ForecastMetrics

ForecastMetrics is a tool to store forecast data from multiple sources in VictoriaMetrics or InfluxDB.

Features:

• Hourly forecast updates written to Influx or VictoriaMetrics for multiple locations
• Http server implementing the prometheus query_range endpoint, allowing it to be used as a Prometheus data source for adhoc forecasts in Grafana.

I currently use VictoriaMetrics as my time series database. Because of that, this project does a few things specifically to support it:

• Writes hourly sunup information
• Uses no retention policies (not supported in VictoriaMetrics)
• No overwrite of metrics
  • Every forecast is a new tag/label, since VictoriaMetrics doesn't support overwriting metrics as Influx does.
  • Past data is written one data point per hour, also because overwriting data is unsupported.
  • An alternative mode can be enabled by setting overwrite_data to true in the config file. In this mode, there will only be one forecast series per source/location.

Currently supported sources:

• National Weather Service (NWS) (US-only)
• VisualCrossing (Global)
• No other sources planned at this time, due to not meeting the below criteria (7 day hourly forecast, reasonably priced or free)
• Open an issue if you find a worthy source!

Usage:

Install

• Download a binary from the latest Release if your architecture is available

  curl -O https://github.com/tedpearson/ForecastMetrics/releases/download/v4.1.0/forecastmetrics-linux-arm64
  

• Make the binary executable

  chmod +x forecastmetrics-linux-arm64
  

• If your architecture is not avaialable, you'll need to build from source:

  • Clone this repo
  • Install Go

  • cd ForecastMetrics
    go build
    

Config

• Get the example application and location configs:

  curl https://raw.githubusercontent.com/tedpearson/ForecastMetrics/master/forecastmetrics.example.yaml > forecastmetrics.yaml
  curl https://raw.githubusercontent.com/tedpearson/ForecastMetrics/master/locations.example.yaml > locations.yaml
  

• Modify the configs with your own values for:

  • locations
  • influxdb/victoriametrics connection
    • if using influxdb, you may set overwrite_data to true, creating only a single series for each source/location.
  • desired influx measurement names (metrics prefixes for victoriametrics)
  • which weather sources to enable
  • add your own key for Visualcrossing, if desired
  • server config for ad-hoc forecasts:
    • Set the port the server should listen on (set to 0 to disable the server)
    • Insert your own Azure Maps Shared Key (requires Azure Maps account. There is a free tier.)

Ad-hoc Forecasts Setup

Since version 4.0, ForecastMetrics supports use as a prometheus data source in grafana for getting ad-hoc weather forecasts for any location.

• Only simple queries are supported, no functions or other features
• Queries should look like this: forecast_metricname{source="nws",location="place"}
• The location tag supports these formats:
  • place name
  • lat,lon
  • place name|nickname
  • lat,lon|nickname
• An optional tag save is also supported. if save="true", ForecastMetrics will add it to locations.yaml and update the metric every hour.
  • The locations.yaml file needs to be writable by the user running the process for this to work.
• To add as a data source to Grafana, add as a Prometheus data source. When you save, there will be an error about "404 Not Found - There was an error returned querying the Prometheus API." You can ignore this error and proceed to configuring a dashboard.

Run

Run the binary like this:

./forecastmetrics --config forecastmetrics.yaml --locations locations.yaml

Grafana Dashboard

I've included definitions for my grafana dashboard in the repo, both for InfluxDB and VictoriaMetrics which I now use. Here are screenshots of each in use. I use this dashboard daily for my local weather forecast.

Influx Dashboard

VictoriaMetrics Dashboard

Rationale behind included/planned sources:

I was looking for a replacement for DarkSky, who were bought by Apple and retired their API in 2021 2022 2023. DarkSky had the best forecasts and a generous free version, with 7 days of forecast data available.

I used the DarkSky data to power my own visualizations of my local forecast in Grafana. I find my Grafana graphs of forecast data much more intuitive than any weather app or website out there. I display the 7 day forecast for temps, precip, wind, and clouds, on the same graph with 7 days of actual data history from my Ambient Weather personal weather station, and also the forecast from 24 hours previous. (You can find my Ambient Weather exporter here.)

So when I went looking for replacements I needed these features:

• At least 7 days of HOURLY forecast data. Daily highs and lows are not very interesting to look at in a graph.
• I preferred Free APIs or APIs allowing at least 1500 forecasts per month, as I only made <200 calls/day to DarkSky, and paying large amounts for my personal forecast dashboard is just silly.
  • This is why visualcrossing is a supported source, because their free tier supports 250 forecasts/day.