the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
eFLaG: enhanced future FLows and Groundwater. A national dataset of hydrological projections based on UKCP18
- 1UK Centre for Ecology & Hydrology, Maclean Building, Benson Lane, Crowmarsh Gifford, Wallingford, Oxon, OX10 8BB, UK
- 2Irish Climate Analysis and Research UnitS (ICARUS), Maynooth University, Ireland
- 3British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK
- 4School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, B15 2TT, UK
- 5British Geological Survey, Maclean Building, Benson Lane, Crowmarsh Gifford, Wallingford, Oxon, OX10 8BB, UK
- 6HR Wallingford, Howbery Park, Crowmarsh Gifford, OX10 8BA
- anow at: British Antarctic Survey, High Cross, Madingley Rd, Cambridge CB3 0ET
- bnow at: Natural Environment Research Council, Polaris House, Swindon, Wilts, SN2 1EU
- 1UK Centre for Ecology & Hydrology, Maclean Building, Benson Lane, Crowmarsh Gifford, Wallingford, Oxon, OX10 8BB, UK
- 2Irish Climate Analysis and Research UnitS (ICARUS), Maynooth University, Ireland
- 3British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK
- 4School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, B15 2TT, UK
- 5British Geological Survey, Maclean Building, Benson Lane, Crowmarsh Gifford, Wallingford, Oxon, OX10 8BB, UK
- 6HR Wallingford, Howbery Park, Crowmarsh Gifford, OX10 8BA
- anow at: British Antarctic Survey, High Cross, Madingley Rd, Cambridge CB3 0ET
- bnow at: Natural Environment Research Council, Polaris House, Swindon, Wilts, SN2 1EU
Abstract. This paper presents an ‘enhanced future FLows and Groundwater’ (eFLaG) dataset of nationally consistent hydrological projections for the UK, based on the latest UK Climate Projections (UKCP18). The hydrological projections are derived from a range of river flow models (Grid-to-Grid, PDM, GR4J and GR6J), to provide an indication of hydrological model uncertainty, as well as groundwater level (Aquimod) and groundwater recharge (ZOODRM) models. A 12-member ensemble of transient projections of present and future (up to 2080) daily river flows, groundwater levels and groundwater recharge were produced using bias corrected data from the UKCP18 Regional (12 km) climate ensemble. Projections are provided for 200 river catchments, 54 groundwater level boreholes and 558 groundwater bodies, all sampling across the diverse hydrological and geological conditions of the UK. An evaluation was carried out, to appraise the quality of hydrological model simulations against observations and also to appraise the reliability of hydrological models driven by the RCM ensemble, in terms of their capacity to reproduce hydrological regimes in the current period. The dataset was originally conceived as a prototype climate service for drought planning for the UK water sector, so has been developed with drought, low river flow and low groundwater level applications as the primary focus. The evaluation metrics show that river flows and groundwater levels are, for the majority of catchments and boreholes, well simulated across the flow and level regime, meaning that the eFLaG dataset could be applied to a wider range of water resources research and management contexts, pending a full evaluation for the designated purpose.
- Preprint
(3095 KB) -
Supplement
(5652 KB) - BibTeX
- EndNote
Jamie Hannaford et al.
Status: final response (author comments only)
-
RC1: 'Comment on essd-2022-40', Anonymous Referee #1, 08 Jul 2022
The title of the paper is misleading, it had me expecting to read about projected changes in flows and groundwater in the UK, when in fact the paper presents an evaluation of hydrological models prior to being forced with future runs. Indeed the opening sentence of the abstract doubles down on this - the paper presents a dataset of natioanlly consistenty hydrological projections for the UK based on the latest UK Climate Projections. The paper doesnt do this and thus needs to be reframed as a paper evaluating hydrological models and RCM ensemble members in capturing observed regime - PRIOR to providing climate change projections for the sector.
Apart from this major oversight, the paper is well written and presented. There are a number of questions I was left with after reading that should be addressed in a revision. First, why the differential sampling of models for flows and groundwater? 3 different hydro models but only one GL model and one recharge model. In addition, why was uncertainty in model parameters not investigated.
