On the use of finite‐difference and neural‐network models to evaluate the impact of underground water overexploitation

The purpose of this study is to include expert knowledge as one part of the modelling system and thereby offer the chance to create a productive interactive system between expert, mathematical model, ASM, and artificial neural networks (ANNs). An attempt to determine outflow‐influencing parameters in order to simulate spring flow is presented. The Bouteldja dune aquifer is fed by rains and streaming water on the sandy argillaceous relieves in the Est. The lateral passage to the gravel of the Bouteldja Plain is marked by numerous bogs that correspond to the piezometric level. These bogs have long been an environment for migratory birds and a natural reserve for many species. However, the continued exploitation of about 30 wells has negatively influenced the hydrodynamic equilibrium of the aquifer and has brought a diminution of the sources' capacity. In this study, we tried by using a hydrodynamic model and the neural network to ascertain the state of the resources and to identify the factors responsible for the decreasing flows of the three principal springs of the area (Bougles, Bourdim and Titteri) by using neural networks. The results obtained show a continued exhaustion of the reserve since 1986 with a large cone of depression. The ANNs show that the decrease in flows of the springs is not only due to the unfavourable climatic conditions, but also to the intensive exploitation of the aquifer. These results show that the groundwater reserves are decreasing over time, thus highlighting the need to take some urgent measures to stop this phenomenon. Copyright © 2006 John Wiley & Sons, Ltd.     

Short term streamflow forecasting using artificial neural networks

The research described in this article investigates the utility of Artificial Neural Networks (ANNs) for short term forecasting of streamflow. The work explores the capabilities of ANNs and compares the performance of this tool to conventional approaches used to forecast streamflow. Several issues associated with the use of an ANN are examined including the type of input data and the number, and the size of hidden layer(s) to be included in the network. Perceived strengths of ANNs are the capability for representing complex, non-linear relationships as well as being able to model interaction effects. The application of the ANN approach is to a portion of the Winnipeg River system in Northwest Ontario, Canada. Forecasting was conducted on a catchment area of approximately 20 000 km². using quarter monthly time intervals. The results were most promising. A very close fit was obtained during the calibration (training) phase and the ANNs developed consistently outperformed a conventional model during the verification (testing) phase for all of the four forecast lead-times. The average improvement in the root mean squared error (RMSE) for the 8 years of test data varied from 5 cms in the four time step ahead forecasts to 12.1 cms in the two time step ahead forecasts.     

Analysis of methods to estimate spring flows in a karst aquifer

Hydraulically and statistically based methods were analyzed to identify the most reliable method to predict spring flows in a karst aquifer. Measured water levels at nearby observation wells, measured spring pool altitudes, and the distance between observation wells and the spring pool were the parameters used to match measured spring flows. Measured spring flows at six Upper Floridan aquifer springs in central Florida were used to assess the reliability of these methods to predict spring flows. Hydraulically based methods involved the application of the Theis, Hantush-Jacob, and Darcy-Weisbach equations, whereas the statistically based methods were the multiple linear regressions and the technology of artificial neural networks (ANNs). Root mean square errors between measured and predicted spring flows using the Darcy-Weisbach method ranged between 5% and 15% of the measured flows, lower than the 7% to 27% range for the Theis or Hantush-Jacob methods. Flows at all springs were estimated to be turbulent based on the Reynolds number derived from the Darcy-Weisbach equation for conduit flow. The multiple linear regression and the Darcy-Weisbach methods had similar spring flow prediction capabilities. The ANNs provided the lowest residuals between measured and predicted spring flows, ranging from 1.6% to 5.3% of the measured flows. The model prediction efficiency criteria also indicated that the ANNs were the most accurate method predicting spring flows in a karst aquifer.     

Improving generalization of artificial neural networks in rainfall–runoff modelling / Amélioration de la généralisation de réseaux de neurones artificiels pour la modélisation pluie-débit

Abstract

Artificial neural networks (ANNs) are general-purpose techniques that can be used for nonlinear data-driven rainfall–runoff modelling. The key issue to construct a good model by means of ANNs is to understand their structural features and the problems related to their construction. Indeed, the quantity and quality of data, the type of noise and the mathematical properties of the algorithm for estimating the usual large number of parameters (weights) are crucial for the generalization performances of ANNs. However, it is well known that ANNs may suffer from poor generalization properties due to the high number of parameters and non-Gaussian data noise. Therefore, in the first part of this paper, the features and problems of ANNs are discussed. Eight Avoiding Overfitting Techniques are then presented, considering that these are methods for improving the generalization of ANNs. For this reason, they have been tested on two case studies—rainfall–runoff data from two drainage basins in the south of Italy—in order to gain insight into their properties and to investigate if there is one that absolutely gives the best performance.     

