Coastal waters around Australia are undergoing rapid change under the combined influence of climate variability and human activity, yet long-term, large-scale assessments of water quality remain limited. Here, we use a physics-informed deep learning approach applied to 20 years (2003–2022) of MODIS Aqua satellite observations to characterise spatial and temporal variability in key optical water-quality constituents. We reveal consistent coastal-to-offshore gradients, with elevated total suspended solids (TSS) and chlorophyll-a (TChla) in northern coastal regions and lower values in offshore waters. Over the study period, TSS shows modest but spatially coherent increases, while dissolved organic carbon (DOC) remains relatively stable except in northern regions. In contrast, TChla declines across most Australian bioregions, coinciding with increasing diffuse light attenuation (Kd(490)), indicating reduced water clarity and photic depth. Statistical analyses, including correlation and Granger causality, suggest that changes in Kd(490) are primarily driven by TSS, and that increased light attenuation constrains phytoplankton biomass in many coastal areas. Together, these results point to a continent-wide trend of “darkening and ungreening” in Australian coastal waters, with potential implications for primary productivity, ecosystem health, and fisheries. This study demonstrates the value of combining satellite observations with machine learning to provide long-term, synoptic insights into coastal ecosystem change at national scales.