on Long Term Forecasting

It’s a common understanding that weather forecasts beyond 3 days into the future are to be taken with a grain of salt. But how much salt is really necessary? I did some experimentation to check:

From the nearly four years of cruising we’ve had so far, I built the impression that forecasting has “gotten it right” for us far more often than it has missed. And I tend to reinforce this empirical impression with a logical thought. Borrowing an excellent expression from New Zealand’s weather guru MetBob : “weather is a mix of pattern and chaos”. As technology advances, so does our ability to collect, store and analyze data, thus better understanding the Pattern aspect of weather, and further enclosing the aleatory element of Chaos. Right? Well … that I can only hope for, and up to now all I had was my impressions gleaned from practical experience.

The exercise of planning for the passage between Fiji to New Zealand last year threw a new light on this matter for me, and I explain.

One key characteristic of this passage is that the worst weather conditions can happen upon the approach to New Zealand, and that is about 6 days after departure. Add to this the fact that it is necessary to allow for at least 2 full days prior to departure in Fiji to comply with clearing-out procedures, restocking etc. That means in practice that we needed a good-confidence forecast for over 10-days into the future. Quite a stretch.

With that in mind, and taking advantage that I was going to be looking at weather anyway, I decided to do some practical experimentation.



The first thing I did was to collect forecasts for three arbitrary spots more or less along the route between Fiji and New Zealand. The data is all from the GFS computer model, acquired via Saildocs.


The charts below show Surface Pressure, Wind Direction and Wind Speed forecasts for a point just south of Fiji on November 25th 2017 (the first point to the left on the diagram above). I started collecting this forecast 13 days in advance (the farthest I can get on GFS) up to the same day. Notice how conditions forecast 13 days before (1,006Hpa, wind from 97 degrees at 18 knots) were not too different from the actual fact, nearly 2 weeks later (1,011Hpa, 115 degrees, 21 knots). Also, looking at the trend lines, look how the forecasts start to converge nicely to the actual conditions.


Below, another view, this time for a point roughly mid-way between Fiji and New Zealand, also for November 25th 2017. And a similar pattern is to be seen: a reasonable “ballpark” forecast of conditions 13 days in advance, nicely converging to the actual ones, especially when getting 7 days and closer to the event.


Now broadening the analysis a little bit, for the next geographic location I looked at the forecast for a 24-hr window over it:


The charts above show weather conditions – pressure, wind speed and direction – for a location just North of New Zealand, during the period between November 25th 06:00AM and November 26th 06:00AM. The solid lines represent the average conditions during the period and the dotted lines show the maximum and minimum values. Note how consistent the forecast has remained, especially from about 9-10 days before the event.



But looking at forecasts for just a spot location can be too narrow, at least for the purpose of this analysis. So, here is another look into it.

The animation above shows the development of the forecast for the 500Mb layer over a macro-region covering Australia, the Tasman Sea, New Zealand and Fiji (look at this post for more information regarding 500mb charts) for the morning of November 20th 2017. Notice that even 16 days prior to the event, the forecast starts to hint that an Upper Level Ridge might establish over the region – bringing benign sailing conditions. The location and general aspect of this feature consolidates further about 8 days prior to the event and this forecast remains stable all the way thenceforth.

A similar story is reflected on the Surface Pressure forecast (above): solid hints of High Pressure establishing on the area between 15 and 9 days prior to the date, and then a well consolidated High is finally forecast over the area starting 8 days before the event, and remaining like that for the rest of the forecast period.



Again, the passage from Fiji to New Zealand is a typical case where a long-term forecast is necessary. Because storms can become violent near NZ, departure from Fiji needs to be planned according to expected weather conditions at arrival. Considering the 7 days necessary to sail there, plus another 2 or 3 to clear out of Fiji, the go / no-go decision has to be made with a 9-to-10 day forecast at hand.

Here is how it looked for us:


The graph above shows the expected wind directions along the passage, each line representing a different forecast. The lightest-dotted line is a forecast pulled 9 days before departure from Fiji, and the solid green line is the actual wind direction during the passage. With the exception of the first forecast, pulled 9 days before departure, all subsequent ones were in pretty good agreement to what happened in reality.

And here is the expected wind speeds along the passage.


With just one exception, all forecasts – pulled 9, 7, 5, 3, 1 days prior to departure – were very consistent with the actual wind speeds observed along the way.

That’s very good prediction, I would say.



As I was putting this post together, a question kept coming to mind: was I lucky for having caught a very stable weather period or region in doing this experiment? Are forecasts normally this consistent, particularly beyond 1 week? Is it only the GFS that produces such results?

The following picture shows Rain and Surface Pressure forecast for a broad area covering the South Atlantic Ocean for January 29th, 21:00 (Brazil Time), using both the GFS and European Models:


Most key weather features are already present 9 days before the event, on both models: the big High Pressure centered to the South East of Santa Helena, and the fronts South of Africa and South East of South America. There is, however, some “indecision” on both models regarding that small Low Pressure System – and the associated Front – just off the coast of Brazil (the swirling feature on the upper left corner of each map).

Now let’s take a look at the Winds forecast for the same area, same date:


Again, most main features already present 9 days before the event, but some fluctuation regarding the aspect of the Low and associated Front off the coast of Brazil, even 5 days before.

And now a closer look at that area on Brazil’s South East coast:


Both models decisively forecast rainy weather – and this is influence of that Low which we saw previously. By the way, I am writing this paragraph on the day of this forecast, and it has been indeed raining cats and dogs for the last 12 hours as … per … forecast !.

It’s been very rainy here the past 12 hours or so … and both models saw that coming. Nine days ago.


But notice how both models change their interpretation of the aspect of the area covered by rain.

This effect gets worse on the Winds forecast:


Now one can say this is pretty much all-over-the-place along the forecasting period and in between both models.

This variability hints there must be something else affecting the forecast in this area, making it difficult for both models to correctly predict that Low and associated fronts off the SE Coast of Brazil. Indeed – if memory serves me well (I grew up sailing in this area) – forecast here has always been less predictable than in the areas we have cruised recently.

With the above information, if I were to plan a passage in this area, I would do more research to understand underlying conditions that affect weather – and forecasting.



Personally, what I take from these experiments is that computer-generated forecasts can offer a reasonable long-term picture for passage planning, with some caveats:

– use of long-term forecasting information is only safe when there is reasonable understanding of the underlying aspects of weather in the area of interest. In other words, one needs to know what her/she is looking for when considering a long-term forecast

– test thoroughly before use. Start looking regularly at the models long before departure, to get a sense whether the models are predicting consistently for the area, or not

– the use of at least two models is a major factor in appreciating long-term forecasts. If they coincide, better are the odds the forecast will remain consistent. The opposite being true as well

– finally, some areas are notoriously tough for models’ prediction, especially when under the influence of specific weather systems (this is true, for instance, for French Polynesia when Fronts of Troughs are stalled over there, or Tonga/Fiji when the ITCZ is hovering over them. I also suspect something of the like must affect model predictability for the SE Coast of Brazil in certain circumstances).


Finally, I can’t emphasize enough the importance of a holistic approach when it comes to passage planning and forecasting. Information from computerized forecasting models, while important, can not be the only aspect considered in the process. Anecdotal information from those familiar with the area, man-made forecasts and analyses from official/ reputed weather sources, and even gut feeling contribute enormously to proper appreciation of the scenario and decision-making.



Well, it’s been a long-winded story, this post. And I hope it helps.

Happy Sailing, and THANK YOU for following us on this Journey !