show the days in the last 5 years where temperature was below 0 F 
from sensor-1234

What would a programing language look like if it were designed for a specific slice of science? What about a query language?

Weather data is most often interacted with through read queries.

• What’s the latest forecast for my location?
• How do the three most recent forecasts for New York City compare against one another?
• What was the cumulative precipitation for this list of locations June 26th, 2020 and how does it compare against each location’s long term average?

N-dimensional data poses a challenge for efficient query. Weather data adds layers of complexity as well, and data providers put a large amount of onus on the consumer to load, process, and query the data in the ways they need.

But perhaps there’s a way to leverage the spatiotemporal semantics of weather data, in combination with nascent file formats optimized for efficient n-dimensional data access, to design a query language that allows consumers to get exactly what they want for their problem at hand.

show the precipitation and temperature aggregated daily 
in the latest forecasts from ecmwf-hres and noaa-gefs

show *variable(s)* in [obs time range or forecast issue time(s)] where [filter] 
from [source]

What are the fundamental operations?

Examples are how existing things do it, not necessarily how wxql should.

• Choose the dataset(s)
  • The FROM clause in SQL. df = pd.read_xxx(…) or ds = xr.open_dataset(…)
• Join
  • Align and combine multiple datasets
  • Could join be result of geometric operation like intersects? eg to join up list of polygon locations with all pixels which touch
• Choose variables
  • The SELECT var1, var2, … clause in SQL. df[["var1", "var2"]] in pandas and xarray.
• Filter
  • WHERE clause in SQL. .loc[] in pandas, .sel() in xarray.
  • WHERE intersects(position, geometry)
• Aggregate
  • Two parts:
    • Define the groups, eg “daily” or all values along one or more dimensions .mean(dim=(“x”, “y”)) to take avg across space.
    • Reduction function across values in each group. eg. mean, max, std dev
  • SQL does this with reduction function in the select clause and groups defined by groupby SELECT max(var) … GROUP BY date_trunc(time, “day”)
  • Pandas: df.groupby(…).max(), or for time dimensions, df.resample(”D”).max()
• Transform
  • Unary operation on each element (eg abs, sqrt), binary operation between pairs of elements (eg a + b), ternary (eg if a then b else c), 4+ (case statement?)
  • Output transformations that operate on the full output of a query, different than value→value. Could be
    • array→bytes (eg. compress values using zstd, plot as png)
    • bytes→bytes (eg. compress image bytes)
    • array→interactive UI element?

Dev reference