Predicting the dissolved oxygen concentration and identifying its driving factors are essential for improved prevention and management of anoxia in estuaries. However, complex hydrodynamic conditions and the limitations in traditional methods result in challenges in the identification of the driving factors for the low dissolved oxygen (DO) phenomenon. The objective of our study is to develop a robust deep learning model using four-year in situ data collected from an automatic water quality monitoring station (AWQMS) in an estuary, for accurate identification and quantification of the driving factors influencing DO levels. Mitigations in hypoxia were observed during the initial two years, but a subsequent decline in DO concentrations was witnessed recently. The periodicity of DO concentrations in the Pearl River Estuary reduced with the increase in the hypoxic intensity. Maximal information coefficient (MIC) and extreme gradient boosting (XGBoost) were employed to determine the significance of input variables, which were subsequently validated by using the long- and short-term memory networks (LSTMs). The driving factors contributing to the hypoxia problem were shown as temperature, pH, conductivity, and NH₄⁺-N concentrations. Notably, the evaluation index values of the hybrid model are MAPE = 0.0887 and R² = 0.9208, which have been improved compared with the LSTM model by about 99.34% in MAPE reduction and 16.56% in R² improvement, indicating that the MixUp-LSTM model was capable of effectively capturing nonlinear relationships between DO and other water quality indicators. Based on existing literature, three traditional statistical methods and four machine learning models were also performed to compare with the proposed MixUp-LSTM model, which outperformed other models in terms of prediction accuracy and robustness. Overall, the successful identification of the driving factors for the deoxygenation phenomenon would have important implications for the governance and regulation of low DO in estuaries.

Utilizing the optimal mass transportation (OMT) technique to convert an irregular 3D brain image into a cube, a required input format for a U-net algorithm, is a brand new idea for medical imaging research. We develop a cubic volume-measure-preserving OMT (V-OMT) model for the implementation of this conversion. The contrast-enhanced histogram equalization grayscale of fluid-attenuated inversion recovery (FLAIR) in a brain image creates the corresponding density function. We then propose an effective two-phase residual U-net algorithm combined with the V-OMT algorithm for training and validation. First, we use the residual U-net and V-OMT algorithms to precisely predict the whole tumor (WT) region. Second, we expand this predicted WT region with dilation and create a smooth function by convolving the step-like function associated with the WT region in the brain image with a $$5\times 5\times 5$$5 × 5 × 5 blur tensor. Then, a new V-OMT algorithm with mesh refinement is constructed to allow the residual U-net algorithm to effectively train Net1–Net3 models. Finally, we propose ensemble voting postprocessing to validate the final labels of brain images. We randomly chose 1000 and 251 brain samples from the Brain Tumor Segmentation (BraTS) 2021 training dataset, which contains 1251 samples, for training and validation, respectively. The Dice scores of the WT, tumor core (TC) and enhanced tumor (ET) regions for validation computed by Net1–Net3 were 0.93705, 0.90617 and 0.87470, respectively. A significant improvement in brain tumor detection and segmentation with higher accuracy is achieved.