Second, was any testing done on the transferability of models to future conditions? In particular the lumped conceptual models (GR and PDM) are held to perform better for modified catchments because the calibration picks this up. However, for future runs for later in the century are these modifcations likely to be stationary and therefore how representative are these models for these catchments likely to be? On a related note, it would be useful to descibe the range of distribubances affecting catchments - is it predominatly abstractions?
IF the purpose of the dataset is to inform adaptation planning why was RCP8.5 selected and why not more scenarios. There is debate in scientific literature presently about the realism of RCP8.5 - its selection here needs to be justified further. I understand the need for crystalising the uncertainty as stated, but this needs to be done based on the objectives or intended uses of the dataset.
Am I correct to conclude that there is no evaluation of model performances for an independent verification period, ie. that all metrics presented are for calibration period? How then can we be sure that calibrated models perform outside of the period used to train them, even during the baseline period, never mind under changed future climates. Please justify NOT using a verficiation period and what this may mean for your results.
Minor comments
Some aspects of the uncertainties explored seem to be missing from Table 1 - eg. The first sentence of section 3 mentions 2 regional climate models, groundwater models are also not included.
Line 196 can you say how this was done for clarity
Line 216 - it would be useful to provide a brief overview of this downscaling method here. It is an important part of the methods and this is the go to paper descibing the dataset.
Line 236 - what does copied down mean here - lacks clarity.
Figure 2 - there are 12 PPE members from text but 15 columns. Not sure I follow, is there an ensemble mean presented?
I like how there is tracability between FFWGL and this dataset in terms of models, catchments etc. This is sensible and useful. I also like how research and industry needs fed into catchment selection. I also like the description of stage 2 assessment and its expectations. Dont often see this as nicely explained.
Line 604 - this concerns me and needs teasing out a bit more, especially given the objective of informing adaptation in the sector. What does to a degree mean? I can understand the need to assess future changes relative to the 'real' present (ie disturbed) for planning but what are the implications and cautions that should be borne in mind?
Figure 4 caption needs to state #catchments and refer to table detailing the skill scores.
Line 644 missing word.
Line 680 full stop missing.
Line 840 - should really discuss the transferability of models and parameter sets from current to future conditions as a key source of uncertainty here, along with the stationarity of disturbances in relative catchments.
-
AC1: 'Reply on RC1', Jamie Hannaford, 29 Sep 2022
The title of the paper is misleading, it had me expecting to read about projected changes in flows and groundwater in the UK, when in fact the paper presents an evaluation of hydrological models prior to being forced with future runs. Indeed the opening sentence of the abstract doubles down on this - the paper presents a dataset of natioanlly consistenty hydrological projections for the UK based on the latest UK Climate Projections. The paper doesnt do this and thus needs to be reframed as a paper evaluating hydrological models and RCM ensemble members in capturing observed regime - PRIOR to providing climate change projections for the sector.
Apart from this major oversight, the paper is well written and presented. There are a number of questions I was left with after reading that should be addressed in a revision. First, why the differential sampling of models for flows and groundwater? 3 different hydro models but only one GL model and one recharge model. In addition, why was uncertainty in model parameters not investigated.
>>>Thank you for the positive assessment of the paper’s communication. The different choice of models reflect availability and suitability of models for both flow and groundwater. Parametric uncertainty was beyond scope, give the high number of existing model runs, and is highlighted as a priority in the future look (sect. 8)
Second, was any testing done on the transferability of models to future conditions? In particular the lumped conceptual models (GR and PDM) are held to perform better for modified catchments because the calibration picks this up. However, for future runs for later in the century are these modifcations likely to be stationary and therefore how representative are these models for these catchments likely to be? On a related note, it would be useful to descibe the range of distribubances affecting catchments - is it predominatly abstractions?
>>> This is always the perennial problem with future runs using calibrated models. However, the purpose of eFLaG is not really to look at such questions of non-stationarity in catchments, abstractions etc, which is a major topic in itself. Rather eFLaG is about running the models under current conditions and exploring future climate impacts. We suggest adding some caveats to the discussion to emphasise this point. Seperately, a follow-up study is already looking at future changes in abstractions/discharges as well as natural flows. We will highlight this avenue for future work in section 8. The range of disturbances are broad and users can explore these through the NRFA ‘Factors Affecting Runoff’ codes, and we will highlight this with a reference. Abstractions are prominent, but in some catchments discharges entirely balance or outweigh abstractions.