Evaluation of spring flow in the uplands of Matalom,Leyte, Philippines

In order to assess the reliability of the springs near Matalom, Leyte, Philippines as a sustainable source of drinking water, we measured precipitation and outflow of five small and two large springs for the region for a period of a year and analyzed the recession spring flow data. Although monthly spring flow follows a similar pattern to that of the rainfall, the regression relationship between both parameters is poor except for the smallest spring. To determine the dry season spring flow behavior, we analyzed the spring flow data with a mechanistic recession flow model originally developed for prediction of stream drought flow in the northeastern U.S. by Brutsaert and Nieber in 1977. The model describes the dry season spring flow well assuming that the aquifer behaves as a linear reservoir. The analysis shows that the flow “half-life” for the springs is about one month. By adding the individual spring flows to derive a watershed outflow we were able to evaluate how well the simple watershed geometry underlying the analysis of Brutsaert and Nieber [Regionalized drought flow hydrographs from a mature glaciated plateau. Water Resour Res 1977;13(3):637–43] applies to the more complex watersheds.     

Conceptual fuzzy neural network model for water quality simulation

Artificial neural networks (ANNs) have been applied successfully in various fields. However, ANN models depend on large sets of historical data, and are of limited use when only vague and uncertain information is available, which leads to difficulties in defining the model architecture and a low reliability of results. A conceptual fuzzy neural network (CFNN) is proposed and applied in a water quality model to simulate the Barra Bonita reservoir system, located in the southeast region of Brazil. The CFNN model consists of a rationally‐defined architecture based on accumulated expert knowledge about variables and processes included in the model. A genetic algorithm is used as the training method for finding the parameters of fuzzy inference and the connection weights. The proposed model may handle the uncertainties related to the system itself, model parameterization, complexity of concepts involved and scarcity and inaccuracy of data. The CFNN showed greater robustness and reliability when dealing with systems for which data are considered to be vague, uncertain or incomplete. The CFNN model structure is easier to understand and to define than other ANN‐based models. Moreover, it can help to understand the basic behaviour of the system as a whole, being a successful example of cooperation between human and machine. Copyright © 2006 John Wiley & Sons, Ltd.     

Genetic programming model for forecast of short and noisy data

Though forecasting of river flow has received a great deal of attention from engineers and researchers throughout the world, this still continues to be a challenging task owing to the complexity of the process. In the last decade or so, artificial neural networks (ANNs) have been widely applied, and their ability to model complex phenomena has been clearly demonstrated. However, the success of ANNs depends very crucially on having representative records of sufficient length. Further, the forecast accuracy decreases rapidly with an increase in the forecast horizon. In this study, the use of the Darwinian theory‐based recent evolutionary technique of genetic programming (GP) is suggested to forecast fortnightly flow up to 4‐lead. It is demonstrated that short lead predictions can be significantly improved from a short and noisy time series if the stochastic (noise) component is appropriately filtered out. The deterministic component can then be easily modelled. Further, only the immediate antecedent exogenous and/or non‐exogenous inputs can be assumed to control the process. With an increase in the forecast horizon, the stochastic components also play an important role in the forecast, besides the inherent difficulty in ascertaining the appropriate input variables which can be assumed to govern the underlying process. GP is found to be an efficient tool to identify the most appropriate input variables to achieve reasonable prediction accuracy for higher lead‐period forecasts. A comparison with ANNs suggests that though there is no significant difference in the prediction accuracy, GP does offer some unique advantages. Copyright © 2006 John Wiley & Sons, Ltd.     

A neural network model for predicting aquifer water level elevations

Artificial neural networks (ANNs) were developed for accurately predicting potentiometric surface elevations (monitoring well water level elevations) in a semiconfined glacial sand and gravel aquifer under variable state, pumping extraction, and climate conditions. ANNs "learn" the system behavior of interest by processing representative data patterns through a mathematical structure analogous to the human brain. In this study, the ANNs used the initial water level measurements, production well extractions, and climate conditions to predict the final water level elevations 30 d into the future at two monitoring wells. A sensitivity analysis was conducted with the ANNs that quantified the importance of the various input predictor variables on final water level elevations. Unlike traditional physical-based models, ANNs do not require explicit characterization of the physical system and related physical data. Accordingly, ANN predictions were made on the basis of more easily quantifiable, measured variables, rather than physical model input parameters and conditions. This study demonstrates that ANNs can provide both excellent prediction capability and valuable sensitivity analyses, which can result in more appropriate ground water management strategies.     

