Orientadores: Secundino Soares Filho, Ivette Luna Huamani / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-25T07:58:05Z (GMT). No. of bitstreams: 1 Lopes_MaianaSantos_M.pdf: 4275882 bytes, checksum: 1bc688bd7ffd93b0d9522e1800541de2 (MD5) Previous issue date: 2014 / Resumo: A previsão de vazões médias mensais é um insumo fundamental para o planejamento da operação das usinas hidrelétricas do Sistema Interligado Nacional (SIN). Durante os últimos anos, diferentes modelos baseados em inteligência computacional têm sido sugeridos para esse problema. A principal contribuição desta dissertação é o desenvolvimento de um sistema de suporte para a previsão e geração de séries sintéticas de vazões mensais, necessárias para o planejamento da operação das usinas do SIN. Este sistema permite analisar o desempenho de modelos clássicos de geração de séries sintéticas e de previsão de vazões, permitindo comparações entre um conjunto específico de modelos clássicos de séries temporais e de inteligência computacional para todas as usinas hidrelétricas do SIN / Abstract: The prediction of monthly average inflows is a fundamental input for the operation planning of the hydroelectric plants of the National Interconnected System (SIN). During the last years, different models based on computational intelligence have been suggested for this problem. The main contribution of this dissertation is the development of a support system for the prediction and generation of synthetic series of monthly average inflows, necessary for planning the operation of the plants of SIN. This system permits to analyze the performance of different models for forecasting and generation of synthetic series of inflows, allowing the comparison between a set of models based on classical time series and computational intelligence for all the hydropower plants of the SIN / Mestrado / Energia Eletrica / Mestra em Engenharia Elétrica

Dynamic expansions and contractions of stream networks can play an important role for hydrologic processes as they can connect different parts of the landscape to the stream channels. However, we know little about the temporal and spatial variations of stream networks during different flow and wetness conditions. This study focuses on the contraction and expansion of stream networks during different flow conditions in the boreal Krycklan catchment, located in Northern Sweden. The stream network and initiation points were extracted from a gridded digital elevation model (DEM) of 5-meter resolution, and then compared with the stream network initiation points (heads) observed during the spring flood (freshet) period in 2012. From the results of the study, it was clearly seen that the observed stream heads and the stream heads appearing in the stream network map extracted from DEM did not agree very well. 49% of the total observed stream heads (49) fell onto the low order stream branches and headwater streams derived from the DEM. Only few of them exactly matched the modeled stream heads. Moreover, the modeled stream network was much denser than the observed stream network, and so the simple raster based dynamic model developed could not well represent the dynamic stream network extension in the real system. Most headwater streams in the study catchment were man-made ditches, which were dug to drain water wetlands and to increase forest productivity. The majority of observed stream heads were formed by seepage from the saturated surrounding soils, while only a few of them were formed by saturation overland flow.  On the other hand, the dynamic stream network derived from the DEM suggested that the number of streams of lower order and their lengths was sensitive to change in streamflow, especially during the high flow episode.

