yuiseki/nilim-disaster-reports

NILIM disaster survey reports: page text of NILIM's own works

Page-level text of disaster field-survey reports published by the National Institute for Land and Infrastructure Management (NILIM, 国土交通省 国土技術政策総合研究所, 国総研), extracted from the text layer of the official PDFs. One JSON line is one PDF page.

What is in, and what is left out. The dataset owner chose to publish only reports whose text NILIM published alone with no other rights notice: rights classes a (no other rights notice found) and b (NILIM's text, with figures or photos credited to others). Every report that was jointly published or co-authored with another organisation (the Public Works Research Institute PWRI, the Building Research Institute BRI, the Port and Airport Research Institute PARI, the National Institute of Maritime, Port and Aviation Technology MPAT, universities, academic societies, the MLIT Housing Bureau and others), every report with an all-rights-reserved notice, and every report whose rights could not be assessed (classes c and d) is left out.

	reports	PDFs	pages
in: class a	14	23	313
in: class b	12	42	570
in: total	26	65	883
out: class c (joint publication, co-authorship or rights notice of another party)	108	503	12,764
out: class d (unclear)	10	61	478
out: total	118	564	13,242

So this is a small part (65 of 629 collected PDFs) of NILIM's disaster survey output. The class of every PDF and the evidence for it are in RIGHTS_REVIEW.tsv; every left-out PDF is listed in EXCLUDE.txt with one reason per report. Class b reports keep their credit lines (for example （山口県提供）, 出典：地理院地図) because they are part of the text; the credited figures and photos themselves are not in this text corpus.

This dataset contains only text. The PDFs, figures, maps and photographs are not included; every line gives the official PDF URL (source_url) and the SHA-256 of the PDF that was read (source_pdf_sha256), so the original can be downloaded and checked. It is not an official NILIM product and is not endorsed by NILIM.

Source and the two files

NILIM is the research institute of the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) for roads, rivers, sabo, coasts, ports, airports, sewerage, buildings and housing. After earthquakes, floods, typhoons, landslides and storm surges it sends researchers to the sites (often as part of TEC-FORCE) and publishes what they found.

config	file	series	reports	PDFs	pages	characters	empty pages	size
tnn	tnn.jsonl	国総研資料 (Technical Note of NILIM)	7	46	781	2,101,307	5	3,547,292 bytes
flash	flash.jsonl	quick field-survey reports (速報)	19	19	102	68,322	0	254,747 bytes
default	both files	both	26	65	883	2,169,629	5	3,802,039 bytes

By series and class:

series	class	reports in	PDFs in	pages in	reports out	PDFs out	pages out
tnn	a	1	10	243	0	0	0
tnn	b	6	36	538	0	0	0
tnn	c	0	0	0	25	420	11,025
tnn	d	0	0	0	3	54	335
flash	a	13	13	70	0	0	0
flash	b	6	6	32	0	0	0
flash	c	0	0	0	83	83	1,739
flash	d	0	0	0	7	7	143

• tnn (国総研資料): numbered technical reports, each published as a set of split PDFs (cover, front matter, chapters, appendices, colophon) linked from the report's page.
• flash: PDFs linked from NILIM's page 「国内で生じた災害の調査報告」 (https://www.nilim.go.jp/lab/bbg/saigai/index.html): short reports written within days of an event. One PDF is one report (report_id is flash- plus 16 hex digits of the URL's SHA-256).

source_type only records which index the PDF came from. Load one file with load_dataset("yuiseki/nilim-disaster-reports", "tnn") or "flash"; the default config reads both. The text is Japanese, with English titles and abstracts on report covers.

Coverage

The catalog of 2026-10-07 has 630 PDF URLs: 521 split PDFs of 35 国総研資料 reports found from NILIM's TEC-FORCE page (https://www.nilim.go.jp/lab/bbg/saigai/TEC-FORCE.html), the structures and foundations division page (https://www.nilim.go.jp/lab/ucg/research/02.html) and two reports added by hand (1320, 1356), plus 109 bulletins. 629 PDFs were downloaded on 2026-10-06/07 (the colophon of report 728 returned 404) and all 629 were reviewed; 65 are published.

