Dossiers / Tokyo · Moscow · Soviet Far East · 1938–1940

Dossier · Japanese cryptanalysis

Moscow, 1940: Soviet weather read in Tokyo

For two years, from August 1938 to October 1940, the Japanese Army’s meteorological service read the Soviet Union’s enciphered weather reports. Each time the Soviets changed their key, it recovered the new one within hours or days and reported to the Minister of War. In 1940 it even reconstructed the key tables of the Moscow broadcast. Those tables are transcribed and checked here: they reproduce exactly the worked example attached to the report.

Dossier by Brice Dudout · transcription and checking with Claude · documents: JACAR archives, twelve reports of the 陸軍氣象部, 1938–1940 · written 28 October 2026

Why encipher the weather

A weather report is a string of figures with no obvious secret: pressure, temperature, wind, cloud. To an army, however, it is intelligence. Most of the weather systems that reach Manchuria and Japan come from the west, across Siberia. Whoever reads the Soviet observations can better forecast the weather for his own air operations, and also learns which stations are working, and where.

Until 1938 the stations of the Soviet Far East broadcast their observations in clear, in the international synoptic code: groups of five figures, always in the same order. On 11 August 1938, the day the ceasefire ending the fighting at Lake Khasan, on the border of Korea and the USSR, came into force, they went over to cipher. Moscow, for its part, also enciphered the section of its broadcast covering European Russia.

On the other side, the 陸軍氣象部, the Japanese Army’s meteorological service, was listening. Its reports, classified “top secret”, went to Vice-Minister of War Tōjō Hideki, then to Ministers Itagaki Seishirō and Hata Shunroku, and finally to Tōjō again, minister from July 1940. They are signed by the two successive heads of the service, 新妻雄 until March 1940, then 三上喜三. Receipt stamps show them circulating in the minister’s secretariat, the Army Aeronautical Department, and the military affairs and defence sections.

Twelve reports

Twelve of these reports are held by the National Institute for Defense Studies (防衛研究所) in Tokyo and can be consulted online through JACAR (Japan Center for Asian Historical Records). Each announces a Soviet key change and gives the method for reading the new cipher.

Where the date of introduction is known, the report goes out the same day or within two days. The Far East system of February 1940 is the only one that held out for nearly a week (15 to 21 February).

Four generations of cipher

All these ciphers start from the same plain text: the international synoptic report, a line of five to seven five-figure groups per station. All rest on the same operation as well: a key is added or subtracted figure by figure, without carrying (9 + 3 gives 2). What changes from one generation to the next is how the key is produced.

1. August 1938, Far East. Groups 1 and 3 are swapped, then 2 and 5, and 3 is added to each figure (4 to the station index). The 7th group is always 00000. From September the additive becomes a group of five different figures (34563, then 75647, then 13524), changed every two to four weeks; only the 5th enciphered group receives a separate additive.
2. February 1940, Far East. The cipher moves to eleven symbols: the ten figures and the cross ×, which marks missing data. The first group is enciphered with a substitution table; once deciphered, it serves as the key for the following groups of the same line.
3. February–June 1940, Moscow. In each group one figure is substituted, and that figure governs what is subtracted from the other four. Eleven pairs of values are therefore enough. The Soviets changed these values four times between late April and early June; each time the Japanese recovered them within one or two days.
4. June–October 1940, Moscow. A true key table, described below.

One detail betrays the Soviet design: in all these keys, from 1938 to 1940, the figure 0 almost never appears (once only, in one of the June tables). A 0 would let the plain text through unchanged.

The Moscow tables

On 10 June 1940 Moscow changed system, and again on 26 June. The report of 16 July gives the method in four operations:

1. find the key group in the broadcast: three times the same figure, then twice the same figure (for example 55533);
2. cut what follows into groups of five: five groups for the 01:00 and 13:00 local-time observations, six for those at 07:00 and 19:00;
3. replace 7 with X and X with 7;
4. in the table of the day, take the row given by the first three figures of the key and the column given by the last two, and from that point subtract, without carrying, the run of figures in the table. The result is the report in clear.

Two tables are in force at once: one for odd days, one for even days, according to the date in Moscow. The report encloses four of them (two for 10–25 June, two for the period beginning 26 June), copied by hand on large sheets. Each table has eleven rows (000, 111 … 999, XXX) and ten columns (00, 11 … 99); a row holds 109 figures.

