Subsurface chlorophyll maxima (SCMs), commonly occurring beneath the surface mixed layer in coastal seas and open oceans, account for main changes in depth-integrated primary production and hence significantly contribute to the global carbon cycle. To fill the gap of previous methods (in situ measurement, remote sensing, and the extrapolating function based on surface-ocean data) for obtaining SCM characteristics (intensity, depth, and thickness), we developed an improved deep neural network (IDNN) model using a Gaussian radial basis activation function to retrieve the vertical profile of chlorophyll a concentration (Chl a) and associated SCM characteristics from surface-ocean data. The annually averaged SCM depth was further incorporated into the bias term and the Gaussian activation function to improve the estimation accuracy of the IDNN model. Based on the Biogeochemical-Argo (BGC-Argo) data acquired for three regions in the northwestern Pacific Ocean, vertical Chl a profiles produced by our improved DNN model using sea surface Chl a and sea surface temperature (SST) were in good agreement with the observations, especially in regions with low surface Chl a. Compared to other neural-network-based models with one hidden layer and a sigmoid activation function, the IDNN model retrieved vertical Chl a profiles well in more eutrophic subpolar regions. Furthermore, the application of the IDNN model to infer vertical Chl a profiles from remote-sensing information was validated in the northwestern Pacific Ocean.

Monitoring the spatiotemporal variability of nutrient concentrations in shelf seas is important for understanding marine primary productivity and ecological problems. However, long time-series and high spatial-resolution nutrient concentration data are difficult to obtain using only on ship-based measurements. In this study, we developed a machine-learning approach to reconstruct monthly sea-surface dissolved inorganic nitrogen (DIN), dissolved inorganic phosphorus (DIP), and dissolved silicate (DSi) concentrations in the Yellow and Bohai seas from 2003–2019. A large amount of in situ measured data were first used to train the machine-learning model and derive a reliable model with input of environmental data (including sea-surface temperature, salinity, chlorophyll-a, and K_d490) and output of DIN, DIP, and DSi concentrations. Then, longitudinal (2003–2019) monthly satellite remote-sensing environmental data were input into the model to reconstruct the surface nutrient concentrations. The results showed that the nutrient concentrations in nearshore (water depth < 40 m) and offshore (water depth > 40 m) waters had opposite seasonal variabilities; the highest (lowest) in summer in nearshore (offshore) waters and the lowest (highest) in winter in nearshore (offshore) waters. However, the DIN:DIP and DIN:DSi in most regions were consistently higher in spring and summer than in autumn and winter, and generally exceeded the Redfield ratio. From 2003–2019, DIN showed an increasing trend in nearshore waters (average 0.14 μmol/L/y), while DSi showed a slight increasing trend in the Changjiang River Estuary (0.06 μmol/L/y) but a decreasing trend in the Yellow River Estuary (–0.03 μmol/L/y), and DIP exhibited no significant trend. Furthermore, surface nutrient concentrations were sensitive to changes in sea-surface temperature and salinity, with distinct responses between nearshore and offshore waters. We believe that our novel machine learning method can be applied to other shelf seas based on sufficient observational data to reconstruct a long time-series and high spatial resolution sea-surface nutrient concentrations.

Nitrogen dioxide (NO₂) is an important precursor of atmospheric aerosol. Forecasting urban NO₂ concentration is vital for effective control of air pollution. This paper proposes a hybrid deep learning model for predicting daily average NO₂ concentrations on the next day, based on atmospheric pollutants, meteorological data, and historical data during 2014 to 2020 in five coastal cities of Shandong peninsula, northern China. A random Forest (RF) algorithm was used to select input variables to reduce data dimensionality trained by the sequence to sequence (Seq2Seq) the model and describe how the Seq2Seq model understands each predictor variable. The hybrid model combining an RF with Seq2Seq network (RF-S2S) was evaluated and achieved a Pearson’s correlation coefficient of 0.93, a Nash–Sutcliffe coefficient (NS) of 0.79, a Root Mean Square Error (RMSE) of 5.85 µg/m³, a Mean Absolute Error (MAE) of 4.50 µg/m³, and a Mean Absolute Percentage Error (MAPE) of 20.86%. Feature selection by an RF model improves the performance of the Seq2Seq model, reducing errors by 19.7% (RMSE), 20.3% (MAE), and 29.3% (MAPE), respectively. Carbon monoxide (CO) and PM10 are two common, important features influencing the prediction of NO₂ concentrations in coastal areas of northern China. The results of RF-S2S models can capture general trends and disruptions more accurately than can long-short term memory (LSTM) models with and without feature selection. The decreasing tendency of NO₂ from 2014 to 2020 illustrated by the empirical mode decomposition (EMD) method is one important obstacle to improving the RF-S2S prediction accuracy. An EMD-based RF-S2S model could help to perform the short-term forecast of NO₂ concentrations efficiently.

Insufficient studies in characterizing vertical structure of Chlorophyll-a (Chl-a) in the ocean critically limit better understanding about marine ecosystem based on global climate change. In this study, we developed a Gaussian-activation deep neural network (Gaussian-DNN) model to assess vertical Chl-a structure in the upper ocean at high spatial resolution. Our Gaussian-DNN model used the input variables including satellite data of sea surface Chl-a and in-situ vertical physics profiles (temperature and salinity) in the northwestern Pacific Ocean (NWPO). After validation test based on two independent datasets of BGC-Argo and ship measurement, we applied the Gaussian-DNN model to reconstruct temporal evolution of 3-D Chl-a structure in the NWPO. Our modelling results successfully explain over 80% of the Chl-a vertical profiles in the NWPO at a horizontal resolution of 1° × 1° and 1 m vertical resolution within upper 300 meters during 2004 to 2022. Moreover, according to our modelling results, the Subsurface Chlorophyll Maxima (SCMs) and total Chl-a within 0-300 m depths were extracted and presented seasonal variability overlapping longer-time trends of spatial discrepancies all over the NWPO. In addition, our sensitivity testing suggested that sea-water temperatures predominantly control 3-D structures of the Chl-a in the tropical NWPO, while salinity played a key role in the temperate gyre of the NWPO. Here, our development of the Gaussian-DNN model may also be applied to craft long term, 3-D Chl-a products in the global ocean.

Based on tropical cyclone (TC) track data and gridded observational rainfall data of CN05.1 during the period of 1961 to 2014, we examine the contribution of TCs on three metrics of summertime rainfall regimes and identify the connection between TC-induced precipitation events and El Niño–Southern Oscillation (ENSO) in middle–lower reaches of Yangtze River Basin (MLYRB). At the regional scale, TCs are responsible for approximately 14.4%, 12.5%, and 6.9% of rainfall events for normal, 75th, and 95th percentile precipitation cases, respectively. There is no evidence of significant long-term trends of the three type events linked with TCs, while their interdecadal variability is remarkable. Fractionally, larger proportions of TC-induced events occur along southeast coastal regions of MLYRB for normal rainfall events, and they are recorded over southwest and central-east MLYRB for 95th percentile cases. Moreover, a larger contribution of 95th percentile precipitation events to summer total rainfall is found than that for 75th percentile cases, suggesting that TCs may exert stronger impacts on the upper tail of summertime precipitation distribution across MLYRB. The TC-induced normal rainfall events tend to occur more frequency over central-west MLYRB during negative phase of ENSO in summer. However, the higher likelihood of TC-induced rainfall for three defined metrics are found over the majority of areas over MLYRB during negative ENSO phase in spring. In preceding winter, La Niña episode plays a crucial role in controlling the frequency of both normal and 75th percentile precipitation events.