The National Water Act (Act 36 of 1998) provides for an ecological Reserve as the quantity (flow) and quality of water needed to protect aquatic ecosystems. While there are methods available to quantify the ecological Reserve in terms of flow, methods of linking flow to water quality are lacking. Therefore, the research presented in this thesis investigated various modelling techniques to estimate the effect of flow on water quality. The aims of the research presented in this thesis were: Aim 1: Can the relationship between flow and water quality be accurately represented by simple statistical models? Aim 2: Can relatively simple models accurately represent the relationship between flow and water quality? Aim 3: Can the effect of diffuse sources be omitted from a water quality model and still obtain realistic simulations, and if so under what conditions? Aim 4: Can models that solely use historical monitoring data, accurately represent the relationships between flow and water quality? In Chapter 3, simple Q-C regressions of flow and water quality were investigated using Department of Water Affairs (DWA) historical monitoring data. It was found that while flow versus salinity regressions gave good regression fits in many cases, the Q-C regression approach is limited. A mechanistic/statistical model that attempted to estimate the point and diffuse signatures of nutrients in response to flow was developed in Chapter 4 using DWA historical monitoring data. The model was verified as accurate in certain case studies using observed point loading information. In Chapter 5, statistical models that link land cover information to diffuse nutrient signatures in response to flow using DWA historical data were developed. While the model estimations are uncertain due to a lack of data, they do provide an estimation of the diffuse signature within catchments where there is flow and land cover information available. Chapter 6 investigates the extension of an existing mass-balance salinity model to estimate the effect of saline irrigation return flow on in-stream salinity. The model gave accurate salinity estimates for a low order stream with little or no irrigation within its catchment, and for a permanently flowing river within a catchment used extensively for irrigation. Chapter 7 investigated a modelling method to estimate the reaction coefficients involved in nitrification using only DWA historical monitoring data. Here, the model used flow information to estimate the residence time of nutrients within the studied river reaches. While the model obtained good estimations of nitrification for the data it was applied to, very few DWA data sets were suitable for the model. Chapter 8 investigated the ability of the in-stream model QUAL2K to estimate nutrient concentrations downstream of point and diffuse inputs of nutrients. It was found that the QUAL2K model can give accurate results in cases where point sources dominate the total nutrient inputs into a river. However, the QUAL2K simulations are too uncertain in cases where there are large diffuse source inputs of nutrients as the load of the diffuse inputs is difficult to measure in the field. This research highlights the problem of data scarcity in terms of temporal resolution as well as the range of constituents measured within DWA historical monitoring data for water quality. This thesis in addition argues that the approach of applying a number of models is preferable to applying one model to investigate the research aims, as particular models would be suited to particular circumstances, and the development of new models allowed the research aims of this thesis to be explored more thoroughly. It is also argued that simpler models that simulate a few key processes that explain the variation in observed data, are more suitable for implementing Integrated Water Resource Management (IWRM) than large comprehensive water quality models. From this research, it is clear that simple statistical models are not adequate for modelling the relationship between flow and water quality, however, relatively simple mechanistic models that simulate a limited number of processes and water quality variables, can provide accurate representations of this relationship. Under conditions where diffuse sources are not a major factor within a catchment, models that omit diffuse sources can obtain realistic simulations of the relationship between flow and water quality. Most of the models investigated in this thesis demonstrate that accurate simulations of the relationships between flow and water quality can be obtained using solely historical monitoring data.

The wet weather pattern since the early 1990's has created two problems for the people living in the Red River Valley (RRV): (1) wet field conditions for farmers and (2) more frequent major spring floods in the Red River system. Farmers in the region are increasingly adopting subsurface drainage practice to remove excess water from their fields to mitigate the first problem. However, it is not clear whether subsurface drainage will deteriorate or mitigate the spring flood situation, the second problem. The Soil and Water Assessment Tool (SWAT) model was applied to evaluate the impacts of tile drainage on the Red River's streamflows. The model was calibrated and validated against monthly streamflows at the watershed scale and against daily tile flows at the field scale. The locations and areas of the existing and potential tile drained (PTD) areas were identified using a GIS based decision tree classification method. The existing and maximum PTD areas were found to be about 0.75 and 17.40% of the basin area, respectively. At the field scale, the range of Nash-Sutcliffe efficiency (NSE) for model calibration and validation was 0.34-0.63. At the watershed scale, the model showed satisfactory performance in simulating monthly streamflows with NSE ranging from 0.69 to 0.99, except that the model under-predicted the highest spring flood peak flows in three years. The results of modeling a 100% tiled experimental field showed that about 30-40% of water yield was produced as tile flow. Surface runoff and soil water content decreased about 34% and 19%, respectively, due to tile drainage. However, the impact of subsurface drainage on evapotranspiration (ET) and water yield was mixed. ET slightly decreased in a wet year and slightly increased in a dry year, while the pattern for water yield was opposite to that of ET. The watershed-scaled modeling results showed that a tiling rate of 0.75-5.70% would not have significant effects on the monthly average streamflows in the Red River at Fargo. For the 17.40% tiling rate, the streamflow in the Red River at Fargo might increase up to 1% in April and about 2% in Fall (September to November), while decreasing up to 5% in the remaining months. This SWAT modeling study helped to better understand the impact of subsurface drainage on the water balance and streamflows in the Red River of the North basin. The findings will also help watershed managers in making decisions for the purpose of managing agricultural drainage development in the RRV and other snow dominated watersheds around the world.

