<![CDATA[The application of Unmanned Aerial Vehicles (UAVs) in the Complete Systematic Land Registration (CSLR) and Participatory Systematic Land Registration (PTSL) programs significantly accelerates land mapping through aerial photogrammetry. This method substantially enhances accuracy in land certification processes, boundary identification, disaster mitigation efforts, and spatial planning efficiency. However, despite its effectiveness, manual intervention remains necessary when ground objects are obstructed or parcel boundaries are visually indistinct. To support more reliable, less subjective parcel delineation, advanced image segmentation methods are needed to streamline the derivation of parcel boundaries from UAV imagery. One technology that can be used to address this issue is Segment Anything Model Geo (SAM-Geo), which enables segmentation of farmland parcels with minimal model retraining. Nevertheless, the “segment anything” capability embedded in SAM‑Geo still requires validation through practical segmentation of surface objects, particularly land parcels in complex field conditions. This study evaluates the effectiveness of SAM-Geo, with an emphasis on segmentation accuracy and efficiency, using UAV‑derived orthophotos from five rice field areas that represent different parcel geometries and boundary qualities. We compare SAM‑Geo against a conventional standard reference method, namely, multiresolution segmentation. The findings show that SAM-Geo generally performs better than multiresolution segmentation in detecting rice field parcels and preserving parcel shapes and boundary clarity, particularly in regular, homogeneous fields with clear earthen embankments. SAM‑Geo achieves higher F1‑scores and Intersection over Union (IoU) than multiresolution segmentation across the five areas, and a paired Wilcoxon signed‑rank test on F1‑scores indicates that this accuracy advantage is statistically significant. At the same time, SAM‑Geo produces a non‑negligible number of false positives at the parcel level in more heterogeneous fields and requires longer processing times than multiresolution segmentation in our implementation. These results suggest that SAM‑Geo offers a promising accuracy‑oriented decision‑support tool for automating farmland parcel identification in favourable conditions, but its outputs still need careful review by cadastral officers, especially in complex landscapes and before being integrated into land certification workflows.Received: 2025-05-08 Revised: 2026-08-11 Accepted: 2026-09-14 Published: 2026-11-02]]>
Copyrights © 2026
