CSTAT+: A GPU-accelerated spatial pattern analysis algorithm for high-resolution 2D/3D hydrologic connectivity using array vectorization and convolutional neural network operators. (October 2019)

• Record Type: Journal Article
• Authors: Yu, Feng; Harbor, Jonathan M.
• Is Part Of: Environmental modelling & software. Volume 120(2019)
• Publication Date: 2019-10
• Subjects: Connectivity statistics -- GPU-Accelerated computing -- Gridded hydrological patterns -- Vectorization -- Convolutional neural network; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2019.104496
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Connectivity is a state-of-the-art concept that improves the current understanding of hydrological processes at multiple scales. Two-point connectivity statistics provide a promising approach to quantify hydrologic connectivity, which measures the probability of any two nodes in a hydrologically relevant pattern that are connected based on their separation distances. However, limited computational capacity has been the main constraint for implementations in large gridded patterns (>1 million nodes). Here, we propose a new algorithm based on array vectorization and convolutional neural network operators (convolution and pooling) that leverages parallel computational capacity of a GPU. Test results suggested that the new algorithm significantly increases the computational efficiency and is also sensitive to the variability of connectivity states and robust to complex topography. We envision that our algorithm can pave the way for investigating behaviors in large-scale (e.g., watershed) processes, based on quantifying the connectivity of gridded hydrological patterns and digital elevation models. Highlights: We provide a detailed review of current algorithms for connectivity statistics. We present revisions, improvements, and test result comparisons for a new algorithm. CSTAT + can be efficiently applied to patterns of large grids (>1 million cells). CSTAT+ is sensitive to variability of connectivity states for spatial patterns. CSTAT+ is robust to topographic flats and depressions without preprocessing.