SHYREG method is a regionalized method for rainfall and flood frequency analysis (FFA). It is bas... more SHYREG method is a regionalized method for rainfall and flood frequency analysis (FFA). It is based on processes simulation. It couples an hourly rainfall generator with a rainfall-runoff model, simplified enough to be regionalized. The method has been calibrated using all hydro meteorological data available at the national level. In France, that represents about 2800 raingauges of the French Weather Service network and about 1800 stations of the hydrometric National Bank network. Then, the method has been regionalized to provide a rainfall and flow quantiles database. An evaluation of the method was carried out during different thesis works and more recently during the ANR project Extraflo, with the aim of comparing different FFA approaches. The accuracy of the method in estimating rainfall and flow quantiles has been proved, as well as its stability due to a parameterization based on average values. The link with rainfall seems preferable to extrapolation based solely on the flow. Thus, another interest of the method is to take into account extreme flood behaviour with help of rainfall frequency estimation. In addition, the approach is implicitly multi-durational, and only one regionalization meets all the needs in terms hydrological hazards characterisation. For engineering needs and to avoid repeating the method implementation, this method has been applied throughout a 50 meters resolution mesh to provide a complete flood quantiles database over the French territory providing regional information on hydrological hazards. However, it is subject to restrictions related to the nature of the method: the SHYREG flows are "natural", and do not take into account specific cases like the basins highly influenced by presence of hydraulic works, flood expansion areas, high snowmelt or karsts. Information about these restrictions and uncertainty estimation is provided with this database, which can be consulted via web access.
<p>In recent years, the number of large-scale hydrological forecasting systems has ... more <p>In recent years, the number of large-scale hydrological forecasting systems has been steadily growing. This may lead to regions having numerous models spatially overlapping each other. Some of these regions have what we will refer to as a regional, more specialized, model for the area that performs generally better than their large-scale counterpart, considering the coarser spatial resolution and sometimes lack of calibration of the latter. Our work explored the possibility of using simple methods to retrieve hydrological information from a large-scale model, namely the National Surface and River Prediction System (NSRPS) that will eventually cover the Canadian territory, in order to improve the forecasts from a local system, namely the Système de Prévision Hydrologique (SPH) that covers most of the province of Quebec. Outputs from the two forecasting systems were thus combined using methods including the simple mean, a weighted average in which the weights are optimized either using the Kling-Gupta Efficiency (KGE) or the Continuous Ranked Probability Score (CRPS) as cost functions, or weights calculated from the residual errors of the forecasts. Bayesian Model Averaging (BMA) was also explored to combine the ensemble forecasts from both systems. The results show that it is possible to improve the local hydrological forecasts by using simple weighted combinations with forecasts from the large-scale system. Performance was assessed using many well-known criteria such as the Nash-Sutcliffe Efficiency (NSE), KGE and CRPS. Results were averaged over the 61 available gauging stations and analyzed at lead times ranging from 3 to 120 hours. We observed improvements in all criteria for lead times over 60 hours as well as no loss in performance at any lead times. Finally, the methods were also used in a leave-one-out setup to simulate performance on ungauged basins. The performance gain for ungauged basins is similar to that of the gauged basins, hinting at the fact that these simple methods could also improve forecasts in more remote territories where no measurements are available.</p>
Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-relate... more Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-related variable that is most commonly used in hydrology, as it expresses the total quantity of water (solid and liquid) stored in the snowpack. Measurements of SWE are, however, expensive and not continuously accessible in real time. This motivates a search for alternative ways of estimating SWE from measurements that are more widely available and continuous over time. SWE can be calculated by multiplying snow depth by the bulk density of the snowpack. Regression models proposed in the literature first estimate snow density and then calculate SWE. More recently, a novel approach to this problem has been developed and is based on an ensemble of multilayer perceptrons (MLPs). Although this approach compared favorably with existing regression models, snow density values at the lower and higher ends of the range remained inaccurate. Here, we improve upon this recent method for determining SWE from snow depth. We show the general applicability of the method through the use of a large data set of 234 779 snow depth-density-SWE records from 2878 nonuniformly distributed sites across Canada. These data cover almost 4 decades of snowfall. First, it is shown that the direct estimation of SWE produces better results than the estimation of snow density, followed by the calculation of SWE. Second, testing several artificial neural network (ANN) structural characteristics improves estimates of SWE. Optimizing MLP parameters separately for each snow climate class gives a greater representation of the geophysical diversity of snow. Furthermore, the uncertainty of snow depth measurements is included for a more realistic estimation. A comparison with commonly used regression models reveals that the ensemble of MLPs proposed here leads to noticeably more accurate estimates of SWE. This study thus shows that delving deeper into ANN theory helps improve SWE estimation.
