the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Nov 2021
Technical note: A revised incoming neutron intensity correction factor for soil moisture monitoring using cosmic-ray neutron sensors
Abstract. The cosmic-ray neutron sensor method of soil moisture measurement is now widely used and is fundamental to the COSMOS-UK soil moisture monitoring network. The method is based on a relationship between a measured flux of neutrons and soil moisture, and requires the neutron count to be adjusted for time variations of atmospheric pressure, humidity and the incoming flux of cosmic-ray neutrons. This note describes an empirical approach to the development of a revised correction factor for the last of these. Using the revised correction factor makes a significant difference to the derived soil moisture at wetter sites. This has implications for quantifying the soil moisture regime at these sites and management decisions that depend on a proper understanding of soil moisture dynamics, such as flood management and the release of greenhouse gases.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(701 KB)
Interactive discussion
Status: closed
-
CC1: 'Comment on hess-2021-564', Heye Bogena, 24 Nov 2021
Based on data from the UK Cosmic-ray Soil Moisture Observing System network the authors found that the correction of incoming neutrons variations caused spurious trends in the soil moisture estimates, especially for a site in Scotland with high organic carbon content.
This is a timely topic, as more and more Cosmic-ray neutron sensors (CRNS) are installed world-wide. Correction for incoming neutron variability is essential for the application of CRNP since in addition to the short-term fluctuations, a strong positive trend in incoming neutrons was observed in 2016 and 2017 due to the solar cycle. Therefore, without appropriate correction, the CRNS data will show spurious trends in soil moisture.
For this purpose, data from neutron monitors are widely used in the CRNS community. However, neutron monitors show differences in the variations of incoming neutrons, mainly due to different cut-off rigidities. It is clear that these differences also propagate in soil moisture estimates from CRNS signals, which requires the necessity of trend adjustment of the neutron monitor data. Up to now, a dedicated study on this topic is missing. Therefore, this paper is of great interest for the growing community of CRNS users.
I have the following general comments:
The authors argue that the spurious trend becomes more obvious at site with high soil moisture values due the non-linear calibration function. But the COSMOS-UK site Sourhope (https://cosmos.ceh.ac.uk/sites/SOURH) only ~160 km away from Gensaugh also features high soil moisture values up to ~70 Vol.% due to high soil porosity (soil density is 0.65 g/cm³), but there is no spurious trend like Glensaugh.
Therefore, I think there is a need to test this method at several sites to exclude the possibility that the spurious trend is actually due to a local hydrological change at the Glensaugh site triggered by the extreme drought in 2018. Such decreasing trends after this drought can be observed also at other sites in Europe.
So far, the authors used only one COMOS-UK site (Glensaugh) for developing their method and another site (Bunny Park) to test it. The COMOS-UK network as well as the recent COSMOS-Europe data paper (Bogena et al., 2021) provide a perfect basis for a better testing of the method.
The proposed method for correcting influences of incoming neutron variations on CRNS readings has the following shortcomings, which should be discussed in more detail:
1) Although it uses standardised differences using the median, the correction factor derived from the relationship is still influenced by soil moisture dynamics (i.e. it would only perfectly work for a permanently wet site or a lake). I suggest excluding dry periods from the analysis to avoid dry bias.
2) The correction factor depends on the measurement period. It is unclear, how many years are need for achieving a reliable correction factor.
Citation: https://doi.org/10.5194/hess-2021-564-CC1 -
AC1: 'Reply on CC1', Magdalena Szczykulska, 06 Dec 2021
We thank the reviewer for appreciation of our work and for the comments. Below are our responses to the general and specific comments.
1. “The authors argue that the spurious trend becomes more obvious at site with high soil moisture values due the non-linear calibration function. But the COSMOS-UK site Sourhope (https://cosmos.ceh.ac.uk/sites/SOURH) only ~160 km away from Glensaugh also features high soil moisture values up to ~70 Vol.% due to high soil porosity (soil density is 0.65 g/cm³), but there is no spurious trend like Glensaugh.”
We did not state a reason why the trend seems to be more obvious at wetter sites. We simply reported our observation that within the COSMOS-UK network, wetter sites show more obvious trend. COSMOS-UK website already utilises the improved intensity correction and hence VWCs from Sourhope shown there will have any trend removed or reduced.
