rows, 11 press-release cards, three publication rows and 140 notice rows. Run 043 screened 371 national rainfall entries. Run 044 screened 21 publications and 231 notices. The resulting **1,164 per-run discovery records are not 1,164 independent resources**: cross-interface overlaps and known-item retrievals remain identifiable in the disposition ledger. Search-register `records_retrieved` counts these distinct native records within each new run; linked candidate IDs count curated families. Earlier run definitions and counts were preserved.

There were 212 anonymous request attempts, including retries, failed routes, metadata listings, historical queries and PDF retrievals. Each attempt retains its actual UTC start/end, method, URL, non-secret parameters, response status, selected headers, received byte count and hash. A query result count sometimes measures readings, while its linked records identify stations; those units are explicitly distinguished in the query export.

## Prospective method and local leads

The three DHM strategies were frozen before their first source requests. The local hydrological-observations collection supplied station IDs and document URLs as known-item leads. Its earlier files were not retrospectively counted as formal discoveries. Older local observation downloads were made before the complete requested interval had elapsed, so different reading counts do not alone establish a provider revision.

During source research, the local collection identified a separate official publisher/interface at `hydrology.gov.np`. A prospective amendment created run 044 before its first request. Fresh 2026 document mastheads identify the Centre for Hydrology and Water Resources Research (CHWRR), Water and Energy Commission Secretariat, Government of Nepal. Older documents retain their actual provenance: the 2018 flood-warning SOP identifies DHM as publisher; the telemetry list identifies station operators without independently establishing its original publisher. A PDF software-author string such as “dhm” does not override the document masthead.

River selection used corridor and tributary names together with the Narayani basin, plus local known-item IDs. This avoids equating similarly named eastern Bhote Koshi stations with the target corridor. Rainfall selection used coordinates within longitude/latitude bounds [84.35, 27.65, 85.75, 28.55] **or** district Rasuwa, Nuwakot, Dhading or Chitwan. It is a broad central-Nepal planning selection, includes surrounding catchments, and is neither a watershed boundary nor an observed flood footprint. It does not establish coverage of the Tibetan source area.

The station forms were submitted with the native seven-day period ending 31 August 2026 (`date=2026-08-31`, `period=4`, station-specific `seriesid`). Transient CSRF tokens were used only where the public form supplied them and are omitted from exports. The rainfall form did not expose such a field; its native fields were followed. Full returned intervals were retained privately. Reading counts were separately charted for **[25 August 2026 00:00 UTC, 1 September 2026 00:00 UTC)**. Nepal time is UTC+05:45; this interval is not seven Nepal calendar days.

Document selection targeted event reports and daily bulletins dated 25 August–2 September 2026, with selected historical flood, station, access and warning-method records retained as supporting context. CHWRR notice selection used the publication date or an explicit matching date in the title/filename, never the expiry date. All exposed listing rows were screened. Bare CHWRR API arrays exposed no pagination; this establishes exhaustion of the returned lists, not a complete institutional historical archive. No language exclusion was applied. English descriptive titles retain original Nepali titles and filenames, with translation review still explicit.

Requests used a 30-second timeout, one retry for transient failure and a 10 MiB response limit. PDF text and small time-series structures were read for metadata and interpretation. No research code, hydrological simulation, rating-curve fitting, rainfall aggregation or independent scientific reproduction was performed. No provider requests, purchases or messages were sent.

## River observations and remaining gaps

The [DHM river-watch map](https://dhm.gov.np/hydrology/river-watch) supplied the national catalogue after the initially selected real-time listing returned HTTP 500. Thirteen selected or locally known pages were assessed. Nine station queries returned **4,849 readings**, all inside the strict UTC event window; two returned empty arrays, one velocity query did not return an observation array, and one known-item page exposed no station metadata.

| Station or route | Returned readings | Coverage or limitation |
|---|---:|---|
| Bhote Koshi at Rasuwagadhi | 162 | 25 August 00:05–26 August 02:55 UTC |
| Langtang Khola at Shyaprubesi | 163 | 25 August 00:05–26 August 03:05 UTC |
| Bhote Koshi at Shyaprubesi | 163 | 25 August 00:05–26 August 03:05 UTC |
| Trishuli Khola at Dhunche | 1,007 | Through 31 August 23:55 UTC; an internal gap is recorded |
| Phalakhu Khola at Betrawati | 166 | Ends 26 August 03:35 UTC |
| Trishuli at Betrawati | 166 | Ends 26 August 03:35 UTC |
| Trishuli at Kali Khola | 1,008 | 25 August 00:05–31 August 23:55 UTC |
| Narayani at Devghat | 1,006 | Through 31 August 23:55 UTC; internal gaps are recorded |
| Trishuli at Galchi | 1,008 | 25 August 00:05–31 August 23:55 UTC; series 36628 |
| Trishuli at Bhorle | 0 | Successful empty array; no inference of a dry river or absent flood |
| Kali Khola rainfall entry on the river map | 0 | River-form variable/route mismatch; the rainfall query returned observations in run 043 |
| Langtang velocity station | Not established | No historical observation array returned |
| Furke Khola/Malekhu, page 5611 | Not established | Current station identity and series route unresolved |

