Sustainable urban disaster response requires Earth observation (EO) resources to be selected according to event demands, environmental conditions, sensing capabilities, and urban targets. This study proposes an Observation Linked Open Data (O-LOD)-based Urban Disaster Observation Task (UDOT) enhancement framework for semantic EO resource configuration. O-LOD organizes heterogeneous data through four dimensions, Event, Context, Subject, and Object, which an instantiation algorithm populates as task graphs. Layered GeoSPARQL queries then match thematic and analytical capabilities, qualify orbit-derived observation opportunities against context and object constraints, and rank feasible alternatives using the Observation Capability Evaluation Model (OCEM). Evaluation on the 2020 Khartoum flood and Bobcat wildfire narrowed 202 satellite–sensor pairs to three flood-capable and four wildfire-capable pairs and returned two Khartoum and eight Bobcat ranked opportunities, with complete queries executing in 3.0 s and 0.07 s. Constraint ablation and resolution sensitivity analyses identified the conditions governing the feasible set. Publicly accessible Sentinel, Landsat, and MODIS products corroborated both retained and excluded results, supporting water-extent and burn-severity mapping where coverage, spectral bands, and image quality met the task requirements. O-LOD therefore provides a traceable semantic link from disaster observation demand to qualified and ranked EO opportunities and subsequent image assessment, supplying task-oriented inputs for downstream scheduling and supporting context-aware sensing for sustainable urban disaster response.

Terraced fields are difficult to delineate from high-resolution imagery because narrow and irregular boundaries, fragmented patches, and visual similarity to adjacent cropland can lead to omission errors and boundary confusion. This study adapts DeepLabV3+ for terraced field extraction from 2 m GF-1 RGB imagery through a task-oriented integration of spatial-detail preservation, contextual modeling, and boundary refinement. The resulting framework incorporates a dilated residual encoder, a gated global–local context module, and a boundary-supervised PointRend refinement head to address spatial-detail loss, background confusion, and boundary uncertainty in terrace mapping. On the re-evaluated test set, the framework achieved 86.68% precision, 87.77% recall, 87.22% F1 score, and 77.34% target-class IoU. Relative to the unmodified DeepLabV3+ baseline, the F1 score and IoU increased by 10.01 and 14.46 percentage points, respectively. These results indicate that task-oriented integration of spatial-detail preservation, contextual modeling, and boundary refinement can improve fine-scale terraced-field delineation from high-resolution optical imagery.