analogs

The analogs package implements a general framework for distance-based neighborhood models, where analogs are selected using climate and/or geographic distance constraints, then summarized via weighted statistics or local regression. Common methods like climate analog impact models, climate velocity, and geographically weighted regression are all specific configurations of this framework.

Every analysis follows the same two-stage pattern: (1) select a neighborhood of analog locations from a reference pool based on climate similarity, geographic proximity, or both; and (2) summarize the neighborhood using counts, weighted means, regression coefficients, or other statistics.

Core Functions

The core engine is analog_search(), which exposes the full flexibility of the framework. Simplified wrapper functions configure it for common analysis types:

• analog_impact() predicts ecological state variables using analog impact models (AIMs)
• analog_regression() fits local weighted regressions across analog neighborhoods
• analog_velocity() finds the nearest geographic analogs under a climate constraint
• analog_similarity() finds the nearest climate analogs within a geographic constraint
• analog_availability() counts analogs that meet climate and distance thresholds
• analog_intensity() computes distance-weighted aggregations of analog properties

Distance metrics

Geographic distance computations support projected coordinates as well as lon-lat coordinates via great circle distances.

Climate similarity is measured using Euclidean or Mahalanobis distance in climate space. Multivariate climate distances are Euclidean by default, but you can use mahalanobis_transform() to implement Mahalanobis distance based on global spatial covariance, or supply x_cov to any analog function to account for site-specific covariance patterns based on historic temporal climate variability.

Computational optimization

The package is designed to maximize performance in various ways:

• The core search architecture is built in optimized C++.
• Parallel processing is available via the n_threads parameter.
• The package uses a lattice search index to efficiently query large candidate pools.
• For repeated queries against the same candidate pool, build_analog_index() can be used to pre-build a reusable search index.
• Optimal index parameters for your specific data can be identified with tune_index_res(), which is used internally by default.
• Memory-safe queries for large raster datasets are available via tiled_analog_search().
• To increase speed at the cost of some precision, you can adaptively downsample your reference data.

Installation

You can install the development version of analogs from GitHub with:

pak::pak("matthewkling/analogs")

Documentation

See the function documentation linked above for detailed usage examples, or browse the full reference documentation.