Applications of soft computing and statistical methods in water resources management.

摘要

Water resource management is the development and use of different techniques for water system planning, development, and operation to overcome problems related to quality and quantity of water. With the increase of pressures on water resources, namely anthropogenic activity and climate change, the ability to accurately predict extreme conditions continues to be a challenge to decision makers and watershed managers. The objectives of this study were to analyze and test the ability of new modeling techniques to find robust and cost-effective models for sustainable water resource managements in both water quantity and quality fields. Water quantity: Stream networks are the blood vessels of terrestrial and aquatic life in a watershed. Therefore, flow decreases during dry seasons and can directly impact the sustainability of ecosystem health. Index flow is the criterion that determines the minimum flow rate, which maintains and protects stream aquatic ecosystems. Therefore, this index was chosen to describe the impacts of water withdrawals on stream ecosystem health. Having the knowledge and the ability to precisely determine water withdrawals within a watershed using index flow is essential for decision makers and watershed managers. In the water quantity part of this study, various new modeling techniques were tested to find more robust approach(s) for estimating the index flow for ungaged streams in the State of Michigan. Four different techniques, linear regression, fuzzy regression, fuzzy expert, and adaptive neuro-fuzzy inference system (ANFIS), were evaluated using a 10-fold cross validation method. Results of the study showed that the fuzzy expert (Mamdani) model was the most robust technique for modeling index flow. Water quality: Sediment is considered the largest surface water pollutant by volume, which needs to be addressed through surface water quality planning and managements. In the planning process, different management scenarios have to be evaluated by watershed managers and stakeholders, which require multiple water quality parameter forecasting and estimation. Physically based models are considered good techniques for sediment estimations; however, they require a large number of parameters and massive calculations, especially during different management scenario evaluations. For the simulation process, the use of new cost-effective modeling approaches to reproduce the results obtained from a physically based (input intensive) models will save time and calculation efforts. In the water quality part of this study, two fusion or blend methods were created to model the sediment load for the Saginaw River Watershed. ANFIS and Bayesian Regression models were tested to find the best alternative(s) to a calibrated physically based model (Soil and Water Assessment Tool - SWAT). For these two models, four different method-types were considered and tested, namely General, Temporal, Spatial and Spatiotemporal. Both techniques, Bayesian Spatiotemporal and ANFIS Spatial models were revealed as good alternatives to the SWAT model for sediment estimations at the watershed scale (global level). However, at the subbasin scale (local level), both Bayesian and ANFIS techniques showed satisfactory results for about 50% of the total of 155 subbasins in the watershed. Transformation of sediment data improved the forecasting capability of both ANFIS and Bayesian techniques even though sediment data still had a bimodal distribution after the transformation.