<p>In recent years, the number of large-scale hydrological forecasting systems has ... more <p>In recent years, the number of large-scale hydrological forecasting systems has been steadily growing. This may lead to regions having numerous models spatially overlapping each other. Some of these regions have what we will refer to as a regional, more specialized, model for the area that performs generally better than their large-scale counterpart, considering the coarser spatial resolution and sometimes lack of calibration of the latter. Our work explored the possibility of using simple methods to retrieve hydrological information from a large-scale model, namely the National Surface and River Prediction System (NSRPS) that will eventually cover the Canadian territory, in order to improve the forecasts from a local system, namely the Système de Prévision Hydrologique (SPH) that covers most of the province of Quebec. Outputs from the two forecasting systems were thus combined using methods including the simple mean, a weighted average in which the weights are optimized either using the Kling-Gupta Efficiency (KGE) or the Continuous Ranked Probability Score (CRPS) as cost functions, or weights calculated from the residual errors of the forecasts. Bayesian Model Averaging (BMA) was also explored to combine the ensemble forecasts from both systems. The results show that it is possible to improve the local hydrological forecasts by using simple weighted combinations with forecasts from the large-scale system. Performance was assessed using many well-known criteria such as the Nash-Sutcliffe Efficiency (NSE), KGE and CRPS. Results were averaged over the 61 available gauging stations and analyzed at lead times ranging from 3 to 120 hours. We observed improvements in all criteria for lead times over 60 hours as well as no loss in performance at any lead times. Finally, the methods were also used in a leave-one-out setup to simulate performance on ungauged basins. The performance gain for ungauged basins is similar to that of the gauged basins, hinting at the fact that these simple methods could also improve forecasts in more remote territories where no measurements are available.</p>
<p>Particle filtering is interesting for snow data assimilation because of its mini... more <p>Particle filtering is interesting for snow data assimilation because of its minimal assumptions. However, implementing a particle filter over a large spatial domain is challenging for many reasons. For instance, the number of required particles rises exponentially as the domain size increases. Another important issue when spatializing a particle filter for snow data assimilation is the creation of spatial discontinuities when resampling the particles at locations where snow observations are available. In this presentation, we will describe how we implemented a spatialized particle filter for snow data assimilation over a large portion of the province of Quebec, Canada (600 000 km<sup>2 </sup>). Two different types of snow observations where assimilated with this particle filter: sporadic manual snow surveys, which measure snow water equivalent directly, and continuous automated snow depth observations, which we converted to snow water equivalent using an ensemble of neural networks. We will then explain how a more frequent data assimilation can create unwanted discontinuities and break the spatial structure of the particles, and how we can remediate that by using an adaptation of the Schaake Shuffle reordering method. We will show that this solution significantly reduces the random noise in the distribution of the particles and decreases the uncertainty associated with the estimation. We emphasize that the proposed spatialized particle framework could also eventually accommodate other types of data, such as citizen science data and gamma monitoring data. Overall, the proposed method allows to obtain improved spatial representation of snow water equivalent compared to the previous operational method used by the government of Quebec.<span> </span></p>
<p>Global or large-scale hydrological forecasting systems covering entire c... more <p>Global or large-scale hydrological forecasting systems covering entire countries, continents and even the entire planet are growing in popularity. As more large-scale hydrological forecasting systems emerge, it is likely that they will co-exist with pre-existing local forecasting systems. It is the case for instance in Canada, where most provinces have their own streamflow forecasting system, while the new NSRPS will eventually cover the whole country using a 1km by 1km grid. Those province, for instance Quebec, built their own forecasting systems on hydrological models configured for river catchments rather than a regular grid. Using this situation as a starting point and a case study, we propose a Bayesian framework for merging the forecasts from two systems. Within this Bayesian framework, the large-scale prior information comes from the NSRPS. This prior information is then updated using forecasts from the government of Quebec and the associated likelihood. In order to account for forecast uncertainty, this work is carried out using a probabilistic approach for both the NSRPS and Quebec’s Système de Prévision Hydrologique (SPH). While SPH produces probabilistic forecasts by default, the preliminary version of the NSRPS that we had access to is deterministic. Consequently, forecasts from the NSRPS had to be dressed into an ensemble in order to use them as prior distribution within the Bayesian merging framework. Alternative prior distributions (climatology, Markov chain) are also considered instead of those obtained from the NSRPS. Since both forecasting systems include ungauged sites, a version of this Bayesian merging framework based on regional statistics was also developed and tested using cross-validation. Our results show that the merged forecasts perform at least as well as the best individual system, for both gauged and ungauged basins. For longer lead times, merged forecasts can even outperform individual systems. Considering that the NSRPS relies on a non-calibrated model with no data assimilation, those results show that there could be important practical gains in merging large scale hydrological forecasts with local scale forecasts.</p>
