Multivariate analysis of watershed health and sustainability in Turkey

Abstract

National hydrological network data between 1970 and 2002 for 96 stations across 25 watersheds were used to monitor annual trends in 13 variables: streamflow rates, water temperature, pH, electrical conductivity (EC), sediment concentration and eight nutrient levels (Na, K, Ca+Mg, CO3, HCO3, Cl, SO4 and boron). The dataset was analysed with multiple linear regression (MLR) models, principal components analysis (PCA) and hierarchical cluster analysis (CA). The Turkish watersheds have experienced a significant increase in pH, K, CO3 and a significant decrease in streamflow rate and sediment concentration between 1970 and 2002, with considerable spatial variations. There was also an increasing trend in streamwater temperature, at a rate of 0.05°C yr–1 (p > 0.05). The MLR models had high r 2 values of 69.6% to 99.9% at p ≤ 0.001 for 12 out of the 13 variables, with r 2 of 42% for boron (p < 0.05). PCA reduced the dimensionality of the dataset to four principal components that explain most (81.7%) of the variance. CA was able to distinguish six geographically associated groupings of watersheds with similar attributes in concordance with the climate zones of Turkey, despite the use of different clustering methods (complete, McQuitty, average, centroid and single linkage methods). Multivariate analyses of dynamic watershed characteristics provide the basis on which preventive and mitigative measures can be tailored to secure and enhance watershed health and sustainability.