2. “Therefore, I think there is a need to test this method at several sites to exclude the possibility that the spurious trend is actually due to a local hydrological change at the Glensaugh site triggered by the extreme drought in 2018. Such decreasing trends after this drought can be observed also at other sites in Europe.
So far, the authors used only one COSMOS-UK site (Glensaugh) for developing their method and another site (Bunny Park) to test it. The COSMOS-UK network as well as the recent COSMOS-Europe data paper (Bogena et al., 2021) provide a perfect basis for a better testing of the method.”
The method has been applied to all sites in the COSMOS-UK network. This is stated and discussed further in section 3 of the technical note. As mentioned in the response above, the COSMOS-UK website already utilises the revised background intensity correction. We very much hope that the methodology described in the technical note is explored by other ‘cosmos’ networks and agree that the COSMOS-Europe data set will provide an interesting opportunity for further analyses. We note that some of the sites in Figure A.3 of Bogena et al. 2021 (https://doi.org/10.5194/essd-2021-325) may display a similar trend to those seen within the COSMOS-UK network (e.g. Harrild, Rollesbroich and Ruraue).
3. “Although it uses standardised differences using the median, the correction factor derived from the relationship is still influenced by soil moisture dynamics (i.e. it would only perfectly work for a permanently wet site or a lake). I suggest excluding dry periods from the analysis to avoid dry bias.”
We agree that the wet periods contribute points with more stable soil moisture. In the attempt to avoid the dry bias, we applied a lower quartile regression which puts more weight towards counts corresponding to the wetter periods while still utilising all the available data. We feel this is more appropriate than attempting to identify wet and dry periods given the temperate maritime climate of the UK. This is discussed in section 2.1. In practice, the reviewer’s suggested approach and the lower quartile regression yield similar results for the COSMOS-UK network sites.
4. “The correction factor depends on the measurement period. It is unclear, how many years are needed for achieving a reliable correction factor.”
We discuss this aspect in section 3. Since the regression is performed between COSMOS-UK and Jungfraujoch (JUNG) neutron monitor counts, it is not so much the record length but the range of available JUNG counts which very much depends on the solar cycle. For instance, in the period between 2018 and 2021, there was not much variation in the JUNG counts. As a result, sites with data for this period only would not have a reliable estimate of G.
Citation: https://doi.org/10.5194/hess-2021-564-AC1
-
AC1: 'Reply on CC1', Magdalena Szczykulska, 06 Dec 2021
-
RC1: 'Comment on hess-2021-564', Anonymous Referee #1, 27 Dec 2021
The comment was uploaded in the form of a supplement: https://hess.copernicus.org/preprints/hess-2021-564/hess-2021-564-RC1-supplement.pdf
-
AC2: 'Reply on RC1', Magdalena Szczykulska, 17 Feb 2022
We thank the reviewer for the comments. Below are our responses.
- “The Authors present a revised incoming correction factor that could be used for improving soil moisture estimated with cosmic-ray neutron sensing (CRNS). The topic is interesting and timing and probably the study fits a technical manuscript in HESSD. The analysis is based on the extensive datasets collected within the COSMOS-UK sites and the manuscript is generally also well structured and written.”
We thank the reviewer for appreciation of our work and for the positive general comment.
- “[1] the study proposes a revised factor (eq.8) and the results are compared to the state-of-the-art (eq. 3). However, as far as I have understood, it is also well known an additional method (eq. 4) that so far it seems to be not widely adopted by the “CRNS community” but only used in some studies e.g., (Howat et al., 2018). So why do the Authors did not test this “not so common” revised method before proposing something new?”
These approaches were tested, but we did not report this in the paper. We will add a comparison of the different correction approaches and how they affect the corrected counts and volumetric water content (VWC) at the COSMOS-UK sites. In particular, we will provide a comparison between our approach and the approaches given in Hawdon et al., 2014 (i.e. gamma based on cut-off rigidity approach and a nearest neutron monitor based approach).
