Scalable Spatial Stream Network (S3N) Models

TL;DR

This paper introduces S3N models that extend Gaussian processes for scalable, efficient modeling of species distributions in complex river networks, enabling large-scale ecological mapping.

Contribution

The paper presents a novel S3N framework that significantly improves computational efficiency of spatial stream network models for large ecological datasets.

Findings

S3N accurately recovers spatial and covariance parameters.

S3N reduces bias and variance compared to standard models.

Demonstrated on 285 fish species in the Ohio River Basin.

Abstract

Understanding how habitats shape species distributions and abundances across spatially complex, dendritic freshwater networks remains a longstanding and fundamental challenge in ecology, with direct implications for effective biodiversity management and conservation. Existing spatial stream network (SSN) models adapt spatial process models to river networks by creating covariance functions that account for stream distance, but preprocessing and estimation with these models is both computationally and time intensive, thus precluding the application of these models to regional or continental scales. This paper introduces a new class of Scalable Spatial Stream Network (S3N) models, which extend nearest-neighbor Gaussian processes to incorporate ecologically relevant spatial dependence while greatly improving computational efficiency. The S3N framework enables scalable modeling of spatial stream networks, demonstrated here for 285 fish species in the Ohio River Basin (>4,000 river km). Validation analyses show that S3N accurately recovers spatial and covariance parameters, even with reduced bias and variance compared to standard SSN implementations. These results represent a key advancement toward large-scale mapping of freshwater fish distributions and quantifying the influence of environmental drivers across extensive river networks.