probabilistic forecasts from observations

SolarArbiter / solarforecastarbiter-core

Core data gathering, validation, processing, and reporting package for the Solar Forecast Arbiter

https://solarforecastarbiter-core.readthedocs.io

MIT License

33 stars 21 forks source link

probabilistic forecasts from observations #195

Closed wholmgren closed 4 years ago

wholmgren commented 4 years ago

some options include:

persistence ensemble (most recent N observations at same time of day)
raw persistence ensemble (most recent N observations)
complete history persistence ensemble (all observations at same time of day)
climatology (all observations)

cwhanse commented 4 years ago

Relevant to our phone discussion about computing percentiles, I recommend we stick with numpy.percentile with the default interpolation=linear. The technique is the default in R's quantile function and is intuitive (connect points on the CDF with straight line segments and interpolate). Alternatives have been discussed at length for numpy e.g. here with no consensus that there is a need to change the default.

wholmgren commented 4 years ago

also relevant to #115

wholmgren commented 4 years ago

@dplarson just wanted to make sure you saw @cwhanse's comment/link above. I neglected to look at link until now - it's helpful.

dplarson commented 4 years ago

Per today's team discussion, I'll take lead on adding reference probabilistic forecasts. My hope is to open a draft PR by next week to outline my general approach and get feedback before I dive into the coding details.