The major benefit of geo-referenced exposure modelling tools is the provision of spatially distributed information on expected environmental concentrations. This allows for identifying local and regional concentration differences in the environment which facilitates the development of efficient mitigation strategies. Predicted substance concentrations in the environment are governed by emission rates and representation of the substances' transport and transformation processes on the one hand and by the description of the spatial environmental heterogeneity and temporal variability on the other hand. The shape of river basins and streamflow variability within them is a product of physiographic and climatic factors like e. g. topography, land use, precipitation, or evapotranspiration. These factors are very variable in space and time. This heterogeneity in river basins may have an impact on surface water concentrations of various substances. In this work a hydrological framework for geo-referenced exposure assessment in river networks has been developed which predominantly addresses spatial heterogeneity of river basins. The theoretical background for parameterising a river network for the application of GREAT-ER (Geo-referenced Regional Exposure Assessment Tool for European Rivers) is elaborated and implemented. Quantity of discharge, flow velocity of river water and depth of river bed have to be determined at any location in a river network for the representation of substance dilution, transport and degradation. Temporal variability is handled by a probabilistic approach which demands choice and parameterisation of probability distribution functions to describe the river network characteristics. It is substantiated that discharge and its variation can be described by a lognormal probability distribution. This distribution can be parameterised by spatially distributed information on effective precipitation and specific low flow discharge from the German Hydrological Atlas. Geoprocessing methods are applied to couple information from these maps and the river network. Evaluation of discharge probability distributions by means of gauging data demonstrates good agreement. River depth and flow velocity are estimated on the basis of spatially distributed river structure data and therefore account for actual river morphology more than former approaches do. A comparison with hitherto used flow velocity and depth estimation shows significant differences which trigger perceivable differences in surface water concentration estimates. Identification of the sensitivity of hydrological parameters in terms of chemical fate estimation attaches importance to spatial explicit consideration of river networks. The main benefit of the presented methods is comprehensive incorporation of geo-referenced river basin characteristics into the data basis for the GREAT-ER model because this provides the basis for successful prediction of surface water concentrations by GREAT-ER.

McKinney Butte, a late Tertiary andesite vent and flow complex, is located near the town of Sisters, Oregon, in the upper Deschutes Basin, and is situated along the structural trend that forms the eastern margin of the High Cascades graben (Sisters fault zone and Green Ridge). Rapid development and over appropriated surface water resources in this area have led to an increased dependence upon groundwater resources. A primary concern of resource managers is the potential impact of expanding groundwater use on stream flows and spring discharge. Two sets of springs (McKinney Butte Springs and Camp Polk Springs) discharge to Whychus Creek along the east flank of McKinney Butte, and during low-flow conditions supply a substantial component of the total flow in the creek. Despite their contribution to Whychus Creek, the springs along McKinney Butte are small-scale features and have received less attention than larger volume (> 2 m³/s) springs that occur in the basin (i.e., Metolius Spring and Lower Opal Springs). This study used discharge measurements in Whychus Creek upstream and downstream of the springs, and mixing models using measurements of electrical conductivity and temperature in the springs and Whychus Creek to determine the contribution of the springs to the creek. Isotopic, thermal, and geochemical signatures for the McKinney Butte and Camp Polk Springs, and local streams (Whychus Creek and Indian Ford Creek) and springs (Metolius Spring, Paulina Spring, Alder Springs, and Lower Opal Spring) were assessed to determine the source(s) of the McKinney Butte and Camp Polk Springs. The discharge and hydrochemical data along with hydraulic head data from local wells were used in the development of a conceptual model of groundwater flow for the McKinney Butte area. Discharge from the McKinney Butte Springs supplies the majority of water to Whychus Creek on the east flank of McKinney Butte (~0.20 m³/s), provides up to 46% of the flow in the creek, and is relatively stable throughout the year. Discharge from the Camp Polk Springs is less than 0.05 m³/s. Isotopic, thermal, and geochemical signatures indicate distinct sources for the McKinney Butte and Camp Polk Springs. Groundwater discharged at the McKinney Butte Springs is depleted in heavy stable isotopes (δD and δ¹⁸O) relative to the Camp Polk Springs. Recharge elevations inferred from stable isotope concentrations are 1800-1900 m for the McKinney Butte Springs and 950-1300 m for the Camp Polk Springs. Elevated water temperature in the McKinney Butte Springs relative to the average air temperature at the inferred recharge elevation indicates the presence of geothermal heat and implies deep circulation in the flow system. The temperature in the Camp Polk Springs is not elevated. The Camp Polk Springs, though not the McKinney Butte Springs, contain elevated concentrations of ions Cl, SO₄, and NO₃ that are indicative of contamination. The study results indicate the source of the Camp Polk Springs is shallow groundwater whereas the McKinney Butte Springs discharge water that has circulated deep in the groundwater flow system. Additionally, the hydrochemical traits of the McKinney Butte Springs are similar to Metolius Spring, suggesting discharge from the McKinney Butte Springs is controlled by the structural trend that forms the eastern margin of the High Cascades graben. The significant difference in discharge between the McKinney Butte Springs and Metolius spring may be related to the size of faults that occur locally.

