Deep reinforcement learning for variability prediction in latent heat flux from low-cost meteorological parameters

2020 
Predicting cropland latent heat flux (LHF) from commonly measured low-cost meteorological parameters (MPs) like net solar radiation, soil and air temperature, vapor pressure deficit, wind speed, and canopy temperature of the crops is essential for modeling crop production and managing water resources economically. In this treatise, we explore the deep reinforcement learning framework for short-term LHF trend estimation from the above MPs. The problem is reformulated as a classification problem, where each MP is acquired for a cost, and the objective is to optimize the trade-off between the predicted trend error and the relative MP acquisition cost. A sequential trend forecasting problem is evaluated via Q-learning with a linear guesstimate and a deep Q-learning scheme via neural network, where the distinct actions are the individual request for the MP values, and each episode is terminated by anticipating a trend. The proposed methodology is validated on the acquired farm-data, collected from the field experiments conducted on the cropland monitoring sites at Bidhan Chandra Krishi Viswavidyalaya State Agricultural University, Kalyani, West Bengal, India. The three non-rice crops, namely the yellow Sarson (mustard), potato, and green-gram, are studied owing to their similar energy balance partitioning patterns.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    0
    References
    0
    Citations
    NaN
    KQI
    []