January 16, 2017 at 1:22 pm #458
This thread reports the discussion on the development of the protocol of the River Memory experiment. Particpants to the experiment were given a draft report of the first results.
During the first step of the discussion, participants are expected to comment of the structure of the experiment (the protocol). The protocol of the experiment is available at http://dl-ng005.xtr.deltares.nl/view/113/. After agreeing the final form of the protocol, the text at the above link will be updated with the final description.
After agreeing the protocol, the experiment leader will prepare a draft paper to be discussed here.
Please take into account that this discussion is public. Confidential remarks should be addressed to the leader by email.
Thanks for your cooperation.
January 26, 2017 at 2:16 pm #496
- This topic was modified 2 years ago by Alberto Montanari.
The analysis of the relationship between correlation and the variables analyzed (basin size, BI and AI) is very interesting and gathers the combined effect of geophysical and climatic drivers in the catchment. I understand that the availability of data in such amount of rivers and catchments plays a critical role in the selection of the variables. Nevertheless, the inclusion of a climatic index in the analysis such as the aridity index of Martonne, or the rainfall index of Lang that only require mean annual rainfall and temperature in the catchment might help with the physical interpretation of the results. Also, in those catchments where topographic data are available it would be interesting to see if there is any relationship with some orographic index that depends on the relief of the catchment.
Cristina AguilarJanuary 26, 2017 at 4:46 pm #497
The analysis indeed begins to be very interesting! I have a couple of remarks or comments:
– regarding the boxplots used to analyze correlations relative to physical and quantified metrics (such as basin area in Fig. 6), do we really need boxplots? I’m wondering if the boxplots are not aggregating too much information and so hiding some results, maybe scatter plots would provide more information (or did you try and see that they were unusable?). And if using boxplots, the identification of trends or patterns can become subjective: have you thought about using objective criteria for this (significance tests for example).
– effect of elevation: this is one of the simplest geophysical characteristics to get, so I think we should have it for every basin. Also, representing the correlations on a map can be misleading: you plot correlations at the outlets, but sometimes it is not clear if the points are on a tributary or on the main river, and for me the mean elevation is more interesting than the outlet altitude. (but I recognize that we already see some nice information on Fig. 13!).
– We can provide for France mean values of P and T if needed.
The basis of the article is already good, thanks!
Guillaume.January 26, 2017 at 8:54 pm #498
Sorry that I missed some comments in my previous post. I have two suggestions on potential additional analyses. Both of them are to be considered in the future as they are time-consuming and thus would delay Any’s paper.
1. Regarding the season identification, I think that we might overcome some of the limitations of the method of Lee et al. (2015) if we combine it with the seasonality space representation of the directional statistics. For example, we could generate and analyze the temporal distribution of peaks within the HFS in order to quantify the consequences of the assumption of symmetrical timing for the high flow season. Similarly, the uniform selection of length (3-months) can be related to the standard deviation. In this way, the method of Lee et al. would remain as the core method to assess the season identification, but we could also take advantage of some of the indicators of the directional statistics at least in those rivers where the assumptions of Lee’s method might be problematic.
2.Regarding this statement in the protocol: “It is interesting to highlight that the state of a catchment, and in particular its storage, is affected by previous precipitation… Therefore, we utilize here previous flows as a proxy for catchment storage instead of rainfall. While the above assumption may be reasonable, one should consider that it may not hold when the river flows are impacted by massive regulation”.
We have daily flow series in two rivers in Andalucía (Spain): The Guadalquivir river (1932-2013, 47000 km2) and a smaller one, the Guadalete river (1960-2013, 3677km2), where we also have the daily precipitation at the catchment generated by a model that considers elevation gradients. Both catchments are quite regulated but at least for one of them we have the restituted series (let’s say the natural flows). So I could apply the methodology considering both, flow and precipitation daily series to see the effect of regulation and also to analyze the “improvement” if we use precipitation as the proxy for catchment storage instead of flow. I need first to do some data standardization and preparation previous to any analysis but my guess is that if I spend some time with these rivers that we have already studied in different projects, where we have a lot of geophysical and meteorological data and know really well, we can use them as a kind of “laboratory” to test some of the assumptions that we make at the different steps of the methodology.