<p>Streamflow prediction in ungauged basins (PUB) is a process generating streamflow time series at ungauged reaches in a river network. PUB is essential for facilitating various engineering tasks such as managing stormwater, water resources, and water-related environmental impacts. Machine Learning (ML) has emerged as a powerful tool for PUB using its generalization process to capture the streamflow generation processes from hydrological datasets (observations). ML’s generalization process is impacted by two major components: data splitting process of observations and the architecture design. To unveil the potential limitations of ML’s generalization process, this dissertation explores its robustness and associated uncertainty. More precisely, this dissertation has three objectives: (1) analyzing the potential uncertainty caused by the data splitting process for ML modeling, (2) investigating the improvement of ML models’ performance by incorporating hydrological processes within their architectures, and (3) identifying the potential biases in ML’s generalization process regarding the trend and periodicity of streamflow simulations.</p><p>The first objective of this dissertation is to assess the sensitivity and uncertainty caused by the regular data splitting process for ML modeling. The regular data splitting process in ML was initially designed for homogeneous and stationary datasets, but it may not be suitable for hydrological datasets in the context of PUB studies. Hydrological datasets usually consist of data collected from diverse watersheds with distinct streamflow generation regimes influenced by varying meteorological forcing and watershed characteristics. To address the potential inconsistency in the data splitting process, multiple data splitting scenarios are generated using the Monte Carlo method. The scenario with random data splitting results accounts for frequent covariate shift and tends to add uncertainty and biases to ML’s generalization process. The findings in this objective suggest the importance of avoiding the covariate shift during the data splitting process when developing ML models for PUB to enhance the robustness and reliability of ML’s performance.</p><p>The second objective of this dissertation is to investigate the improvement of ML models’ performance brought by Physics-Guided Architecture (PGA), which incorporates ML with the rainfall abstraction process. PGA is a theory-guided machine learning framework integrating conceptual tutors (CTs) with ML models. In this study, CTs correspond to rainfall abstractions estimated by Green-Ampt (GA) and SCS-CN models. Integrating the GA model’s CTs, which involves information on dynamic soil properties, into PGA models leads to better performance than a regular ML model. On the contrary, PGA models integrating the SCS-CN model's CTs yield no significant improvement of ML model’s performance. The results of this objective demonstrate that the ML’s generalization process can be improved by incorporating CTs involving dynamic soil properties.</p><p>The third objective of this dissertation is to explore the limitations of ML’s generalization process in capturing trend and periodicity for streamflow simulations. Trend and periodicity are essential components of streamflow time series, representing the long-term correlations and periodic patterns, respectively. When the ML models generate streamflow simulations, they tend to have relatively strong long-term periodic components, such as yearly and multiyear periodic patterns. In addition, compared to the observed streamflow data, the ML models display relatively weak short-term periodic components, such as daily and weekly periodic patterns. As a result, the ML’s generalization process may struggle to capture the short-term periodic patterns in the streamflow simulations. The biases in ML’s generalization process emphasize the demands for external knowledge to improve the representation of the short-term periodic components in simulating streamflow.</p>

Among other sources of uncertainties in hydrologic modeling, spatial rainfall variability, channel hydraulic variability, and model parameter uncertainty were evaluated. The Monte Carlo and Harr methods were used to assess 90% certainty bounds on simulated flows. The lumped watershed model, Hydrologic Simulation Program FORTRAN ? HSPF, was used to simulate streamflow at the outlet of the Luxapallila Creek watershed in Mississippi and Alabama. Analysis of parameter uncertainty propagation on streamflow simulations from 12 HSPF parameters was accomplished using 5,000 Monte Carlo random samples and 24 Harr selected points for each selected parameter. Spatial rainfall variability propagation on simulated flows was studied using six