François Anctil

Daily reservoir inflow forecasting using artificial neural networks with stopped training approach

In this paper, an early stopped training approach (STA) is introduced to train multi-layer feed-forward neural networks (FNN) for real-time reservoir inflow forecasting. The proposed method takes advantage of both Levenberg–Marquardt Backpropagation (LMBP) and cross-validation technique to avoid underfitting or overfitting on FNN training and enhances generalization performance. The methodology is assessed using multivariate hydrological time series from Chute-du-Diable hydrosystem in northern Quebec (Canada). The performance of the model is compared to benchmarks from a statistical model and an operational conceptual model. Since the ultimate goal concerns the real-time forecast accuracy, overall the results show that the proposed method is effective for improving prediction accuracy. Moreover it offers an alternative when dynamic adaptive forecasting is desired.

Artificial neural network modeling of water table depth fluctuations

Three types of functionally different artificial neural network (ANN) models are calibrated using a relatively short length of groundwater level records and related hydrometeorological data to simulate water table fluctuations in the Gondo aquifer, Burkina Faso. Input delay neural network (IDNN) with static memory structure and globally recurrent neural network (RNN) with inherent dynamical memory are proposed for monthly water table fluctuations modeling. The simulation performance of the IDNN and the RNN models is compared with results obtained from two variants of radial basis function (RBF) networks, namely, a generalized RBF model (GRBF) and a probabilistic neural network (PNN). Overall, simulation results suggest that the RNN is the most efficient of the ANN models tested for a calibration period as short as 7 years. The results of the IDNN and the PNN are almost equivalent despite their basically different learning procedures. The GRBF performs very poorly as compared to the other models. Furthermore, the study shows that RNN may offer a robust framework for improving water supply planning in semiarid areas where aquifer information is not available. This study has significant implications for groundwater management in areas with inadequate groundwater monitoring network.

Two decades of anarchy? Emerging themes and outstanding challenges for neural network river forecasting

This paper traces two decades of neural network rainfall-runoff and streamflow modelling, collectively termed ‘river forecasting’. The field is now firmly established and the research community involved has much to offer hydrological science. First, however, it will be necessary to converge on more objective and consistent protocols for: selecting and treating inputs prior to model development; extracting physically meaningful insights from each proposed solution; and improving transparency in the benchmarking and reporting of experimental case studies. It is also clear that neural network river forecasting solutions will have limited appeal for operational purposes until confidence intervals can be attached to forecasts. Modular design, ensemble experiments, and hybridization with conventional hydrological models are yielding new tools for decision-making. The full potential for modelling complex hydrological systems, and for characterizing uncertainty, has yet to be realized. Further gains could also emerge from the provision of an agreed set of benchmark data sets and associated development of superior diagnostics for more rigorous intermodel evaluation. To achieve these goals will require a paradigm shift, such that the mass of individual isolated activities, focused on incremental technical refinement, is replaced by a more coordinated, problem-solving international research body.

Downscaling Precipitation and Temperature with Temporal Neural Networks

The issues of downscaling the outputs of a global climate model (GCM) to a scale that is appropriate to hydrological impact studies are investigated using a temporal neural network approach. The time-lagged feed-forward neural network (TLFN) is proposed for downscaling daily total precipitation and daily maximum and minimum temperature series for the Serpent River watershed in northern Quebec (Canada). The downscaling models are developed and validated using large-scale predictor variables derived from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP– NCAR) reanalysis dataset. Atmospheric predictors such as specific humidity, wind velocity, and geopotential height are identified as the most relevant inputs to the downscaling models. The performance of the TLFN downscaling model is also compared to a statistical downscaling model (SDSM). The downscaling results suggest that the TLFN is an efficient method for downscaling both daily precipitation and temperature series. The best downscaling models were then applied to the outputs of the Canadian Global Climate Model (CGCM1), forced with the Intergovernmental Panel on Climate Change (IPCC) IS92a scenario. Changes in average precipitation between the current and the future scenarios predicted by the TLFN are generally found to be smaller than those predicted by the SDSM model. Furthermore, application of the downscaled data for hydrologic impact analysis in the Serpent River resulted in an overall increasing trend in mean annual flow as well as earlier spring peak flow. The results also demonstrate the emphasis that should be given in identifying the appropriate downscaling tools for impact studies by showing how a future climate scenario downscaled with different downscaling methods could result in significantly different hydrologic impact simulation results for the same watershed.

Impact of the length of observed records on the performance of ANN and of conceptual parsimonious rainfall-runoff forecasting models

Abstract Although attractive to hydrologists, artificial neural network modeling still lacks norms that would help modelers to create and train efficient rainfall-runoff models in a systematic way. This study focuses on the impact of the length of observed records on the performance of multiple-layer perceptrons (MLPs), and compare their results with those of a parsimonious conceptual model equipped with an updating scheme. Both models were assessed for 1-day-ahead stream flow predictions. Ninety-two different model scenarios were obtained for 1-, 3-, 5-, 9-, and 15-year time sub-series created from a 24-year training set, shifting by a 1-year sliding window. All the model scenarios were verified against the same 7-year test set. The results revealed that MLP stream flow mapping was efficient as long as wet weather data were available for the training; the longer series implicitly guarantee that the data contain valuable information of the hydrological behavior; the results were consistent with those reported for conceptual rainfall-runoff models. The physical knowledge in the conceptual models allowed them to make much better use of 1-year training sets than the MLPs. However, longer training sets were more beneficial to the MLPs than to the conceptual model. Both types shared best performance about evenly for 3- and 5-year training sets, but MLPs did better whenever the training set was dominated by wet weather. The MLPs continued to improve for input vectors of 9 years and more, which was not the case of the conceptual model.