国総研資料 reports, all 35 (in: classes a and b; out: c and d):

report	title	PDFs	pages	class	other parties / reason	in dataset
115	平成１５年５月２６日宮城県沖地震被害に係わる現地調査報告書	17	80	d	no text layer, authors and notices unreadable	out
133	平成15年7月26日宮城県北部地震被害に係わる現地調査報告書	16	88	d	no text layer, authors and notices unreadable	out
233	平成15年（2003年）十勝沖地震被害に係わる現地調査報告書	21	167	d	no text layer, authors and notices unreadable	out
248	平成16年（2004年）新潟県中越地震被害に係わる現地調査概要	38	214	c	joint survey with PWRI and BRI	out
250	平成15年7月九州豪雨災害に関する調査報告会	18	141	c	proceedings with PWRI and five universities	out
273	2004年の高潮・高波災害	10	102	b	credits: 気象庁, 菜生海岸災害調査検討委員会	in
415	平成19年（2007年）能登半島地震建築物被害調査報告	14	170	c	co-published with BRI	out
430	平成19年（2007年）新潟県中越沖地震建築物被害調査報告	13	210	c	co-published with BRI	out
438	平成19年（2007年）能登半島地震被害調査報告	16	250	c	co-published with PWRI and BRI	out
439	平成19年（2007年）新潟県中越沖地震被害調査報告	15	253	c	co-authored with PWRI	out
481	平成20年(2008年)岩手・宮城内陸地震建築物被害調査報告	16	206	c	co-published with BRI	out
486	平成20年（2008年）岩手・宮城内陸地震被害調査報告	15	246	c	co-published with PWRI and BRI	out
636	平成23年（2011年）東北地方太平洋沖地震調査研究（速報）	22	577	c	co-published with BRI	out
646	平成23年（2011年）東北地方太平洋沖地震 土木施設災害調査速報	17	497	c	co-published with PWRI	out
658	2011年東北地方太平洋沖地震津波による海岸保全施設の被害調査	1	45	b	credits: 電子国土, Google Earth	in
680	平成23年（2011年）東北地方太平洋沖地震による仙台空港の舗装に関する被害報告	1	26	c	co-authored by a PARI researcher	out
703	平成24年（2012年）5月6日に茨城県つくば市で発生した建築物等の竜巻被害調査報告	13	89	c	co-published with BRI	out
726	平成23年(2011年)東北地方太平洋沖地震による強震記録	10	243	a	-	in
728	平成23年（2011年）紀伊半島台風12号土砂災害調査報告	23	219	c	co-published with PWRI	out
758	平成24年7月九州北部豪雨災害に関する調査	13	101	b	credits: 九州地方整備局, 江口希之建築都市研究所	in
781	2011年東北地方太平洋沖地震津波による海岸保全施設の被害調査（その2）	1	56	b	credits: 国土地理院, Google Earth, one photographer	in
798	平成23年(2011年)東北地方太平洋沖地震による港湾施設等被害報告	43	3,156	c	co-published with PARI; all rights reserved	out
814	平成23年（2011年）東北地方太平洋沖地震による道路橋等の被害調査報告	12	767	c	co-published with PWRI	out
868	平成26年（2014年）北海道地方被災橋梁等調査報告	15	85	c	co-published with PWRI; all rights reserved	out
875	平成25年（2013年）台風26号伊豆大島土砂災害調査報告	16	104	c	co-published with PWRI; all rights reserved	out
936	平成25年山口・島根豪雨災害及び台風第18号災害に関する調査	10	96	b	credits: 山口県, 萩市, 地理院地図	in
967	平成28年（2016年）熊本地震土木施設被害調査報告	15	363	c	co-published with PWRI	out
980	平成28年（2016年）12月22日に発生した新潟県糸魚川市における大規模火災に係る建物被害調査報告書	14	93	c	co-published with BRI	out
1042	平成28年（2016年）熊本地震による港湾施設等被害報告	1	54	c	co-published with MPAT; all rights reserved	out
1069	平成28年（2016年）台風10号等に伴う豪雨による北海道地方被災橋梁等調査報告	11	88	c	co-published with PWRI	out
1111	令和元年（2019年）房総半島台風および東日本台風による土木施設・建築物等災害調査報告	15	417	c	co-published with PWRI and BRI	out
1151	令和2年（2020年）7月豪雨 道路災害調査報告	15	164	c	co-published with PWRI	out
1202	道路橋の洗掘による被災リスク要因に関する研究 －被災事例の統計分析等による検討及び令和3年(2021年)の豪雨災害調査－	1	138	b	credits: 地理院地図, 地図・空中写真閲覧サービス	in
1320	令和６年能登半島地震土木施設被害調査等報告	21	740	c	co-published with PWRI	out
1356	令和６年能登半島地震建築物被害調査等報告	21	1,896	c	co-published with BRI; university and company authors	out