random grid point sets of Next Generation Weather Radar (NEXRAD) rainfall data (i.e., 109, 86, 58, 29, 6, and 2 grid points) from the baseline scenario (115 NEXRAD grid points). Uncertainty in channel hydraulic properties was assessed comparing the baseline scenario (USGS FTABLE) versus the EPA RF1 FTABLE scenario. The difference between the baseline scenario and the remaining scenarios in this study was evaluated using two criteria: the percentage of observed flows within the HSPF 90% certainty bounds (Reliability) and the width of the HSPF 90% certainty bounds (Sharpness). Daily observed streamflow data were clustered into three groups to assess the model performance by each class: below normal, normal, and above normal flows. The parameter uncertainty propagation results revealed that the higher the model Sharpness the lower the model Reliability. The model Sharpness and Reliability results using 2 NEXRAD grid points were markedly different from those results using the remaining NEXRAD data sets. The hydraulic property variability of the main channel affected storm event paths at the watershed outlet, especially the time to peak flow and recessing limbs of storm events. The comparison showed that Harr?s method could be an appropriate initial indicator of parameter uncertainty propagation on streamflow simulations, in particular for hydrology models with several parameters. Parameter uncertainty was still more important than those sources of uncertainty accomplished in this study because all of the median relative errors of model Reliability and Sharpness were lower than +/- 100%.

Societal developments are associated with land use change, and with urbanization in particular. Urbanization can influence hydrological processes by decreasing evapotranspiration and infiltration as well as by increasing streamflow, peak flow and overland flow. This causes higher risks of flooding. Although several studies have investigated the impacts of urbanization on streamflow over the last decades, less is known about how urbanization affects the hydrological processes in peri-urban areas characterized by a complex mosaic of different land uses. This study aimed to model the impact of land use change, or more specifically urbanization, on the hydrological responses of the small peri-urban Ribeira dos Covões catchment (6.2 km2) located in central Portugal. The catchment has undergone rapid land use change since the mid- 1950s associated with conversion of agricultural fields (decreased from 48 to 4%) into woodland and urban areas, which increased from 44 to 56% and from 8 to 40%, respectively. For the study, the hydrological modelling system MIKE SHE was used. Parameters and data of climate, vegetation and soil types were used as input. There were also land use maps and daily streamflow values available for the hydrological years 2008/09 to 2012/13, which were used to calibrate and validate the model. The statistics from the calibration and validation both indicated that the model simulated the streamflow well. The model was designed to examine both how past land use change might have affected the streamflow, and to investigate the impacts on hydrology if the urban area was to be increased to cover 50% of the catchment. It was not only the importance of the urban cover’s size that was tested, but also the placement of additional urban areas. Three future scenarios were run, all with a 50% urban cover, but distributed differently within the catchment. The study did not indicate that an increase in urbanization leads to higher peak flow or streamflow. Neither could any decrease in infiltration be seen. All three scenarios however gave an increase in overland flow of approximately 10% and a decrease in evapotranspiration by 55%, regardless of where the urban areas were added. The reliability of the models can be enhanced by additional climate, soil and vegetation data. This would improve the results and make them more useful in decision making processes in the planning and management of new urban areas. / Samhällets ständiga utveckling medför förändringar i markanvändning. Utvecklingen och förändringarna är framför allt associerade med urbanisering som kan påverka ett avrinningsområdes hydrologiska processer genom att exempelvis reducera dess evapotranspiration och infiltration samt öka vattenföringen, högsta flödet och ytavrinningen. Det i sin tur ökar risken för översvämning. Trots att många studier har undersökt urbaniseringens inverkan på vattenföring de senaste decennierna saknas viss kunskap om dess påverkan på hydrologin i stadsnära avrinningsområden, kännetecknade av flera olika typer av markanvändning. Denna studie syftade till att modellera hur förändringar i markanvändning, eller mer specifikt urbanisering, påverkar