IF the purpose of the dataset is to inform adaptation planning why was RCP8.5 selected and why not more scenarios. There is debate in scientific literature presently about the realism of RCP8.5 - its selection here needs to be justified further. I understand the need for crystalising the uncertainty as stated, but this needs to be done based on the objectives or intended uses of the dataset.
>>> This is a fact of the climate projections – the regional projections are chosen as they are spatially coherent and transient but there is only one scenario. This is a limitation of that product. We do highlight this in the caveats, but will strengthen it and highlight the need to look at other projections. We will also make the fact clear in the earlier ‘UKCP data’ section.
Am I correct to conclude that there is no evaluation of model performances for an independent verification period, ie. that all metrics presented are for calibration period? How then can we be sure that calibrated models perform outside of the period used to train them, even during the baseline period, never mind under changed future climates. Please justify NOT using a verficiation period and what this may mean for your results.
>>>We could in theory have split the observed flows into a calibration and unseen validation period (used for calculating the statistics only) and then later recalibrated on the entire period to get the best calibration for use with RCMs. But, given we typically have many years in our calibration period, this would not make much difference. The paper by Harrigan et al (2018) using GR4J (see table 2 especially) shows this. hess-22-2023-2018.pdf
Minor comments
Some aspects of the uncertainties explored seem to be missing from Table 1 - eg. The first sentence of section 3 mentions 2 regional climate models, groundwater models are also not included.
>> this is just the wording, what is in the brackets is the GCM and RCM separately, we will reword to avoid ambiguity. We will add the groundwater to the table.
Line 196 can you say how this was done for clarity
>>not sure what is not clear, here we are saying what we are not doing (infilling to get to 365 like the Prudhomme 2012 study).
Line 216 - it would be useful to provide a brief overview of this downscaling method here. It is an important part of the methods and this is the go to paper descibing the dataset.
>>>we omitted this for brevity since it is a standard method used in many previous studies and well described in these references. In response to this comment we will add the following: ‘The bias-corrected precipitation products were then downscaled from 12km to 1km based on the distribution of the observed Standard-period Average Annual Rainfall (SAAR) over the period 1961-1990, as in previous studies (Bell et al., 2007; Kay & Crooks, 2014). This involved calculating the ratio of the observed SAAR at 1km to the observed SAAR averaged over 12km, and then multiplying RCM values by this ratio. This ensured that the spatial variability of rainfall was captured, but the total rainfall across the original 12km RCM grid cell remained unchanged.’
Line 236 - what does copied down mean here - lacks clarity.
>>Agreed, we will describe this more clearly, adding ‘by simply setting all 1km grid cells to the value of the containing 12km grid cell.’
Figure 2 - there are 12 PPE members from text but 15 columns. Not sure I follow, is there an ensemble mean presented?
>>there are 12 columns, but the labelling is not 01 – 12 but 01 – 15, with some missing (e.g. 14). This is just the nomenclature of the RCM runs. We will make this clear In the caption.
I like how there is tracability between FFWGL and this dataset in terms of models, catchments etc. This is sensible and useful. I also like how research and industry needs fed into catchment selection. I also like the description of stage 2 assessment and its expectations. Dont often see this as nicely explained.
>>Thank you very much!
Line 604 - this concerns me and needs teasing out a bit more, especially given the objective of informing adaptation in the sector. What does to a degree mean? I can understand the need to assess future changes relative to the 'real' present (ie disturbed) for planning but what are the implications and cautions that should be borne in mind?
>>>We agree that ‘to a degree’ is ambiguous and warrants fuller explanation. As highlighted above, that is exactly our objective, climate changes relative to the real (disturbed) present – as with the original, widely used FFGWL product, which also did not take account of human influences.
We will add a paragraph earlier in the methods, and then another sentence or two to this discussion, to highlight that this is an obvious simplification – entirely justifiable for most planning purposes, but one which needs to be considered in other applications where users may be interested in future changes in such influences. As noted, having G2G is an advantage as it does allow one naturalised run for comparisons.
