François Brissette

Comparison of Stochastic Optimization Algorithms in Hydrological Model Calibration

AbstractTen stochastic optimization methods—adaptive simulated annealing (ASA), covariance matrix adaptation evolution strategy (CMAES), cuckoo search (CS), dynamically dimensioned search (DDS), differential evolution (DE), genetic algorithm (GA), harmony search (HS), pattern search (PS), particle swarm optimization (PSO), and shuffled complex evolution–University of Arizona (SCE–UA)—were used to calibrate parameter sets for three hydrological models on 10 different basins. Optimization algorithm performance was compared for each of the available basin-model combinations. For each model-basin pair, 40 calibrations were run with the 10 algorithms. Results were tested for statistical significance using a multicomparison procedure based on Friedman and Kruskal-Wallis tests. A dispersion metric was used to evaluate the fitness landscape underlying the structure on each test case. The trials revealed that the dimensionality and general fitness landscape characteristics of the model calibration problem are impo...

Assessing the limits of bias‐correcting climate model outputs for climate change impact studies

Bias correction of climate model outputs has emerged as a standard procedure in most recent climate change impact studies. A crucial assumption of all bias correction approaches is that climate model biases are constant over time. The validity of this assumption has important implications for impact studies and needs to be verified to properly address uncertainty in future climate projections. Using 10 climate model simulations, this study specifically tests the bias stationarity of climate model outputs over Canada and the contiguous United States (U.S.) by comparing model outputs with corresponding observations over two 20 year historical periods (1961–1980 and 1981–2000). The results show that precipitation biases are clearly nonstationary over much of Canada and the contiguous U.S. and where they vary over much shorter time scales than those normally considered in climate change impact studies. In particular, the difference in biases over two very close periods of the recent past are, in fact, comparable to the climate change signal between future (2061–2080) and historical (1961–1980) periods for precipitation over large parts of Canada and the contiguous U.S., indicating that the uncertainty of future impacts may have been underestimated in most impact studies. In comparison, temperature bias can be considered to be approximately stationary for most of Canada and the contiguous U.S. when compared with the magnitude of the climate change signal. Given the reality that precipitation is usually considered to be more important than temperature for many impact studies, it is advisable that natural climate variability and climate model sensitivity be better emphasized in future impact studies.

Stochastic Multisite Generation of Daily Precipitation Data Using Spatial Autocorrelation

Abstract There are a number of stochastic models that simulate weather data required for various water resources applications in hydrology, agriculture, ecosystem, and climate change studies. However, many of them ignore the dependence between station locations exhibited by the observed meteorological time series. This paper proposes a multisite generation approach of daily precipitation data based on the concept of spatial autocorrelation. This theory refers to spatial dependence between observations with respect to their geographical adjacency. In hydrometeorology, spatial autocorrelation can be computed to describe daily dependence between the weather stations through the use of a spatial weight matrix, which defines the degree of significance of the weather stations surrounding each observation. The methodology is based on the use of the spatial moving average process to generate spatially autocorrelated random numbers that will be used in a stochastic weather generator. The resulting precipitation pr...

Structural and Non-Structural Climate Change Adaptation Strategies for the Péribonka Water Resource System

This paper discusses the possibility for a privately managed hydro-power system to adapt to a projected increase in water flow in their central-Québec watersheds by adding power generation potential. Runoffs simulated by a lumped rainfall-runoff model were fed into a stochastic dynamic programming (SDP) routine to generate reservoir operating rules. These rules were optimized for maximum power generation under maximal and minimal reservoir level constraints. With these optimized rules, a power generation simulator was used to predict the amount of generated hydropower. The same steps, excluding calibration, were performed on 60 climate projections (from 23 general circulation models and 3 greenhouse gas emission scenarios) for future horizons 2036–2065 and 2071–2100. Reservoir operation rules were optimized for every climate change projection for the 3 power plants in the system. From these simulations, it was possible to determine hydropower numbers for both horizons. The same steps were performed under a modified system in which an additional turbine was added to each power plant. Results show that both the non-structural (optimizing reservoir rules) and structural (adding turbines) adaptation measures allow for increased power production, but that adapting operating rules is sufficient to reap the most of the benefits of increased water availability.

A versatile weather generator for daily precipitation and temperature

Stochastic daily weather generators are often used to generate long time series of weather variables to drive hydrological and agricultural models. More recently, they have also been used as a downscaling tool for studying the impacts of climate change. This article describes a versatile stochastic weather generator (WeaGETS) for producing daily precipitation, and maximum and minimum temperatures (Tmax and Tmin). WeaGETS regroups several options of other weather generators into one package, such as three Markov models to produce precipitation occurrence, four distributions to generate wet day precipitation amount, and two methods to simulate Tmax and Tmin. More importantly, a spectral correction approach is included in WeaGETS for correcting the underestimation of interannual variability, which is a problem common to all weather generators. The performance of WeaGETS is demonstrated through a comparison against two well-known weather generators (WGEN and CLIGEN) with respect to the generation of precipitation, Tmax, and Tmin for two Canadian meteorological stations. The results show that the widely used first-order Markov model is adequate for producing precipitation occurrence, but it underestimates the longest dry spell for the low-precipitation station. The higher-order models have positive effects. The mixed exponential and skewed normal Pearson III distributions are consistently better than the exponential and gamma distributions at generating precipitation amounts. The two-component mixed exponential distribution is better at representing extreme precipitation events than the other three distributions. WeaGETS is consistently better than WGEN and CLIGEN at producing Tmax and Tmin. Both WGEN and CLIGEN underestimate the monthly and interannual variances of precipitation and temperatures. However, WeaGETS successfully preserves the observed low-frequency variability and autocorrelation functions of precipitation and temperatures. Overall, WeaGETS is consistently better than the other two weather generators (WGEN and CLIGEN) for producing precipitation, Tmax, and Tmin. The Matlab freeware allows for easy modification of all routines, making it easy to add additional weather variables to simulate.

