Evaluating the influence of landscape spatial upscaling on the performance of river water quality models

• Published in: Ecological indicators 2024-01, Vol.158, p.111607, Article 111607
• Main Authors: Aalipour, Mehdi, Wu, Naicheng, Fohrer, Nicola, Jabbarian Amiri, Bahman
• Format: Article
• Language: English
• Subjects: Generalisation; Landscape metrics; Landscape structure; LULC; Spatial scale; Upscaling; Water quality model
• ISSN: 1470-160X; 1872-7034
• DOI: 10.1016/j.ecolind.2024.111607

•Scaling up landscape patches impacts the performance of water quality models.•Models at 1:25,000 scale reveal higher performance for water quality models.•1:50,000 models offer reliable insights into the performance of water quality models.•Coarser scales show lower performance of water quality models. This study examines the effect of the upscaling of landscape structure, which arise from generalizing land use and land cover (LULC) maps, on river water quality prediction. The research uses spatial data from 36 river catchments in the southwestern Caspian Sea region. Regression models were developed to understand the links between river water quality and the landscape structure of the catchments through the generalization of LULC maps on four scales ranging from 1:25,000 to 1: 250,000. These models sought to establish relationships between various landscape structure metrics, which include shape index (SHP), fractal dimension index (FDI), perimeter-to-area ratio (PARA), related circumscribing circle (RCC), and contiguity index (CI), and several water quality variables, including total dissolved solids, electrical conductivity, sulphate, bicarbonate, calcium, chlorine, magnesium, and sodium. The results of the study revealed that upscaling, which results in the altering the shape of landscape patches and consequently increasing their irregularity, had diverse effects on the performance of the predicting water quality models. The research findings indicate that the models developed on a scale of 1:25,000 showed substantial performance in prediction, accounting for varying from 68 % to 82 % (p > 0.05) of the total variation in bicarbonate, electrical conductivity, sulphate, calcium, and magnesium. Similarly, the models developed on a scale of 1:50,000, with a predictive performance varying from 68 % to 72 % (p > 0.05), were relatively reliable in describing the variation in sodium, total dissolved solids, and chlorine. In comparison, the predictive performance of the models developed on scales of 1:100,000 and 1:250,000 was comparatively lower, with explanatory power ranging from 53 % to 63 % and 39 % to 61 %, (p > 0.05), respectively, for the corresponding water quality variables. These models demonstrated weaker associations with water quality variables than those developed at finer scales.