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.     

Evaporation from boreal reservoirs: A comparison between eddy covariance observations and estimates relying on limited data

Hydrological models used for reservoir management typically lack an accurate representation of open-water evaporation and must be run in a scarce data context. This study aims to identify an accurate means to estimate reservoir evaporation with simple meteorological inputs during the open-water season, using long-term eddy covariance observations from two boreal hydropower reservoirs with contrasting morphometry as reference. Unlike the temperate water bodies on which the majority of other studies have focused, northern reservoirs are governed by three distinct periods: ice cover in the cold season, warming in the summer and energy release in the fall. The reservoirs of interest are Eastmain-1 (52°N, mean depth of 11 m) and Romaine-2 (51°N, mean depth of 42 m), both located in eastern Canada. Four approaches are analysed herein: a combination approach, a radiation-based approach, a mass-transfer approach, and empirical methods. Of all the approaches, the bulk transfer equation with a constant Dalton number of 1.2 x 10⁻³ gave the most accurate estimation of evaporation at hourly time steps, compared with the eddy covariance observations (RMSE of 0.06 mm h⁻¹ at Eastmain-1 and RMSE of 0.04 mm h⁻¹ at Romaine-2). The daily series also showed good accuracy (RMSE of 1.38 mm day⁻¹ at Eastmain-1 and RMSE of 0.62 mm day⁻¹ at Romaine-2) both in the warming and energy release phases of the open-water season. The bulk transfer equation, on the other hand, was incapable of reproducing condensation episodes that occurred soon after ice breakup. Basic and variance-based sensitivity analyses were conducted, in particular to measure the variation in performance when the bulk transfer equation was applied with meteorological observations collected at a certain distance (~10–30 km) from the reservoir. This exercise illustrated that accurate estimates of open water evaporation require representative measurements of wind speed and water surface temperature.     

Hydrologic evaluation of spatially disaggregated global ensemble rainfall forecasts

A statistic–stochastic multi‐fractal downscaling technique was evaluated from a hydrologic point of view. Ensemble hydrologic forecasts with a time step of 3 h were performed for original and disaggregated ensemble rainfall forecasts issued by the Canadian Global Ensemble Prediction System in its 2009 operational version. This hydro‐meteorological operational forecasting chain was conducted using the hydrological model SWMM5. The model was implemented on a small 6‐km² urban catchment located in the Québec City region. The hydrological evaluation was based on the comparison of forecasted flows to the observed ones, calculating several deterministic and probabilistic scores, and drawing rank histograms and reliability diagrams. Disaggregated products led to a better representation of the ensemble members' dispersion. This disaggregation technique represents an interesting way of bridging the gap between the meteorological models' resolution and the high degree of spatial precision sometimes required by hydrological models in their precipitation representation. Copyright © 2013 John Wiley & Sons, Ltd.     

Modelling of Seasonal Evapotranspiration from an Agricultural Field Using the Canadian Land Surface Scheme (CLASS) with a Pedotransfer Rule and Multicriteria Optimization

The performance of the Canadian Land Surface Scheme (CLASS 3.5) was assessed using turbulent fluxes derived from data recorded at two micrometeorological stations located in a potato field in Quebec, Canada. The minimum stomatal resistance, the maximum leaf area index, and the initial water content of the third soil layer were optimized using the Non-Dominated Sorting Genetic Algorithm-II and the mean square error of the latent heat flux. With respect to benchmark solutions, the optimization improved the sensible and latent heat fluxes by 31 and 23%, respectively. The use of a pedotransfer rule in adjustment of the water content of mineral soils having small percentages of organic matter provided better estimates of the evapotranspiration during the growing stage. However, like the original version of the model (without the pedotransfer rule), it underestimated evapotranspiration throughout the maturity stage. It is noteworthy that the original version produced a good estimate of cumulative evapotranspiration over the entire season as a result of over- and underestimates at the beginning and maturity stage of the growing season, respectively.     