We keep in touch. Cheers!
CristinaJanuary 30, 2017 at 11:14 am #501
Thank you very much for the draft report, which is indeed very interesting.
A couple of ideas for potential additional analysis:
1) Regarding the inclusion of climatic indices, a first easy step could be the use of a climate classification map, as it doesn’t require gathering any rainfall or temperature data. For instance, the updated map of the Köppen-Geiger climate classification of Peel et al. (2007). The map in raster format can be downloaded from the supplementary materials: Peel, M.C., Finlayson, B.L., McMahon, T.A. Updated world map of the Köppen-Geiger climate classification (2007). Hydrology and Earth System Sciences, 11 (5), pp. 1633-1644.
2) We could explore the topography differences between all catchments, and beyond elevation. An open source Digital Elevation Model with global coverage such as SRTM could be used to delineate the catchment boundaries and calculate topographic indices. This would involve some extra work and time, so maybe it could be considered in the future. Alternatively, we could calculate topographic indices only in the Austrian catchments with the DEM that has already been used.
3) Regarding future work, it could also be interesting to have a look at the paper of Serinaldi and Kilsby (2016). They analyze daily stream flow fluctuations exceeding high thresholds, and explore the relationship between the occurrence and magnitude of extreme flow values and the properties of the complete time series. Serinaldi, F., Kilsby, C.G. Understanding persistence to avoid underestimation of collective flood risk (2016) Water (Switzerland), 8 (4), art. no. 152.
Looking forward to seeing how the work develops!
MariaJanuary 30, 2017 at 1:07 pm #502
The analysis and presentation of results is interesting. Regarding the protocol I have next comments:
I believe that we should try to gather additional data to enhance the physical interpretation of the results (if we have time to focus on this?). At the moment, effect of basin size, baseflow index and aridity index is available for all selected catchments but other potential controls are limited to some of the catchments (e.g., glaciers, soil, lakes). Especially effect of elevation could be analysed also for other catchments (and not just for Austrian catchments). If the digital elevation model is not available for some specific areas we could use SRTM digital elevation model (as already proposed). Further, we might even focus on some other topographic indices.
I also agree with previous posts and I could provide rainfall (average annual rainfall amounts or maximum 12-h or 24-h rainfall amounts for different return periods) and air temperature (monthly or annual) data for the Slovenian catchments.
I agree with Thirel that scatter plots would maybe be better option than boxplots (e.g., Fig. 6) and using colours or different point types we could add additional information to these figures (e.g., in case of Fig. 6 colours could be used to separate catchments based on geographic location-country).
Thank you for great work.
NejcJanuary 31, 2017 at 8:08 am #503
this is a very interesting analysis and a very clearly written report. thank you.
I have a few suggestions which can be considered to complement the analysis and interpretations:
1) looking on the location of the basins, they cover very different climate and physiographic conditions in Europe, so perhaps some more interpretation of results along a north-south transect will be interesting. this may include a link of results with above mentioned climate (precipitation, climate class, etc) characteristics. Do we have basin boundaries for all analysed basins? (for Austrian we can provide them). if yes, we can use e-obs dataset to extract some climate related characteristics for the basins.
2) I see some links with our flood change experiment. maybe be some of the patterns can be explained by trends in peak-over threshold flood time series, which are analysed in our experiment.
3) for the interpretation of flood correlations (at least in Austria), some assessment of soil moisture (and its changes) will be interesting. maybe there are some open soil moisture datasets for Europe, which might be worth to look at (this is just a suggestion)
thank yoy again.
I look forward to a draft of the paper…
dpJanuary 31, 2017 at 1:46 pm #506
Thank you for this interesting analysis and for the draft report which is clear and well-structured. I agree with previous posts for the possible improvements in the investigation of the physical predictors, especially for the possibility of including climate indices, and eventually seasonal precipitation statistics. I would like to stress a few points and propose possible complementary elements:
1 – I agree with Guillaume Thirel that it could be interesting to see also the scatterplots of the seasonal correlations against the physical predictors. The investigation by scatterplots could allow visualizing possible “threshold” effects that could be hidden in the box-plots (see for instance Figures 6, 7 and 8). In particular, I think that it would be interesting to further investigate the effect of the basin size on the river memory. The box-plots in Figure 6 correspond to groups of basins with a very wide range of catchment areas, especially for the group of largest catchments (from 1557 to 70091 km2). An additional graphical analysis of the effect of basin size for the largest catchments could validate the expectation of finding an enhanced seasonal river memory for larger scale systems, possibly emerging above a threshold size.