Marine Layer Stratus Study

The intent of this study is to develop a better understanding of the behavior of late spring through early fall marine layer stratus and fog at Vandenberg Air Force Base, which accounts for a majority of aviation forecasting difficulties. The main objective was to use Leipper (1995) study as a starting point to evaluate synoptic and mesoscale processes involved, and identify specific meteorological parameters that affected the behavior of marine layer stratus and fog. After identifying those parameters, the study evaluates how well the various weather models forecast them. The main conclusion of this study is that weak upper-air dynamic features work with boundary layer motions to influence marine layer behavior. It highlights the importance of correctly forecasting the surface temperature by showing how it ties directly to the wind field. That wind field, modified by the local terrain, establishes the low-level convergence and divergence pattern and the resulting marine layer cloud thicknesses and visibilities.     

Uncertainty and multiple objective calibration in regional water balance modelling: case study in 320 Austrian catchments

We examine the value of additional information in multiple objective calibration in terms of model performance and parameter uncertainty. We calibrate and validate a semi‐distributed conceptual catchment model for two 11‐year periods in 320 Austrian catchments and test three approaches of parameter calibration: (a) traditional single objective calibration (SINGLE) on daily runoff; (b) multiple objective calibration (MULTI) using daily runoff and snow cover data; (c) multiple objective calibration (APRIORI) that incorporates an a priori expert guess about the parameter distribution as additional information to runoff and snow cover data. Results indicate that the MULTI approach performs slightly poorer than the SINGLE approach in terms of runoff simulations, but significantly better in terms of snow cover simulations. The APRIORI approach is essentially as good as the SINGLE approach in terms of runoff simulations but is slightly poorer than the MULTI approach in terms of snow cover simulations. An analysis of the parameter uncertainty indicates that the MULTI approach significantly decreases the uncertainty of the model parameters related to snow processes but does not decrease the uncertainty of other model parameters as compared to the SINGLE case. The APRIORI approach tends to decrease the uncertainty of all model parameters as compared to the SINGLE case. Copyright © 2006 John Wiley & Sons, Ltd.     

Estimation of mean residence times of subsurface waters using seasonal variation in deuterium excess in a small headwater catchment in Japan

We measured deuterium excess (d = δD ? 8δ¹⁸O) in throughfall, groundwater, soil water, spring water, and stream water for 3 years in a small headwater catchment (Matsuzawa, 0·68 ha) in the Kiryu Experimental Watershed in Japan. The d value represents a kinetic effect produced when water evaporates. The d value of the throughfall showed a sinusoidal change (amplitude: 6·9‰ relative to Vienna standard mean ocean water (V‐SMOW)) derived from seasonal changes in the source of water vapour. The amplitude of this sinusoidal change was attenuated to 1·3–6·9‰ V‐SMOW in soil water, groundwater, spring water, and stream water. It is thought that these attenuations derive from hydrodynamic transport processes in the subsurface and mixing processes at an outflow point (stream or spring) or a well. The mean residence time (MRT) of water was estimated from d value variations using an exponential‐piston flow model and a dispersion model. MRTs for soil water were 0–5 months and were not necessarily proportional to the depth. This may imply the existence of bypass flow in the soil. Groundwater in the hillslope zone had short residence times, similar to those of the soil water. For groundwater in the saturated zone near the spring outflow point, the MRTs differed between shallow and deeper groundwater; shallow groundwater had a shorter residence time (5–8 months) than deeper groundwater (more than 9 months). The MRT of stream water (8–9 months) was between that of shallow groundwater near the spring and deeper groundwater near the spring. The seasonal variation in the d value of precipitation arises from changes in isotopic water vapour composition associated with seasonal activity of the Asian monsoon mechanism. The d value is probably an effective tracer for estimating the MRT of subsurface water not only in Japan, but also in other East Asian countries influenced by the Asian monsoon. Copyright © 2006 John Wiley & Sons, Ltd.     

基于遗传算法优化的ENSO指数的动力预报模型反演

基于NCEP/NCAR提供的1958～1995年全球月平均海温距平场再分析资料,采用动力系统反演思想和遗传算法途径,进行了El Nino/La Nina指数的动力预报模型的参数优化和模型反演,从上述海温资料中重构了Nino3海温距平指数的非线性动力模型.模型预报试验结果表明,遗传算法具有的全局搜索和并行计算优势能够客观、有效地反演海温指数的动力预报模型,对Nino3海温指数和El Nino/La Nina事件进行较为客观准确的预测,为El Nino/La Nina预测提供有益的研究参考.     

