Abstract:The role of clouds in regulating temperature and humidity is pivotal for various sectors such as agriculture, urban development, and aviation. Despite this, the precise forecasting of cloud cover remains a formidable challenge. This paper introduces a novel cloud prediction model that leverages the time-adaptive method, dynamic variable parameter method, and random forest method, along with data from the GRAPES (Global/Regional Assimilation and Prediction System) model and FY-4A satellite. The objective is to enhance the precision of cloud amount predictions. The study reveals that the daily fluctuations in total cloud cover in the Jiuquan region and its vicinity are relatively minor, yet they exhibit pronounced seasonal patterns, peaking in spring and summer and diminishing in autumn and winter. A spatial analysis indicates a gradient from fewer clouds in the northern areas to a higher concentration in the south. The cloud cover at various grid points is influenced by a multitude of factors. Employing the dynamic variable parameter method, this paper dynamically selects forecast variables based on their correlation with cloud cover at different grid points, which are then used to refine the random forest model. The model demonstrates an accuracy range of 0.55 to 0.80, thereby significantly elevating the precision of cloud amount forecasts. Furthermore, the incorporation of a time-adaptive method allows for the continuous updating of the random forest model. However, the cloud prediction accuracy hovers around 0.55, which may be attributed in part to the scarcity of comprehensive data, thereby impacting the model’s predictive capabilities for cloud amounts.
