← Trozovka (John Winsome Munar)

Weather by Email at Sea

How I get a 10-day weather forecast on a ship that only has company email

By John Winsome Munar (Trozovka), Master Mariner · November 2026 · Also on GitHub

This is a write-up of how I solved a real problem on board: getting a useful 10-day forecast with isobars and wind, when the weather fax is broken, the charterer's weather service is unreliable, and the only link to the outside world is the ship's company email.

I built a working system for my own use. The tool itself stays private. This page explains the ideas behind it, in enough detail that another seafarer (or a developer helping one) can understand every part and build their own.

Contents

  1. The problem
  2. What I needed
  3. Why GRIB files
  4. Where the forecast comes from
  5. Try the free service first: Saildocs
  6. My setup in one picture
  7. The mailbox is the memory
  8. Commands by email, and keeping strangers out
  9. A forecast area that follows the ship
  10. Keeping the files small
  11. Tropical cyclones
  12. Weather fax style charts
  13. Getting it through the ship's email
  14. Sending orders from the ship
  15. Reading it with no internet
  16. Living on 300 MB a week
  17. Cost and limits
  18. What I learned
  19. Disclaimer

1. The problem

I command a tramp ship. That means no fixed route: the next port can change while we are in the middle of an ocean. Good weather planning matters more on a tramp than on a liner, because I often have to decide a route a week ahead, and change it when a low or a typhoon is coming.

On board I had:

Apps like Windy and Ventusky are excellent, but they need a real internet connection. On my ship they were useless.

2. What I needed

I wrote the requirements down before writing any code:

3. Why GRIB files

GRIB is the standard file format for weather model data. A GRIB file is not a picture. It is a grid of numbers: pressure and wind at every point, for every forecast hour.

That sounds less friendly than a picture, but for a ship it is much better:

GRIB file Picture (screenshot or fax)
Size for 10 days about 100 KB 30 to 1000 KB per picture
Forecast times every 12 or 24 hours, with a slider one moment per picture
Zoom, units zoom anywhere, Beaufort or knots fixed
Works offline yes, once received yes

You open GRIB files in free chart software such as OpenCPN (with its GRIB plug-in) or XyGrib. These show isobars, wind barbs and a time slider, all offline.

Later I also added black-and-white chart pictures, because a picture can be read on a phone in two seconds. More on that in section 12.

4. Where the forecast comes from

Two weather models are free and allowed for this use:

Why two models matter. At first I compared my own weather app with an email service, and then noticed both used GFS. That is not a real comparison: if GFS is wrong about a typhoon track, both are wrong together. A true second opinion needs a different model. When GFS and ECMWF agree, I trust the forecast more. When they disagree, I plan with extra margin. Professional weather routers work the same way.

The fields I use are only three: mean sea level pressure (for isobars) and the two wind components at 10 metres (for wind speed and direction).

5. Try the free service first: Saildocs

Before building anything, try Saildocs. It is a free service run since 1999 by the people behind Sailmail. You send a short request by email and it replies with a GRIB file. It can also send a forecast every day by itself (a subscription).

A request looks like this (sent as plain text, signature removed):

send gfs:0N,30S,30W,10E|1,1|0,12..240|PRMSL,WIND

That means: GFS model, a box from 0° to 30°S and 30°W to 10°E, a 1-degree grid, every 12 hours out to 240 hours (10 days), pressure and wind. To get one every day, start the line with sub instead of send.

Saildocs' website shows a "not secure" warning in the browser because it uses plain http. That has nothing to do with the email service and you never give a password there.

For many ships, Saildocs is enough. I built my own because I wanted three things it does not do: a forecast area that moves with my passage plan, two models side by side, and several recipients (ship email and my own email) from one setup.

6. My setup in one picture

 Me (phone or laptop)
   │  email: passage plan, position, commands
   ▼
 A mailbox used only by the program
   ▲
   │  reads commands, sends results
 A small program run on a free schedule in the cloud
   │  every 3 hours: answer new commands
   │  once a day:   download GFS + ECMWF, cut my area,
   │                list tropical cyclones, draw charts
   ▼
 Email to my recipients: ship email + my own email
   ├─ Forecast email: GRIB files + text
   └─ Charts email:   black-and-white pictures

Nothing runs on the ship except email. Nothing runs at home either: the program runs on a free scheduled job in the cloud (I use GitHub Actions, in a private repository). My laptop can be switched off for months.

