<![CDATA[Mangrove ecosystem monitoring in tropical coastal zones is a task fraught with layered complexity, stemming from highly dynamic environmental fluctuations, the inherent inadequacy of conventional survey methods in capturing rapid change, and escalating anthropogenic pressures that continue to reshape coastal landscapes. Against this backdrop, the present study introduces a Multilevel Data Fusion Framework grounded in Digital Twin (DT) architecture, developed specifically to monitor Rhizophora apiculate stands within the Ngurah Rai Grand Forest Park (Tahura Ngurah Rai), Bali, Indonesia (coordinates: 8.73°S / 115.19°E). Drawing on the Joint Directors of Laboratories (JDL) fusion model, the framework operates across five hierarchical processing levels (L0–L4) and ingests a 576-day NASA/POWER climate record spanning January 2024 through July 2025. A four-stage preprocessing pipeline eliminated all 94 detected sentinel values, reducing T2M standard deviation by 98.9%. The Weighted Average Fusion (WAF)-derived Environmental Index (EI) yielded a mean of 0.536, with zero days classified under Critical condition throughout the observation window. The Growth Rate Index (GRI) averaged 0.771, corresponding to an aboveground biomass increment of +1.14 kg/m² and an estimated carbon sequestration rate of approximately 10.6 TC/ha/year. An EI–GRI Pearson correlation of r = 0.891 (R² = 0.794, p < 0.001) demonstrates a strong end-to-end predictive association within the pipeline. ONNX classification models embedded within a Unity AR platform achieved an F1-score of 0.913 at inference times consistently below 5 ms, confirming real-time operational viability]]>
