Reconstruction of Sea Surface Chlorophyll-a Concentration in the Bohai and Yellow Seas Using LSTM Neural Network

Reconstruction of Sea Surface Chlorophyll-a Concentration in the Bohai and Yellow Seas Using LSTM Neural Network

Blog Article

In order to improve the spatiotemporal coverage of satellite Chlorophyll-a (Chl-a) concentration products in marginal seas, a physically constrained deep learning model was established in this work to reconstruct sea surface Chl-a concentration in the Bohai and Yellow Seas using a Long Short-Term Memory (LSTM) neural network.Adopting the Mascara permutation feature importance method, time sequences of several geographical and physical variables, including longitude, latitude, time, sea surface temperature, salinity, sea level anomaly, wind field, etc., were selected and integrated to the reconstruction model as input parameters.Performance inter-comparisons between LSTM and other machine learning or deep learning models was conducted based on OC-CCI (Ocean Color Climate Change Initiative) Chl-a product.

Compared with Gated Recurrent Unit, Random Forest, XGBoost, and Extra Trees models, the LSTM model exhibits the highest accuracy.The average unbiased percentage difference (UPD) of reconstructed Chl-a concentration is 11.7%, which is 2.9%, 7.

6%, 10.6%, and 10.5% smaller than that of the other four models, respectively.Over the majority of the study area, the root mean square error is less than 0.

05 mg/m3 and the UPD is below 10%, indicating that SAMSUNG RS8000 RS68N8330S9/EU American-Style Fridge Freezer - Matte Stainless the LSTM model has considerable potential in accurately reconstructing sea surface Chl-a concentrations in shallow waters.