The 19 published bulletins (flash), by path under https://www.nilim.go.jp/lab/bbg/saigai/:

• class a (13): sabo surveys of 2013-2014 (h26/140612yokosuka.pdf, h26/140820hiroshima_yagi.pdf, h26/140821hiroshima_yagi.pdf, h26/140822hiroshima_heli.pdf, h26/140822hiroshima_kabehigashi.pdf, h25typhoon26/131018izuoshima.pdf, h25typhoon26/131018izuoshima_betten.pdf, h25akita/130813akita_senboku.pdf), h22shimane/h22shimane_saigai.pdf (2010 rockfall), flood surveys (h25typhoon18/130927_suigai.pdf, h25yamaguchi/20130805suigai_result.pdf, h21hyougo/h21hyougo3.pdf) and the 2016 Kumamoto sewerage survey (h28/20160421kumamotogesuidou.pdf).
• class b (6): h27/20150722takamatsu.pdf (storm surge, JMA figures), h25typhoon18/130924fukui_obama.pdf (電子国土 base map), h24shimane/h24shimane.pdf (gust; photos from a school, Google map), h22yamaguchi/h22yamaguchi.pdf and h22gihu/h22gihu.pdf (floods; JMA rainfall), surugawan/surugawan.pdf (2009 earthquake; NIED K-NET/KiK-net records).

The 90 bulletins left out are written jointly by NILIM with BRI (most building-damage bulletins, 2003-2026), PWRI (most sabo bulletins), PARI, academic societies, universities or the MLIT Housing Bureau, or surveyed with them (class c, 83), or have unreadable text or a passage that may be quoted from others (class d, 7). Each is in EXCLUDE.txt with its reason.

Files and order

• tnn.jsonl, flash.jsonl: UTF-8 JSON Lines, one page per line, separated by \n only. Documents are in the order of the source catalog, pages in PDF order. Each line is the per-PDF extraction line copied byte for byte; nothing was re-serialised. Split on \n, not on every Unicode line break, if you read the files yourself.
• MANIFEST.json: lines, documents, bytes and SHA-256 of each file; pages, characters, empty pages and U+FFFD counts; the excluded PDFs with their class and reason; the class summary.
• provenance.yaml: source indexes, catalog date and SHA-256, fetch and extraction dates, tool versions, exclusion rule, and what was changed (nothing in the text).
• RIGHTS_REVIEW.tsv: the rights review, one row per collected PDF (629 rows, published or not).
• EXCLUDE.txt: the 564 source_urls left out, grouped by report, each group with its class, other parties, basis, quoted evidence and, for four reports, the all-rights-reserved notice.
• LICENSE: licence, exact quotation of the NILIM terms, attribution.

Hugging Face converts the JSONL to Parquet automatically for the viewer and datasets; no Parquet is published separately.

Schema

All 22 fields are present on every line.

field	type	meaning
document_id	string	SHA-256 (hex) of source_url. Same for every page of one PDF.
source_type	string	tnn (国総研資料) or flash (bulletin from the disaster index page). Index of origin only.
report_id	string	国総研資料 number (e.g. 758) for tnn; flash- plus the first 16 hex digits of document_id for flash.
report_title	string	Title as written on the index or report page. For 18 of the 19 flash PDFs it ends with the link text, e.g. （PDF ファイル：490KB ）.
event	string	rainfall (report 758, 13 PDFs) or 要調査 ("to be checked") for the other 52 PDFs. Not inferred from the text; read the title.
section	string	Link text of the PDF on the report page (e.g. 表紙, 奥付, a chapter title, 全文); for flash the same as the title.
source_url	string	Official URL of the PDF on www.nilim.go.jp.
index_url	string	Official page that links the PDF (the report page for tnn, the disaster index for flash).
source_pdf_path	string	Path of the PDF in the compiler's local collection (pdf/ks075801.pdf, pdf/flash/<24 hex>.pdf). The PDFs are not in this dataset.
source_pdf_sha256	string	SHA-256 of the PDF that was downloaded and read.
source_pdf_bytes	int	Size of that PDF in bytes.
page_number	int	1-based physical page index in the PDF. Not the printed page number.
text	string	Text of the page from pdftotext -layout, as extracted. Empty string for a page without a text layer.
characters	int	Length of text in Unicode code points.
non_whitespace_characters	int	Code points of text that are not whitespace.
has_text	bool	non_whitespace_characters > 0.
needs_ocr_review	bool	not has_text. Marks empty pages only; it does not detect bad text layers.
replacement_characters	int	Number of U+FFFD in text (0 on every line).
extraction_method	string	Always pdftotext-layout.
extractor_version	string	Always pdftotext version 24.02.0.
license_policy_url	string	Always https://www.nilim.go.jp/aboutlink.htm, the NILIM terms page.
rights_review	string	Note in Japanese written at collection time, before the rights review (原則PDL1.0、共同著作・第三者素材・個別条件は要調査 or, for report 758, 原則PDL1.0、個別素材・共著部分は要確認). Superseded by RIGHTS_REVIEW.tsv.

