Leaf wetness duration (LWD) models have been proposed as an alternative to in situ LWD measurement, as they can predict leaf wetness using physical mechanism and empirical relationship with meteorological conditions. Applications of advanced machine learning (ML) algorithms in the development of empirical LWD model can lead to improvements in the LWD prediction. The current study developed LWD model using extreme learning machine, random forest method, and a deep neural network. Additionally, performances of these ML-based LWD models are evaluated and compared with existing models. Observed LWD and meteorological variable data are obtained from nine farms in South Korea. Temporal and geographical information were also used. Additionally, the priorities of the employed variables in the development of the ML-based LWD models were analyzed. As a result, the ML-based LWD models outperformed the existing models; the random forest led to the best performance for LWD prediction among the tested LWD models. Strengths of associations between input variables and leaf wetness were relative humidity, short wave radiation, air temperature, hour, latitude, longitude, and wind speed in descending order. Uses of the geographical and time information in development of LWD model can improve the performance of LWD model.
Effects of water-saving pruning on the growth and water consumption of jujube (Ziziphus jujuba Mill.) in a loess hilly region with deep soil desiccation