Figure 4 caption needs to state #catchments and refer to table detailing the skill scores.
>>>agreed, will change
Line 644 missing word.
>>>agreed, will change
Line 680 full stop missing.
>>>agreed, will change
Line 840 - should really discuss the transferability of models and parameter sets from current to future conditions as a key source of uncertainty here, along with the stationarity of disturbances in relative catchments.
>>>agreed, as above, we will strengthen coverage of this point throughout the paper.
-
AC4: 'Reply on RC1', Jamie Hannaford, 29 Sep 2022
This was missed from our initial reply:
The title of the paper is misleading, it had me expecting to read about projected changes in flows and groundwater in the UK, when in fact the paper presents an evaluation of hydrological models prior to being forced with future runs. Indeed the opening sentence of the abstract doubles down on this - the paper presents a dataset of natioanlly consistenty hydrological projections for the UK based on the latest UK Climate Projections. The paper doesnt do this and thus needs to be reframed as a paper evaluating hydrological models and RCM ensemble members in capturing observed regime - PRIOR to providing climate change projections for the sector.
>>>we feel this is not entirely correct as we do provide a nationally consistent dataset of river flow projections, this is exactly what eFLaG is, and as is made clear from the open dataset. The paper is a data paper and so of course much of the material is about the development of the data and this includes the evaluation (forced with past data as well as with the RCM data used for future runs). We do not feature projections themselves as that is the topic of a separate scientific paper (though note later replies to reviewer 2 that we will feature some lmited exemplare results from the eFLaG portal, a viewer that has been released subsequent to this paper). We suggest modifying the title e.g. ‘ The eFLaG dataset: developing nationally consistent projections of future flows and groundwater based on UKCP18’ but we must still emphasise that this is about a dataset of hydrological projections
-
AC1: 'Reply on RC1', Jamie Hannaford, 29 Sep 2022
-
RC2: 'Comment on essd-2022-40', Anonymous Referee #2, 02 Aug 2022
Review of “eFLaG: enhanced future FLows and Groundwater. A national dataset of hydrological projections based on UKCP18” by Hannaford et al.
This manuscript deals with a dataset of multimodel hydrological projections across the UK. It presents in a detailed way how this dataset has been produced, but it lacks a description of the dataset itself. I would therefore recommend a major revision for the authors to bring in an overview of what’s actually in this (otherwise important) dataset. I would therefore also appreciate more detailed comments on how to use it, notably for interpreting hydrological projections where models have been calibrated against anthropogenically disturbed streamflow observations.
Main comments
- Calibration period: Despite (too) long descriptions of hydrological model set-ups, there are important missing information on the calibration of conceptual models. First, I haven’t seen the calibration period explicitly mentioned in the text. I assume that is the complete historical 1961-2018 period covered by simobs simulations, but this is to be made explicit.
- Parameter transferability: the previous comment calls for another question on the transferability of parameters of conceptual models, which has been shown for some time as an important issue when dealing with climate change (see e.g. Thirel et al., 2015). This issue is unfortunately completely absent from the manuscript, even in the discussion part. This issue is often dealt with by using more or less advanced split-sample set-ups, but many alternative propositions have been made over the recent years (e.g. TodoroviÄ et al., 2022). This issue should therefore be dealt with in the manuscript, at the very least as a comment in the discussion part.
- Calibration on catchments with anthropogenic disturbances: This issue is seemingly considered as a rather light one in the manuscript (L604-615). I rather disagree here, as the main underlying hypothesis is not even mentioned here: models calibrated on influenced data will deliver hydrological projections of influenced streamflow in which the amount and seasonality of anthropogenic disturbances (abstractions, reservoir management and so on) is equal to those during the calibration period. Which is clearly an unrealistic feature of the future. This makes the understanding and the use of such projections quite difficult for water managers and stakeholders. I hope that communication around the EFLaG dataset can handle this issue, but it is definitely not convincing in the manuscript.
- Description of dataset: as already pointed out by Anonymous Referee #1, this manuscript describes how the eFLaG dataset has been built (which is definitely commendable), but it does not describe the dataset itself. Indeed, there is no e.g. (1) overall summary statistics (temporal or spatial) of present-day-period simulated streamflows or groundwater levels, (2) overall summary statistics of projected evolution or changes over the UK, and (3) no case study example of the numerous time series produced. All these features are essential to get a grasp of what’s in the dataset and are therefore in my view a required feature of a data paper.
