PREDICTION AND DETECTION OF HONEY HARVESTS FROM REMOTE SENSING AND WEATHER DATA

There have been many efforts to use remotely sensed data to map or predict the production of honey. Many studies recommend that weather data are incorporated into the model. Here we assess the ability of satellite and weather data to predict the volume of honey produced from Corymbia calophylla (Myrtaceae) in southwest Australia. Utilising honey harvest data over 8 years, it was found that January NDVI could predict a ‘good' harvest (more than 40 kg of honey harvested per hive) to 79% accuracy. Poor harvests (less than 20 kg of honey) and moderate harvests (between 20 and 40 kg of honey) were not distinguishable. Assessing weather for January and February showed that the weather data from January was highly influential. Good harvests occurred after a cool, dry January, moderate harvests after a warmer, wetter January and poor harvests associated with warmer, drier January. Using a decision-tree approach, the combination of January weather and NDVI classified good harvests to 90% accuracy. Classification into the three quality levels achieved 69% accuracy from the overlapping data for poor and moderate years. This study used monthly weather data. Addition of daily weather data and apiary health variables may improve the predictive accuracy.