Hydrology and Earth System Sciences Discussions, 2020
Abstract. Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is t... more Abstract. Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-related variable that is most commonly used in hydrology, as it expresses the total quantity of water (solid and liquid) stored in the snowpack. Measurements of SWE are, however, expensive and not continuously accessible in real time. This motivates a search for alternative ways of estimating SWE from measurements that are more widely available and continuous over time. SWE can be calculated by multiplying snow depth with the bulk density of the snowpack. Regression models proposed in the literature first estimate snow density and then calculate SWE. More recently, a novel approach to this problem has been developed and is based on an ensemble of multilayer perceptrons (MLPs). Although this approach compared favourably with existing regression models, snow density values at the lower and higher ends of the range remained inaccurate. Here, we improve upon this recent method for de...
Small catchments are largely under-represented in gauged catchments samples used to calibrate nat... more Small catchments are largely under-represented in gauged catchments samples used to calibrate nationwide flood quantiles estimation methods. For instance, in France, catchments of less than 10 km2 represent 3% of the gauged catchments but 40% of the sites of interest. Those small basins can generally be found upstream of a gauging station. This size gap should be fulfilled by regionalisation techniques. Those methods aim to transfer information from gauged basins to ungauged ones, but they rarely address the size gap issue. The work presented her aims to question about what happens inside the gauged catchments. Are the model calibrated at gauged catchments scale still valid at a finer one or is it useful to implement some downsizing method?
Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-relate... more Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-related variable that is most commonly used in hydrology, as it expresses the total quantity of water (solid and liquid) stored in the snowpack. Measurements of SWE are, however, expensive and not continuously accessible in real time. This motivates a search for alternative ways of estimating SWE from measurements that are more widely available and continuous over time. SWE can be calculated by multiplying snow depth by the bulk density of the snowpack. Regression models proposed in the literature first estimate snow density and then calculate SWE. More recently, a novel approach to this problem has been developed and is based on an ensemble of multilayer perceptrons (MLPs). Although this approach compared favorably with existing regression models, snow density values at the lower and higher ends of the range remained inaccurate. Here, we improve upon this recent method for determining SWE from snow depth. We show the general applicability of the method through the use of a large data set of 234 779 snow depth-density-SWE records from 2878 nonuniformly distributed sites across Canada. These data cover almost 4 decades of snowfall. First, it is shown that the direct estimation of SWE produces better results than the estimation of snow density, followed by the calculation of SWE. Second, testing several artificial neural network (ANN) structural characteristics improves estimates of SWE. Optimizing MLP parameters separately for each snow climate class gives a greater representation of the geophysical diversity of snow. Furthermore, the uncertainty of snow depth measurements is included for a more realistic estimation. A comparison with commonly used regression models reveals that the ensemble of MLPs proposed here leads to noticeably more accurate estimates of SWE. This study thus shows that delving deeper into ANN theory helps improve SWE estimation.
The objective of flood frequency analysis (FFA) is to associate flood intensity with a probabilit... more The objective of flood frequency analysis (FFA) is to associate flood intensity with a probability of exceedance. Many methods are currently employed for this, ranging from statistical distribution fitting to simulation approaches. In many cases the site of interest is actually ungauged, and a regionalisation scheme has to be associated with the FFA method, leading to a multiplication of the number of possible methods available. This paper presents the results of a wide-range comparison of FFA methods from statistical and simulation families associated with different regionalisation schemes based on regression, or spatial or physical proximity. The methods are applied to a set of 1535 French catchments, and a k-fold cross-validation procedure is used to consider the ungauged configuration. The results suggest that FFA from the statistical family largely relies on the regionalisation step, whereas the simulation-based method is more stable regarding regionalisation. This conclusion emphasises the difficulty of the regionalisation process. The results are also contrasted depending on the type of climate: the Mediterranean catchments tend to aggravate the differences between the methods.
