This is a comparison between the new PredictWind weather router and seven other widely used routers.

Weather routing with a grid based, highly optimized algorithm

Uses high resolution grib data (wind and currents) when- and wherever available

Provides both optimal weather routing and comfort routing

Advanced boat polar inter- and extrapolation

Allows a course to have both waypoints and roundings

Multithreaded execution with multiple weather sources including the ECMWF ensemble day 34 day forecast

For each weather source, display of a corridor within which the “Estimated Time En route” (ETE) can be near to optimal

Display of the overlap of such corridors to show passages where the ensemble of weather forecasts agree that it is optimal to pass

Numerous data available along the route: wind, current, waves, speed, distances, angles, times, air pressure, temp, rain, CAPE index

Continuous interpolation of these data along the route

Presentation also of motoring time in comfort routing

A sample routing result is shown below. The top graph shows the optimal / fastest route for a Beneteau First 40.7 from Tampa (Florida) to Cancun (Mexico), considering wind (but not current) with four different wind forecast sources (PWG, PWE, ECMWF and GFS). The blue areas around the optimal routes represent alternate points you might pass without more total time loss than 0.1%. Where these areas for different forecasts overlap, there is forecast consensus that these are good points to pass. The more consensus in the ensemble of forecasts, the darker blue the area.

The lower graph is the same routing display, but this time also currents are considered (Mercator forecast with 0.1 degree resolution). The optimal routes and the optimal corridor are now narrower and moved a little to the West to avoid the Gulf stream counterflow.

The new PredictWind weather router has been compared with the weather routers of Adrena, Expedition, Fastseas, OpenCPN, Squid, and the original PredictWind router. Fifteen different courses were used. A Volvo 65 boat polar was used for four courses, TP52 for five, and First 40.7 for six. Currents and waves were not considered in these benchmarking test cases. GFS wind forecasts with 0.25 degree resolution and 14 days duration were used by all the routers (except Fastseas, which uses GFS 1 degree) for the comparisons. Comfort routing was not used. The tests were run during May-August, 2020.

In each case, the start and finish points were set well off land, not to disqualify routers with poor land avoidance. To avoid bias, the courses were selected before running any tests, and then not changed. No routings were excluded, or rerun, because of results. Hence, the results should be independent of any bias.

Weather routing is a complex, meta heuristic process. No router can be expected to be the best in all cases. In each case, the fastest among the routes was ranked the most optimal and for each other route the percentage ETE (Estimated Time En route) above the optimal was computed. The router that generated the biggest number of optimal routes, and the smallest mean ETE above optimal, was judged to be the best. In order to include failures in the evaluation, a ranking was also done, where a failure was deemed inferior to finding a route. In each case, the router with the shortest ETE was given rank 1, next to the shortest rank 2 etc. If one router failed it was given rank 8 for that case, if two failed they got rank 7.5 etc. For each router the mean was taken for these ranks over all cases. Please see the ‘Mean rank’ in the results table below.

In order to eliminate time-en-route differences which solely depend on the different ways the various routers do their internal calculations, a “simulation” was done for each route. The ETE was computed for the same start time, the same wind grib forecast (GFS 0.25 degree), identical modes of interpolation of grib data, the same boat polar and the same techniques to inter- and extrapolate boat polar data. Still, the comparison between these simulated ETEs may not be fair, since when time shifts are introduced along the route, the individual router could have chosen another more optimal route according to a dislocated wind forecast. To make comparisons as fair as possible, both the ETE estimated by the router itself and the simulated ETE were used, with the same importance, i.e, with fifteen different courses there were 30 sets of ETEs to compare.

Standard settings, with no currents and no motoring, were used for all the routers. For isochronal routing, the standard timestep setting was one hour.

For Adrena, the Pro edition was used. The same set of grib files as used by the PredictWind routers was used.

Also for Expedition, the same set of wind grib files as for PredictWind and Adrena was used. Routing was primarily run with ‘Avoid vector chart land’ and ‘Avoid ENC land set, but without considering CMap (to get reasonable response times). Land avoidance problems were very common. Applying CMap land avoidance seldom helped. Instead, whenever necessary, land to avoid was manually specified as “race note areas”. Those were always respected by the grid-based, and mostly by the isochronal router.

Fastseas uses the GFS 1 degree resolution wind forecasts (there is no other option).

OpenCPN can use only one wind grib file. The GFS 0.25 degree resolution, 3 hour timestep, 14 days file was downloaded via Saildocs.
For Squid, the GFS 25 km wind forecast was also used.

Results are summarized in this table, followed by some general comments, and then the graphs with some comments for each case. “Orig PW” is used to mean the old PredictWind router, “Exp grid” for Expedition with a grid based method, “Exp iso” ditto with an isochronal method.

No router was the best in all cases, but only the new PredictWind router had no routing failures, always avoided land, and was the leader in finding the fastest routes.

Generally, weather routing is volatile and unreliable when there are wind speeds below some 5 knots, making results of comparisons fairly random in those cases. When so many as eight routers are compared it is very hard to be better than all in more than 50% of the cases. If one router is better than each one of the others in 91% of the cases it will (as a mean) be better than all other routers in 50% of the cases. Similarly to the fact that the best sailor does not win all races.

Land avoidance was a major problem in routing with Expedition. Specification of large detailed “race note areas” (this process takes a long time) most often helped the grid-based router, but not seldom made the isochronal router go from creating a route crossing land to failing to find a route.

The 15 routing cases are shown in the following graphs. Short comments are included after each graph. There is also a table of the own and simulated ETEs of each router, ranked in order of increasing average own / simulated ETE.

Case 1. Transpac, from Los Angeles to Honolulu. Boat polar: TP52.

Expedition isochronal failed crossing land.

Case 2. From Heraklion to Palma Mallorca. Boat polar: TP52.

Expedition isochronal failed when ‘race note areas’ were specified to avoid land crossings.

Case 3. From Cape Verde to Barbados. Boat polar: Volvo 65.

Case 4. From Cape Town to Melbourne. Boat polar: Volvo 65.

OpenCPN failed. Fastseas took the shortest and (in this case) slowest route.

Case 5. Chicago Mac (from Chicago to Mackinac island). Boat polar: TP52.

Case 6. Sydney Hobart race. Boat polar: TP52.

Both Expedition routers failed despite careful specification of “race note areas” for mainland Australia and Tasmania. Adrena also failed. The Squid route goes cross land.

Case 7. Ocean race leg: from Hague to Genoa (with a waypoint in the Gibraltar strait). Boat polar: Volvo 65.

Original PredictWind, Adrena, Expedition isochronal, Fastseas and OpenCPN failed. Expedition isochronal route got into NW African land despite activated map avoidance and “race note area” specification.

Case 8. From Alicante to Cape Verde. Boat polar: Volvo 65.

OpenCPN and original PredictWind routers failed.

The All New PredictWind Weather Router