GIS code and data_Procedural grassland scene construction

Realistically modelling grasslands presents a persistent challenge because of their inherent structural complexity, extensive spatial coverage, and high component density. Previous research prioritized surface appearances through texture tilling, fundamentally neglecting the internal organizational principles that govern species compositions and component distributions. This work introduces a procedural framework for generating geographically plausible grassland scenes. We first integrate parametric scene modelling with visual perception theory to ensure visual realism and faithful reproductions of the observed colour distributions at the individual component level. Moreover, a deep neural network is developed to learn the underlying spatial distribution patterns of grassland components. This network enables the generation of new grassland imagery that preserves the learned statistical properties of colour and spatial attributes derived from reference samples. By coupling the framework with prepared 3D models, it can achieve a geographically consistent generation process across arbitrary spatial extents, effectively mitigating visual repetition and randomness. Validations demonstrate significant visual realism improvements across diverse grasslands. Our core findings establish that explicit modelling of component-level attributes and their spatial distribution patterns is essential for producing high-fidelity grassland scenes, providing novel insights for advancing toward the theory and practice of virtual geographical environments.