- Length of model descriptions: I guess that the manuscript is currently quite lengthy because of the extent to which hydrological model set-ups are described, at the expense of the more general and important issues listed above. A new balance should be reached in the revised version of the manuscript.
Specific comments
- Table 1: I disagree with the partitioning of uncertainties here: “model structure” or “model choice” (or here “hydrological models” like “climate models”) are equivalent. GR4J and GR6J are indeed different models, probably a bit similar to each other than to PDM for example, but “model structure uncertainty” is commonly used as opposed to “model parameter uncertainty” in common uncertainty decomposition of hydrological projections (see e.g. Christierson et al., 2012 for the UK). This relates to one of my main comment on parameter temporal transferability.
- L178-189: Please recall a reference for UKCP18
- L190-196: This issue with Hadley Centre models has been around for the last 15 years at least… But here it means that simrcm streamflow series also have 360 days per year? This is what I can see from the data files, but this is not discussed or even mentioned in the manuscript. I wonder what a water manager would say when looking at those files… It would therefore be necessary to rise the issue in the manuscript and also provide some advice for water managers and stakeholders on how to use such unusual time series, in order to prevent any misuse of even rejection based on lack of credibility.
- L215-217: This spatial disaggregation step is not clear enough. Plus, what is the standard-period in SAAR? And why is HadUK-Grid not used here? All these choices are not enough commented and justified.
- L228-236: Could you give here a simple description of the PET formula (e.g. “Penman-Monteith-like”)? Indeed, the choice of PET formula may have strong consequences especially for low-flow changes (see e.g. Lemaitre-Basset et al., 2022). This choice would also deserve a comment in the manuscript.
- Figure 2: The bias-correction factors are quite high for some month/model. This should also deserve a comment, especially with the somewhat overlapping issues of model weighting versus internal variability.
- L291-294: In relation to one of my main comments, I could not find in the eFLaG_Station_Metadata.xlsx file any flag indicating a near-natural catchment (e.g. belonging to UKBN2) or borehole that would help identifying locations where streamflow/groundwater projections are natural streamflow/groundawater projections. This lack of flag (I noted the FARL field, but this is far from being the only relevant source of disturbances) makes me uncomfortable with this definitely rather non-homogeneous dataset.
- Figure 3: The text is very small and makes maps difficult to read.
- L395: The CHESS version cited here has been superseded.
- Table 3: It is necessary to have this table in the main text?
- L555-559: How is SGI used for evaluation? This is unclear until Figure 5 a few pages later when we learn about the NSE_SGI.
- L634: already written above.
- Figure 5, caption: “NSE_SGI”
- Figure 7: This figure is definitely too small for it to be correctly read and interpreted (see e.g. L698-700). What about using instead a metrics (or a very few metrics) based on the correspondence between FDCs? This would allow showing all locations in the main text.
- Figure 9: This is a poor choice of color scale which makes contrasts much too difficult to read. I would highly recommend using one of the scales available at https://colorbrewer2.org and recommended by the IPCC (2018), and reserve high values for darker colors, for them to be emphasized. Plus, the legend is repeated.
- Figure 10: Same comments as above. Impossible to distinguish between 2-5, 5-10, and 10-20 classes.
Technical corrections
- L105: “EdgE”
- L126: “a,b,c”
- Table 1: Please define “PPE” (Perturbed Physics Ensemble)
- L274: Please define “EIDC”
- L644: missing “models”
- L694: “re”?
References
Christierson, B. v., Vidal, J.-P. and Wade, S. D.: Using UKCP09 probabilistic climate information for UK water resource planning. J. Hydrol., 424-425, 48-67, https://doi.org/10.1016/j.jhydrol.2011.12.020, 2012.
IPCC WGI Technical Support Unit: IPCC Visual Style Guide for Authors. https://www.ipcc.ch/site/assets/uploads/2019/04/IPCC-visual-style-guide.pdf, 2018
Lemaitre-Basset, T., Oudin, L., Thirel, G., and Collet, L.: Unraveling the contribution of potential evaporation formulation to uncertainty under climate change, Hydrol. Earth Syst. Sci., 26, 2147–2159, https://doi.org/10.5194/hess-26-2147-2022, 2022.
