Value of Spatially Distributed Rainfall Design Events—Creating Basin-Scale Stochastic Design Storm Ensembles
Round 1
Reviewer 1 Report
I think it is not a novel manuscript, but a good application on real events.
I suggest to incorporate also other rainfall events in term of intensity and type.
good use of language, no recommendation or concern about
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Review: Value of spatially distributed rainfall design events – Creating
basin scale stochastic design storm ensembles
by Ville Lindgren, Tero Niemi, Harri Koivusalo and Teemu Kokkonen
The manuscript describes the application of a stochastic rainfall model over a macro-scale catchment in south-western Finland. The model is trained and validated with radar data for two independent rainfall events. Analyses is conducted using various statistical methods and measures and on different spatial scales.
Overall the study design and analysis is adequate. In general, the manuscript is well written, however, for some parts it would be good to provide more details. Nevertheless I have a few concerns which I would like to raise below.
Major concerns:
Description of events: The motivation for the study is to generate a series of design storms using a stochastic rainfall model. In my understanding a design event is a heavy rainfall event that has a certain probability to occur e.g. once in 10years. I assume that the two events were selected in order to represent a typical heavy rainfall event of more convective (event 1) and frontal (event 2) nature. Given the fact that there are 10+ stations in the catchment it should be possible to give an orientation of how frequent the selected events are.
Validation of model: Generally the model seems to perform very well. However, I am wondering how surprising this is, since the model seems to be trained more ore less with the same measures as are analysed in the validation section. Maybe the authors can clarify this and potentially validate against more independent measures.
Variability across scales: Focusing the analysis on different spatial scales is useful, however, the conclusion out of this section that variation is larger for smaller scales events is somehow not surprising. The question is more, if all of the ensemble members have the same likelihood to occur. Furthermore I would find it useful to also show the spread of the respective radar events over all the 3rd scale catchments in order to see if the respective spread in the ensemble is somehow physically justified.
Different averaging methods: I am wondering if the conclusion that the maximum Rtot is underestimated if only the 3 nearest grid boxes at the station locations are considered is generally true or if is due to the specific station location. In order to test this, I would suggest to randomly vary the positions of the stations (but keeping the density of the network) and analyse the frequency of cases, the maximum Rtot is underestimated when only the stations are considered.
Minor issues:
Fig 4&5: is there in both cases no rain in the smaller subcatchments (e.g. panels c to e )
Fig 6 & 7: I would suggest to change the color scale – wetter conditions should be depicted in blue, dryer regions in red.
Table 1. I assume the last column should be the cumulative areal mean rainfall. Same also in section 3.2.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper describes a framework to generate space-time stochastic fields of rainfall that are suitable for use as design storms in hydrological studies. This represents an important innovation on the current standard practice of using either uniform or very simple patterns of rainfall as design storms.
The paper is well written, the results are novel and interesting, and will be suitable for publication once the following points have been considered by the authors:
I understand that the main point of the paper is to emphasise the need to include a better representation of the spatial and temporal distribution of rainfall in a design storm. However, the use of the pySTEPS framework to generate the stochastic rainfall fields is sufficiently novel that I feel that more details on the process should be presented in the paper.
I missed seeing some evidence that the various models that generate the time series of model parameters were appropriate. Having a table that lists the values of some of the more important parameters for the various components of the framework would be informative.
I very much like to see the rain fields accumulated over the storm duration as a qualitative measure that gives confidence in the suitability of the framework to generate realistic rain fields. This is because the accumulations depend on all aspects of the framework: the field advection, the temporal evolution of the field, and the spatial distribution of the instantaneous field. Perhaps this could be added at the start of Section 2.6 as part of the overall (qualitative) evaluation of the model.
Figure 3(d). How important do you think the covariance between the two advection dimensions is? The simplest (starting) assumption is to assume that the dimensions are independent, but the observed (blue) trace suggests some sort of temporal evolution, do you think that it matters?
Line 418. The consequence of using a single advection vector is that the domain of the simulation should be restricted to about 250 km, since rotation in the advection becomes increasingly evident as the domain scale is increased. Unfortunately, I do not have any references to back this up, but this has been my experience.
Line 445. “unit conversion explains most of ensemble variation” is unlikely to be true. You are using a single Z-R relationship so that is not a source of the variance. The variation at the large scale between ensemble members probably comes from the long memory of the broken-line process for the mean areal rainfall relative to the storm duration.
Line 448 “simplification of a complex natural phenomenon”, likely the impact of topography on the distribution of rain during a specific extreme event needs to be included in the analysis somehow.
(Very) Minor points:
Line 36: Dense rain gauge networks are actually quite expensive to run, mainly due to the cost of maintaining a large number of instruments in the field that require regular visits.
Figure 4,5: The yellow median line is nearly invisible. I joke with my students that they should only use yellow for the lines that they don’t want anyone to see…
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I am currently out of office with only limited possibilities to go online. Based on the author response, however, it seems that my major concerns have been sufficiently tackled. Hence I would suggest to accept the manuscript,
