Residual autocorrelation plots for precipitation model without adjustments and including lag/logged lag/lag residuals (lag1) to control for autocorrelation using the NYC dataset.

Autocorrelation for a lag period of 20 days (x-axis) was first assessed without further adjustment (a) using the partial autocorrelation function (PACF). Autocorrelation was strongest at lag of day 1. Including the COVID-19 case count of the previous day (Yt-1) (b) or the logged COVID-19 case count of the previous day (log(Yt-1)) (c) resulted in comparable reductions of autocorrelation among model residuals. Inclusion of lagged residuals of daily case count of day 1 (d) also led to some reduction of autocorrelation but the effect was considerably less pronounced. (TIFF)