Simulation And Prediction Of Nutrient Loads In A Non-point Source Pollution River Based On LOADEST And PA/GRNN Models

Tangpu reservoir as a drinking water source in Shaoxing city,Serving a population of nearly 3000000 people, the water quality of reservoir is excellent, but the total nitrogen, total phosphorus exceed the standard, which belong to medium nutrition level, blue-green algae was outbreak in 2010, burying hidden trouble for Shaoxing city living water.This article is in view of Tangpu reservoir basin,based on 2009～2011 monthly monitoring data of hydrology and water quality in reservoir upstream river-Shuangjiang river and Wanghua river,carring out simulation and prediction of research water source TN、TP loads. The core contents include principal component analysis of water quality index、TN、TP loads simulation under different time scales based on LOADEST model and the prediction of TN、 TP monthly loads based on based on PA/GRNN model. The main conclusions are as follows:Analysing the reservoir water quality indicators by principal component analysis, the results show that the first three principal components accountes for 77.87% and 74.31% of the total variance in Shuangjiang river and Wanghua river,respectively,which can explain most of the reason of the overall water quality change.the average monthly water cluster analysis and principal component analysis show that the basin has obvious dry season and wet season,nitrate nitrogen in the field inflows into the river by surface runoff and groundwater during wet season which causes agricultural non point source pollution, the ammonia nitrogen comes from domestic pollution sources is the main factor affecting the water quality during dry season.Establishing a LOADEST model using 2009～2010 bimonthly water quality data and continuous daily flow data of Shuangjiang river and Wanghua river section which in the upper reaches of Tangpu basin, the results show that seven parameters model of Shuangjiang river TN load and six parameters model of Wanghua river TP load are satisfactory,validation period of R2 of TN、TP loads between measured values and the model value reached 0.98 and 0.91 respectively, the Nash-Sutcliffe coefficients are respectively 0.97 and 0.92,the simulation precision is high,simulatting the different time scales(day/month/season/year) of 2009～2011 Shuangjiang river TN、TP loads;the results show that the months of the maximum TN、TP loads are march、june、august,the main control factors of river nutrient load in addition to rainfall runoff, human activities (fertilization,irrigation) also have an important impact.Using wavelet analysis analying Shuangjiang river monthly flow and TN monthly concentration simulated by LOADEST model,explores the relations. between the load、flow and concentration,the results show that the main cycle fluctuations of both flow and TN concentration are close,amplitude fluctuations and wave energy of flow are far more fierce than which of TN concentration, so the TN load (the product of flow rate and concentration) depends on the sharply fluctuating flow mainly,which covers up smoothly fluctuating concentration,explaining the reason why the relationship between the flow and calculated load is good but the correlation between the concentration is less obvious.In order to further the research of predict of river nutrient load,this study tries to combine PA model with GRNN model together,proposes PA/GRNN load forecasting model,the water quality time series are divided into deterministic component and random component in PA/GRNN model firstly,using the PA model and GRNN model respectively to predict the deterministic cycle components and random error components of water quality time series,and their sums are TN、 TP monthly forecasting loads,compared with the actual loads,TN、TP month loads forecasting average relative error are 16.39%、18.20%, results are satisfactory, this study shows that application of PA/GRNN model for water quality prediction not only can effectively reveal the periodic variation characteristics and internal random variation characteristics of water quality time series,but also can broaden the water quality prediction index, prolong the prediction length and improve the precision of prediction, which has important theoretical and practical significance to grasp the future total input of pollutants of river and warning of water eutrophication.