ocular

ocular provides tools for obtaining high quality spectral indices from reusable remote sensing pipelines.

Installation

o <- options() # store original options

options(pkg.build_vignettes = TRUE)

if (!require("pak")) {
  install.packages("pak")
}

pak::pak("AAGI-AUS/ocular")
options(o) # reset options

Given a point and a date range, ocular provides spectral indices commonly used in crop monitoring and agricultural research, enables batched retrieval of vegetative indices (VI) values for multi-site datasets, and offers reusable pipelines for bundling pre-processing, sample-free field boundary delineation, and other applied steps.

Retrieval and processing

Landsat and Sentinel-2 satellite imagery are retrieved from the Microsoft Planetary Computer Data Catalog and stored as SpatioTemporal Asset Catalog (STAC) scenes in an ocular object. This allows the quality and temporal coverage of the retrieved data to be optimised before it is transformed to a data frame or raster.

Cloud masking and cloud cover filtering are set during retrieval, while other processing steps and output stages are configured and applied in reusable pipelines. Sample-free field boundary delineation in ocular consists of four modular functions. In custom pipelines, these functions can be reordered or reconfigured. For a default sequence, use boundary_delineation().

Experimental features include:

• Landsat-MODIS data fusion, which can increase temporal coverage (leave-one-out validation against held-out Landsat scenes is available to assess the viability of the fused estimates).
• Fields of The World (FTW) global field boundary data incorporated as an optional prior to guide field boundary delineation, provided they are supplied in a compatible GeoParquet file.

Sample-free field boundary delineation

ocular field boundary delineation uses a multi-temporal spectral feature stack. It minimises the need for labelled training data by using the supplied point as a reference (seed). The four modular functions are:

• segment_area(): initial search to classify a plausible field area using a constrained breadth-first search flood fill.
• trace_perimeter(): perimeter tracing for refining irregular or noisy field boundaries.
• segment_interior(): within-field segmentation to remove areas whose VI values fall outside the user-specified bounds or differ from the seed signature for the targeted crop production zone.
• split_area(): field splitting applicable where a targeted zone is surrounded by other cropping areas, or multiple adjacent zones are required in a single mask.

Placing these functions at different stages produces different results, in addition to modifying the parameter settings. ocular pipelines offer a convenient way to test and apply post-retrieval steps.

Global field boundary data from Fields of The World (FTW) can be incorporated as an optional prior to guide delineation, provided they are supplied in a compatible field boundary GeoParquet file. Whether this improves accuracy requires independent validation. As such, it is considered an experimental feature.

Landsat-MODIS data fusion

Landsat-MODIS data fusion uses daily MCD43A4 NBAR data to estimate additional VI values not covered by Landsat. Leave-one-out validation is available against held-out Landsat scenes. This provides reportable metrics for assessing the viability of fused estimates for your workflow. However, these metrics do not establish general accuracy or replace independent validation. Landsat-MODIS data fusion is therefore also marked as experimental.