Estimating Sensible and Latent Heat Fluxes over an Inland Water Body Using Optical and Microwave Scintillometers

Boundary-Layer Meteorology - Observations of turbulent heat fluxes over inland water bodies are scarce despite being critical to adequate lake parametrization for numerical weather forecast and...     

How do potential evapotranspiration formulas influence hydrological projections?

ABSTRACT

This paper evaluates the sensitivity of hydrological projections to the choice of potential evapotranspiration formulas on two natural sub-catchments, in Canada and Germany. Twenty-four equations, representing a large range of options, are applied for calibration over the whole observation time series and for future conditions. The modelling chain is composed of dynamically downscaled climatic projections and a 20-member (ensemble) hydrological model, along with a snow module. The roots of the sensitivity and its propagation within the hydrological chain are evaluated to show influences on climate change impact conclusions. Results show large differences between the 24 simulated potential evapotranspiration time series. However, these discrepancies only moderately affect the calibration efficiency of hydrological models as a result of adaptation of parameters. Choice of formula influences hydrological projections and climate change conclusions for both catchments in terms of simulated and projected values, and also in the magnitude of changes during important dynamic periods such as spring and autumn high flows and summer low flows. Spread of the hydrological response is lower for the combinational formulas than for temperature-based or radiation-based equations. All the results reveal the importance of testing a large spectrum of potential evapotranspiration formulas in a decision-making context, such as water resources management.     

On the Use of Hydrological Models and Satellite Data to Study the Water Budget of River Basins Affected by Human Activities: Examples from the Garonne Basin of France

Natural and anthropogenic forcing factors and their changes significantly impact water resources in many river basins around the world. Information on such changes can be derived from fine scale in situ and satellite observations, used in combination with hydrological models. The latter need to account for hydrological changes caused by human activities to correctly estimate the actual water resource. In this study, we consider the catchment area of the Garonne river (in France) to investigate the capabilities of space-based observations and up-to-date hydrological modeling in estimating water resources of a river basin modified by human activities and a changing climate. Using the ISBA–MODCOU and SWAT hydrological models, we find that the water resources of the Garonne basin display a negative climate trend since 1960. The snow component of the two models is validated using the moderate-resolution imaging spectroradiometer snow cover extent climatology. Crop sowing dates based on remote sensing studies are also considered in the validation procedure. Use of this dataset improves the simulated evapotranspiration and river discharge amounts when compared to conventional data. Finally, we investigate the benefit of using the MAELIA multi-agent model that accounts for a realistic agricultural and management scenario. Among other results, we find that changes in crop systems have significant impacts on water uptake for agriculture. This work constitutes a basis for the construction of a future modeling framework of the sociological and hydrological system of the Garonne river region.     

Hydrological Evaluation of the Canadian Meteorological Ensemble Reforecast Product

This paper evaluates the quality of the ensemble hydrological reforecasts obtained using the 18-year ensemble meteorological reforecast dataset available from the Canadian Centre for Meteorological and Environmental Prediction (CCMEP). This study focuses on four large watersheds in the province of Quebec. A distribution-based scaling (DBS) post-processing method is used to correct the 18-year ensemble precipitation reforecasts. An Ensemble Kalman Filter (EnKF) assimilation technique is also assessed to improve the initial conditions of the hydrologic model. There is a slight improvement in performance and reliability after applying the DBS approach to precipitation reforecasts, but this technique induces a reduction in the spread. The impact of the integration of the post-processed precipitation into the hydrologic model is also quite marginal. However, the addition of an EnKF provides better ensemble hydrological reforecasts with high performance, reliability, and skill, especially in the first reforecast horizons. The best results are, however, generally obtained when using DBS and an EnKF together. Combining DBS and an EnKF, hydrological forecasts for the next two weeks are obtained using the CCMEP reforecast and also the second generation Global Ensemble Forecast System (GEFS v2) reforecast, which is considered a reference. Forecasts of comparable skill and spread are obtained, with CCMEP-based forecasts showing better spread during the first week, and GEFS v2–based reforecasts showing better skill and spread during the second week. Finally, it is shown that the two meteorological reforecast products assessed in this study have similar economic value for hydrological forecasting applications based on the cost–loss model.