Remote Sensing With Machine Learning to Monitor Seagrass Meadows: Long‐Term Trends in Tauranga Harbour, New Zealand

Abstract

Seagrass meadows play critical roles in coastal ecosystems by stabilising sediments, supporting biodiversity and sequestering carbon, yet monitoring their recovery at scale is often constrained by the lack of scalable and repeatable approaches. This study applies an open‐source remote‐sensing framework to track long‐term changes in Zostera muelleri in Tauranga Harbour, New Zealand, from 1959 to 2024. Using supervised machine learning (Random Forest) applied to Landsat and Sentinel‐2 imagery, combined with regression‐based estimates of seagrass density, we generated maps of seagrass extent and condition with overall accuracies of 0.88 for Sentinel‐2 and 0.77 for Landsat across all habitat classes. Beyond total area, the approach resolved changes in meadow structure, including patch size, connectivity, fragmentation and interannual variability, providing a landscape‐scale perspective on seagrass dynamics. Comparisons with manual aerial‐photo delineations supported the reliability of the broad harbour‐wide patterns. The combined record showed a 49% decline in seagrass extent between 1959 and 1990, followed by strong recovery since 2016, with extent exceeding 4,000 ha in 2023 and accompanied by recent subtidal expansion. The observed recovery occurred alongside documented management actions and variation in sediment‐related conditions, while preceding‐season temperature was positively associated with summer subtidal extent. The framework provides managers and practitioners with a repeatable means of monitoring long‐term seagrass recovery and informing adaptive coastal management.

Bibliographic Information