hydrologin i det lilla stadsnära avrinningsområdet Ribeira dos Covões (6,2 km2) i centrala Portugal. Avrinningsområdet har genomgått snabba markanvändningsförändringar sedan mitten av 1950-talet i samband med en omvandling av åkrar (täckningsarean har minskat från 48 till 4 %) till skogsmark och urbaniserade områden, vilkas storlek har ökat från 44 till 56 % respektive 8 till 40 %. För att uppfylla syftet har den hydrologiska modellen MIKE SHE använts. Parametrar avseende klimat samt vegetations- och jordegenskaper användes som indata till modellen. Det fanns också tillgång till en markanvändningskarta över området samt dagliga flödesvärden mellan de hydrologiska åren 2008 och 2013. Dessa användes för att kalibrera och validera modellen. Statistiken för både kalibreringen och valideringen indikerade en fullt acceptabel modell. Modellen var avsedd att undersöka dels hur tidigare förändring i markanvändning kan ha påverkat vattenföringen, dels för att studera effekten på hydrologin om urbaniseringen fortgår tills dess täckning är 50 % av avrinningsområdet. Det var inte bara betydelsen av de urbana ytornas storlek som testades, utan även placeringen av dem. Tre framtidsscenarier togs fram, alla med en urban yta på 50 % fördelad olika inom avrinningsområdet. Studien indikerade inte att ytterligare urbanisering ökar vare sig flödet eller det högsta flödet. Inte heller gav de någon minskning av infiltration. Alla tre scenarierna gav emellertid en ökning av ytavrinningen med cirka 10 % och en minskning av evapotranspirationen med 55 %, oavsett placering av de urbana ytorna. Modellernas tillförlitlighet skulle kunna förbättras med hjälp av ytterligare klimat-, vegetations- och jordindata. Det skulle förbättra resultaten och göra dem användbara i beslutsfattanden vid planering och utveckling av nya urbana områden.

Submitted in fulfilment of the requirements of the degree of Master of Engineering , Durban University of Technology, Durban, South Africa, 2016. / Reservoirs are designed to specific volume called the dead storage to be able to withstand the quantity of particles in the rivers flowing into it during its design period called its economic life. Therefore, accurate calculation of the quantities of sediment being transported is of great significance in environment engineering, hydroelectric equipment longevity, river aesthetics, pollution and channel navigability. In this study different input combination of monthly upstream suspended sediment concentration and upstream flow dataset for Inanda Dam for 15 years was used to develop a model for each month of the year. The predictive abilities of each of the developed model to predict the quantity of suspended sediment flowing into Inanda Dam were also compared with those of the corresponding developed Sediment Rating Curves using two evaluation criteria - Determination of Coefficient (R2) and Root-Mean-Square Error (RMSE). The results from this study show that a genetic programming approach can be used to accurately predict the relationship between the streamflow and the suspended sediment load flowing into Inanda Dam. The twelve developed monthly genetic programming (GP) models produced a significantly low difference when the observed suspended sediment load was compared with the predicted suspended sediment load. The average R2 values and RMS error for the twelve developed models were 0.9996 and 0.3566 respectively during the validation phase. The Genetic Programming models were also able to replicate extreme hydrological events like predicting low and high suspended sediment load flowing into the dam. Moreover, the study also produced accurate sediment rating curve models with low RMSE values of between 0.3971 and 11.8852 and high R2 values of between 0.9833 and 0.9962. This shows that sediment rating curves can be used to predict historical missing data of the quantity of suspended sediment flowing into Inanda Dam using existing streamflow datasets. The results from this study further show that the predictions from the Genetic Programming models are better than the predictions from the Sediment Raring Curve models, especially in predicting large quantities of suspended sediment load during high streamflow such as during flood events. This proves that Genetic Programming technique is a better predictive tool than Sediment Raring Curve technique. In conclusion, the results from this study are very promising and support the use of Genetic Programming in predicting the nonlinear and complex relationship between suspended sediment load and streamflow at the inlet of Inanda Dam in KwaZulu-Natal. This will help planners and managers of the dam to understand the system better in terms of its problems and to find alternative ways to address them.