Estimation of the spatial variation of stream flow by neural models and surface algorithms

Abstract

The present study tried to estimate spatial variation of stream flow in face of climatic uncertainty. Seven neurogenetic models with different but related input variables were used to predict stream flow of both gauged and ungauged sub-basins of two river networks of Eastern India's intratropical region. The models were validated and the better model was selected for estimation of stream flow. PARITYCGD was found to be the better model due to its lower RMSE and higher efficiency than any other considered models. The PARITYCGD model was then compared with three conceptual hydrologic models, and here also it was selected as the better model due to higher efficiency and reliability and lower RMSE and uncertainty than the other considered conceptual models. The radial basis surface interpolation was now used to generate spatial variation of stream flow within the two river networks due to the generated weather scenarios by PRECIS climate models. According to the results, the degradation of catchment, which is generally interpreted from high magnitude of stream flow, was observed in north-west and north-east region within the two river networks as per the surface diagram generated from model predictions due to observed rainfall and land-use data. For the future climatic data, spatial variation of stream flow was found to be concentrated in the same two regions only the areas showed an increasing trend, which was more in A2 scenario than in the case of B2 scenario of climate change. The model output was applied to generate spatial variation of water quality and pollution, which are explained in Chapters 10 and 11. © 2010 Springer Science+Business Media B.V.