The observed modal spacing of returned river timestamps was ten minutes. No duplicate timestamps, null values, negative values or nonnumeric values were found in those arrays. This is limited structural charting, not instrument quality assurance: missing intervals, truncation, sensor failure and undocumented quality flags remain separate concerns.

Water level is presented in metres against the source reference. The gauge datum and its relationship to a common vertical datum were not established. In particular, Galchi’s numerically different reference must not be treated as equivalent to another station’s gauge height. Stage is not discharge or flood depth. Current warning thresholds/status labels are mutable station metadata rather than historical event observations. A telemetry endpoint is not a demonstrated flood-arrival time, and a maximum among available readings is not necessarily the true event peak.

## Rainfall observations

All 66 selected [DHM rainfall stations](https://dhm.gov.np/hydrology/rainfall-watch-map) returned historical arrays: **65,404 native readings**, of which **65,344** fall inside the strict UTC window. The 60 outside-window readings remain accounted for rather than silently removed from native counts. No duplicate timestamps, null values, negative values or nonnumeric values were identified in the retrieved arrays. Unequal station counts alone are not evidence of duplicates.

The station export records each actual start/end, observed spacing, largest internal gap and counts. Some stations have substantial missing intervals or incomplete edges. The source chart labels rainfall in millimetres, but its seven-day plotting code retains only points above 0.2. This assessment uses the **unfiltered returned array**, including zeros; it does not count only visible plotted points. The accumulation duration and instrument quality-control definitions were not established for every historical value. No basin-average rainfall, intensity, event total or rainfall attribution of the cascade was calculated.

## Official documents and their interpretation

The 27 document links comprise 15 DHM supporting/historical documents and 12 CHWRR-hosted documents. Twenty-six PDF bodies were retrieved; one station-list PDF returned HTTP 403. Meaningful text was extracted from 22 documents. Four retrieved PDFs lacked meaningful extractable text and remain untranscribed; retrieval and page counting are not full content review. Nepali extraction can also contain imperfect glyphs, so detailed translation remains a follow-up where precision matters.

The fresh [CHWRR publication list](https://hydrology.gov.np/cm/api-public/publications) includes a nine-page technical report on the 26 August event. Its listing calls the report preliminary, while its title calls it technical. Page 4 reports a peak discharge of approximately 5,850 m³/s around 16:00 at Devghat, assumes a 2,600 m³/s base flow, and estimates 57.40 million m³ total volume and 19.96 million m³ above that base flow for 14:10–18:00. These are **provider-reported estimates**, not values calculated or independently validated by the catalogue. The rating curve, uncertainty and underlying discharge series remain unresolved. [CHWRR technical report](https://hydrology.gov.np/cm/files/Bhotekoshi-Trishuli-Narayani_Flood_Report_CHWRR_1%20_1787809133090.pdf)

The same report contains station-identity conflicts: page 5 identifies Rasuwagadhi as index 446.42, whereas the current DHM page gives 446.22. Page 1 labels an 08:40 last reading as Bhote Koshi Syabrubesi while using thresholds also present at Rasuwagadhi. Both statements remain visible rather than being silently merged. Agency-reported warning or observation times do not establish individual message receipt, comprehension or evacuation. The report’s chronology was not used to fit a causal flood-travel model.

The printed volume figures also leave an unresolved integration interval. Total minus excess is 37.44 million m³, equivalent to four hours at the stated constant base flow of 2,600 m³/s. The printed 14:10–18:00 interval is three hours and fifty minutes. This arithmetic relationship is recorded as a request for clarification of the interval or integration convention; it does not replace the source estimates or reproduce their underlying observations.

The [CHWRR notice list](https://hydrology.gov.np/cm/api-public/notice) returned the 27 August press release and eight daily bulletins for 25–28 August, 30–31 August and 1–2 September. **29 August was not recovered from this listing snapshot**; this is not a claim that no bulletin was issued. The two-page press release has no extractable text. The bulletins combine reported morning river status, three-day forecasts and advisories. Their 08:00 issue labels, PDF creation timestamps and actual dissemination times are distinct. Forecast days are defined in the documents as 24 hours starting at 10:00. Forecasts are not observed inundation boundaries.

