Deep Learning-based Reconstruction Of Chlorophyll A Data And Its Spatial And Temporal Variation Characteristics

Chlorophyll-a is the primary photosynthetic pigment of marine phytoplankton and an important indicator of water quality.It reflects the primary productivity of the ocean,and its spatiotemporal characteristics contain basic ecological information of the sea area,which is closely related to various marine environmental factors and has significant research implications.Satellite remote sensing technology,with its large scale,wide coverage,and strong continuity,has become an effective way to obtain information about chlorophyll-a concentration.However,the sensors carried by satellites have limited coverage and are easily affected by meteorological factors such as clouds and sun glares,resulting in large areas of missing data in satellite remote sensing chlorophyll-a data,greatly reducing the utilization of the data.Due to the trade-off between time and spatial resolution of satellites,the spatial resolution of chlorophyll-a data from high-time-resolution satellites is often low,making it difficult to analyze the detailed spatial variation of chlorophyll-a at small area scales.To address these issues,this study first used the DINCAE method to reconstruct the missing chlorophyll-a data,evaluated the reconstruction accuracy of DINCAE in all aspects,and compared analysis it with the DINEOF method.Then the Random Forest algorithm was used to downscale the chlorophyll-a images in the nearshore area to improve its spatial resolution.Finally,the reconstructed complete chlorophyll-a data were used for spatiotemporal change analysis,and the influencing factors of chlorophyll-a concentration changes were explored.The research results are as follows:(1)The R of chlorophyll-a data reconstructed by the DINCAE algorithm and the original data is 0.87.The DINCAE algorithm reconstruction can retain more small-scale detail features of the original image,with high restoration degree to the original data.The reconstruction accuracy of both DINCAE and DINEOF algorithms is greatly affected by the missing rate.The reconstruction bias of DINCAE decreases faster with the increase of the missing rate than the reconstruction bias of DINEOF,indicating that the reconstruction accuracy of the DINCAE algorithm is more sensitive to the data missing rate.(2)The chlorophyll-a data generated by the downscaling reconstruction model can well reflect the spatial variation characteristics of chlorophyll-a concentration in the nearshore sea area.Comparative analysis with the data of the measurement stations shows that the chlorophyll-a data generated by the downscaling model has high consistency with the measured data,with an R2 of 0.82.Through sensitivity analysis,it is concluded that atmospheric correction plays an important role in the model downscaling reconstruction,indirectly determining the model accuracy.(3)The spatial distribution of chlorophyll-a concentration in the northern part of the South China Sea presents a pattern of high in the shallow water areas near the land and low in the offshore water areas,decreasing step by step with the water depth.On the annual scale,the concentration of chlorophyll-a shows a declining trend year by year.On the seasonal scale,the concentration of chlorophyll-a is highest in winter and lowest in spring.(4)The concentration of chlorophyll-a is negatively correlated with sea surface temperature and salinity.As temperature and salinity increase,the concentration of chlorophylla gradually decreases.The seasons with the highest correlation between sea surface temperature and salinity and chlorophyll-a concentration are winter and summer,respectively.