Uncertainty Methods for Time Series

[john-waczak]

Given a time series of measurements, we want to characterize measurement uncertainty three ways

1. Instrument uncertainty (provided by the sensor manufacturer)
2. Representativeness Uncertainty
3. Temporal Variogram Nugget.

For 1, @lakithaomal , @prabudewage , @Rittik047 and anyone else working on making the Central Nodes / LoRa Nodes should put together a CSV with the associated uncertainties for each device

For 2, We can establish a rolling window of width $\Delta t$ over which we estimate the representativeness uncertainty as the standard deviation or mean deviation for measurements over that window.

For 3, we want to compute the temporal semi-variogram for each time series and identify an additional estimate of the uncertainty via the so called nugget, i.e. the y intercept of the variogram with the $\Delta t=0$ line.

[ashenfernando1]

Does $\Delta t$ change for different timescales @davidlary?

@john-waczak , were you thinking of folding this as a calculated column into the influxdb itself?