I am currently building a model that has an operational reservoir that I'm struggling with. This reservoir is operated to hold back snowmelt during the spring, and then this water is slowly released through the summer – but unfortunately that is all I know about it. There are no records of stage/discharge and there are no design drawings. I’ve got it set up as its own HRU and its at the outlet of a sub-basin. I’m wondering if you have any advice on specific parameters that you would include (e.g. :VariableWeirHeight) and how you might go about calibrating it? I see in the manual there is reference to approximate rule curves, but I'm not sure what this means and if they would be relevant. If there are any existing templates of example input files that would also be very appreciated.
Thanks for all your help,
Stefan
setting up and calibrating operational reservoirs with minimal data
-
sgronsdahl
- Posts: 7
- Joined: Sun Sep 22, 2019 9:13 pm
-
rchlumsk
- Posts: 156
- Joined: Fri Sep 16, 2016 3:18 pm
Re: setting up and calibrating operational reservoirs with minimal data
Hi Stefan
Do you have any idea of what kind of outlet control there is? That will help to understand how best to represent the outflows.
In any case, here are a few ideas of how to approach it. Keep in mind that the uncertainty around the structures at the reservoir and uncertainty in human operations will play a large role, and will generally mean that any interpretation of results should be taken with a grain of salt.
0. Lake representation: there is an option to represent the reservoir as a lake, suitable for small bodies without specified outflow characteristics, but where some attentuation is still included in the model (see page 71 in the manual). If this is not suitable as there is a large amount of retained storage, see points 1-3.
1. Stage-Storage curve: you will need to build a relationship between elevation and area/storage. If you have no drawings or estimates of this, then contours with areas at various elevations can be used to build it (assuming the data is available, check various sources such as Land Information Ontario). If you don't have an idea of the storage characteristics or a good guess, the reservoir will be next to impossible to model.
2. Outflow: at a basic level you will need an idea of what controls the outflow, and add an outflow column to the relationship (i.e. stage-outflow relationship). For example, if you know it is a weir of a given height and length, you can estimate the outflow with stage. If the outflow is a combination of multiple structures, you can estimate the total outflow based on a number of structures and the sum of the outflows (similar to variable weir heights command). I have not explored too many of the options for outflow in Raven other than the basic stage-outflow curve, but presumably if you have more information on the type of structure, you can pick an outflow relationship accordingly.
3. Seasonal curves: a useful feature is the ability to specify the stage-stage-outflow curves on a seasonal basis, which can be used to mimic the seasonal operations of the reservoir outflows. I don't believe this is in the manual, but I will post an example of this on the the thread at a later time.
4. Calibration: this will be difficult without information, as typically either the reservoir inflows, outflows, or observed stage (or all three) would be needed for calibration/validation. In absence of this, the best you can likely do is use anecdotal data for if/when the reservoir has overtopped or spilled (e.g. known to have not spilled in last 5 years, overtopped in 2015, etc.) to calibrate the model to reproduce that behaviour. Of course, this will be a very approximate calibration with a lot of asterisks around the results, but possibly the best that can be done without more info.
Hope that helps, please post back if you have any more notes that will help in building this model and any more questions. I will dig up that example of seasonal curves for you to look at as well.
Cheers,
Rob
Do you have any idea of what kind of outlet control there is? That will help to understand how best to represent the outflows.
In any case, here are a few ideas of how to approach it. Keep in mind that the uncertainty around the structures at the reservoir and uncertainty in human operations will play a large role, and will generally mean that any interpretation of results should be taken with a grain of salt.
0. Lake representation: there is an option to represent the reservoir as a lake, suitable for small bodies without specified outflow characteristics, but where some attentuation is still included in the model (see page 71 in the manual). If this is not suitable as there is a large amount of retained storage, see points 1-3.
1. Stage-Storage curve: you will need to build a relationship between elevation and area/storage. If you have no drawings or estimates of this, then contours with areas at various elevations can be used to build it (assuming the data is available, check various sources such as Land Information Ontario). If you don't have an idea of the storage characteristics or a good guess, the reservoir will be next to impossible to model.
2. Outflow: at a basic level you will need an idea of what controls the outflow, and add an outflow column to the relationship (i.e. stage-outflow relationship). For example, if you know it is a weir of a given height and length, you can estimate the outflow with stage. If the outflow is a combination of multiple structures, you can estimate the total outflow based on a number of structures and the sum of the outflows (similar to variable weir heights command). I have not explored too many of the options for outflow in Raven other than the basic stage-outflow curve, but presumably if you have more information on the type of structure, you can pick an outflow relationship accordingly.
