Coastal sea temperature deep learning forecasting method and system based on multi-source data

Abstract

The invention discloses a coastal sea temperature deep learning forecasting method and system based on multi-source data. The method comprises the steps of obtaining MODIS satellite remote sensing data of a target sea area, ERA5 meteorological reanalysis data and sea surface temperature observation data of a sea surface buoy site; inputting the MODIS satellite remote sensing data and the ERA5 meteorological reanalysis data into a pre-trained spatial random forest model to obtain a sea surface temperature field of a kilometer-by-kilometer spatial resolution of the target sea area at the current moment; inputting the sea surface temperature field and the longitude and latitude information corresponding to the sea surface temperature field into a pre-trained time sequence prediction model of a corresponding month to obtain a sea surface temperature prediction value of a position corresponding to the sea surface temperature field hour by hour in future n hours; and outputting the hour-by-hour sea surface temperature prediction value in the future n hours, where n is greater than or equal to 1. According to the method and the system, the problem of low forecasting precision caused by the influence of tide, river input and weather varied factors on an offshore area is solved, and the accuracy and the reliability of short-term sea temperature forecasting are improved.

Images