Wavelet Analysis of the Interannual Variability in Southern Québec Streamflow

The objectives of this study are to describe the local interannual variability in southern Quebec, Canada, streamflow, based on wavelet analysis, and to identify plausible climatic teleconnections that could explain these local variations. Scale-averaged wavelet power spectra are used to simultaneously assess the interannual and spatial variability in 18 contiguous annual streamflow time series. The span of available observations, 1938- 2000, allows depicting the variance for periods up to about 12 yr. The most striking feature, in the 2-3-yr band and in the 3-6-yr band—the 6-12-yr band is dominated by white noise and is not considered further—is a net distinction between the timing of the interannual variability in local western and eastern streamflows, which may be linked to the local climatology. This opens up the opportunity to construct two regional time series using principal component (PC) analysis. Then, for each band, linear relationships are sought between the regional streamflow and five selected climatic indices: the Pacific-North America (PNA), the North Atlantic Oscillation (NAO), the Northern Hemisphere annular mode (NAM), the Baffin Island-West Atlantic (BWA) and the sea surface temperature anomalies over the Nino-3 region (ENSO3). The correlation analysis revealed the presence of a change point in the streamflow time series, as reported by others, occurring around 1970. For example, the west and east 2-3-yr bands are positively correlated to PNA since 1970, which was not the case prior to that change point. The proposed regional east-west divide is particularly evident prior to 1970, with a negative NAM correlation for the west and a positive NAM (and negative ENSO3) for the east. The picture for the less energetic 3-6-yr band is mixed, with alternating dominance of teleconnection patterns, but the 1970 change point holds.

Neural network estimation of air temperatures from AVHRR data

Multilayer feed-forward (MLF) neural networks were employed to estimate air temperatures in Southern Québec (Canada) using Advanced Very High Resolution Radiometer (AVHRR) images. The input variables for the networks were the five bands of the AVHRR image, surface altitude, solar zenith angle, and Julian day. The estimation was carried out using a dataset collected during the growing season from June to September 2000. Levenberg--Marquardt back-propagation (LM-BP) was used to train the networks. The early stopping method was applied to improve the LM-BP and to generalize the networks. Bands 4 and 5, which are used for retrieval of surface temperature, were the most critical components for the estimation. The contribution of Julian day to the precision of estimated air temperature was much superior to that of altitude and solar zenith angle for the dataset of inter-seasonal air temperatures. The network using all five bands, Julian day, altitude, and solar zenith angle provided the best results, with 22 nodes in the hidden layer. In the time series of estimated and station air temperatures, the difference between the temperatures was generally maintained within 2°C on various canopies, even during steep variations in August and September.

A recurrent neural networks approach using indices of low‐frequency climatic variability to forecast regional annual runoff

This paper evaluates the potential of using low-frequency climatic mode indices to forecast regional annual runoff in northern Quebec and the Labrador region. The impact of climatic trends in the forecast accuracy is investigated using a recurrent neural networks (RNN) approach, time-series of inflow to eight large hydropower systems in Quebec and Labrador, and indices of selected modes of climatic variability: El Nino-Southern Oscillation (ENSO), North Atlantic Oscillation (NAO), Pacific-North American (PNA), Baffin Island-West Atlantic (BWA) and sea-level pressure (SLP) at Iceland. A wavelet analysis is used to show that the selected climatic patterns are related to annual runoff from 1950 to 1996 in northern Quebec. The forecast results indicate that the use of BWA, PNA and ENSO indices results in better forecast skill than the use of SLP or NAO. Overall, the use of the BWA index is found to provide the best forecast improvement (38% on average), whereas the use of PNA provides 28% of improvement on average. Using the SLP index improves the forecast accuracy by 4%, and the use of an ENSO indicator leads to an improvement of 6%. The NAO index used here is found to provide only a modest improvement.

Accelerating shear velocity in gravel-bed channels

Abstract The behaviour of the shear velocity along a gravel-bed channel is investigated experimentally in the presence of a negative pressure gradient (accelerating flow). Different methods of estimation of the shear velocity, derived from vertical profiles of the mean longitudinal point velocity, are examined and a new method is proposed. Results show that the proposed method of estimation is comparable to the St Venant and Clausers methods. At a specific cross section, for constant bottom slope and relative roughness, shear velocity increases with discharge.

Evaluation of streamflow simulation by SWAT model for two small watersheds under snowmelt and rainfall

Abstract The degradation of the river water quality in Canadian rural catchments is of concern. In these catchments, the Soil Water Assessment Tool (SWAT) model can help better understand the problems related to diffuse pollution. The numerous documented applications of SWAT have been dominated by areas uniquely driven by rainfall. Given that Canadian hydroclimatic conditions differ due to the presence of a seasonal snowpack of long duration, evaluation of the hydrological performance needs to be performed prior to attempting any water quality simulations. The objective of the present work is to evaluate the hydrological behaviour of the SWAT model under snowmelt and rainfall for two small watersheds located in southeastern Canada. Different calibration schemes are evaluated including seasonal effects. One-year calibration gave satisfactory daily performances measured with Nash-Sutcliffe efficiency (NS) ranging between 61 and 83% and deviations of volume (Dv ) between −10 and 1%, while in validation, NS was 40–73% and Dv between −20 and −3%. The SWAT model has difficulties in reconciling both seasons. When winter and summer data are used separately to calibrate the model, the model performance is still much better for the winter season than for the summer one. However, the latter is considerably improved when only summer observations are provided for calibration. Conversely, calibration based strictly on the winter observations provides no real advantage over that based on all available data. A two-step composite calibration, which optimizes the SWAT snow accumulation and melt-related parameters on the winter data, after all other model parameters have been optimized on the summer data, provides a compromise.