To add for clarity, Howat et al., 2018 give gamma for a particular site in Greenland using a neutron monitor (THUL) also located in Greenland, so their result is for that particular case study.
- “[2] as far as I have understood, the revised factor (eq.8) converges to what was currently available but not widely implemented (eq.4) not only in the form but, even more important, to the actual parameter i.e., G = 1.2 vs. gamma = 1.19 (See (Howat et al., 2018)). So, did I misunderstand or should the Authors be already satisfied by using eq. 4 without the need to propose a revised factor?"
Howat et al., 2018 obtain the value of gamma = 1.19 (or beta as stated in the paper) for a particular site in Greenland without explanation beyond ‘based on regressions to the global neutron monitor dataset’. Our study details how to derive G values at specific COSMOS-UK sites (there are 48 of them), so site specific incoming neutron intensity corrections. These site specific G values vary between 1 and 1.5 for the COSMOS-UK network. While the mean value across sites seem to converge to 1.2, it is not the case at individual sites.
We will clarify this aspect in the manuscript.
- “for the development of the new revised factor (eq.8) the Authors compare incoming neutrons from RNMS (e.g., Jungfraujoch) to CRNS neutrons locally collected at a soil moisture site (SMS) during period where it is expected low variability due to soil moisture changes. Namely, removing local influences due to soil moisture, variability in the neutron counts should then be related to incoming fluctuations. The Authors then compare these local fluctuations to the RNMS. It is well discussed that, if these fluctuations are not the same, on a longer term, should be due to different cutoff rigidity and altitude between the RNMS and SMS. But since also eq.4 was developed to account for these factors, from my understanding it should be not a surprise that this revised method converge to eq. 4. So, overall, it seems to me that the Authors simply analyzed some time series and found empirically what is already know and addressed in literature. I might be wrong but, if this is the case, I encourage the Authors to clarify and improve the manuscript to better convey the novelty of the study.”
The local fluctuations at SMS are different from RNMS due to the differences in cut-off rigidity and altitude, but also likely due to other differences, such as detector differences and site conditions. Whilst the CRNS detectors each have nominally the same response function as each other (but different to the RNMS), it is the interplay between the CRNS response function and the local SMS neutron environment (neutron energy spectrum) which can lead to site specific values of G (or gamma), even for sites at the same cut-off rigidity. For example Weimar et al., 2020 note that “The relative thermal contribution of the signal of the standard CRNP is in particular large for moist soil.” It is also worth noting that none of the detectors deployed for COSMOS-UK have any thermal shield, so will be susceptible to thermal neutron leakage. In the revised manuscript, we will properly discuss these and other effects (e.g. wet soil generated neutrons) which may affect G, and in turn suggest possible physical interpretations of G.
Eqn. 4 accounts for some of the differences between SMS and RNMS by normalising neutron intensities at time of interest to a reference value (I/Iref), but also via gamma. Hawdon et al., 2014 relate gamma to cut-off rigidities of locations of interest. This does not take into consideration the other above mentioned differences which do not necessarily cancel out via I/Iref. In Howat et al 2018, both the cosmic ray neutron sensor and the reference neutron monitor are located in Greenland. The authors are primarily concerned with elevation differences and find gamma = 1.19 for that specific case study.
COSMOS-UK sites and the Jungfraujoch neutron monitor have very different location characteristics when compared to the study conducted by Howat et al., 2018. We also do not try to say that G has a single value for the whole COSMOS-UK network, but rather we investigate each individual site and find that G varies between 1 and 1.5, and we only report in L 144 that the average value of G is 1.2. We will clarify this point.
The novelty is that we calculate G for each individual COSMOS-UK site using directly neutron count data from the local site of interest (so we do not have to make assumptions about what causes the differences). We take a data-centric approach, and find the relationship directly from the data. While these neutrons are subject to soil moisture variability, we use a lower quartile regression fitting to put more weight on the wetter points (lower count rates) which have more constant soil moisture in the case of the UK. This method successfully reduces the unphysical drying trend in the CRNS soil moisture observations at wetter COSMOS-UK sites, for instance Glensaugh, when G=1 and provides better performance in this case when compared to G based on cut-off rigidity as given by Hawdon et al., 2014. This is an indicator of other influences which, as above, we will discuss further in our revised manuscript.