It is good to use quantiles for separating the catchment set in groups having the same sample size; however, this approach leads to a very wide range of characteristics within each group, especially for the classes including extreme values. Maybe the use of a larger number of classes for separating the catchment set could be useful? I agree also on the interest of using a statistical non-parametric test for the assessment of significance of differences between classes.
2 – For a potential additional analysis for improving the physical interpretation: an estimation of the characteristic response time of the catchment could be considered, as the time to peak. The estimation could be based on different approaches, either by using streamflow data only or by cross-correlation between streamflow and precipitation (if P series are available, as for the French basins). This additional analysis would aim at investigating whether the seasonal river memory for the high flow season is enhanced for slow-response catchments, as it could be intuitively expected, and quantify this possible trend.
3 – I think that the analysis by country is less significant than the use of physical descriptors, and could be reduced. For example, I was wondering whether it is meaningful to show the box-plots by country (e.g. Figure 3b and 5b) or if it would be sufficient to report the geographical distribution by a map as in Figure 2. By the way, by looking at the map in fig. 2a, we can see a geographical trend within France, since the highest correlation coefficients are located in the northern part of France, which is subject to oceanic climate and slower catchment response.
Also, I agree with the comment in the text of the draft on section 4.8.3 (Country-level verification) saying that this part seems less meaningful, and redundant with other parts.
These are only complementary suggestions. The general protocol of the report is already very good. I look forward to see and discuss the developments of the work. Thank you!
AndreaJanuary 31, 2017 at 5:00 pm #508
María J. PoloParticipant
Thank you for the draft report; I think it is a clear representation of the goals of this experiment and I found the overall results really interesting and concluding on the main focus of the work. I also welcome this open discussion initiative, which makes it richer not only the discussion-process itself but also the final work to be done. I find the whole protocol reliable and clear to apply elsewhere, which is also a really worthy result of the work. Thanks!
Previous posts have been really interesting. Some additional comments follow:
• Despite it is a large set of data, I miss some additional description of the cases included in the analysis. Table 1 a good example of resuming key information for this, but I think that information about relevant climate descriptors of the contributing areas to each case, grouped up to shorten the table, is really needed to offer a clear scenario of the overall variability of examples of included, and the likely facts beyond the differences found (or not). Topographic-hydrological descriptors may be useful too.
• I agree with some of our colleagues’ posts about the inclusion of scatterplots of seasonal correlations against physical predictors. The physical interpretation underlying the results of the analysis is one of the motivations I find most appealing when going through this kind of experiments.
• I found the section devoted to the PCA description a bit lengthy, but maybe because we’re really used to applying this analysis. I’m not sure if a shorter text and some references to read further would be welcome by the reader or not.
• There seems to be some word missing and some typo in line 450.
• I wonder about rivers in southern Europe and I apologize for not including additional datasets on this. I understand this may arrive too late to say and I assume my fault!
Thank you again for the work and putting all together, I’m really looking forward to going on this experiment and final work. Thank you!
María J.February 8, 2017 at 9:37 pm #513
Thank you everyone for your comments and suggestions!
Your points are very interesting and I will try to include them in the manuscript.
Unfortunately, we probably do not have enough time for extensive additional analysis (collecting data for all of the rivers and performing again the basis of the analysis) as we should be discussing a final form of the manuscript in about a month’s time.
However, including some climatic information for the rivers is feasible and I will try to do that.
Also, if you already have actual daily rainfall series for some of the rivers (showing significant correlation) of length >50 years, could you please email them to me? If length is greater (say around 100 years), we could check persistent effects there too. In general, they need not be many, as this would be included only as a case study in order to comment at some extent on the effect of rainfall/hydroclimatic processes in river memory.
In any case, I will provide an improved version of the manuscript based on your suggestions in approximately a month, and we can move on from there.
Thanks a lot again for this exciting collaboration!
The forum ‘Experiment Forum’ is closed to new topics and replies.