Creating a simple weather application with Phoenix LiveView

In this article we will discuss our experience building an online weather application in Elixir using Phoenix LiveView. We created a real-time weather application that allows users to see the past, current, and forecast temperature and precipitation data for any UK postcode. The goals of building this app were:

to further familiarise ourselves with Phoenix LiveView
to investigate some of the libraries available for displaying graphs in LiveView

Our reason for displaying both temperature and precipitation data simultaneously was to test the capabilities of the libraries in question, as some rather complex graph configurations are required to display two lines on one y-axis (minimum and maximum temperature) and an additional line on a second, independent y-axis (precipitation).

We wrote our app’s front-end using a combination of simple HTML attributes and Phoenix LiveView’s built-in hooks to create dynamic behaviour without any JavaScript. For example, when the user inputs a postcode and submits the form, a new graph for the temperature and precipitation in the given area is instantly generated.

We investigated two libraries in order to generate our graphs: Contex and Vega-Lite. Contex is a simple library for generating charts and plots in Elixir by generating SVG output. Vega-Lite is a more general tool, a high-level language (based on Vega, a declarative language for creating interactive visualisations) with various functionality for different common graph types, where the visualisation is created as a JSON object.

Using APIs and creating an input form

Before we could work with any graph libraries, our first task was to use an open-source API to retrieve the weather data we required for the graph. We began by getting the current temperature for a fixed “default” postcode, to ensure it was working correctly. It was not difficult to find an API that suited our purposes. We soon came across Open-Meteo, which contained all the information we needed: the current temperature, the maximum and minimum temperature for a given day, the seven-day forecast, and the precipitation.

However, this API came with a limitation: it was only able to retrieve the weather for a given latitude and longitude, rather than for a given postcode. Due to this, we had to find a second open-source API, which could fetch the latitude and longitude for a given postcode. We ultimately landed on the API provided by postcodes.io for this purpose. We then followed this up with a call to Open-Meteo, so that when the application was fully set up inserting a postcode would seamlessly fetch the relevant temperature data for that location. The following code shows the function to retrieve this information and add it to the LiveView socket:

elixir
def assign_temp_and_location(socket, location) do
  {coordinates, admin_ward} = Weather.get_location(location)
  temperature = Weather.get_weather(coordinates)

  assign(socket,
    temperature: temperature,
    admin_ward: admin_ward
  )
end

Having achieved this, our next task was simple: creating an input field to allow the user to specify a postcode that we would then fetch the weather data for. We were able to implement this using a simple HTML form element which, upon submission, queries the APIs and generates a new graph based on the data received.

Creating a graph with Contex

At this stage we had a barebones implementation of the current temperature through an input field. The next step of the process was expanding this from a basic plaintext temperature display to a more detailed graph containing the forecast, maximum and minimum temperature, and precipitation. We needed to make use of an external library to accomplish this, and originally found Contex, which allows for the creation and rendering of SVG charts through Elixir. This initially seemed like the right call, as the Contex charts were neat and legible, and the code needed to create them was relatively simple:

elixir
defp assign_chart_svg(%{assigns: %{chart: chart, admin_ward: admin_ward}} = socket) do
  assign(socket,
    :chart_svg,
    Contex.Plot.new(700, 400, chart)
    |> Contex.Plot.titles("Daily maximum and minimum temperature in #{admin_ward}", "")
    |> Contex.Plot.axis_labels("Date", "Temperature")
    |> Contex.Plot.plot_options(%{legend_setting: :legend_right})
    |> Contex.Plot.to_svg()
  )
end

However, problems soon arose with attempting to use this library for our purposes. Firstly, Contex would not allow for the setting of custom intervals on a line graph, meaning when trying to include both the seven-day history and seven-day forecast the x-axis would be abbreviated, with an interval every two days instead of every day. Secondly, our desire to make use of multiple y-axes to display both the maximum and minimum temperature and the precipitation simultaneously was not possible.

Exacerbating this problem was Contex’s documentation, which was very limited, particularly for line charts, a relatively recent addition to the library. This meant that it was difficult to ascertain whether there were solutions to our problems. Unable to achieve what we set out for with Contex, we opted to investigate different libraries.

From Contex to Vega-Lite

We were recommended to look at Vega-Lite, which has very thorough documentation for the JSON specification. By combining this with the documentation for Vega-Lite Elixir bindings, we had the possibility to generate graphs with much greater functionality than Contex had provided.

Vega-Lite is very powerful, and using it allowed us to easily display both the seven-day history and the seven-day forecast data received from the API. We were also able to show the precipitation in addition to the temperature data, each with its own independent y-axis. We could also modify the colours of the lines, add axis labels and modify their angles for optimum visual appeal. We were also able to add a vertical line in the middle of the graph, indicating the data points for the current date.

It’s worth noting however, that when using the Vega-Lite Elixir bindings, all of the options have been normalised to snake-case atom keys. For example, in axis (for colouring the axes labels), the field is title_colour:, rather than `titleColor:` as given in the JSON specification. This caused us some brief trouble when at first we used the camel-case version, and the options were not displaying. The following is a partial excerpt of the code used for the graph:

elixir
new_chart = Vl.new(width: 700, height: 400, resolve: [scale: [y: "independent"]])

chart =
  Vl.data_from_values(new_chart, dataset)
  |> Vl.layers([
    Vl.new()
    |> Vl.layers([
      Vl.new()
      |> Vl.mark(:line, color: "#FF2D00")
      |> Vl.encode_field(:x, "date", type: :ordinal, title: "Date", axis: [label_angle: -45])
      |> Vl.encode_field(:y, "max", type: :quantitative, title: "Maximum temperature"),

In order to render the Vega-Lite graphs to a usable format (in our case SVG) we needed the VegaLite.Export functions. Unfortunately for SVG, PDF, and PNG exports, these functions rely on npm packages, meaning we had to add Node.js, npm, and some additional Vega and Vega-Lite dependencies to our project. Exporting the graph as an SVG was the best option as it allowed us to reuse code from the Contex implementation and display the graph in our HTML page render as we had before, but if we had wanted to avoid installing the npm packages, it would also have been possible to export the graph directly to HTML or as a JSON instead.

The final process

At the end of the process, we had broadly succeeded in creating what we set out to create: a Phoenix LiveView app that could display the weather and precipitation data for a week on either side of the current date for any given UK postcode, in a neat and colour-coded line graph. We came away from the process with both a better understanding of LiveView and a good idea of the strengths and weaknesses of the two libraries we utilised.

Contex provides a simpler and lightweight functionality, making it easy to create basic graphs. However, its comparatively limited library and its insufficient documentation provide obstacles to using it for more complex graphs, and as such it ultimately proved unsuitable for our purposes. Meanwhile, Vega-Lite is thoroughly documented and contains more intricate and advanced functionality, allowing us to create the application as outlined. Despite this, it does also have several drawbacks: its language-agnostic documentation occasionally made it slightly confusing to implement in Elixir, and the packages necessary to export the graphs create a significant JavaScript footprint in the application. When working on a project that could require one of these two libraries, it might help to consider these strengths and weaknesses in order to determine which would be the best fit.