Magnetotelluric inversion for azimuthally anisotropic resistivities employing artificial neural networks

An extension of an artificial neural network (ANN) approach to solve the magnetotelluric (MT) inverse problem for azimuthally anisotropic resistivities is presented and applied for a real dataset. Three different model classes, containing general 1-D and 2-D azimuthally anisotropic features, have been considered. For each model class, characteristics of three-layer feed forward ANNs trained through an error back propagation algorithm have been adjusted to approximate the inverse modeling function. It appears that, at least for synthetic models, reasonable results would be obtained by applying the amplitudes of the complex impedance tensor elements as inputs. Furthermore, the Levenberg-Marquart algorithm possesses optimal performance as a learning paradigm for this problem. The evaluation of applicability of the trained ANNs for unknown data sets excluded from the learning procedure reveals that the trained ANNs possess acceptable interpolation and extrapolation abilities to estimate model parameters accurately. This method was also successfully used for a field dataset wherein anisotropy had been previously recognized.     

新一代地震预报专家系统NGESEP预报效能的初步研究

利用４０个地震震例对新一代地震预报专家系统ＮＧＥＳＥＰ的预报效果进行了检验。比较了使用信度分布图和信度等值线图两种方法未来发震地点的效能，分析了系统对地震三要素的预报结果，表明该系统具有较高的预报效能。     

Real-time correction of water stage forecast during rainstorm events using combination of forecast errors

This study proposes a real-time error correction method for the forecasted water stage using a combination of forecast errors estimated by the time series models, AR(1), AR(2), MA(1) and MA(2), and the average deviation model to update the water stage forecast during rainstorm events. During flood forecasting and warning operations, the proposed real-time error correction method takes advantage of being individually and continuously implemented and the results not being updated to the hydrological model and hydraulic routings so as to save computational time by recalibrating the parameters of the proposed methods with real-time observation. For model validation, the current study adopts the observed and forecasted data on a severe typhoon, Morakot, collected at eight water level gauges in Southern Taiwan and provided by the flood forecast system FEWS_Taiwan, which is linked with the reliable quantitative precipitation forecast (QPF) at 3 h of lead time provided by the Center Weather Bureau in Taiwan, as the model validation. The results of numerical experiments indicate that the proposed real-time error correction method can effectively reduce the errors of forecasted water stages at the 1-, 2-, and 3-h lead time and so enhance the reliability of forecast information issued by the FEWS_Taiwan. By means of real-time estimating potential forecast error, the uncertainties in hydrology, modules as well as associated parameters, and physiographical features of the river can be reduced.     

Drainage in heterogeneous sand columns with different geometric structures

This paper discusses multi-step drainage experiments in two heterogeneously packed sand columns (10 × 10 × 20 cm³). Different packing structures were generated using two different sand types. One purpose of the study was to test the influence of packing structures on the movement of water. The second purpose was to assess the quality of predictions for the outflow curves in both columns made with an upscaled model. The heterogeneous structures of the columns can be considered as two opposing extremes. The first column was packed with a random arrangement of two sand types that is not stochastically homogeneous and where a cluster running through the column exists for both materials. The second column was packed with a periodic pattern of coarse-sand inclusions in a fine-sand background and has a clearly defined unit cell. The depth-averaged (2D) spatial distribution of the water content in the columns was monitored during the whole multi-step outflow experiment using neutron radiography. The 3D water content was measured at the steady states by neutron tomography. The experimental results are compared with the model predictions of an upscaled model derived with the homogenization theory. The parameters for the upscaled model are calculated from the hydraulic parameters of the two sand types. These hydraulic parameters were first identified in independent measurements on samples of the two individual sand types, separately. Additionally, the hydraulic parameters of both sands were identified by fitting a numerical model to the measured outflow curves. The different column structures showed a significant effect on water retention and the effective retention function, as water was trapped in the coarse-sand inclusions of the periodic structure. We included this trapping effect in the effective retention function of the upscaled model with an apparent air entry pressure. Contrary to the retention, the different packing structures had no large effect on the dynamic behavior of the outflow. The effective conductivity of the columns is therefore not significantly influenced by the structure. The upscaled models predicted the movement of the averaged water content in the two columns well. This confirms the applicability of upscaled models even if the underlying requirements are not strictly met.     

