Modelling nitrogen loads at the catchment scale under the influence of land use

• Published in: Physics and chemistry of the earth. Parts A/B/C Vol. 29; no. 11; pp. 811 - 819
• Main Authors: Payraudeau, S, Cernesson, F, Tournoud, M.G, Beven, K.J
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 2004; Elsevier [2002-....]
• Subjects: Earth Sciences; Hydrology; Land use; Model sensitivity; Prediction uncertainty; Sciences of the Universe; Water quality model; France; MED, Western Mediterranean; Water quality model; Model sensitivity; Prediction uncertainty; Land use
• ISSN: 1474-7065; 1873-5193
• DOI: 10.1016/j.pce.2004.05.008

Land use data are essential for water quality models. Pollutant inputs to streams are indeed a direct function of human activities that can be represented, at least approximately, in terms of land use. Remote sensing is a valuable data source to determine the land use on a catchment. However the land use data obtained by this kind of information are subject to significant uncertainties, including misclassification or categorical uncertainty. This paper presents a method to analyse the impact of the land use categorical uncertainty on the responses of a nitrogen load model at the outlet of a catchment. We use the POL model, a semi-distributed event-based model on a French Mediterranean rural catchment and we focus on agricultural land use. First, the sensitivity analysis realised by simulations considering a uniform land use on the catchment, shows a great sensitivity of the estimated load to the land use change. Second, the categorical land use uncertainty is analysed on a total nitrogen load prediction set calculated with randomly generated land use maps consistent with the confusion matrix that characterizes misclassification of land use. Thus, from 1% to 10% of misclassified agricultural area results in a variation of almost 40% on nitrogen loads for the three studied events. Misclassified areas explain from 46% to 75% of the variance of the estimated nitrogen load. These first results illustrate the importance of sensitivity and uncertainty analyses to improve the confidence of a water quality model and need to be extended to other input data sets.