An automated scheme, based on statistical regression, uses four predictors derived from a single-station radiosonde profile to forecast precipitation for a 12 h period. Testing shows good correlation between predictions and observations but also the need for adjusting systematic errors.