DEVELOPMENT OF LOCALIZED IN-ORCHARD WEATHER PREDICTION MODELS FOR APPLE CROP MANAGEMENT

Karisma Yumnam

doi:10.7273/000006984

In-orchard microclimates such as air temperature, relative humidity, and wind speed have become increasingly critical for precise and timely orchard management. This hyperlocal weather is critical for decision support tools such as growing degree days, cold hardiness, and fruit surface temperature models. Such decision-support tools for in-orchard management are derived from open-field weather station data with an assumption that there is minimal to no difference between open-field and in-orchard weather. Architecture type, growth stage, and various management practices can cause different weather conditions inside the orchard, that may lead to bias and uncertainty in the weather-based models. This study quantified orchard management effects on air temperature, relative humidity, solar radiation, and wind speeds at different time scales using two seasons of data from six commercial apple orchards. Additionally, the in-orchard AT, RH, and WS prediction models were developed to translate open-field data to in-orchard conditions. Typically, orchards have 1.9 to 4.4 ℃ cooler air temperature and 9.2 to 27.5% higher relative humidity due to the evident impacts of tree transpiration. Also, in-orchard microclimates stations recorded lower solar radiation (267.8 to 483.2 W m-2) and wind speed (2.2 to 3.7 m s-1). Monthly averages data revealed the dependence of orchard effects on the phenological stages of apple canopies. Wind resistance and reduced air mixing caused drier microclimate during winter (RH offset: 8.7 %) and spring (RH offset: 7.7 %). Orchard training systems and height do significantly (p < 0.05) affect the hourly air temperature and relative humidity offsets during summer. Overhead sprinklers enhanced the reduction in air temperature (4.6 ℃) and increase in relative humidity (16.2 %) inside the orchard, and the effects tend to linger during evening and night hours. One statistical (multiple linear regression) and two machine learning (k-nearest neighbor, random forest) based in-orchard prediction models were developed to account for these effects. The random forest based in-orchard prediction model outperformed the other models with root mean square error of 0.51 ℃, 3.14 %, and 0.21 m s-1 for air temperature, relative humidity, and wind speed predictions respectively. Correspondingly, the models reduced the overall open-field and in-orchard offsets error by 62 %, 63 %, and 85 %.

DEVELOPMENT OF LOCALIZED IN-ORCHARD WEATHER PREDICTION MODELS FOR APPLE CROP MANAGEMENT

Files and links (1)

Abstract

Metrics

Details