I was hoping to find out the frequency for when the daily/hourly forecasts are updated. I see a request last year by @rmerigan1 in the feature request section: Time of forecast - but it went unanswered. Is it three times a day? I don’t see an update epoch time in the API. Hoping to determine how often I should persist the results if they’re not changing that often.
Thanks!
_DS
if you gather enough data, you should be able to figure it out. It might differ per region (I don’t know, but it might)
Yep - I have plenty of data - but it’s not reflecting the “3 times a day” comment - nor can I see any regular demarcation lines where a regular update to something like temperature happens:
(Also seeing that a “low 0-day forecast” that jumps between 50 degrees and 42 degrees doesn’t seem like a very good forecast.)
That’s just the daily forecasts. I was specifically trying to determine the hourly forecasts as the difference in computing 10 dailies is different than 240 hourlies.
If I had to guess - I’d say something happens around 3:45 am (local) and 2:45 pm (local) - but it’s not consistent. I’ve also only focused on 0-day forecasts for this secondary review.
If I look at, for example, a 4-day forecast - still no regular changes that I can pick up.
Thanks!
I’m not sure what I’m looking at. Are you plotting the hourly forecast for some specific time, or is it the hourly forecast 1 hour in the future (that would shift with time).
anyway it changes a lot apparently.
It might be that the data on which the forecast is based, is in fact continuously adjusted by the measured value. Do you see any correlation with that?
I poll the forecast API every minute to get the current conditions. Because that response includes everything - I have access to all of the daily and hourly forecast segments (which I want anyway). The graphs here show the 0-day forecast from the JSON object. Specifically:
forecast.daily[0].air_temp_high
forecast.daily[0].air_temp_low
Using [1,2,3,4, etc] will let me see the other day’s future forecast. This timeline here shows when I pull the data - which is every 60 seconds. So at that time on the graph, that is when WeatherFlow is telling me the forecast for that specific day (0) or for a future day (1-9). (Easier to visually represent a single day’s change in the forecast to see data update frequencies than trying to show the hourly changes - but I did that first).
Because I’m populating my own dashboards with historical data (and since I can’t ask the API for historical forecast data) I store it myself and was curious what a comparison of deviations in forecast quality would look like.
I also use it to reimport stored JSON logs if I happen to want to do something else with the data in InfluxDB that requires resetting the datastore - or testing variations of my dashboards.
The difference here is that the epoch times for the hourly forecasts certainly are “static” but the data is changing (apparently) and so I’m trying to understand when I should overwrite my existing forecast data before time reaches “now” for each of the forecasted days/hours. Every time I rerun an hourly analysis - which data do I care about? The data that was forecasted at midnight? at 4 pm? at 11:59 pm? 
If it’s 11:59 pm - it makes no sense to run it at any other moment of the day as the time periods get overwritten (for my log importing use case - it makes sense for future forecast updates which I do every 15 minutes anyway).
the forecast itself is a slow calculation that perhaps is only run a few times per day. But to me it makes sense to somehow correct that forecast a bit, based on the difference between the measured temperature and the forecasted one.
it would be interesting to see, as an example, starting at 08.00 in the morning, the forecast for the temperature at 18.00 that day, and see how that changes during the day. With next to it the actual measured change in the temperature during the day. If the forecast has relatively fast changes, it might be because they do a correction to the forecast based on the current temperature and how far the current temperature is off from the forecast temperature for the current time.
You might see some correlation. You might not see a correlation, but they still might adjust the forecast based on the current temperature, when the forecast is actually a continuous curve and they do the adjustment based on the difference between the current temperature and the continuous curve.
I have no idea, but it would be nice if you could figure it out.
My first example shows that use case specifically. I show the 9 days of forecasts aligned with the actuals but I’m only running the hourly forecasts through at two points during the day. What is collected at noon and midnight. And they’re certainly different forecasts for the same day.
Can someone from WeatheFlow shed some light on how often and when the forecasts are updated?
Thanks!
I think what I’m asking for is a graph of a specific, fixed time in the feature and see when that forecast changes.
and below that a graph of difference between the current temperature, and the predicted temperature for “now” or close to “now”. If there is a correlation, than great, if there isn’t any, than we still don’t know.