Skip to content

NEWS.md

Latest commit

 

History

History
92 lines (73 loc) · 4.03 KB

NEWS.md

File metadata and controls

92 lines (73 loc) · 4.03 KB
Raw

ocular 0.1.0

Parameter handling

  • Explicit settings supplied through rs_params(), including values equal to the defaults, take precedence over calibration baselines.
  • An omitted interior_sensitivity or split_sensitivity inherits the stored setting; an explicit NULL disables the corresponding stage.
  • Removed the pre-publication categorical calibration extent and optional circular growth constraint. Self-calibration now expands a complete, seed-centred square until its precision condition is met or the feature grid prevents further symmetric expansion. Initial area growth has no hidden radial constraint.

Runtime behaviour

  • Field splitting now defaults to two consecutive measurable divergent line-of-sight profiles and evaluates each axis provisionally. An axis is rolled back if it would remove any cell in the finite, 8-connected seed-signature component containing the supplied point. Prepared ordinary synthetic regressions cover this software invariant, not field identity or delineation accuracy.
  • Successful field boundary delineation stages are quiet by default rather than printing internal search diagnostics. segment_area() warns once when no candidate field is identified and returns an empty mask.
  • FTW queries warn when an unavailable local icu extension prevents a source year from being read; the field boundary can still provide undated calibration support.
  • Removed unused tuning and visual-cleanup controls, obsolete aliases and convenience wrappers, and an unreachable MOD13Q1 retrieval route before the first public release.

Documentation

  • Revised the package and API reference documentation to distinguish initial area growth, perimeter refinement, within-field segmentation, and field splitting, and to state the validation limits of experimental FTW prior and Landsat-MODIS fusion features.
  • Reorganised the vignettes around three user tasks: a first retrieval and output workflow, field boundary delineation with reusable multi-site pipelines, and experimental features with their validation requirements.

ocular 0.0.93

Rename

  • Renamed the package from oculaR to ocular.
  • New objects use ocular and ocular_pipeline as their primary S3 classes.

oculaR 0.0.92

User-facing changes

  • Added explicit source = selection to get_rs(). Specify "landsat-8" or "landsat-5" when a Landsat workflow is required, including Landsat-MODIS fusion and fusion validation.
  • Standardised the search_windows, vi_sensitivity, and vi_threshold parameter names.
  • Added is_fused to identify synthetic fused scenes. Fusion diagnostics report anchor_mae and anchor_nmae separately from held-out fusion error.
  • Added optional field boundary priors from compatible GeoParquet files through add_ftw_prior() and diagnose_against_ftw().
  • Added resumable, cached multi-site processing with batch_rs().

Reliability and validation

  • Strengthened validation of dates, geometry, sources, masks, aggregation functions, and delineation parameters; invalid inputs now fail early with clearer messages.
  • Improved calibration, cache-key construction, and cache-clearing safeguards. Custom batch pipelines require an explicit cache key before their results are cached.
  • Added offline tests across the public API and updated package documentation.
  • Live STAC and remote field boundary GeoParquet queries, as well as scientific accuracy against independent reference data, remain separate validation tasks.

oculaR 0.0.91

Field-boundary reference support

  • Added diagnose_against_ftw() for pixel-level agreement statistics (IoU, F1, precision, and recall) between a delineation and a reference polygon.
  • Added the ability to use an attached field polygon as calibration support. Agreement with the same polygon is an in-sample diagnostic, not an independent accuracy assessment.

oculaR 0.0.90

  • Initial feature development: STAC retrieval, sample-free field boundary delineation, optional coarse-fine fusion, and leave-one-out fusion evaluation.