Treffer: A learning-based parallel geocomputation engine for heterogeneous spatial domain.

The growing complexity of geocomputations presents significant challenges in efficient resource utilization. The current expert-driven formulation paradigm for modeling the computational characteristics of heterogeneous spatial domains frequently results in suboptimal load distribution. Here, the study presents the learning-based parallel geocomputation engine (LPGE), a novel framework that leverages advancements in machine intelligence to enhance computing performance. LPGE operates across three levels: meta-intelligence, perceptual intelligence, and cognitive intelligence. At the meta-intelligence level, LPGE introduces a comprehensive computational feature representation for multimodal spatial data, encompassing morphological structure, aggregate quantity, spatial distribution, and topological association features. At the perceptual intelligence level, LPGE incorporates innovative artificial intelligence models proposed in this study, enabling the automatic modeling of computational features and computational intensity (CI) across heterogeneous spatial domains. At the cognitive intelligence level, LPGE employs a CI-augmented hybrid scheduling system incorporating CI perception and adaptive task stealing to balance computational workloads. Evaluated on large-scale spatial intersection on vector data and viewshed analysis based on digital elevation model data, LPGE demonstrates a 20× improvement in load balancing over conventional methods, demonstrating its potential to advance intelligent geocomputing and provide scalable solutions for large-scale geoprocessing challenges. [ABSTRACT FROM AUTHOR]