Thirel, G., Andréassian, V., Perrin, C., Audouy, J.-N., Berthet, L., Edwards, P., Folton, N., Furusho, C., Kuentz, A., Lerat, J., Lindström, G., Martin, E., Mathevet, T., Merz, R., Parajka, J., Ruelland, D. & Vaze, J. : Hydrology under change: an evaluation protocol to investigate how hydrological models deal with changing catchments. Hydrolog. Sci. J., 60, 1184-1199, https://doi.org/10.1080/02626667.2014.967248, 2015.
TodoroviÄ, A., Grabs, T. and Teutschbein, C.: Advancing traditional strategies for testing hydrological model fitness in a changing climate. Hydrolog. Sci. J., https:// 10.1080/02626667.2022.2104646, 2022.
-
AC2: 'Reply on RC2', Jamie Hannaford, 29 Sep 2022
We thank the reviewer for these helpeuland constructive comments wich will guide the revised manuscript. We reply to these comments in turn below.
This manuscript deals with a dataset of multimodel hydrological projections across the UK. It presents in a detailed way how this dataset has been produced, but it lacks a description of the dataset itself. I would therefore recommend a major revision for the authors to bring in an overview of what’s actually in this (otherwise important) dataset. I would therefore also appreciate more detailed comments on how to use it, notably for interpreting hydrological projections where models have been calibrated against anthropogenically disturbed streamflow observations.
>>> We describe the dataset in section 9. For how to use the data, this is the subject of a series of separate demonstrators we describe briefly, but which are being written up elsewhere. For this data paper we cannot highlight all use cases, and we have already highlighted some of the uses of the original FFGWL data. We agree that we could add a few short paras highlighting uses of the data and appropriate caveats. We will clarify the issue with anthropogenic disturbances, as highlighted by reviewer 1 too.
Main comments
- Calibration period: Despite (too) long descriptions of hydrological model set-ups, there are important missing information on the calibration of conceptual models. First, I haven’t seen the calibration period explicitly mentioned in the text. I assume that is the complete historical 1961-2018 period covered by simobs simulations, but this is to be made explicit.
>> also noted in response to R1, we will clarify.
- Parameter transferability: the previous comment calls for another question on the transferability of parameters of conceptual models, which has been shown for some time as an important issue when dealing with climate change (see e.g. Thirel et al., 2015). This issue is unfortunately completely absent from the manuscript, even in the discussion part. This issue is often dealt with by using more or less advanced split-sample set-ups, but many alternative propositions have been made over the recent years (e.g. TodoroviÄ et al., 2022). This issue should therefore be dealt with in the manuscript, at the very least as a comment in the discussion part.
>>> we will clarify in the method and also in the discussion, and will add the references highlighted as good examples, thank you.
- Calibration on catchments with anthropogenic disturbances: This issue is seemingly considered as a rather light one in the manuscript (L604-615). I rather disagree here, as the main underlying hypothesis is not even mentioned here: models calibrated on influenced data will deliver hydrological projections of influenced streamflow in which the amount and seasonality of anthropogenic disturbances (abstractions, reservoir management and so on) is equal to those during the calibration period. Which is clearly an unrealistic feature of the future. This makes the understanding and the use of such projections quite difficult for water managers and stakeholders. I hope that communication around the EFLaG dataset can handle this issue, but it is definitely not convincing in the manuscript.
>>> we will clarify in the method and also in the discussion. Of course, we recognise that disturbances will change in future. But it is not really our intention to accurately model catchment flows to the 2080s – rather, to look at potential climate futures for each catchment given current conditions, for planning purposes, and as widely adopted in many other appplications (including the original FFGWL, the model for eFLaG). We will modify this accordingly in the intro and the discussion, as also noted in our replies to R1.
