Assessing performance of spatial statistical stream network models in predicting alkalinity: A comparative study of two river basins in Türkiye

Abstract

Developing and implementing water quality models that utilize sophisticated processes and plenty data in large river basins can be a challenging task. Factors such as sampling design, prediction at unsampled sites, and derivation of new spatial information in monitoring studies can restrict the methodologies applicable to water quality modeling. In recent years, spatial stream network (SSN) models, incorporating hydrologic distance and topological data structures, have emerged as a novel approach for predicting water quality in extensive study areas. These models have integrated conventional geostatistical methodologies and have demonstrated their effectiveness in analyzing water quality within large river basins. In this study, we aimed to investigate the feasibility of using SSN to predict alkalinity through a comparative analysis of two distinct river basins in Türkiye. We evaluated the prediction performance of SSN and the spatial characterization of the basins using spatial data sets as model components. The models are based on moving-average (MA) functions and constructed by evaluating alkalinity observations at 90 locations in Kucuk Menderes River Basin and 224 locations in the Coruh River Basin, using torgegram, correlogram, principal component analysis, and exhaustive search algorithm for covariate selection. In this research, we evaluated the optimal combination of covariates obtained from 20 different variables within the context of five spatial data set classes: topography, geology, climate, land use, and anthropology. The outcomes of the model indicate that impact of anthropogenic activities on autocorrelation based on hydrologic distance, impact of spatial covariates on variance, and root mean square prediction error (RMSPE) values of the predictions in Kucuk Menderes are more effective than those in the Coruh. The findings from the model predictions show that modeling water quality on stream network scale by using topological relationships and autocorrelation based on hydrologic distance can be recommended for rivers affected by complex and detailed processes.