An Improved Stochastic Weather Generator for Hydrological Impact Studies

A stochastic weather generator based on the WGEN model has been tested on 13 meteorological stations in Quebec, Canada. The generator, called WeaGETS, accounts for longer persistence of wet and dry spells by including second and third order Markov chain models. It also includes regional correction factors to adjust the precipitation percentile values as simulated by the WGEN model with respect to observed precipitation. This is a first step toward the development of a model to construct basin scale projections of future changes in climate intended for hydrological impact studies. A direct validation of the generator using selected extreme indices of precipitation has shown that the modified generator generally performed better than WGEN at simulating daily precipitation distribution, quantity and occurrence. Some discrepancies still remained or were amplified which appear to be season-related, suggesting recourse to seasonal correction factors. However, because the generator is aimed at developing climate change projections, no additional parameters were introduced in the model to keep it as parsimonious as possible. WeaGETS was indirectly validated by conducting a series of hydrological modelling experiments on the Châteauguay River Basin located in southern Quebec. Results of the simulations show that WeaGETS was able to adequately represent the duration of summer low flow events as well as the annual direct runoff. However an overestimation of the peak flows was observed for the more extreme flood events with return periods exceeding 50 years. Whether or not such an overestimation is solely caused by the generator overestimating extreme precipitation events and/or consistent combinations of precipitation and temperature needs to be further addressed through additional modelling experiments on various watersheds and with more observed climatic data before drawing definitive conclusions.

Depositional mechanics of turbulent nuées ardentes (surges) from their grain sizes

The ability of turbulent nuées ardentes (surges) to transport coarse pyroclasts has been questioned on the basis that settling velocities of coarse fragments in the deposits are much too high for them to have been supported by turbulence in a dilute gas suspension. A computer model is used to evaluate the settling velocity of pyroclasts in suspensions of varying concentration and temperature. Since suspension of grains in low-concentration surges occurs if the shear velocity exceeds the settling velocity, the shear velocities related to the 16th and 84th percentiles, and the mean of the grain-size distribution are compared in surge deposits of the Vulsini, with the shear velocity necessary to move the coarsest grain on the bed surface (the Shields criterion). The results show that the settling velocities do not vary significantly in gaseous suspensions having volume concentrations lower than 15%, and that an increase in concentration to 25% is not sufficient to decrease the settling velocity of the coarser fraction, if it represents flow shear velocity. It is shown that the settling velocity of the mean grain size (Mz) best depicts the shear velocity of a dilute turbulent suspension. Applying the results to the May 1902 paroxysmal nuées ardentes of Mount Pelée shows that the estimated mean velocities are well within the observed velocities, and sufficient to support all the clasts in dilute, turbulent suspensions.

Can Precipitation and Temperature from Meteorological Reanalyses Be Used for Hydrological Modeling

AbstractThis paper investigates the potential of reanalyses as proxies of observed surface precipitation and temperature to force hydrological models. Three global atmospheric reanalyses (ERA-Interim, CFSR, and MERRA), one regional reanalysis (NARR), and one global meteorological forcing dataset obtained by bias-correcting ERA-Interim [Water and Global Change (WATCH) Forcing Data ERA-Interim (WFDEI)] were compared to one gridded observation database over the contiguous United States. Results showed that all temperature datasets were similar to the gridded observation over most of the United States. On the other hand, precipitation from all three global reanalyses was biased, especially in summer and winter in the southeastern United States. The regional reanalysis precipitation was closer to observations since it indirectly assimilates surface precipitation. The WFDEI dataset was generally less biased than the reanalysis datasets. All datasets were then used to force a global conceptual hydrological model...

Using Natural Variability as a Baseline to Evaluate the Performance of Bias Correction Methods in Hydrological Climate Change Impact Studies

AbstractPostprocessing of climate model outputs is usually performed to remove biases prior to performing climate change impact studies. The evaluation of the performance of bias correction methods is routinely done by comparing postprocessed outputs to observed data. However, such an approach does not take into account the inherent uncertainty linked to natural climate variability and may end up recommending unnecessary complex postprocessing methods. This study evaluates the performance of bias correction methods using natural variability as a baseline. This baseline implies that any bias between model simulations and observations is only significant if it is larger than the natural climate variability. Four bias correction methods are evaluated with respect to reproducing a set of climatic and hydrological statistics. When using natural variability as a baseline, complex bias correction methods still outperform the simplest ones for precipitation and temperature time series, although the differences ar...

Water resources optimization method in the context of climate change

AbstractThis paper describes a method for water resources optimization in the context of climate change. The method takes into account the midterm variability or seasonality of inflows as well as the uncertainty in the climate change and resulting flows. The objective of the optimization algorithm is to find a compromise between the long-term planning of water resources systems and the midterm operations for optimum hydropower production. The proposed algorithm consists of the midterm dynamic programming formulation coupled with the use of the expected value of the cost-to-go function between two consecutive long-term periods. Future climate projections and transition probabilities between projections represent the stochastic nature of inflows and the nonstationarity of climate. The performance of the method was evaluated through the simulation of inflow projections for the Manicouagan River basin in Quebec, Canada. The results showed that the algorithm was able to adapt the operating policy to the climat...