The [DHM downloads list](https://dhm.gov.np/downloads) supplied historical station lists, locations/catchment areas, a monitoring-network map, procurement/verification guidance, request forms and historical charges. The BS 2082 annual report was published before the August 2026 cascade, and the Bhote Koshi update listed on 10 July 2025 concerns an earlier event. Both are explicitly historical. Station altitude is not gauge-zero datum, published record periods need not be continuous, and a network map does not establish current instrument operation.

Several download rows displayed “0 bytes” despite returning nonempty PDFs. The original listing size and actual received size are kept separately; a listing placeholder is not evidence of an empty document. The public station-table route returned HTTP 404, and the DHM bulletin section did not expose a populated current list. These gaps keep run 042 partial. Purchased historical data and official verification services have separate request/use conditions; no purchase, request or redistribution permission is implied by access to their guidance.

## Names, assessments and counting

All 40 earlier viewer-facing search titles were rewritten as descriptions of the searched material, source and event context. Two new display titles were clarified after their original strategies were frozen. The **42-row title ledger** preserves each prior title. Historical strategy IDs, dates, exact criteria, results and counts were not rewritten. The original wording remains available in the frozen strategies and dated exports.

Release preparation further distinguishes the five river records ending on 26 August: their asset titles now state **25–26 August 2026 (UTC; record ends early)** instead of suggesting observations through 31 August. The requested seven-day window remains in the metadata. Asset pages display readable station, variable/unit, temporal-coverage and query-result fields, alongside attributed report estimates and unresolved interpretation questions.

Each new asset retains its original name or filename, source identity, access evidence, language/translation state, scientific role, temporal interpretation, technical metadata, rights state and limitations. The combined register now has **6,312 automated asset assessments**, preserving 6,206 earlier assessments and their evidence dates. Of these, 660 have recorded follow-up; all 106 new assets retain a reuse-scope follow-up because no component-specific open redistribution licence was established. Additional flags cover incomplete series, station identity, denied access and unreadable text. A completed assessment can therefore contain unresolved scientific or rights questions.

River and rainfall services are each one family; individual station/query pages are child assets. Daily bulletins form one dated series. Reports and historical/supporting collections are grouped separately. Document links, source rows, reading counts and curated resource families must not be summed as though they were interchangeable PRISMA units. Automated discovery dispositions are not final manuscript exclusion reasons.

A separate relationship ledger documents the same station identifier appearing in two viewers, companion event documents, and three potential alternate listings of historical reports. Potential matches are based on listing metadata; no identical-byte claim or new independent resource is created. The annual report’s publication row and download row state different May 2026 dates, which remain preserved. Original discovery dispositions are retained alongside these later curation notes.

## Reproducibility files

- [Phase summary and integration counts](2026-09-08-hydrology-summary.json)
- [Per-run outcomes and actual UTC dates](2026-09-08-hydrology-outcomes.csv)
- [Prospectively frozen strategies and amendment](2026-09-08-hydrology-strategies.json)
- [Exact queries, parameters, counts and record relationships](2026-09-08-hydrology-queries.json)
- [All 212 source request attempts and response hashes](2026-09-08-hydrology-requests.json)
- [All native discovery records and screening dispositions](2026-09-08-hydrology-dispositions.csv)
- [Relationships between station routes and document listings](2026-09-08-hydrology-relationships.csv)
- [Station-series coverage and structural findings](2026-09-08-hydrology-stations.csv)
- [Document inventory, dates, page counts and text availability](2026-09-08-hydrology-documents.csv)
- [Asset assessments with limitations and follow-ups](2026-09-08-hydrology-assessments.json)
- [One assessment row per asset](2026-09-08-hydrology-asset-assessments.csv)
- [Source-linked metadata fields](2026-09-08-hydrology-metadata.csv)
- [Descriptive asset titles and original names](2026-09-08-hydrology-asset-titles.csv)
- [Search-title revision ledger](2026-09-08-hydrology-search-titles.csv)
- [Source-level follow-ups](2026-09-08-hydrology-follow-ups.json)

Full source responses and decoded observations are retained privately with their request hashes. This public catalogue exports metadata and links, not raw station readings, scanned documents, session tokens, source error traces or personal/payment details. The record supports the search-reporting and evidence-charting portions of a PRISMA-informed review; independent screening, deduplication decisions across the wider review and final study/resource inclusion remain separate work.

## Next source searches

The strongest next step is a separately planned **BIPAD search for historical river observations and dated public flood alerts**, using the locally recorded Malekhu/Galchi and alert-service leads. It should reconcile station identifiers and distinguish message issuance from receipt. Other gaps include authoritative gauge datums, rating curves, quality flags, a readable press-release transcription and the missing bulletin date. Follow-up routes are leads, not executed searches or promised available data.