Evaluation of three prevalent global riverine nutrient transport models

Abstract

Abstract Global nitrogen (N) and phosphorus (P) transport models provide a quantitative assessment of riverine nutrient source, transport, accumulation and depletion processes. By conducting a meta-analysis of the modeled results and accuracy metrics from previous studies, this study evaluated the accuracies and their influencing factors of three prevalent global riverine N and P transport models (Global NEWS, IMAGE-GNM and WorldQual). The Global NEWS model exhibited higher accuracies in predicting riverine dissolved organic nitrogen (DON; R2 = 0.58, NSE = 0.57) and dissolved organic phosphorus (DOP; R2 = 0.59, NSE = 0.59) yields compared to riverine dissolved inorganic nitrogen (DIN; R2 = 0.56, NSE=-0.80) and dissolved inorganic phosphorus (DIP; R2 = 0.33, NSE=-0.12) yields. The DIN and DIP sub-models of Global NEWS were applicable for basins with areas greater than 2.2×104 km2 and 3.2×104 km2, respectively. The IMAGE-GNM model demonstrated satisfactory accuracies in predicting riverine total nitrogen (TN; R2 = 0.56, NSE = 0.53) and total phosphorus (TP; R2 = 0.59, NSE = 0.48) concentrations, particularly in European basins. The IMAGE-GNM model performed better for simulation of riverine TN concentration when data set was longer than 21 years and for regions above 54°N, and for simulation of riverine TP concentration when data set was longer than 22 years and for regions above 55°N. The WorldQual model demonstrated relatively poor performance in simulating riverine TN (R2 = 0.76, NSE = 0.34) and TP (R2 = 0.71, NSE=-0.25) concentrations. For model improvements in future, the Global NEWS and WorldQual would benefit from more detailed in-stream nutrient retention/release and transformation modules, while improved chemical weathering dynamics could further enhance the Global NEWS. For the IMAGE-GNM, modification of the soil erosion module is warranted to enhance efficiency in basins outside Europe. Consideration of legacy effects is required to improve these three models. The results of this study provide valuable guidance for the model selecting and improvement for specific needs.