Flood frequency analyses (FFAs) are needed for flood risk management. Many methods exist ranging ... more Flood frequency analyses (FFAs) are needed for flood risk management. Many methods exist ranging from classical purely statistical approaches to more complex approaches based on process simulation. The results of these methods are associated with uncertainties that are sometimes difficult to estimate due to the complexity of the approaches or the number of parameters, especially for process simulation. This is the case of the simulation-based FFA approach called SHYREG presented in this paper, in which a rainfall generator is coupled with a simple rainfall-runoff model in an attempt to estimate the uncertainties due to the estimation of the seven parameters needed to estimate flood frequencies. The six parameters of the rainfall generator are mean values, so their theoretical distribution is known and can be used to estimate the generator uncertainties. In contrast, the theoretical distribution of the single hydrological model parameter is unknown; consequently, a bootstrap method is applied to estimate the calibration uncertainties. The propagation of uncertainty from the rainfall generator to the hydrological model is also taken into account. This method is applied to 1112 basins throughout France. Uncertainties coming from the SHYREG method and from purely statistical approaches are compared, and the results are discussed according to the length of the recorded observations, basin size and basin location. Uncertainties of the SHYREG method decrease as the basin size increases or as the length of the recorded flow increases. Moreover, the results show that the confidence intervals of the SHYREG method are relatively small despite the complexity of the method and the number of parameters (seven). This is due to the stability of the parameters and takes into account the dependence of uncertainties due to the rainfall model and the hydrological calibration. Indeed, the uncertainties on the flow quantiles are on the same order of magnitude as those associated with the use of a statistical law with two parameters (here generalised extreme value
SHYREG method is a regionalized method for rainfall and flood frequency analysis (FFA). It is bas... more SHYREG method is a regionalized method for rainfall and flood frequency analysis (FFA). It is based on processes simulation. It couples an hourly rainfall generator with a rainfall-runoff model, simplified enough to be regionalized. The method has been calibrated using all hydro meteorological data available at the national level. In France, that represents about 2800 raingauges of the French Weather Service network and about 1800 stations of the hydrometric National Bank network. Then, the method has been regionalized to provide a rainfall and flow quantiles database. An evaluation of the method was carried out during different thesis works and more recently during the ANR project Extraflo, with the aim of comparing different FFA approaches. The accuracy of the method in estimating rainfall and flow quantiles has been proved, as well as its stability due to a parameterization based on average values. The link with rainfall seems preferable to extrapolation based solely on the flow. Thus, another interest of the method is to take into account extreme flood behaviour with help of rainfall frequency estimation. In addition, the approach is implicitly multi-durational, and only one regionalization meets all the needs in terms hydrological hazards characterisation. For engineering needs and to avoid repeating the method implementation, this method has been applied throughout a 50 meters resolution mesh to provide a complete flood quantiles database over the French territory providing regional information on hydrological hazards. However, it is subject to restrictions related to the nature of the method: the SHYREG flows are "natural", and do not take into account specific cases like the basins highly influenced by presence of hydraulic works, flood expansion areas, high snowmelt or karsts. Information about these restrictions and uncertainty estimation is provided with this database, which can be consulted via web access.
