Estuaries and shallow lagoons are the most productive marine systems in the world and contribute to the maintenance of coastal biodiversity by providing various unique habitats for aquatic species. However, due to the inputs of nutrients, sediments and pollutants from the surrounding landscape, light availability and primary productivity can be strongly restricted, which may trigger rapid ecological change over time and cause substantial loss of coastal flora and fauna. This thesis explores the use of Sentinel-2 remote sensing imagery to monitor water properties of shallow intertidal tidal estuaries, for the purposes of managing the estuaries of New Zealand. The first goal (Chapter 2) was to develop a method to remove the influence of water bottom substrate reflectance in order to detect the true water colour (represented by dominant wavelength) and diffuse attenuation coefficients from Sentinel-2 imagery, in the case study estuary of Tauranga Harbour. The new methodology used direct measurement of bottom reflectance of intertidal areas while exposed, and used a regression estimator to derive subtidal bottom reflectance from particle size (developed using the intertidal properties). The method required the water depth to be known (from LiDAR) and reflectance observations from multiple water depths (either extracted along transects or at the same location at different tides). The method was applied to all available Sentinel-2 images and showed seasonal fluctuations and strong correlations with chlorophyll-a, suspended sediments and coloured metal ions collected by the regional council monitoring programme. This methodology was then applied to 12 estuaries and the results were interpreted in terms of ecological changes (Chapter 5). The second goal (Chapter 3) was to detect the distribution or density of seagrass/sandflats (where microphytobenthos (MPB) thrive) and use this as a basis to estimate their gross primary productivity (GPP). The new methodology combined Sentinel-2 imagery, machine learning and literature-derived photosynthesis-irradiance (P – I) curves. The machine-learning model included (1) supervised classification with random forest to delineate seagrass and sandflat areas (2) and three machine learning regressions (artificial neural network (ANN), support vector machine (SVM) and random forest regression (RFR)) to predict the density of seagrass. The result showed ANN was the optimal algorithm to predict seagrass coverage. By adjusting the input water depth and light intensity, the methodology could be further developed to predict the response of seagrass and MPB to sea level rise. To counteract the negative impact of increased turbidity caused by coastal erosion and sediment resuspension due to sea level rise and climate change, controlling sediment loading in coastal waters could be an effective solution for maintaining current productivity throughout the entire harbour. Considering monitoring suspended sediment concentration (SSC) is essential for understanding the resilience of coastal wetlands, a prediction model in Chapter 4 consisting of satellite imagery, numerical simulation and machine learning was developed to enable continuous estimation of SSC. The prediction model included two steps: (1) comparing the Delft3D-derived SSC with corrected satellite data and using K-means classification to categorise the differences into classes; (2) developing a random forest regression model for each class to predict the satellite-derived SSC using Delft3D-derived SSC and other physical parameters. Comparison of the prediction model with in situ measurement showed high accuracy, and the model provided the basis for estimating the accumulated sediment in wetlands. The results showed strong sensitivity to different ways of accounting for incoming SSC supply coastal wetlands. Therefore, employing a model based on real-time observations, like the one developed here, would substantially improve sediment budget estimates in coastal wetlands. Dominant wavelength and diffuse attenuation coefficients (Kd) can form the basis of ecologically relevant indicators. Therefore, in Chapter 5, classification based on these two indicators was developed to cluster New Zealand estuaries with similar states into groups as a basis for management. The dominant wavelength and Kd were derived from the Sentinel-2 images using the seabed correction model developed in Chapter 2. Three groups, including less impacted, moderately impacted and highly impacted, were created which were in broad agreement with other in situ measurements and indicator models such as those focused on light availability and benthic health. Therefore, satellite-derived dominant wavelength and Kd can be two good indicators to reflect estuarine water properties for management.