Circuitscape in Julia: High Performance Connectivity Modelling to Support Conservation Decisions

Conservation applications

• Connectivity maps inform resource evaluations and risk assessments involving population dynamics, genetic diversity, disease, wildfire, and land protection under climate change.The cited applications include decisions about where to invest in land protection or restoration to support species range shifts.Source: §1 · p. 2

Representing landscapes

• Resistance grids assign each landscape cell a relative resistance to movement and can be abstracted as graphs for quantifying ecological distance.Resistance weights may be developed iteratively using expert opinion and information on species’ life histories and movement behaviors [Spear et al., 2010, Zeller et al., 2012].Source: §1 · p. 2

Connectivity approaches

• Least-cost distance models identify a single best route, whereas isolation by resistance models potential movement across multiple pathways as random walks.The single-route approach depends strongly on start and end points and is less suited to comparing restoration or protection options across multiple habitat patches and pathways.Source: §1 · p. 2

Circuitscape

• Circuitscape calculates effective resistance and current flow across heterogeneous landscapes, with flexible inputs for structural and species-specific functional connectivity.Its applications span gene flow, animal movement, fire, water, and disease spread, and its outputs have been used to inform conservation decisions.Source: §1 · p. 2

Computational demands

• The Python package handled problems up to tens of millions of nodes, below increasingly common workloads of hundreds of millions of nodes and dozens of focal pairs.The authors identify these newer community demands as the reason Circuitscape needed an upgrade.Source: §2 · p. 2

Existing alternatives

• GFLOW delivered state-of-the-art performance on large clusters, but its Windows incompatibility and complicated binary builds constrained accessibility and integration.The authors also identify difficulty integrating GFLOW into decision-support tools and workflows as a barrier for practitioners.Source: §2 · p. 2

Julia implementation

• The authors chose Julia for an open-source Circuitscape implementation intended to be maintainable, high performance, accessible, and cross-platform.They describe Julia as combining scripting-language readability with the performance of statically compiled languages.Source: §2 · p. 3
• Julia’s sparse-matrix and factorization support enabled multiple solvers, while its JIT compiler specializes generic code for precision, index types, and platforms.The authors state that these capabilities supported new user-facing features with relatively little effort in a straight reimplementation.Source: §2 · p. 3

Goals and expected use

• The rewrite aims to support relatively fine-resolution modeling of gene flow and species movement at continental or global scales while remaining open source and accessible.The authors expect faster processing of large datasets to encourage high-resolution data use, sensitivity testing, and dynamic simulations.Source: §2 · p. 3

Paper scope

• The paper proceeds from algorithm and numerical-methods overviews to new features, benchmarks, and discussion of collaboration with conservation managers and stakeholders.The stated benchmark coverage includes synthetic problems and real user data.Source: §2 · p. 3

Inputs

• Circuitscape takes a raster landscape as input, with each cell’s value representing habitat, dispersal-route, or movement-barrier qualities.Resistance values are problem-specific and may be derived empirically from movement or genetic data or from expert opinion.Source: §3 · p. 3

Solving

• For each focal-node pair, Circuitscape forms a current-source and sink setup and solves the graph-Laplacian system to obtain nodal voltages.The algorithm then computes effective resistance from the solution.Source: §3 · p. 4

Graph construction

• The model represents grid cells as graph nodes and assigns edge weights proportional to movement probabilities or migrant exchange between neighboring cells.Focal points identify landscape locations of interest, such as habitat patches, protected areas, or populations to connect.Source: §3 · p. 4

Outputs

• Branch currents represent movement probabilities along graph edges, and their accumulation produces nodal-current maps written as raster ASCII grids for GIS.The algorithm also writes nodal voltages to disk.Source: §3 · p. 4

4 Overview of Numerical Methods

• Circuitscape’s main computation applies Ohm’s and Kirchhoff’s laws to large resistive grids by solving a sparse linear system.The system uses the graph Laplacian as the conductance matrix, with current sources and nodal voltages as its other quantities.Source: §4 · p. 5
• Cholesky factorization suits smaller study areas with many focal nodes because one factorization can serve multiple pairs through back substitution.It scales poorly in memory for larger matrices because factorization can produce factors less sparse than the original matrix.Source: §4 · p. 5

4.1 Preconditioner Implementation

• For large problems, the authors use preconditioned conjugate gradients with an algebraic multigrid preconditioner.Their Smoothed Aggregation AMG variant is suited to matrices generated by elliptic partial differential equations, including planar-graph Laplacians.Source: §4 · p. 5, §4.1 · p. 5

5.1 New Solver Mode

• The new Cholesky-based solver supports landscape tiling and sliding-window analyses that yield fine-scale, localized results.It uses the CHOLMOD library; sparse direct solvers can run out of memory on larger matrices because factorization factors are often less sparse than the original matrix.Source: §5.1 · p. 5

5.2 Generic Programming

• A 437-million-pixel Mojave desert tortoise dataset crashed the old version but ran smoothly to completion after the upgrade.The previous Python package’s hardcoded 32-bit matrix indexing caused crashes on simulations with hundreds of millions of nodes; Circuitscape.jl offers a runtime option for 64-bit indexing.Source: §5.2 · pp. 5–6

5.3 Improved Parallel Computations

• Circuitscape.jl extends parallelism to more problem types and can run multiple Circuitscape instances in parallel across landscape tiles.These parallel features work on Linux, macOS, and Windows.Source: §5.3 · p. 6

6.1 Experimental Setup

• The experiments used an Intel Xeon Silver 4114 CPU at 2.20 GHz with 384 GB RAM, Julia v1.1.0, and Circuitscape v5.5.0, compared with Circuitscape v4.0.5.Source: §6.1 · p. 6

6.2 Standard Benchmarks

• The standard benchmark tests used synthetic problems available from the Circuitscape/BigTests repository.Source: §6.2 · p. 6

6.3 Benchmarks on Real Data

• Nearly 18x speedup reduced the Sonoran Desert all-to-one computation from over 2 days to less than 3 hours.The prior version did not support parallelism for this mode; the Julia version supported it with little effort.Source: §6.3 · p. 6

• The Julia upgrade means users no longer have to coarsen analyses because of tooling limitations.Source: §7 · p. 6
• Faster runs enable sensitivity analysis of individual resistance-map layers and rapid prototyping of resistance maps.The paper says previous Circuitscape versions made practical sensitivity testing prohibitively slow.Source: §7 · p. 6
• Cutting execution time from weeks to days or days to hours enables stronger stakeholder engagement and faster conservation-decision turnaround.The authors describe this as improving the feedback loop with policy makers.Source: §7 · p. 6
• Julia’s ease of use and mathematical syntax support fast scripting, while its modern compiler lets prototypes scale to large datasets.Source: §7 · p. 6

• Up to 1800% improvement in computation time and the ability to solve landscapes with billions of pixels are reported for the Julia upgrade.Source: §8 · p. 7
• The upgrade is intended to let researchers analyze ecological processes over large landscapes at fine resolutions.Source: §8 · p. 7
• Circuitscape.jl is open source under the MIT license, with binaries available for download and installation through the Julia package manager.Source: §8 · p. 7