The program is written in Python. It uses the ecCodes library (from ECMWF) to read and write GRIB files, and matplotlib with Cartopy to draw the charts.

7. The mailbox is the memory

The program keeps no database. Every time it runs, it reads all recent command emails in its mailbox, oldest first, and replays them:

At the end of the replay it knows the current state. This has two big advantages:

  1. No ship position is stored anywhere in the code. The code repository could be made public by mistake one day and there would still be nothing in it about where my ship is or was. In risky waters this matters.
  2. Nothing to break. There is no state file to corrupt. If something goes wrong, the next run rebuilds everything from the emails.

After a command email is answered, the program puts a label on it in the mailbox, so it sends only one confirmation per command.

8. Commands by email, and keeping strangers out

The commands are plain lines of text. For example:

GRIBPOST PLAN
WP 14-35.2S 048-12.7W 2026-12-01 06:00
WP 34-21.0S 018-28.0E 2026-12-10 12:00
TO captain@example.com
END
CODE xxxxxx
GRIBPOST POS 12-15.4S 080-02.1E 2026-12-05 08:00
CODE xxxxxx

Positions are written the way a navigator writes them: degrees, then minutes with a decimal, then the hemisphere. All times are UTC. I never type these by hand: a planner page writes them for me (section 14).

The program ignores lines it does not understand, and it ignores quoted text, so company disclaimers, signatures and old replies do not confuse it.

Keeping strangers out. Anyone who learns the program's address could send it a fake plan. Two checks stop that:

  1. Allowed senders. Only addresses on an allow list are obeyed. The list starts with my own email. I can add the ship's email by sending an "allow" command from an address that is already allowed.
  2. A command code. Every command email must contain a secret code line. Email "From" addresses can be faked; the code cannot be guessed.

If an allowed sender gets the code wrong, the program replies and says why ("no code line found" or "code does not match"). Strangers get no reply at all.

The mailbox password. The program logs in to its mailbox with an app password (Gmail calls it that). An app password opens the whole mailbox, so the program must have a mailbox of its own that holds nothing else. Never give a program the password of your personal email. Two traps I hit:

The passwords live in the cloud service's secret store, never in the code.

9. A forecast area that follows the ship

This is the part that makes the forecast useful on a tramp ship.

The passage plan. Before departure I send the waypoints with ETAs. To save typing, the planner can fill in all ETAs from a departure time and a planned speed, using great-circle distances between waypoints. It can also import a route file exported from OpenCPN (GPX) or from an ECDIS (RTZ).

Where is the ship now? Between waypoints the program interpolates the position by time. That gives a planned position for any moment.

Correcting with a position report. Ships are early or late. When I send a position, the program finds the closest point on the planned route, looks up the time the plan expected the ship to be there, and works out the difference:

delay = time of my report − planned time at the closest point on the route

It then shifts the whole remaining schedule by that delay. If I report being two days late at a waypoint, every later ETA moves two days. If I skip reporting for a few days, the plan alone still gives a reasonable estimate.

How often to report. The plan gives a dead-reckoning style estimate: position by time along the route. Because the forecast box has 8° to 10° of margin and covers three days ahead, an estimate that is a few hours off doesn't change the forecast. My routine:

Situation What I send
Before departure The passage plan (once)
Normal passage, on schedule Noon position once a day if convenient, at least every 2 to 3 days
More than about 6 hours ahead of or behind the plan A position report straight away
Route changed (new port, weather diversion) A new passage plan, not just a position

A position report only corrects the timing along the planned track. If the ship leaves that track, the forecast box must follow a new route, so a new plan is the right message. Each report is two lines and a few hundred bytes.

The forecast box. Each day the program takes the estimated track for the next 72 hours, draws a box around it, and adds margin: 8° of latitude and 10° of longitude on each side, so weather systems are seen before they arrive. The box is kept at least 20° by 25° and at most about 40° by 50°, and is snapped to whole degrees.

