Statistical methodology

The experimental design and the statistical methodology deployed in the analysis of ART’s trials of the Atlant technology were developed in our early trials 20072008. They have been fine-tuned since with an overall consistency in approach since the first South Australian Trial in 2008.

This work has been done by independent econometricians, mathematicians and spatial statisiticians at Analytecon and the National Institute for Applied Statistics Research Australia at the University of Wollongong (NIASRA).

To deal with the problems inherent in detecting any impact of technologies aimed at the enhancement of rainfall the following key issues were addressed:

  • Eliminate “noise” in the data and isolate any actual rainfall enhancement.
  • Remove bias (in particular omitted variable bias) and avoid incorrectly assigned effects.
  • Deal with spatiotemporal correlation in the data to eliminate overstating of statistical confidence.

Methods deployed included:

  • Deployment of extensive meteorological data to more effectively and efficiently define target and control areas with the consequent use of dynamic corridors.
  • Randomised crossover scheduling of operations.
  • Exhaustive modeling of meteorological and topographic covariates to increase signal-to-noise ratio to sweep out day-to-day differences in average rainfall. This eliminates/reduces the bias of unmeasured meteorological conditions that are correlated with rainfall.
  • An instrumental variable to predict naturally occurring downwind rainfall, as well as day and gauge-level random effects.
  • Counterfactual predictions of rainfall without any enhancement effect.
  • Spatiotemporal resampling (the Block Bootstrap) to avoid understating standard errors.
  • Non-parametric permutation testing

Collage of statistical graphs