@franzstein, that is a good summary. Currently OS supports OWM using forecast data for the next 24 hours as the basis for a Zimmerman calculation as the standard service. But for users that are prepared to host a local Weather Service within their own network then other options become available. Firstly, a DarkSky option is in pilot that uses the last 24 hours as the basis for the Zimmerman calc. For users with a WU compatible PWS then the local Weather Service option can use that to calculate Zimmerman based on yesterday’s temp/humidity/precip plus today’s precip. The later option is the most closely aligned to the old WU method. So you can see that there are a few options based on slightly different approaches. This is all new work as a result of the recent WU changes and is going through a bit of revision and refinement so things may evolve
To break it down a bit more. There are basically two variables here; 1) the choice of weather provider, and 2) the “window” of data used. Unfortunately nothing is simple in this world and the best option is very dependent on the user’s situation.
In terms of the weather provider, then the choice goes along the lines of: 1) the PWS in your backyard is generally the best option, 2) the next best choice would be a nearby PWS but unfortunately the change in WU api has ruled this out, 3) then we need to look at an online weather provider that best matches your local conditions i.e. either OWM and DarkSky. For some OWM will provide better results but for others DarkSky may be more representative so a bit of research is needed. The last option, 4) is to go old-school and revert to manual watering levels!
In terms of the “window” of data used i.e. historic or forecast, then that is harder to answer. The theory goes that in a perfect world using forecast weather is better as it avoids stressing the plants. To put this a different way, the forecast approach is about “giving the plants enough water to see them through an upcoming dry patch” whereas the historic approach is about “giving the plants a load of water that they have been desperate for all day”. Unfortunately, the logic breaks down a bit as data is not perfect. Unfortunately, historic data is pretty accurate whereas forecast data is prone to error. So whilst the forecast approach may be better in principle it may not be better in practice. It comes down to the sophistication of the weather provider’s algorithms and their source information.
So at the moment there are a few options available to users but the best one is probably down to individual circumstances and requires a bit of trial and error.