Dark Sky Temperature Jumps

Does anyone know why my dark sky temperature sensor has a lot of jumps in the data and isn’t smoother?

Do other people see this in their data?

image

What is your refresh_interval set to? In the attached picture, I am using the Dark Sky temperature sensor set to 300 seconds. At every 10 minutes I do get a bit of a jaggy graph but it is certainly usable. It also appears that you are using ºF Mean. Are you post processing the data before graphing?

Here is the lovelace card settings for the attached pic.

  • type: history-graph
    title: ‘Temperature Graph’
    refresh_interval: 300
    hours_to_show: 24
    entities:
    • entity: sensor.dark_sky_temperature
      name: Outside Temp
    • entity: sensor.kit_temp_sensor_226
      name: Kitchen
    • entity: sensor.tstat1_temp
      name: Downstairs
    • entity: sensor.tstat2_temp
      name: Upstairs
      temp_graph

I didn’t intend to do mean. But it looks similar in my lovelace card.

Below is my lovelace card. It just looks odd that there are jumps up and then down by so much, it isn’t smooth like it would be in reality. Parts of the curve are but others not…

Below is my configuration

Weather prediction

sensor:

  • platform: yr
  • platform: darksky
    api_key: !secret darksky_api_key
    forecast:
    • 0
      monitored_conditions:
    • temperature

I see it too, not just on temperature, but on relative humidity as well. The steps seem to be aligned between the two measurements. I’ve poked around the 'net, but I have yet to find any explanation.

I know there isn’t a fix but I am using the filter sensor to smooth out the data, seems to work decently.

It appears DarkSky uses some sort of interpolation (perhaps spline fitting) to estimate weather parameters between measurement sites. I’m guessing the update intervals cause the jumps. Personally, it’s not too bothersome because it’s just an artifact of what DarkSky does to get hyper-local current and forecast values.

If you do filtering, e.g. a moving average, there will be a time displacement of filtered data depending upon the sample interval. You could shift the data at plot time to account for that, but you wouldn’t have a full set of data for the most recent interval.