Predicting water quality using partial least squares regression of land use and morphology (Danjiangkou Reservoir, China)

Abstract

According to previous hydrology studies, both the land-use type and the spatial pattern of landscapes affect the quality of water in river basins. However, demonstrating and quantifying the effects of landscape patterns on water quality remain challenging. The present study was conducted in the water source protection zone of Danjiangkou in Hubei Province, China. Based on locations of water quality monitoring stations in the area of the reservoir and the hydrological analysis module in the ArcGIS software, the study area was divided into 10 sub-basins. The “source-sink” landscape pattern obtained from land use/land cover (LULC) analysis of the sub-basins was then used as the medium. Binary data characterized as foreground and background required for morphological spatial pattern analysis (MSPA) corresponded to source and sink landscape types. Partial least squares regression (PLSR) analysis of water quality and the MSPA data were integrated to comprehensively evaluate the impacts of land-use types and landscape pattern changes on water quality in the basin. The results show that land use, landform and spatial landscape pattern have a important impact on the water quality of the watershed. When using the PLSR model based on the “source-sink” morphological spatial pattern to predict water quality, compared to the model analysis solely based on LULC, the MSPA based PLSR improves the principal component grouping, x-variable interpretation rate, y-variable interpretation rate (R2), model prediction rate (Q2), and model prediction accuracy. This study shows that that landscapes is demonstrated to affect water quality and that the use of multiple methods - PLSR, LULC and MSPA data - is a useful complement to hydrological literature. PLSR modeling combined with LULC analysis and MSPA data can accurately predict the water quality of the basin. The use of this improved model for water quality prediction is the focus of this study.