<p>In recent years, the number of large-scale hydrological forecasting systems has ... more <p>In recent years, the number of large-scale hydrological forecasting systems has been steadily growing. This may lead to regions having numerous models spatially overlapping each other. Some of these regions have what we will refer to as a regional, more specialized, model for the area that performs generally better than their large-scale counterpart, considering the coarser spatial resolution and sometimes lack of calibration of the latter. Our work explored the possibility of using simple methods to retrieve hydrological information from a large-scale model, namely the National Surface and River Prediction System (NSRPS) that will eventually cover the Canadian territory, in order to improve the forecasts from a local system, namely the Système de Prévision Hydrologique (SPH) that covers most of the province of Quebec. Outputs from the two forecasting systems were thus combined using methods including the simple mean, a weighted average in which the weights are optimized either using the Kling-Gupta Efficiency (KGE) or the Continuous Ranked Probability Score (CRPS) as cost functions, or weights calculated from the residual errors of the forecasts. Bayesian Model Averaging (BMA) was also explored to combine the ensemble forecasts from both systems. The results show that it is possible to improve the local hydrological forecasts by using simple weighted combinations with forecasts from the large-scale system. Performance was assessed using many well-known criteria such as the Nash-Sutcliffe Efficiency (NSE), KGE and CRPS. Results were averaged over the 61 available gauging stations and analyzed at lead times ranging from 3 to 120 hours. We observed improvements in all criteria for lead times over 60 hours as well as no loss in performance at any lead times. Finally, the methods were also used in a leave-one-out setup to simulate performance on ungauged basins. The performance gain for ungauged basins is similar to that of the gauged basins, hinting at the fact that these simple methods could also improve forecasts in more remote territories where no measurements are available.</p>
Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-relate... more Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-related variable that is most commonly used in hydrology, as it expresses the total quantity of water (solid and liquid) stored in the snowpack. Measurements of SWE are, however, expensive and not continuously accessible in real time. This motivates a search for alternative ways of estimating SWE from measurements that are more widely available and continuous over time. SWE can be calculated by multiplying snow depth by the bulk density of the snowpack. Regression models proposed in the literature first estimate snow density and then calculate SWE. More recently, a novel approach to this problem has been developed and is based on an ensemble of multilayer perceptrons (MLPs). Although this approach compared favorably with existing regression models, snow density values at the lower and higher ends of the range remained inaccurate. Here, we improve upon this recent method for determining SWE from snow depth. We show the general applicability of the method through the use of a large data set of 234 779 snow depth-density-SWE records from 2878 nonuniformly distributed sites across Canada. These data cover almost 4 decades of snowfall. First, it is shown that the direct estimation of SWE produces better results than the estimation of snow density, followed by the calculation of SWE. Second, testing several artificial neural network (ANN) structural characteristics improves estimates of SWE. Optimizing MLP parameters separately for each snow climate class gives a greater representation of the geophysical diversity of snow. Furthermore, the uncertainty of snow depth measurements is included for a more realistic estimation. A comparison with commonly used regression models reveals that the ensemble of MLPs proposed here leads to noticeably more accurate estimates of SWE. This study thus shows that delving deeper into ANN theory helps improve SWE estimation.
<p>In recent years, the number of large-scale hydrological forecasting systems has ... more <p>In recent years, the number of large-scale hydrological forecasting systems has been steadily growing. This may lead to regions having numerous models spatially overlapping each other. Some of these regions have what we will refer to as a regional, more specialized, model for the area that performs generally better than their large-scale counterpart, considering the coarser spatial resolution and sometimes lack of calibration of the latter. Our work explored the possibility of using simple methods to retrieve hydrological information from a large-scale model, namely the National Surface and River Prediction System (NSRPS) that will eventually cover the Canadian territory, in order to improve the forecasts from a local system, namely the Système de Prévision Hydrologique (SPH) that covers most of the province of Quebec. Outputs from the two forecasting systems were thus combined using methods including the simple mean, a weighted average in which the weights are optimized either using the Kling-Gupta Efficiency (KGE) or the Continuous Ranked Probability Score (CRPS) as cost functions, or weights calculated from the residual errors of the forecasts. Bayesian Model Averaging (BMA) was also explored to combine the ensemble forecasts from both systems. The results show that it is possible to improve the local hydrological forecasts by using simple weighted combinations with forecasts from the large-scale system. Performance was assessed using many well-known criteria such as the Nash-Sutcliffe Efficiency (NSE), KGE and CRPS. Results were averaged over the 61 available gauging stations and analyzed at lead times ranging from 3 to 120 hours. We observed improvements in all criteria for lead times over 60 hours as well as no loss in performance at any lead times. Finally, the methods were also used in a leave-one-out setup to simulate performance on ungauged basins. The performance gain for ungauged basins is similar to that of the gauged basins, hinting at the fact that these simple methods could also improve forecasts in more remote territories where no measurements are available.</p>