We will improve the manuscript to stress where the novelty of this study lies.
- “[4] my last comment is related to the general assumption that incoming neutron counts from a RNMS adequately represents the relevant incoming neutron flux at the SMS and the revised factor accounts for some additional differences (L61-63). Based on that, the Authors conclude and suggest (L165-171) several research activities that could be performed for further improvements. Indeed I agree that using incoming fluctuation from RNMS is a first order correction that has to be considered also for CRNS applications. This assumption has however two shortcomings that should be considered. First, time series at RNMS need also several corrections that are still under investigations and the focus of current research activities and improvements. Thus, these time series are not error-free. Second, some local incoming fluctuations at SMS are not detected by RNMS. Thus, these time series could not well inform local incoming fluctuations even in the case they were error-free. For these reasons, personally I do not see a good suggestion to push much effort in improving a method that is based on input (i.e., the time series at RNMS) that has these drawbacks. In contrast, I have seen that the use of alternative detectors installed directly at the SMS for the detection of incoming fluctuations has been suggested in literature using e.g., muons detectors (Stevanato et al., 2019; Stowell et al., 2021) or neutron spectrometers (Cirillo et al., 2021; Fersch et al., 2020). Personally, I believe that improving and working with these approaches could be much more valuable suggestions for further studies and developments instead of improving the manipulation of no error-free and non-representative time series from RNMS. Alternative, a discussion of the added value of the present revised factor in comparison to abovementioned approaches should be reported.”
We thank the reviewer for this helpful comment – yes we agree that (as above) there will be neutron energies (epithermal and thermal) which the CRNS detects, but not the RNMS. Indeed, for future work, the promising development of local reference detectors should be considered as a significant advancement, where affordable. We will add a brief discussion of these approaches.
However, given that for COSMOS-UK alone, we now have more than 5 years of data for 48 sites, with no in situ reference measurements, it is of great value to improve the long-term bias correction of the derived soil water content of these historical datasets, using the RNMS approach. Furthermore, our analysis of G across the whole of the COSMOS-UK network is valuable to highlight what improvements may be required (such as the deployment of reference detectors), and their relative cost-benefit.
Citation: https://doi.org/10.5194/hess-2021-564-AC2
-
AC2: 'Reply on RC1', Magdalena Szczykulska, 17 Feb 2022
-
RC2: 'Comment on hess-2021-564', Anonymous Referee #2, 20 Jan 2022
This manuscript "Technical note: A revised incoming neutron intensity correction factor for soil moisture monitoring using cosmic-ray neutron sensors" deals with the problem of incoming neutron radiation correction for cosmic-ray soil moisture sensors.
The authors claim to derive a "revised correction factor" using an amplitude scaling technique that has already been proposed in the literature before, while suggesting the use of median instead of reference count rates. In the same study, the authors find that the proposed median count rate often is insignicifantly different from the reference count rate, that it turns the whole approach to be highly dependent on the individual measurement period, and that it is less generalizable. Good agreement has been found by comparing the overall average at tens of sites with findings from other studies, but the reader is still left with no clear idea about the added value provided by this study.
While the problem statement is relevant for the CRNS community, the novelty of the approach as well as the quality of the explanations are low. However, there is a potential to accept the manuscript with major revisions, particularly by further elaborating on the scientific foundations, on the statistical quality, and on the demonstration of measurement results.
# Major concerns
- The HESS manuscript type "Technical notes" requires to "report new developments, significant advances, and novel aspects of experimental and theoretical methods and techniques which are relevant for scientific investigations within the journal scope." The addressed problem and topic are relevant for studies published in HESS, but it is not clear how to identify the significant advancement and novel aspect of this study. While the manuscript "should be short (a few pages only)", the authors elaborate unnecessarily on less related aspects, such as the importance of soil moisture for hydrology or drying and wetting characteristics of individual sites. I would suggest to provide a more concise document with a strong focus only on the *technical* aspect of incoming neutron correction:
- what is the expected sensitvity of CRNS to these corrections (cite e.g. Baroni et al. 2018 (JoH)),
- provide fundamental derivation of your approach and discuss differences to previous approaches, and
- statistically sound evaluation (using all your individual site data, proper statstical measures, and uncertainty analysis).
