Observations

Robust observations and accurate error covariances are essential for successful calibration. When calibrating climate models, it is advisable to use long-term climate statistics, such as monthly or seasonal averages, to reduce the influence of internal variability. This results in a more stable and representative target for inversion.

EnsembleKalmanProcesses.jl provides several containers for managing observations, with documentation provided here. As inputs to a calibration, observations can consist of a Vector, an EKP.Observation (a single observation), or an EKP.ObservationSeries (many observations).

To iterate through an EKP.ObservationSeries, you must provide a minibatcher. This package provides two helper functions to faciliate the creation of simple batches:

• ClimaCalibrate.minibatcher_over_samples takes in samples or (a number of samples) and a batch size and returns a minibatcher which divides the samples into the batch size, dropping remaining samples.
• ClimaCalibrate.observation_series_from_samples takes in a vector of Observations and a batch size and returns an ObservationSeries with a minibatcher.