How the text was made

1. The PDF URLs were collected from the indexes above into a fixed catalog (2026-10-07, 630 URLs). Each PDF was downloaded once, sequentially and slowly, recording its size and SHA-256 (2026-10-06 10:27 to 22:24 UTC).
2. Each PDF was converted with Poppler pdftotext -enc UTF-8 -layout, which reads the PDF's own text layer. The output was split on form feeds into pages and the page count was checked against pdfinfo. No new OCR was run, and no correction, normalisation (width, old kanji, punctuation), header removal or de-hyphenation was applied. Whitespace from the layout is kept.
3. The rights of all 629 PDFs were reviewed (below). The per-PDF JSONL of the class a and b reports was concatenated into one file per source_type and checked line by line against the per-PDF files.

Quality caveats

The text is the PDF text layer as it is. It is not cleaned training text.

• 5 pages are empty (has_text = false): 2 in report 726 and 3 in report 1202.
• -layout keeps the physical layout. Two-column pages put both columns on one text line, separated by runs of spaces. Running heads, page numbers, captions, table cells and text inside figures are mixed into the page text. Report 726 is mostly tables of strong-motion records.
• report_title of most bulletins contains index link text such as （PDF ファイル：490KB ）, and event is 要調査 (unknown) for 52 of 65 PDFs.
• A PDF is not always body text: covers, tables of contents, author lists and colophons of the 国総研資料 are included as they are.
• No U+FFFD and no garbled text layer was found in the published files, but has_text = true does not mean every character is right.

Rights review

The page text of all 629 collected PDFs was scanned for signs of another rights holder or licence: copyright notices (©, Copyright, 著作権), reservation and reprint restrictions (All rights reserved, 無断転載, 承認を得ずして), joint authorship or publication (other institutes, universities, societies), credits (提供, 出典, 撮影, 転載, 引用, Google, ZENRIN) and Survey Act or GSI map notices (測量法, 承認番号, 国土地理院, 地理院地図, 電子国土). The covers, front matter, author lists and colophons of every report were read, and the hits were read in context. Each report was put in one class: a no other rights notice found; b figures or photos credited to others only; c a rights notice or reprint restriction covering the whole report, or joint authorship or publication with another organisation; d unclear. RIGHTS_REVIEW.tsv has one row per PDF with the class, other parties, basis, a quoted line of evidence (file, page, text), counts of lines matching each signal, its first hit, and whether it is excluded.

What it found in the left-out reports:

• Joint publications. Most left-out 国総研資料 reports are also published in a co-publisher's series (土研資料 of PWRI, 建築研究資料 of BRI, 港湾空港技術研究所資料 of PARI or MPAT), and their colophons carry the co-publisher's mark, e.g. 1151 ks1151_15.pdf page 1 「編集・発行 ©国土技術政策総合研究所 ©国立研究開発法人土木研究所」. Reports 248 and 250 are joint surveys or meeting proceedings with PWRI, BRI and universities.
• All rights reserved. 798 (with PARI), 868 and 875 (with PWRI) and 1042 (with MPAT) state that all rights are reserved and that no part may be reprinted or copied without written approval, e.g. 868 ks086801.pdf page 2: 「Copyright © (2015) by N.I.L.I.M. and P.W.R.I.」「All rights reserved.」「本報告書の全部又は一部の転載、複製は、…文書による承認を得ずしてこれを行ってはならない。」
• Co-authors and joint surveys. Report 680 has a PARI researcher among its four authors (ks0680.pdf page 2). 83 bulletins name BRI, PWRI, PARI, academic societies, universities or the MLIT Housing Bureau as authors or survey members next to NILIM, e.g. R8/kumamototekkinkonkurito.pdf page 1 「国土交通省国土技術政策総合研究所 / 国立研究開発法人建築研究所」.
• Unclear. Reports 115, 133 and 233 have no text layer; six 2016 Kumamoto bulletins have unreadable text layers; h21hyougo/h21hyougo1.pdf page 4 contains an uncredited passage that reads like quoted press text.

