Abstract: The Yellow River Delta Nature Reserve is a designated national wetland nature reserve in China. The effective monitoring of wetland resources within the reserve is of paramount importance for ecological restoration and achieving the United Nations Sustainable Development Goal (SDG). In this study, the simple non-iterative clustering (SNIC) segmentation algorithm integrated with random forest classifier (SNIC-RF) was used to extract wetland information from 2000 to 2020 at 5-year intervals. Both centroid and transfer matrices were used to explore the spatiotemporal variations and conversions among wetland types. The results showed that: (1) Wetland classification based on SNIC-RF achieved overall accuracies greater than 0.89, with Kappa coefficient exceeding 0.87. Compared with the results using only RF, the SNIC-RF method improved the overall accuracy by 2.7%-14.1% and the Kappa coefficient by 3.5%-16.4%, respectively. SNIC-RF demonstrated excellent accuracy in delineating artificial wetlands such as reservoirs and ponds, where distinguishing spectral information was challenging. (2) The wetland area fluctuated from 2000 to 2020, with an increasing trend in the intensity of fluctuations during the period from 2010 to 2020. Natural wetland areas diminished continuously, whereas the proportion of artificial wetlands steadily increased. (3) Before 2010, wetland conversions primarily occurred in natural wetlands, with minimal involvement of artificial wetlands. After 2010, the conversions predominantly shifted from various wetlands to non-wetland types, with paddy fields being the primary source. The expansion of artificial wetlands mainly originated from the conversion of non-wetlands and mudflats. (4) Wetland migration in the Dawenliu Nature Reserve was more complex and intense than that in the Yiqianer Nature Reserve. The centroids of the artificial wetlands in both regions continuously migrate eastward, reflecting the development of the ecological economy in Dongying City.

Key words: estuary wetland, Yellow River Delta Nature Reserve, simple non-iterative clustering, random forest classifier, dynamic change