<p>Particle filtering is interesting for snow data assimilation because of its mini... more <p>Particle filtering is interesting for snow data assimilation because of its minimal assumptions. However, implementing a particle filter over a large spatial domain is challenging for many reasons. For instance, the number of required particles rises exponentially as the domain size increases. Another important issue when spatializing a particle filter for snow data assimilation is the creation of spatial discontinuities when resampling the particles at locations where snow observations are available. In this presentation, we will describe how we implemented a spatialized particle filter for snow data assimilation over a large portion of the province of Quebec, Canada (600 000 km<sup>2 </sup>). Two different types of snow observations where assimilated with this particle filter: sporadic manual snow surveys, which measure snow water equivalent directly, and continuous automated snow depth observations, which we converted to snow water equivalent using an ensemble of neural networks. We will then explain how a more frequent data assimilation can create unwanted discontinuities and break the spatial structure of the particles, and how we can remediate that by using an adaptation of the Schaake Shuffle reordering method. We will show that this solution significantly reduces the random noise in the distribution of the particles and decreases the uncertainty associated with the estimation. We emphasize that the proposed spatialized particle framework could also eventually accommodate other types of data, such as citizen science data and gamma monitoring data. Overall, the proposed method allows to obtain improved spatial representation of snow water equivalent compared to the previous operational method used by the government of Quebec.<span> </span></p>
<p>Global or large-scale hydrological forecasting systems covering entire c... more <p>Global or large-scale hydrological forecasting systems covering entire countries, continents and even the entire planet are growing in popularity. As more large-scale hydrological forecasting systems emerge, it is likely that they will co-exist with pre-existing local forecasting systems. It is the case for instance in Canada, where most provinces have their own streamflow forecasting system, while the new NSRPS will eventually cover the whole country using a 1km by 1km grid. Those province, for instance Quebec, built their own forecasting systems on hydrological models configured for river catchments rather than a regular grid. Using this situation as a starting point and a case study, we propose a Bayesian framework for merging the forecasts from two systems. Within this Bayesian framework, the large-scale prior information comes from the NSRPS. This prior information is then updated using forecasts from the government of Quebec and the associated likelihood. In order to account for forecast uncertainty, this work is carried out using a probabilistic approach for both the NSRPS and Quebec’s Système de Prévision Hydrologique (SPH). While SPH produces probabilistic forecasts by default, the preliminary version of the NSRPS that we had access to is deterministic. Consequently, forecasts from the NSRPS had to be dressed into an ensemble in order to use them as prior distribution within the Bayesian merging framework. Alternative prior distributions (climatology, Markov chain) are also considered instead of those obtained from the NSRPS. Since both forecasting systems include ungauged sites, a version of this Bayesian merging framework based on regional statistics was also developed and tested using cross-validation. Our results show that the merged forecasts perform at least as well as the best individual system, for both gauged and ungauged basins. For longer lead times, merged forecasts can even outperform individual systems. Considering that the NSRPS relies on a non-calibrated model with no data assimilation, those results show that there could be important practical gains in merging large scale hydrological forecasts with local scale forecasts.</p>
Hydrology and Earth System Sciences Discussions, 2020
Abstract. Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is t... more Abstract. Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-related variable that is most commonly used in hydrology, as it expresses the total quantity of water (solid and liquid) stored in the snowpack. Measurements of SWE are, however, expensive and not continuously accessible in real time. This motivates a search for alternative ways of estimating SWE from measurements that are more widely available and continuous over time. SWE can be calculated by multiplying snow depth with the bulk density of the snowpack. Regression models proposed in the literature first estimate snow density and then calculate SWE. More recently, a novel approach to this problem has been developed and is based on an ensemble of multilayer perceptrons (MLPs). Although this approach compared favourably with existing regression models, snow density values at the lower and higher ends of the range remained inaccurate. Here, we improve upon this recent method for de...
Small catchments are largely under-represented in gauged catchments samples used to calibrate nat... more Small catchments are largely under-represented in gauged catchments samples used to calibrate nationwide flood quantiles estimation methods. For instance, in France, catchments of less than 10 km2 represent 3% of the gauged catchments but 40% of the sites of interest. Those small basins can generally be found upstream of a gauging station. This size gap should be fulfilled by regionalisation techniques. Those methods aim to transfer information from gauged basins to ungauged ones, but they rarely address the size gap issue. The work presented her aims to question about what happens inside the gauged catchments. Are the model calibrated at gauged catchments scale still valid at a finer one or is it useful to implement some downsizing method?
Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-relate... more Canada's water cycle is driven mainly by snowmelt. Snow water equivalent (SWE) is the snow-related variable that is most commonly used in hydrology, as it expresses the total quantity of water (solid and liquid) stored in the snowpack. Measurements of SWE are, however, expensive and not continuously accessible in real time. This motivates a search for alternative ways of estimating SWE from measurements that are more widely available and continuous over time. SWE can be calculated by multiplying snow depth by the bulk density of the snowpack. Regression models proposed in the literature first estimate snow density and then calculate SWE. More recently, a novel approach to this problem has been developed and is based on an ensemble of multilayer perceptrons (MLPs). Although this approach compared favorably with existing regression models, snow density values at the lower and higher ends of the range remained inaccurate. Here, we improve upon this recent method for determining SWE from snow depth. We show the general applicability of the method through the use of a large data set of 234 779 snow depth-density-SWE records from 2878 nonuniformly distributed sites across Canada. These data cover almost 4 decades of snowfall. First, it is shown that the direct estimation of SWE produces better results than the estimation of snow density, followed by the calculation of SWE. Second, testing several artificial neural network (ANN) structural characteristics improves estimates of SWE. Optimizing MLP parameters separately for each snow climate class gives a greater representation of the geophysical diversity of snow. Furthermore, the uncertainty of snow depth measurements is included for a more realistic estimation. A comparison with commonly used regression models reveals that the ensemble of MLPs proposed here leads to noticeably more accurate estimates of SWE. This study thus shows that delving deeper into ANN theory helps improve SWE estimation.
The objective of flood frequency analysis (FFA) is to associate flood intensity with a probabilit... more The objective of flood frequency analysis (FFA) is to associate flood intensity with a probability of exceedance. Many methods are currently employed for this, ranging from statistical distribution fitting to simulation approaches. In many cases the site of interest is actually ungauged, and a regionalisation scheme has to be associated with the FFA method, leading to a multiplication of the number of possible methods available. This paper presents the results of a wide-range comparison of FFA methods from statistical and simulation families associated with different regionalisation schemes based on regression, or spatial or physical proximity. The methods are applied to a set of 1535 French catchments, and a k-fold cross-validation procedure is used to consider the ungauged configuration. The results suggest that FFA from the statistical family largely relies on the regionalisation step, whereas the simulation-based method is more stable regarding regionalisation. This conclusion emphasises the difficulty of the regionalisation process. The results are also contrasted depending on the type of climate: the Mediterranean catchments tend to aggravate the differences between the methods.
Flood frequency analyses (FFAs) are needed for flood risk management. Many methods exist ranging ... more Flood frequency analyses (FFAs) are needed for flood risk management. Many methods exist ranging from classical purely statistical approaches to more complex approaches based on process simulation. The results of these methods are associated with uncertainties that are sometimes difficult to estimate due to the complexity of the approaches or the number of parameters, especially for process simulation. This is the case of the simulation-based FFA approach called SHYREG presented in this paper, in which a rainfall generator is coupled with a simple rainfall-runoff model in an attempt to estimate the uncertainties due to the estimation of the seven parameters needed to estimate flood frequencies. The six parameters of the rainfall generator are mean values, so their theoretical distribution is known and can be used to estimate the generator uncertainties. In contrast, the theoretical distribution of the single hydrological model parameter is unknown; consequently, a bootstrap method is applied to estimate the calibration uncertainties. The propagation of uncertainty from the rainfall generator to the hydrological model is also taken into account. This method is applied to 1112 basins throughout France. Uncertainties coming from the SHYREG method and from purely statistical approaches are compared, and the results are discussed according to the length of the recorded observations, basin size and basin location. Uncertainties of the SHYREG method decrease as the basin size increases or as the length of the recorded flow increases. Moreover, the results show that the confidence intervals of the SHYREG method are relatively small despite the complexity of the method and the number of parameters (seven). This is due to the stability of the parameters and takes into account the dependence of uncertainties due to the rainfall model and the hydrological calibration. Indeed, the uncertainties on the flow quantiles are on the same order of magnitude as those associated with the use of a statistical law with two parameters (here generalised extreme value
Uploads
Papers by Jean Odry