3. Seasonal curves: a useful feature is the ability to specify the stage-stage-outflow curves on a seasonal basis, which can be used to mimic the seasonal operations of the reservoir outflows. I don't believe this is in the manual, but I will post an example of this on the the thread at a later time.
4. Calibration: this will be difficult without information, as typically either the reservoir inflows, outflows, or observed stage (or all three) would be needed for calibration/validation. In absence of this, the best you can likely do is use anecdotal data for if/when the reservoir has overtopped or spilled (e.g. known to have not spilled in last 5 years, overtopped in 2015, etc.) to calibrate the model to reproduce that behaviour. Of course, this will be a very approximate calibration with a lot of asterisks around the results, but possibly the best that can be done without more info.
Hope that helps, please post back if you have any more notes that will help in building this model and any more questions. I will dig up that example of seasonal curves for you to look at as well.
Cheers,
Rob
Robert Chlumsky
rchlumsk@uwaterloo.ca
rchlumsk@uwaterloo.ca
-
sgronsdahl
- Posts: 7
- Joined: Sun Sep 22, 2019 9:13 pm
Re: setting up and calibrating operational reservoirs with minimal data
Hi Rob - thanks for the response.
After talking to folks more familiar with the way this reservoir is managed, it seems that its operations are not predictable from year to year. On the plus side, there is a continuous streamflow record not far from the outlet. Originally I was hoping to use this record for calibration, but instead I've decided to use this record to override the reservoir outflow. Thankfully I have other gauged locations that I can calibrate to. During the calibration process I will try to make sure that the long-term stage of the reservoir stays consistent, so that it is oscillating regularly from year to year. Some initial manual adjustments to model parameters makes me think this seems like a promising solution. Do you foresee any issues with this approach?
A second question has also popped up. In the WatershedStorage.csv output file I've noticed that the Cumulative Outflow [mm] column decreases on the order of ~3,000 mm/day (which is highly unrealistic), while the Cumulative Input [mm] increases at a more reasonable rate of ~2 mm/day. This translates to a large MB Error [mm] that changes by around -3,000 mm/day. However, all the hydrographs look good and all the other model storage compartments appear to fluctuate in a reasonable manner through time. I.E. - There doesn't seem to be any glaring mass balance issues with the model. Am I interpreting this WatershedStorage file incorrectly, should I be concerned with this MB Error?
Thanks again for your help,
Stefan
After talking to folks more familiar with the way this reservoir is managed, it seems that its operations are not predictable from year to year. On the plus side, there is a continuous streamflow record not far from the outlet. Originally I was hoping to use this record for calibration, but instead I've decided to use this record to override the reservoir outflow. Thankfully I have other gauged locations that I can calibrate to. During the calibration process I will try to make sure that the long-term stage of the reservoir stays consistent, so that it is oscillating regularly from year to year. Some initial manual adjustments to model parameters makes me think this seems like a promising solution. Do you foresee any issues with this approach?
A second question has also popped up. In the WatershedStorage.csv output file I've noticed that the Cumulative Outflow [mm] column decreases on the order of ~3,000 mm/day (which is highly unrealistic), while the Cumulative Input [mm] increases at a more reasonable rate of ~2 mm/day. This translates to a large MB Error [mm] that changes by around -3,000 mm/day. However, all the hydrographs look good and all the other model storage compartments appear to fluctuate in a reasonable manner through time. I.E. - There doesn't seem to be any glaring mass balance issues with the model. Am I interpreting this WatershedStorage file incorrectly, should I be concerned with this MB Error?
Thanks again for your help,
Stefan
-
rchlumsk
- Posts: 156
- Joined: Fri Sep 16, 2016 3:18 pm
Re: setting up and calibrating operational reservoirs with minimal data
Hi Stefan
I think that overriding the reservoir outflow with observed values in order to calibrate the rest of the model is a great solution. Depending on what you need from the model you will still need to predict the reservoir behaviour into the future (or use a known operational release for short-term predictions).
I assume the stage-storage is still unknown, in which case you will likely want to setup a relationship for a stage volume curve and consider that as a direct calibration parameter, or at least in some calibration runs (i.e. larger volume will mitigate large changes in the curves, reduce spillage, etc.). You will definitely need some realism checks to ensure that the volumes and other parameters (such as ET) are reasonable in the calibration and not forcing 100% ET To avoid reservoir spillage; if you have any reasonable bounds on the reservoir volume, those would be good to include in the stage-storage calibration. Another option is to iteratively set a relationship for stage-storage, calibrate the model, check the results and consider manually tweaking the storage relationship to avoid too much uncertainty in any given calibration run. In any case, overriding the outflows with gauged data is a great approach to constrain your calibration.
