The significance of CO₂ emissions at the water–air interface from inland water bodies in the global carbon cycle has been recognized and is being studied more and more. Although it is important to accurately assess CO₂ emission flux in a catchment, little research has been carried out to investigate the spatio-temporal variations in CO₂ emissions in view of a water continuum. Here, we systematically compared the differences and control factors of CO₂ degassing across the water–air interface of a spring–river–lake continuum in the discharge area of Baotuquan Spring in July 2017, which is a typical temperate karst spring area in Jinan city, northern China, using hydrogeochemical parameters, stable carbon isotope values, and CO₂ degassing flux. Affected by the pCO₂ concentration gradient between the water and ambient air, the spring water showed a high CO₂ degassing flux (166.19 ± 91.91 mmol/(m² d)). After the spring outlet, the CO₂ degassing flux in the spring-fed river showed a slight increase (181.05 ± 155.61 mmol/(m² d)) due to river flow rate disturbance. The river flow rate was significantly reduced by the “blockage” of the lake, which promoted the survival and reproduction of phytoplankton and provided favorable conditions for aquatic plant photosynthesis, increasing the plankton biomass in the lake to 3383.79 × 10⁴/L. In addition, the significant decrease in the dissolved inorganic carbon (DIC) concentration and the increase in the δ¹³C_DIC values in the lake also indicated that the photosynthesis of the lake’s aquatic plants resulted in a significant decrease in the pCO₂ concentration, thus limiting the amount of CO₂ off-gassing (90.56 ± 55.03 mmol/(m² d)).

Accurate joint estimation of heterogeneous hydraulic conductivity fields and time-varying contaminant source parameters in groundwater systems constitutes a challenging high-dimensional inverse problem, particularly under sparse observational conditions and high computational demands. To alleviate this limitation, this study proposes an autoregressive depthwise convolutional neural network (AR-DWCNN) as a lightweight surrogate model for coupled groundwater flow and contaminant transport simulations. The proposed model employs depthwise separable convolutions and dense connectivity within an encoder–decoder framework to capture nonlinear flow and spatiotemporal transport dynamics while reducing model complexity and computational demand relative to conventional convolutional architectures. The AR-DWCNN is further integrated with an enhanced Iterative Local Updating Ensemble Smoother incorporating Levenberg–Marquardt regularization, enabling efficient joint inversion of high-dimensional hydraulic conductivity fields and multi-period contaminant source strengths. Numerical experiments conducted on a synthetic two-dimensional heterogeneous aquifer demonstrate that the surrogate-assisted inversion framework achieves posterior estimates that closely match those obtained using the numerical forward model, while significantly improving computational efficiency. These results indicate that the AR-DWCNN-based inversion method provides an effective and scalable solution for high-dimensional groundwater contaminant transport inverse problems, offering practical potential for uncertainty quantification and remediation design in complex subsurface systems.