Stratospheric ozone levels have been closely monitored since the signing of the Montreal Protocol which banned the ozone depleting substances. Detection of ozone recovery has been complicated by changes in the natural variability. Deriving the trend in ozone attributable to the reduction in the ozone depleting gases individual ground based monitoring sites using multi-linear regression models requires careful selection of the model predictors that are used as proxies for the natural variability. Here we show the trends in the NOAA ground and satellite based instrumental record using an optimized LOTUS regression model. We find that adding proxies in the optimized model for the decadal oscillations, transport of ozone vertically and horizontally, and the stratospheric circulation strength decreases uncertainty by explaining the natural variability. Trends become more positive (but not significantly so) in the high latitudes. Positive trends with high statistical certainty are found at two of the ten Dobson monitoring sites. We will show that positive trends are found in the zonal means in both hemispheres’ mid- latitudes. The optimization of the regression model increases confidence in the trend results. If the ozone recovery trend continues, this method could lead to an earlier detection of ozone recovery in places where no ozone trend is too small compared to the fluctuations to be confirmed so far.