In terms of the MB error, I know there were previously some potential bugs around the mass balance pertaining to reservoirs. I assume you are using the latest Raven version 2.9.2? If you are not I would try it with the latest version, otherwise (and if the issue persists in the latest version) I will pass this one to Dr. Craig as he is far more familiar with the latest MB calculations.
Cheers,
Rob
I think that overriding the reservoir outflow with observed values in order to calibrate the rest of the model is a great solution. Depending on what you need from the model you will still need to predict the reservoir behaviour into the future (or use a known operational release for short-term predictions).
I assume the stage-storage is still unknown, in which case you will likely want to setup a relationship for a stage volume curve and consider that as a direct calibration parameter, or at least in some calibration runs (i.e. larger volume will mitigate large changes in the curves, reduce spillage, etc.). You will definitely need some realism checks to ensure that the volumes and other parameters (such as ET) are reasonable in the calibration and not forcing 100% ET To avoid reservoir spillage; if you have any reasonable bounds on the reservoir volume, those would be good to include in the stage-storage calibration. Another option is to iteratively set a relationship for stage-storage, calibrate the model, check the results and consider manually tweaking the storage relationship to avoid too much uncertainty in any given calibration run. In any case, overriding the outflows with gauged data is a great approach to constrain your calibration.
In terms of the MB error, I know there were previously some potential bugs around the mass balance pertaining to reservoirs. I assume you are using the latest Raven version 2.9.2? If you are not I would try it with the latest version, otherwise (and if the issue persists in the latest version) I will pass this one to Dr. Craig as he is far more familiar with the latest MB calculations.
Cheers,
Rob
Robert Chlumsky
rchlumsk@uwaterloo.ca
rchlumsk@uwaterloo.ca
-
sgronsdahl
- Posts: 7
- Joined: Sun Sep 22, 2019 9:13 pm
Re: setting up and calibrating operational reservoirs with minimal data
Thanks for these tips, this will be really helpful.
Cheers
Stefan
Cheers
Stefan
-
konhee93
- Posts: 3
- Joined: Thu Sep 07, 2017 9:41 pm
Re: setting up and calibrating operational reservoirs with minimal data
Hello Stefan,
Just giving you some additional option.
One method that's worked really well for a site with something close to a rule curve is to use :ReservoirTargetStage
Basically you provide Raven with a time series of elevation to follow. Based on the inflow to the model, Raven will figure out the outflow required to meet the target elevation. You can then calibrate to the outflow data you have. See Section 5.3 of the Raven manual.
To use the :ReservoirTargetStage, you need two things
1. Stage-Volume relationship (no need for flow as flow is calculated for Raven). As Rob has advised earlier, I would take the area of the Lake based on GIS and assume the area does not change too much over the range of reservoir operation.
2. Continuous time series of target stage (which you can derive from your rule curve)
This method worked really well for me at various controlled reservoirs that do not follow the natural hydrograph.
As for your mass balance issue, in the raven_error.txt file, does it say your reservoir has dried out? This indicates that you are generating outflow (through overriding reservoir outflow) when there is not enough water in your reservoir. This could be because your stage-storage is too small, or your stage does not cover a wide enough range of your reservoir elevation.
-Konhee
Just giving you some additional option.
One method that's worked really well for a site with something close to a rule curve is to use :ReservoirTargetStage
Basically you provide Raven with a time series of elevation to follow. Based on the inflow to the model, Raven will figure out the outflow required to meet the target elevation. You can then calibrate to the outflow data you have. See Section 5.3 of the Raven manual.
To use the :ReservoirTargetStage, you need two things
1. Stage-Volume relationship (no need for flow as flow is calculated for Raven). As Rob has advised earlier, I would take the area of the Lake based on GIS and assume the area does not change too much over the range of reservoir operation.
2. Continuous time series of target stage (which you can derive from your rule curve)
This method worked really well for me at various controlled reservoirs that do not follow the natural hydrograph.
As for your mass balance issue, in the raven_error.txt file, does it say your reservoir has dried out? This indicates that you are generating outflow (through overriding reservoir outflow) when there is not enough water in your reservoir. This could be because your stage-storage is too small, or your stage does not cover a wide enough range of your reservoir elevation.
-Konhee