Published in Scientific Papers. Series B, Horticulture, Vol. LXVIII, Issue 1
Written by Boryana TSENOVA, Veska GEORGIEVA, Metodi DINEV
In the present study a scheme for damaging frost occurrence forecast in Bulgaria is presented. It is based on Random Forest technique and uses the regional numerical weather prediction (NWP) model ALADIN output as predictor. Initially, the statistical model is trained with measured data with three-hours frequency at 5 representative meteorological stations in Bulgaria during April and May for the period 1991-2020. Using parameters from the regional NWP model production as predictors gives possibility to forecast frost probability 72 hours ahead. The performance of the scheme is evaluated. Results for 27 synoptic stations during April (2021-2023) show a probability of detection above 0.85 and a false alarm rate below 0.1 independently of the remoteness of the forecast. Most of the considered cases were correctly discriminated by the scheme as “frost” and “non-frost” cases, which would not be the case if only considering the forecasted minimum temperature. Our results show that frost could be forecasted by the presented scheme 3 days before its occurrence, which should be enough to react to minimize damage caused in the agricultural sector.
[Read full article] [Citation]