- Description of dataset: as already pointed out by Anonymous Referee #1, this manuscript describes how the eFLaG dataset has been built (which is definitely commendable), but it does not describe the dataset itself. Indeed, there is no e.g. (1) overall summary statistics (temporal or spatial) of present-day-period simulated streamflows or groundwater levels, (2) overall summary statistics of projected evolution or changes over the UK, and (3) no case study example of the numerous time series produced. All these features are essential to get a grasp of what’s in the dataset and are therefore in my view a required feature of a data paper.
- Length of model descriptions: I guess that the manuscript is currently quite lengthy because of the extent to which hydrological model set-ups are described, at the expense of the more general and important issues listed above. A new balance should be reached in the revised version of the manuscript.
>>>We feel it is more important for users of the data paper to understand the provenance (i.e. methodology) rather than the outcomes, which are the subject of multiple papers and reports.
Specific comments
- Table 1: I disagree with the partitioning of uncertainties here: “model structure” or “model choice” (or here “hydrological models” like “climate models”) are equivalent. GR4J and GR6J are indeed different models, probably a bit similar to each other than to PDM for example, but “model structure uncertainty” is commonly used as opposed to “model parameter uncertainty” in common uncertainty decomposition of hydrological projections (see e.g. Christierson et al., 2012 for the UK). This relates to one of my main comment on parameter temporal transferability.
>We are following the logic and nomenclature of the Smith et al. 2018 paper here. We see GR46J and GR6J as two structures. We will add a line on parametric uncertainty but note we do not sample it (see replies to R1) – this was on an earlier version but omitted.
- L178-189: Please recall a reference for UKCP18
>>will do, we refer to Murphy et al. at line 55 but will reiterate here
- L190-196: This issue with Hadley Centre models has been around for the last 15 years at least… But here it means that simrcm streamflow series also have 360 days per year? This is what I can see from the data files, but this is not discussed or even mentioned in the manuscript. I wonder what a water manager would say when looking at those files… It would therefore be necessary to rise the issue in the manuscript and also provide some advice for water managers and stakeholders on how to use such unusual time series, in order to prevent any misuse of even rejection based on lack of credibility.
>>>We will highlight this in the text, and also in the discussion section (as noted with R1 replies, we will add a short ‘applications’ section. The issue of 360 day years is also highlighted in the demonstrators, with appropriate recommendations.
- L215-217: This spatial disaggregation step is not clear enough. Plus, what is the standard-period in SAAR? And why is HadUK-Grid not used here? All these choices are not enough commented and justified.
>> This is a very standard approach but we will add extra text to clarify this, as detailed in replies to R1. SAAR is used as a standard, not subject to change as HadUK is.
- L228-236: Could you give here a simple description of the PET formula (e.g. “Penman-Monteith-like”)? Indeed, the choice of PET formula may have strong consequences especially for low-flow changes (see e.g. Lemaitre-Basset et al., 2022). This choice would also deserve a comment in the manuscript.
>> This follows the CHESS Method as highlighted with a reference so we did not expand, but we will add a short additional sentence to make clear. We agree PE formulation can be important and we highlight this already in Section 8, L849. Thanks for the additional reference.
- Figure 2: The bias-correction factors are quite high for some month/model. This should also deserve a comment, especially with the somewhat overlapping issues of model weighting versus internal variability.
>>> we will add a comment about the high bias correction factors
L291-294: In relation to one of my main comments, I could not find in the eFLaG_Station_Metadata.xlsx file any flag indicating a near-natural catchment (e.g. belonging to UKBN2) or borehole that would help identifying locations where streamflow/groundwater projections are natural streamflow/groundawater projections. This lack of flag (I noted the FARL field, but this is far from being the only relevant source of disturbances) makes me uncomfortable with this definitely rather non-homogeneous dataset.
>>>We do highlight UKBN2 membership in the spreadsheet (Column I). We do not refer to other sources such as NRFA descriptions or FAR (Factors affecting riunofff) codes, but we will add a sentence to note these sources are available.
- Figure 3: The text is very small and makes maps difficult to read.
>> we would hope this would be made a large map in the paper. We cannot really make the text much bigger without making it too cluttered.
- L395: The CHESS version cited here has been superseded.
>>we will highlight this
- Table 3: It is necessary to have this table in the main text?
>>> yes, it is necessary to allow the reader to look at the metrics close to where they are cited. This would be our preference.
