Relationships between water quality of a long-distance inter-basin water diversion project and air pollution emissions along the canal: Distributions, lag effects, and nonlinear responses.

Understanding and quantifying the influences and contributions of air pollution emissions on water quality variations is critically important for surface water quality protection and management. To address this, we created a five-year daily data matrix of six water quality indicators-permanganate index (CODMn), NH3-N, pH, turbidity, conductivity, and dissolved organic matter (DOM)-and six air pollution indicators-O3, CO, NO2, SO2, 2.5 µm particulate matter (PM2.5), and inhalable particles (PM10)-using data from seven national monitoring stations along the world's longest water-diversion project, the Middle Route of the South-to-North Water Diversion Project in China (MR-SNWD). Multivariate techniques (Mann-Kendall, Spearman's correlation, lag correlation, and Generalized Additive Models [GAMs]) were applied to examine the nonlinear relationships and lag effects of air pollution on water quality. Air pollution and water quality exhibited marked spatial heterogeneity along the MR-SNWD, with all water quality parameters meeting Class I or II national standards and the air pollution indicators exceeding those thresholds. Except for CODMn and DOM, the other water quality and air pollution indicators exhibited significant seasonal differences. Air pollution exhibited significant lag effects on water quality at the northern stations, with NO2, SO2, PM2.5, and PM10 being highly correlated with changes in pH, with an average lag of 17 d. Based on the GAMs, lag effects enhanced the significant nonlinear relationships between air pollution and water quality, increasing the average deviance explained for CODMn, NH3-N, and pH by 93%, 24%, and 41%, respectively. These findings provide a scientific basis for protecting water quality along the long-distance inter-basin water-diversion project under anthropogenic air pollution.

Boosting fine-tuning via Conditional Online Knowledge Transfer.

Fine-tuning is an effective technique to enhance network performance in scenarios with limited labeled data. To achieve this, recent methods exploit the knowledge mined in the source model (e.g., feature maps) to construct an extra regularization signal (RS), collaboratively supervising the target model along with target labels. However, these RSs are generated independently from the target information or are generated from the rough assistance of the target information, resulting in biased supervision different from the target task. In this paper, we propose a Conditional Online Knowledge Transfer (COKT) framework that finely utilizes the target information to construct robust and target-related RS. Specifically, we train a target-dominant RS branch that online supervises the target model in a knowledge distillation manner. The target information dominates the RS branch from three aspects: sample-wise conditional attention, residual feature fusion, and target task loss. With such a target-oriented framework, we can effectively exploit target-related prior knowledge of the source model. Extensive experiments demonstrate that COKT significantly outperforms the fine-tuning baselines, especially for dissimilar target tasks and small datasets. Moreover, different from most of the fine-tuning methods that are restricted to the vanilla fine-tuning scenario, COKT can be easily extended to cross-model and multi-model fine-tuning scenarios.