The increased availability and use of sensor data in environmental monitoring has led to a vastly increased demand for tools that can process, analyze and report on important environmental events in real or near-real time. These data have special characteristics, being high dimensional, recorded in space and time, and with potential missing values / sparsity issues. This thesis concerns itself with the building of statistical and machine learning methods for data of this type. Statistical methods can play an important role in inferring valuable features of the latent data-generating processes of the sensor data, while properly taking into account the uncertainty. Machine learning methods can play a significant role by offering an automatic framework to unearth complex patterns underlying the data, being scalable and dealing with challenges such as sparsity in an innovative way. This dissertation contributes to our understanding of how to monitor the environment, focusing on Dublin Bay as a proof of concept. It examines sea-level rise, water turbidity, and how to manage large environmental datasets that change over time and space with missing values. Firstly, it updates us on Dublin’s sea level record to produce a processed data product from 1938 to 2016. It utilizes a new statistical approach to make better estimates of average sea levels and finds that sea levels have been rising more quickly in recent times. Then it investigates how human activities, such as dredging and dumping, affect water turbidity in Dublin Bay. It employs a new statistical model called VARICH to model the variability in water turbidity over two years and at different locations, finding that weather conditions like wind speed significantly impact most locations. Meanwhile, dredging operations show lower impacts, and dumping operations have a significant impact only at greater water depths. Lastly, the dissertation addresses the challenge of missing values in large-scale environmental datasets by proposing two innovative models for multivariate spatio-temporal forecasting that offer competitive performance without the need for imputation. These models are a transformer-based model, SERT, and a simpler, interpretable model, SST-ANN. Altogether, the chapters of this dissertation provide new insights into environmental trends and offer novel methods for analysing environmental data. This is all taking place against the backdrop of significant global issues such as climate change, loss of biodiversity, and increasing pollution. This thesis stands as an example of how data analysis is advancing in the field of environmental monitoring.