Yield Estimation of Hemerocallis citrina in Yunzhou District by Coupling STNLFFM Spatiotemporal Fusion with the CASA Model

To address the challenges posed by fragmented Hemerocallis citrina fields, the brief duration of key phenological stages, and the difficulty of obtaining remote-sensing observations with both high spatial and temporal resolution, this study developed a yield-estimation framework coupling the Spatial and Temporal Nonlocal Filter-based Fusion Model (STNLFFM) with the Carnegie–Ames–Stanford Approach (CASA). STNLFFM was used to reconstruct daily 10 m NDVI throughout the growing season, and the fused NDVI and meteorological data were used to drive CASA for county-scale yield estimation. The results showed that the STNLFFM-fused NDVI retained the spatial detail of Sentinel-2A well, with correlation coefficients ranging from 0.536 to 0.921 (mean = 0.747) and structural similarity index (SSIM) values ranging from 0.726 to 0.922 (mean = 0.833); fusion accuracy reached approximately 0.92 during the peak flowering and production period in July. Compared with MODIS NDVI, the STNLFFM time series better captured the phenological progression from sprouting to senescence. Cumulative NPP driven by STNLFFM NDVI agreed well with the field-derived aboveground carbon reference (R2 = 0.8142; RMSE = 20.67 gC·m−2). Yield estimates driven by STNLFFM NDVI showed substantially better agreement with field observations than MODIS-based estimates, with R2 increasing from 0.1371 to 0.7405 and RMSE decreasing from 677.61 to 370.45 kg·hm−2, a 45.3% reduction. These findings demonstrate the potential of the STNLFFM-CASA framework for yield estimation in fragmented Hemerocallis citrina fields under the conditions of Yunzhou District in 2023 and provide technical support for local precision management and yield assessment.