In the published reports: the 国総研資料 colophons carry NILIM's own mark and a contact for reprint requests (「編集・発行 ©国土技術政策総合研究所」「本資料の転載・複写の問い合わせは」). This names NILIM, the default rights holder under its terms, and asks for no approval, so it was not treated as another rights notice. Class b credit lines name the owners of figures and photos (GSI, JMA, prefectures and cities, Google, NIED, a regional bureau of MLIT, a private architect's office for one questionnaire figure in 758) that are not reproduced here; a few tables in 273 are JMA data cited with 「出典：気象庁 HP」. No 測量法 (Survey Act) approval notice was found.

The review is a reading of the text, not legal advice, and it can miss notices that exist only as images or on pages without text.

Getting and checking the original PDF

source_url is the official PDF. Download it and compare its SHA-256 with source_pdf_sha256 (and its size with source_pdf_bytes). If they differ, NILIM has replaced the file since 2026-10-06 and the text may not match. page_number is the page index in that PDF.

curl -sO "https://www.nilim.go.jp/lab/bcg/siryou/tnn/tnn0758pdf/ks075801.pdf"
sha256sum ks075801.pdf   # compare with source_pdf_sha256 of its lines

Licence

CC BY 4.0. The dataset owner chose to publish only NILIM's own works (classes a and b): reports NILIM published alone with no rights notice of another party. Jointly published and co-authored reports (with PWRI, BRI, PARI, MPAT, universities, societies and others) and unclear reports are left out (118 reports, 564 PDFs, 13,242 pages; see EXCLUDE.txt and RIGHTS_REVIEW.tsv), so the published text is NILIM content covered by NILIM's own terms.

NILIM's terms (「リンク・著作権等について」, https://www.nilim.go.jp/aboutlink.htm, read on 2026-10-07, Last-Modified 2025-12-19) say, quoted exactly (the duplicated 「は、の」 is in the original):

国土技術政策総合研究所ウェブサイト及び国土技術政策総合研究所のソーシャルメディアアカウントで掲載・発信している情報（以下「コンテンツ」といいます。）は、の著作権は、特記されていない限り国土技術政策総合研究所に帰属し、権利表記の記載がない限り「公共データ利用規約（第1.0版）」（PDL1.0）に準拠した利用条件の下で、利用することができます。

(In English: unless otherwise stated, the copyright of content on the NILIM website belongs to NILIM, and unless there is a rights notice it may be used under the Public Data License 1.0, https://www.digital.go.jp/resources/open_data/public_data_license_v1.0.) PDL1.0 is compatible with CC BY 4.0, which is why the card uses cc-by-4.0. The same page asks for the source in the form 「出典：「○○動向調査」（国土技術政策総合研究所） （当該ページのURL）（○年○月○日に利用）」 and, when content is edited, a separate statement that it was edited, without presenting the result as made by the government. The full section is quoted in LICENSE.

Credit lines in class b reports name the owners of figures and photos that are not included; the licence of this dataset does not extend to those images or to any third-party material quoted in the text.

Attribution

In the form requested by the NILIM terms (出典 and a note on editing):

出典：「国総研資料（災害調査報告）」（国土技術政策総合研究所）（https://www.nilim.go.jp/lab/bbg/saigai/TEC-FORCE.html）（2026年10月6日〜7日に利用） 出典：「国内で生じた災害の調査報告」（国土技術政策総合研究所）（https://www.nilim.go.jp/lab/bbg/saigai/index.html）（2026年10月6日〜7日に利用） 国土技術政策総合研究所の各PDF（各行の source_url）の文字層から pdftotext で本文を抽出し、yuiseki がページ単位の JSONL に収録したもの（Hugging Face: yuiseki/nilim-disaster-reports）を加工して作成。本文の訂正・正規化は行っていない。

In English: Source: NILIM disaster survey reports (国総研資料) and field-survey bulletins, National Institute for Land and Infrastructure Management, MLIT (index URLs above). Text extracted with pdftotext from each PDF's text layer and compiled into page-level JSONL by yuiseki (yuiseki/nilim-disaster-reports); not corrected or normalised; not made or endorsed by NILIM. When you cite one report, cite the report itself (authors, title, report number, year) and give its source_url.