Transcribing them shows what the Japanese had understood: the eleven rows of a table are the same sequence of 109 figures, merely shifted. The key group, sent in clear in the broadcast, only says where to start in this cycle. Each station enciphered with a different key yields another fragment of the same cycle; the fragments overlap, and a few days or weeks of listening are enough to rebuild the whole: the tables of 26 June are complete in the report of 16 July.

The example of 6 July 1940, redone

The report gives an example: two stations received on 6 July 1940 at 04:00 Japan time. A marginal note explains that in Moscow it is still the 5th, an odd day, so table 3 applies. The key is 22244: row 222, column 44.

First figure: X becomes 7, and 7 − 9 without carrying gives 8. The key for station 2 continues that of station 1 in the cycle. We transcribed the tables before touching the example, then drew the key from the transcribed tables: it is identical, figure for figure, to the one in the report, and the decipherment gives back both Japanese clear texts (60 figures out of 60). The example and the tables are on different images of the file, so the test does not reuse what served for the transcription.

The clear text reads with the international code: station 841, cloud, present weather, wind, pressure, temperature… We have not yet been able to identify these stations, for lack of a Soviet station list of 1940.

What these documents tell us

The Moscow system of February–June 1940 is not unknown. Brian Audric, of the Met Office’s IDA unit at Dunstable, described it in 2000 with an example: it is the “Russian code” that Bletchley Park and Dunstable were reading in 1940. The Japanese reports describe the same system, with the substituted figure in the same position, and Audric’s example, with the key he gives, deciphers by the same method. Two services with no known link between them were thus attacking the same traffic at the same time.

What we have found nowhere, in Japanese or in English, is the Japanese work itself: the continuous reading by the Army meteorological service, report after report, from 1938 to 1940, and above all the Moscow key tables reconstructed in 1940. Studies of Japanese intelligence against the USSR deal with Red Army ciphers, not with the weather. To our knowledge, these tables are published here for the first time.

The documents also show the pace of the duel: in a little over two years, a dozen key changes and four successive systems, each more elaborate, and each read within hours or days. The weakness is always the same: an entirely predictable plain text (known station indexes, values in narrow ranges, a final 00000 group, observations at the same hours as those of Japanese and Manchurian stations), and a cipher that keeps the structure of the report.

What remains uncertain

• The June–July 1940 tables are secure: of 4,796 figures transcribed, none disagrees with the cyclic structure, and only 3 illegible figures had to be restored from it. The two October 1940 tables, written in a cursive hand and blurred on the photograph, give only one figure per cell; they cannot be checked in the same way, and the odd-day table has one figure too many. Brice Dudout re-read on the image eight of the cells read as “5”: all are indeed 5s, so the discrepancy lies in another cell or in a 2.
• Re-read on the image by Brice Dudout: the 5th additive of 12 November 1938 is 34235 (not 34135); the subtractor cell for cipher figure 6 on 29 April 1940, badly stamped, reads more probably as 8 (6 remains possible).
• The tables could be tested only on the examples in the reports themselves. We found no enciphered Soviet broadcast of 1940 online; the British files on Russian weather ciphers (The National Archives, HW 54/4 and HW 40/31) are not digitised.
• The search for earlier work covered what is accessible online. Still to be consulted: Ken Kotani, Japanese Intelligence in World War II; the official war history series (戦史叢書); and a US bulletin of 1945 (CINCPAC-CINCPOA 5-45) in which a captured Japanese meteorologist reportedly spoke of the breaking of Allied weather codes.

References

1. Archives

   National Institute for Defense Studies (防衛研究所), Tokyo, via JACAR (jacar.go.jp): Ref. C01004558500, C01004559200, C04120597300, C01004559500, C01004816800, C01004816500, C01004816400, C01004816900, C01004816700, C01004832500, C01004832600 (example: image 4; tables: images 6–7 and 9–10), C01004818600 (tables: image 2). The JACAR images, all rights reserved, are not reproduced; tables and example are transcribed.

2. British testimony

   Brian Audric, The Meteorological Office Dunstable and the IDA Unit in World War II, Royal Meteorological Society, Occasional Papers on Meteorological History no. 2, 2000.

3. British archives

   The National Archives, Kew, HW 54/4, “Correspondence on Russian meteorological cyphers” (1939–1940); HW 40/31, “Exploitation of Russian meteorological communications by Axis Powers” (1944–1945).

4. Context

   宮杉浩泰 (Miyasugi Hiroyasu), “島内志剛日記にみる対ソ通信情報活動”, Intelligence, no. 22, on the Japanese Army’s breaking of Red Army ciphers in 1939–1941.