Functional networks in real-time flood forecasting—a novel application

Functional networks were recently introduced as an extension of artificial neural networks (ANNs). Unlike ANNs, they estimate unknown neuron functions from given functional families during the training process. Here, we applied two types of functional network models, separable and associativity functional networks, to forecast river flows for different lead-times. We compared them with a conventional artificial neural network model, an ARMA model and a simple baseline model in three catchments. Results show that functional networks are flexible and comparable in performance to artificial neural networks. In addition, they are easier and quicker to train and so are useful tools as an alternative to artificial neural networks. These results were obtained with only the simplest structures of functional networks and it is possible that a more detailed study with more complex forms of the model will improve even further on these results. Thus we recommend that the use of functional networks in discharge time series modelling and forecasting should be further investigated.     

An interactive multi-objective optimization framework for groundwater inverse modeling

The groundwater inverse problem of estimating heterogeneous groundwater model parameters (hydraulic conductivity in this case) given measurements of aquifer response (such as hydraulic heads) is known to be an ill-posed problem, with multiple parameter values giving similar fits to the aquifer response measurements. This problem is further exacerbated due to the lack of extensive data, typical of most real-world problems. In such cases, it is desirable to incorporate expert knowledge in the estimation process to generate more reasonable estimates. This work presents a novel interactive framework, called the ‘Interactive Multi-Objective Genetic Algorithm’ (IMOGA), to solve the groundwater inverse problem considering different sources of quantitative data as well as qualitative expert knowledge about the site. The IMOGA is unique in that it looks at groundwater model calibration as a multi-objective problem consisting of quantitative objectives – calibration error and regularization – and a ‘qualitative’ objective based on the preference of the geological expert for different spatial characteristics of the conductivity field. All these objectives are then included within a multi-objective genetic algorithm to find multiple solutions that represent the best combination of all quantitative and qualitative objectives. A hypothetical aquifer case-study (based on the test case presented by Freyberg [Freyberg DL. An exercise in ground-water model calibration and prediction. Ground Water 1988;26(3)], for which the ‘true’ parameter values are known, is used as a test case to demonstrate the applicability of this method. It is shown that using automated calibration techniques without using expert interaction leads to parameter values that are not consistent with site-knowledge. Adding expert interaction is shown to not only improve the plausibility of the estimated conductivity fields but also the predictive accuracy of the calibrated model.     

An Objective Approach for Prediction of Daily Summer Monsoon Rainfall over Orissa (India) due to Interaction of Mesoscale and Large-scale Synoptic Systems

Orissa State, a meteorological subdivision of India, lies on the east coast of India close to north Bay of Bengal and to the south of the normal position of the monsoon trough. The monsoon disturbances such as depressions and cyclonic storms mostly develop to the north of 15° N over the Bay of Bengal and move along the monsoon trough. As Orissa lies in the southwest sector of such disturbances, it experiences very heavy rainfall due to the interaction of these systems with mesoscale convection sometimes leading to flood. The orography due to the Eastern Ghat and other hill peaks in Orissa and environs play a significant role in this interaction. The objective of this study is to develop an objective statistical model to predict the occurrence and quantity of precipitation during the next 24 hours over specific locations of Orissa, due to monsoon disturbances over north Bay and adjoining west central Bay of Bengal based on observations to up 0300 UTC of the day. A probability of precipitation (PoP) model has been developed by applying forward stepwise regression with available surface and upper air meteorological parameters observed in and around Orissa in association with monsoon disturbances during the summer monsoon season (June-September). The PoP forecast has been converted into the deterministic occurrence/non-occurrence of precipitation forecast using the critical value of PoP. The parameters selected through stepwise regression have been considered to develop quantitative precipitation forecast (QPF) model using multiple discriminant analysis (MDA) for categorical prediction of precipitation in different ranges such as 0.1–10, 11–25, 26–50, 51–100 and >100 mm if the occurrence of precipitation is predicted by PoP model. All the above models have been developed based on data of summer monsoon seasons of 1980–1994, and data during 1995–1998 have been used for testing the skill of the models. Considering six representative stations for six homogeneous regions in Orissa, the PoP model performs very well with percentages of correct forecast for occurrence/non-occurrence of precipitation being about 96% and 88%, respectively for developmental and independent data. The skill of the QPF model, though relatively less, is reasonable for lower ranges of precipitation. The skill of the model is limited for higher ranges of precipitation. accepted September 2006