Reproducibility package for "Beyond field edges: Human-validated boundary-signal representation for reference-blind UAV characterisation of fragmented agricultural landscapes"

Version 1.1.0 update

Version 1.1.0 extends the frozen v1.0.0 reproducibility release with a post-selection internal sensitivity analysis of the physical boundary-representation scale. The added analysis reuses frozen RGB maximum-gradient crop-level outputs and evaluates boundary-to-local-background contrast across the five pre-existing candidate radii (0.20, 0.30, 0.40, 0.50 and 0.75 m), together with leave-one-landscape-out sensitivity of radius preference.

No UAV orthomosaics were reprocessed, no signal model was refitted, and the previously frozen 0.40-m representation was not altered. The sensitivity analysis showed a broader high-response neighbourhood around approximately 0.30–0.50 m, while leave-one-landscape-out radius preference remained within 0.30–0.40 m. These results support interpreting 0.40 m as a dataset-specific representation scale rather than a universal physical boundary dimension.

The sensitivity analysis evaluates internal dependence on candidate radius and reference-landscape composition. It does not establish invariance to UAV platform, camera characteristics, flight altitude, image overlap, ground sampling distance, illumination or photogrammetric processing because these acquisition metadata were not consistently preserved.

This reproducibility package supports the associated manuscript, “Beyond field edges: Human-validated boundary-signal representation for reference-blind UAV characterisation of fragmented agricultural landscapes.”

The package contains frozen analysis scripts and configurations, qualified human-reference derivatives, boundary-signal and surface-model selection outputs, derived structural metrics for 40 UAV acquisition footprints, integrated structural-analysis outputs, machine-scale spatial-configuration profiles, robustness analyses, environment information, and a SHA-256 provenance manifest.

The computational workflow explicitly separates reference-guided representation diagnosis and selection from subsequent reference-blind deployment. Human functional-boundary references were used to diagnose the spatial correspondence between annotated boundary axes and UAV image evidence and to select the boundary-signal representation under predefined paired criteria. The selected physical scale and reference-blind surface formulation were then frozen before deployment across the 40 UAV acquisition footprints. Reference geometries, reference masks, crop centres, and distance-to-reference surfaces were not used to construct the full-landscape deployment surfaces.

The deposited release contains 81 scientific files excluding the SHA-256 manifest and corresponds to version 1.1.0 of the reproducibility package. Version 1.1.0 preserves the frozen primary analyses from version 1.0.0 and adds only the post-selection physical-scale sensitivity script, derived sensitivity tables and version documentation. The package provides computational provenance underlying the analyses reported in the associated manuscript; it is not intended to constitute an independently validated parcel-vectorisation or machinery-performance dataset.

Raw UAV orthomosaics are not included because uniform redistribution rights for the source imagery are not established in the analytical archive and because the original imagery is substantially larger than the derived reproducibility products. Shareable derived reference products, analytical tables, configurations, and scripts required to inspect the reported workflow are included.

Machine-scale spatial-configuration scores are relative UAV-observed structural descriptors within the analysed corpus. They represent structural screening profiles derived from fragmentation, operational continuity, spatial regularity, and boundary-structure intensity under predefined operation-specific weights. They should not be interpreted as measurements or validated predictions of machinery suitability, realised machinery performance, soil trafficability, bearing capacity, field capacity, harvest losses, or economic profitability. Rank-stability intervals reported for these profiles derive from 5,000 Dirichlet perturbations of the predefined weight magnitudes; the 40 acquisition footprints were not bootstrap-resampled.

Version: 1.1.0
Version DOI: 10.5281/zenodo.22206411
SHA-256 of deposited ZIP: 996c63a64d1ad4cc90a0cd001e6a6c139ec9a6602c34325974a10dca85bc26d2