As already mentioned in the first review ( and by the other reviewers) assessing the potential of CMLs for rain retrieval in a new region (here Brasil ) could be a great addition to the work that has already been provided on this topic by several groups in Europe, Israel and Africa. Compared to other studies on the same subject and in the sub-tropics the data set available here is much richer (both in terms of available links and gauges – with also a nearby disdrometer) and of great potential interest for demonstrating the advantages and limits of CMLs based rainfall measurement.
The present work is unfortunately far from delivering the full potential of the available data set.
Instead of the authors ‘encouraging future work ‘ on the data set and listing in their conclusion some of the many things that ‘could be done’, one feels like encouraging the authors themselves to take the current analysis a step further in order to take better advantage of the data set and draw some convincing conclusions to ‘’shed some light on the suitability of CMLs ‘ . This would be useful for the CML and hydrometeorology community.
The authors have added some additional work compared to their discussion paper but most of the reviewers’ comments are far from being accounted for in this new version.
The main problem is that the results of the CML-gauge comparison as presented (for instance Fig 6) are mostly showing that the method, as applied here, fails to reproduce the 30’ rainfall time series satisfactorily. And because most of the links are in practice unusable and unreliable for Quantitative precipitation estimate, the use of a CML network to obtain high-resolution rainfields and quantify intense rainfall (for urban hydrology, landsliding risk detection etc.) as proposed in the introduction is a very bad idea !. The CML technique may be used to have a very rough (and quite biased) estimate of average rainfall over a city (cf Fig 3).
Some links (3?) behave well at least for a few events (Fig 4), however the ‘well-functionning’ links are detected a posteriori, thanks to the gauges, so not very usable in practice.
The authors have dismissed the suggestions by several reviewers of the discussion paper, to understand better why so many links are in disagreement with the gauges and decompose and quantify the problem step by step (miss/false-detection ? why ? ) .
One of Rainlink’s step is to compare the consistency of a link with its neighbors – Couldn’t this feature be further exploited to detect the consistency among links and understand the problem - before comparison with gauges ?
Here the assessment is performed using the 30’ time series, would results be better -in terms of detection at least- at the daily time step ?
Other more minor point :
I also regret that the authors dismissed the suggestion made by more than one reviewers to provide not only global statistics but also statistics by rain classes, or at least quantifying if the links perform well in heavy or light precipitation. Once again, given that the authors put forward hydrology, floods, land-sliding as applications, and given the stress put in testing the method in sub-tropical climate with intense rainfall rather then in The Netherland, an analysis of the performance in heavy rainfall is very relevant.
Once AND IF a serions effort for improving the content is done , the english text will also need revising.
The presentation of the data, quality control, signal processing (outliers elimination etc.) and results still lack precision and clarity. (for instance still some confusion between R=ak^b and k=aR^b although already commented for in first review ) –
Some of the processing choices or data filtering appear quite arbitrary and should be better argued for and their impact quantified.
See detailed comments below.
DETAILED COMMENTS :
Section 1 Intro - Introduction
p2 l4
‘backscattring (i.e. reflectivity) is not the only way to measure rainfall with radar, polarimetric radar may propagation parameters such as specific differential phase shift for accurate measurement of heavy rainfall. As a matter of fact the city of Sao Paulo is equipped with (at least) one polarimetric radar. “
p3 l24 : Sahel may be semi arid but rainfall in this region also falls as intense events (cf many recent articles on floods and rainfall intensification in Sahel) and is associated with deep convection – the sentence is misleading. What may oppose the region is the terrain and oceanic influence in SP while flat/continental environment in Sahel.
Section 2 DATA
P4 l14 : do you mean that for ER you have only received and not transmitted power ? Please clarify as this in an important point for attenuation processing.
P4 l18-19 : doubt on the length of links - couldn’t this information been checked on site ? from your brasilian partners ?
Some free internet resources as for instance the site
http://telecocare.teleco.cl9.com.br/telebrasil/erbs/
provides exact locations of RF antennas from all operators in Brasil …….you may check some of the links displayed in Fig 1.
P4 l 28-29 : “closest two gauges’ – Given the density of the network I assume it means very close (less than 1 km ? ) in absolute term. But please provide an indication of the max range considered here.
P5 l 1 : what is the rationale for these threshold values (bias and r2) – they seem arbitrary unless you explain why they were chosen.
P5 – L17 : CML operating frequencies range from 7 to 80 Ghz (at least) depending on regions, regulations, length etc… b is not equal to 1 for the whole range. Please be more specific.
P5 l28 : here you use R=ak^b and in (1) k=aR^b. Please be careful – these inconsistencies in k-R vs R-k were already pointed out in the first review.
P6 1rst paragraph – Figure 3
‘it is clear from the figure that there certainly are differences … ‘ - Please provide a more quantitative assessment of these diffrences between the curves and between frequency – and add on figure or provide in text the values of the a,b coefficients for comparison.
P6 paragraph 2 – Rainlink algo description :
1) at which time scale is done the comparison with nearby links to assess dry/wet ? 15 minutes ? please clarify
3)outlier removal : what do you mean exactly by ‘deviates too much’ ? how do you accumulate specific attenuation over 24 hours ?
4) what is the rationale for the value 2.3 dB ? is it applied what ever the frequency of the links ?
p7
Section 3 – results
P8 l1 – ‘Such a small difference (in 3 month accumulations) suggests that the gauge data set is reliable’ – As well know by the authors agreement in terms of bias over a 3 month period does not mean he series is reliable as a validation data set used at the 30’ time step. Please be more serious in the assessment of the gauge data set.
P8 l12 – ‘three best performing CMLs’ - How exactly was that assessed ? are these the best performing CMLs over the whole period ? best performing in terms of which cretiria ?
P8l17 ‘ the fig shows that the se three CMLs capture reasonably well tow of the rainiest events’ – NO it doesn because on 1 CML is shown for the 2 events !!
Fig 5 = why aren’t the 3 links shown for the 2 events ? this would be much informative.
The lines for the 2 ‘upscaled series’ should be made more visible – as this is what we actually want to compare.
P8 L25 : I very much doubt that this is has any effect on the short links presented in Fig 4 and for 30’ average.
I suggest using the dense available gauge network to check what the spatial decorrelation of the 30’ average rainfall actually is for the SP.
P8 l29 : the authors should make the effort to quantify this point (relative impact of wet antenna vs rain attenuation along the way) using the present or/and their other data sets…. Is there evidence that this bias in attenuation/rainfall is more present just after than just before the storm ?
CONCLUSION
The conclusion will have to re-written once the necessary additional analysis, which is suggested but not performed by the authors, has been done. |