Using genetic algorithms to calibrate a water quality model.

With the increasing concern over the impact of diffuse pollution on water bodies, many diffuse pollution models have been developed in the last two decades. A common obstacle in using such models is how to determine the values of the model parameters. This is especially true when a model has a large number of parameters, which makes a full range of calibration expensive in terms of computing time. Compared with conventional optimisation approaches, soft computing techniques often have a faster convergence speed and are more efficient for global optimum searches. This paper presents an attempt to calibrate a diffuse pollution model using a genetic algorithm (GA). Designed to simulate the export of phosphorus from diffuse sources (agricultural land) and point sources (human), the Phosphorus Indicators Tool (PIT) version 1.1, on which this paper is based, consisted of 78 parameters. Previous studies have indicated the difficulty of full range model calibration due to the number of parameters involved. In this paper, a GA was employed to carry out the model calibration in which all parameters were involved. A sensitivity analysis was also performed to investigate the impact of operators in the GA on its effectiveness in optimum searching. The calibration yielded satisfactory results and required reasonable computing time. The application of the PIT model to the Windrush catchment with optimum parameter values was demonstrated. The annual P loss was predicted as 4.4 kg P/ha/yr, which showed a good fitness to the observed value.

An investigation into the inputs controlling predictions from a diffuse phosphorus loss model for the UK; the Phosphorus Indicators Tool (PIT).

A simple catchment scale model simulating diffuse phosphorus (P) loss from agricultural land to water, the Phosphorus Indicators Tool (PIT), has been developed. Previous research has shown that this model worked well in simulating the average annual P lost from two catchments: Windermere and Windrush, but it was not known which drivers in the model had the greatest control on predicted P delivery to water from agricultural land. In order to simulate the P export from each catchment source via each hydrological pathway specified individually, 108 coefficients are used in the model code. A univariate sensitivity analysis was conducted to evaluate which coefficient exerted the greatest control on the model output. Results from the univariate analysis suggest that the model is sensitive to a number of coefficients, but importantly, not all of the coefficients that were varied in the sensitivity analysis, altered the model output. The PIT model has been calibrated by optimizing results from the univariate analysis against observed data in the Windermere catchment. The simulated results from model calibration fit the observed data well, at the 95% level. This paper describes the methodology developed for the univariate analysis and evaluates the model calibration procedure against observed data from the Windermere catchment.

Evaluating colloidal phosphorus delivery to surface waters from diffuse agricultural sources.

Colloid-facilitated phosphorus (P) delivery from agricultural soils in different hydrological pathways was investigated using a series of laboratory and field experiments. A soil colloidal P test was developed that yields information on the propensity of different soils to release P attached to soil colloids. The relationship between turbidity of soil extracts and total phosphorus (TP) was significant (r2 = 0.996, p < 0.001) across a range of agricultural soils, and a strong positive relationship (r2 = 0.86, p < 0.001) was found between "colloidal P" (H2O-CaCl2 extracts) and turbidity. Linear regression of the proportion of fine clay (<2 microm) for each soil type evaluated against the (H2O-CaCl2) colloidal P fraction gave a weak but positive relationship (r2 = 0.38, p = 0.082). The relative contribution of different particle-size fractions in transporting P in agricultural runoff from grassland soils was evaluated using a randomized plot experiment. A significant difference (p = 0.05) in both TP and reactive phosphorus (RP) in subsurface flow was recorded for different particle-size fractions, with most TP transferred either in association with the 2-microm fraction or with the 0.001-microm or smaller fractions. Total P concentrations in runoff were higher from plots receiving P amendments compared with the zero-P plots; however, these differences were only significant for the >0.45-microm particle-size fractions (p = 0.05), and may be evidence of surface applications of organic and inorganic fertilizers being transferred through the soil either as intact organic colloids or attached to mineral particles. Our results highlight the potential for drainage water to mobilize colloids and associated P during rainfall events.