Accurately measuring and estimating trends and variations in nutrient levels is a significant part of managing emerging eutrophic lakes in developing countries. This study developed an integrated approach containing Seasonal Trend Decomposition using Loess (STL) and a dynamic nonlinear autoregressive model with exogenous input (NARX) network to decompose and estimate the nutrient concentrations in Lake Erhai, a preliminary eutrophic lake in China. The STL decomposition results indicated that total nitrogen (TN) concentration of Lake Erhai progressively descended from 2006 to 2014, except for some agriculture area. The total phosphorus (TP) concentration showed an increasing trend from 2006 to 2013 and then decreased in 2014, but in the area near the tourist attractions, TP increased continuously from 2011 to 2014. Seasonal variations in TN and TP indicated that the lowest water quality of Lake Erhai occurred from July to October. Based on results obtained with STL, TP was selected as the sensitive parameter, as it showed a significant deterioration trend, and the area near the tourist attractions was selected as the sensitive area. Three variables (DO, pH, and water temperature) were selected as input parameters to estimate TP using the dynamic NARX model. The NARX modeling results demonstrated that it can accurately estimate TP concentrations with low root-mean-square error (0.0071 mg/L). The study establishes a new approach to better understand trends and variations in nutrient levels and to better refine estimates by identifying more easily accessible physical parameters in a preliminary eutrophic lake.