The date line. A route that crosses 180° must not turn into a box that wraps the whole world. Longitudes are "unwrapped" along the track (so 179°E is followed by 181°E instead of 179°W), the box is computed, and only at the end converted back. Most GRIB data is stored from 0° to 360°, while navigators use 180°W to 180°E, so this conversion is a classic source of bugs. I tested it on purpose with boxes across the date line.

Once a week, and when a new plan arrives, a second, coarser file covers the whole remaining route (2° grid), so I also see the big picture.

10. Keeping the files small

Size decides whether the forecast gets through a ship's email at all. What I do:

Typical sizes per day:

File Size
GFS, my area, 10 days about 110 KB
ECMWF, my area, 10 days about 75 KB
ECMWF, whole route, 2° (weekly) about 40 KB
Four chart pictures about 160 KB

I also tried zipping the GRIB files (bzip2). It saved almost nothing, because GRIB is already packed tightly. Not worth the trouble for the viewer.

11. Tropical cyclones

At the top of every forecast email there is a short tropical cyclone check:

This is a heads-up, not a warning service. Official tropical cyclone warnings still come from GMDSS: SafetyNET and NAVTEX.

12. Weather fax style charts

GRIB files need chart software. Sometimes I just want to look at the weather on my phone in two seconds, like a weather fax. So the program also draws black-and-white surface charts and sends them in a separate email.

Each chart page has two panels, one above the other, which fits a phone screen. There are 4 pages a day: analysis and +24 hours, +48 and +72, +96 and +120, +168 and +240 (day 7 and day 10).

What makes them useful:

The title of every panel gives the model, the forecast hour, the valid time and the estimated ship position, so a page is never confused with another day.

13. Getting it through the ship's email

Ship email systems often have size limits and filters. Some things that help:

14. Sending orders from the ship

At sea my laptop has no internet. The ship's company email runs on the bridge PC, and my phone has a small Starlink allowance. So the tool to write commands must work completely offline.

The planner page. A single HTML file that runs in any browser with no internet. It has:

It checks the input (minutes must be below 60, ETAs must go forward in time, and so on) and then either opens the email app with the message ready, or copies the message so I can paste it into any email program, including the company email on the bridge PC.

The phone app. On Android I wrapped the same page in a tiny app. It requests no permissions at all, not even internet. It only hands the finished message to the phone's email app (or the clipboard). The message is a few KB, so sending it over Starlink costs almost nothing.

The command code and the draft plan are saved only on that device.

15. Reading it with no internet

A common misunderstanding: OpenCPN does not need internet to show GRIB files. It needs internet only to download them, and I get them by email instead.

At sea I:

  1. open the forecast email on the phone and save the attachments,
  2. copy them to the laptop with a USB cable,
  3. open them in OpenCPN or XyGrib, offline.

The chart pictures I simply look at on the phone.

16. Living on 300 MB a week

A tiny forecast is pointless if the phone burns the whole data allowance in the background. On my Android phone I use a firewall app (Rethink DNS + Firewall, which needs no root) with two modes:

Two lessons from setting this up:

The daily forecast plus charts is about 300 to 400 KB, roughly 2.5 MB a week: under 1% of a 300 MB allowance.

17. Cost and limits

Cost: nothing. The weather data is free, the scheduled cloud job is within the free allowance of a private repository (a daily run takes about two minutes), and a normal free email account is the mailbox.

Limits:

18. What I learned

19. Disclaimer

This is how I solved my own problem. It is not a certified weather service and not a replacement for official forecasts and warnings (GMDSS, SafetyNET, NAVTEX) or for a professional weather routing service where one is required. Always use your own judgement as a seafarer.

Data credits: NOAA/NCEP GFS (public domain). ECMWF open data, CC BY 4.0. GDACS (UN/EC). NHC (NOAA).


About the author. John Winsome Munar (Trozovka) is a Master Mariner who builds practical software for seafarers, including WX weather apps for slow ship internet, PocketMOB Alert (man-overboard alarm), PocketVDR (voyage data recorder) and OpenCPN GPS Server. More at trozovka.github.io.

Text © 2026 John Winsome Munar. Shared under CC BY 4.0: you may share and adapt it with credit.