- L555-559: How is SGI used for evaluation? This is unclear until Figure 5 a few pages later when we learn about the NSE_SGI.
>>>SGI is used as the basis for comparison of the observed and simulated data, with the various metrics used to establish performance. We will make this clearer.
- L634: already written above.
>>not sure what is meant here, there is a general point made and then a specific one wrt Rudd et al. 2017.
- Figure 5, caption: “NSE_SGI”
>>agreed we need to make the caption and text consistent
- Figure 7: This figure is definitely too small for it to be correctly read and interpreted (see e.g. L698-700). What about using instead a metrics (or a very few metrics) based on the correspondence between FDCs? This would allow showing all locations in the main text.
>>we prefer to show for individual catchments. We imagine this would be a large image in the paper (see also appendices).We do in fact show metrics based on correspondence of specific FDC quantiles, e.g. Figure 9 for Q90 and then the various equivalents in the Supplementary info.
- Figure 9: This is a poor choice of color scale which makes contrasts much too difficult to read. I would highly recommend using one of the scales available at https://colorbrewer2.org and recommended by the IPCC (2018), and reserve high values for darker colors, for them to be emphasized. Plus, the legend is repeated.
>>>We agree this could be revisted, thanks for the suggestion. We can use a palette from colorbrewer2, but to be honest, it'll still be difficult to distinguish between adjacent categories -- that's the nature of these palettes, and we feel that it does not effect the interpretation. Re: Legends, there are two legends that are almost identical but one is for Q90 (for the 4 HMs) and the other is for L90 (for g/w)
- Figure 10: Same comments as above. Impossible to distinguish between 2-5, 5-10, and 10-20 classes.
Technical corrections
- L105: “EdgE”
- L126: “a,b,c”
- Table 1: Please define “PPE” (Perturbed Physics Ensemble)
- L274: Please define “EIDC”
- L644: missing “models”
- L694: “re”?
>>> thanks, we will make these changes
-
AC3: 'Additional comment', Jamie Hannaford, 29 Sep 2022
IN addition to the previous reply, the top reply was truncated for some reasons:
The title of the paper is misleading, it had me expecting to read about projected changes in flows and groundwater in the UK, when in fact the paper presents an evaluation of hydrological models prior to being forced with future runs. Indeed the opening sentence of the abstract doubles down on this - the paper presents a dataset of natioanlly consistenty hydrological projections for the UK based on the latest UK Climate Projections. The paper doesnt do this and thus needs to be reframed as a paper evaluating hydrological models and RCM ensemble members in capturing observed regime - PRIOR to providing climate change projections for the sector.
>>>we feel this is not entirely correct as we do provide a nationally consistent dataset of river flow projections, this is exactly what eFLaG is, and as is made clear from the open dataset. The paper is a data paper and so of course much of the material is about the development of the data and this includes the evaluation (forced with past data as well as with the RCM data used for future runs). We do not feature projections themselves as that is the topic of a separate scientific paper (though note later replies to reviewer 2 that we will feature some lmited exemplare results from the eFLaG portal, a viewer that has been released subsequent to this paper). We suggest modifying the title e.g. ‘ The eFLaG dataset: developing nationally consistent projections of future flows and groundwater based on UKCP18’ but we must still emphasise that this is about a dataset of hydrological projections
Jamie Hannaford et al.
Data sets
Hydrological projections for the UK, based on UK Climate Projections 2018 (UKCP18) data, from the Enhanced Future Flows and groundwater (eFLaG) project Hannaford, J.; Mackay, J.; Ascot, M.; Bell, V.; Chitson, T.; Cole, S.; Counsell, C.; Durant, M.; Facer-Childs, K.; Jackson, C.; Kay, A.; Lane, R.; Mansour, M.; Moore, M.; Parry, S.; Rudd, A.; Simpson, M.; Turner, S.; Wallbank, J.; Wells, S.; Wilcox https://catalogue.ceh.ac.uk/documents/1bb90673-ad37-4679-90b9-0126109639a9
Jamie Hannaford et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,174 | 469 | 34 | 1,677 | 72 | 19 | 20 |
- HTML: 1,174
- PDF: 469
- XML: 34
- Total: 1,677
- Supplement: 72
- BibTeX: 19
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1