This paper highlights that while the integrity of water quality data (WQD) is crucial for scientific decision-making and environmental monitoring for ecological protection, significant amounts of data are often missing from water quality monitoring systems due to issues such as sensor failure and communication delays. This missing data results in high-dimensional and sparse (HDS) data. Existing data imputation methods fail to adequately represent potential dynamics and capture deep data features, resulting in unsatisfactory imputation performance. Therefore, this paper proposes a nonlinear low-dimensional representation model (NLR) utilizing convolutional neural networks (CNNs) to impute missing WQDs. The CNN implements two key ideas: a) fusion of temporal features to model the temporal dependence of data across time slots, and b) extraction of nonlinear interactions and local patterns to mine high-order relationship features and achieve deep fusion of multidimensional information. Experimental studies on three real-world water quality datasets demonstrate that the proposed model significantly outperforms existing state-of-the-art data imputation models in terms of estimation accuracy. This provides an effective method for handling water quality monitoring data in complex, dynamic environments.
