Ocean color data is crucial for our understanding of biological and ecological phenomena and processes, and is also an important input source for physical and biogeochemical ocean models. Chlorophyl-a (Chl-a) is a key variable for ocean color in marine environments. Satellite remote sensing quantitative retrieval is the main method for obtaining large-scale ocean Chl-a. However, data loss is the main limitation of Chl-a products based on satellite remote sensing, mainly due to the influence of cloud cover, solar low light pollution, and high satellite perspective. The common methods for reconstructing (filling) missing data usually only consider the spatiotemporal information of the initial image, such as empirical orthogonal functions for data interpolation, optimal interpolation, kriging interpolation, and extended Kalman filtering. However, these methods perform poorly in the presence of large-scale missing values in the image, and ignore the potential of other information about missing pixels in data reconstruction. Considering the environmental variables related to the growth and distribution of marine phytoplankton, we developed a convolutional neural network (CNN) called OCNET for the reconstruction of Chl-a concentration data in open waters. Select sea surface temperature (SST), salinity (SAL), photosynthetically active radiation (PAR), and sea surface pressure (SSP) as input variables for OCNET from reanalysis data and satellite observation data, and correlate them with the environment and phytoplankton population. The developed OCNET model has achieved good performance in reconstructing global ocean Chl-a concentration data and captured the temporal variations of these features. The global Chl-a dataset covers the period from 2001 to 2021, with a temporal resolution of days and a spatial resolution of 0.25 °.
The Ocean Color Climate Change Initiative (OCCCI) 5th edition and the National Oceanic and Atmospheric Administration Multi Sensor DINEOF Global Gap Data (NOAA MSL12) are two Chl-a products used for training OCNET models. The data sources of OCCCI include the European Space Agency's Medium Resolution Imaging Spectroradiometer (MERIS) sensor, NASA's SeaWiFS (Wide Field of View for Ocean Observations) and MODIS Aqua (Medium Resolution Imaging Spectroradiometer Aqua) sensors, and the National Oceanic and Atmospheric Administration's VIIRS sensor (Visible and Infrared Imaging Radiometer Suite). Select sea surface temperature (SST), salinity (SAL), photosynthetically active radiation (PAR), and sea surface pressure (SSP) as input variables for OCNET from reanalysis data and satellite observation data, and correlate them with the environment and phytoplankton population.
Given that CNN is suitable for satellite remote sensing images and climate model data, we have constructed a global OCNET model consisting of 405 regional CNNs.
The data quality is good.
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.ZENODO.DB6456.2024
10.5281/zenodo.10011908