- The fundamental explanation of the involved processes and physics have been marginally addressed. The authors seem to treat the correction approach as yet-another function without discussing the meaning of the analytical form or the physical processes involved. For exampe:
- Why should detected low-energy neutrons in any way behave proportional to incoming high-energy neutrons? What fraction of the detected signal is direct and indirect radiation and how does this change with site conditions? The authors you are citing have answers to that. This would help readers to actually understand your approach and this may have direct implications on the interpretation of the performance of your results at individual sites.
- Can we expect this approach to be different for different regions on Earth? What has this to do with latitude or cut-off rigidity? The answers may have implications on the performance of your distributed sensors and should be discussed using the empirically obtained values for G.
- What could be the physical reason for the necessity of an amplitude scaling in the form of G*(I/I_m-1)? This has implications on the choice of G and its meaing, and could answer whether situations could occur with G<1, too.
- Also put the whole idea into context and note that it is about correcting for solar activity fluctuations and mention that the scale of those variations can be from years (solar cycle) to days (FDs, GLEs). This has direct implications on the conclusions that you could and could not draw from your analysis, e.g. by choosing a "monthly average", with which one would have no chance to resolve any short-term variations.
- The authors determine gamma empirically, while there are high chances of fitting through soil moisture variations. A more "safe" approach would be to calibrate gamma on quite periods (low solar activity) and at sites with constant soil moisture (desert, concrete, or lakes). Some authors (which you have cited) have already traveled along this path and it could be helpful to report on advantages and disadvantages of the various approaches.
- The authors spend two pages to derive their concept, and end up with an equation which is equal to the one from literature which has been presented already in the introduction. It is claimed that the new equation is now "revised", for two reasons:
- it uses median count rates instead of a reference count rate (although this option has already been suggested by Zreda et al. 2012 and others). Interestingly, in your own study you tell that the difference between the use of I_m or I_ref often is insignificant...
- the scaling factor G now is empirical (although this has been done already by other authors which are cited elsewhere in this very study), while they are hardly compared to values from existing approaches. In addition to that, even the relationships to the cut-off rigidity, as proposed by Hawdon 2014, for instance, are empirical. So instead of empirically building on their theory, you reduce the complexity of finding G even more by applying a purely empirical fit. This comes with the risk of low transferability and generalizability of the results.
- Under these circumstances, I strongly suggest to properly discuss previous literature and not to sell the current development as a "revised correction approach". This study for me looks rather like a case study, where some of the existing suggestions have been tested on the UK network data. It is surely important to do that, but it would require a different storyline.
- L143: If the results are dependent on the variation of the incoming radiation, then it should be easy to evaluate this by testing the approach during high and low solar activity. Something that could be done at the COSMOS-UK Network.
- L147: "Preliminary studies show no simple relationship with ... latitude, altitude, or rigidity"
In the last sentence before the conclusions you are mentioning the most important aspect, in my view, which should be central to this study. Consider elaborating on building a theoretical basis for your approach, which involves confrontation with the mentioned quantities. This would help to develop this approach further towards a generalized and transferable method.
# Specific comments:
- Common and vague language is used quite often, rather that clear and factual language. E.g., L17 "perhabs less obvious", L85 "gives a gradient G, not of 1, but 1.5", L122: "change is very much greater".
- L33: the Desilets equation is semi-empirical, the paramters have been derived from observations only. In general, this paragraph can be much shorter. The choice of the conversion function and vegetatiion or organic carbon corrections are of lesser importance here.
- Fig 3: The comparison of time serieses with different G is not insightful. Please provide reference data, e.g. from the neutron monitor (to demonstrate a changing correlation to their signals) or soil moisture data (to demonstrate better match).
- Fig 4: Now you show TDT data, but only with G=1, so there is no chance to compare the approaches and to evaluate the performance of your idea. Moreover, consider showing a weighted average using different depths, instead of only 10 cm data.
Citation: https://doi.org/10.5194/hess-2021-564-RC2 -
AC3: 'Reply on RC2', Magdalena Szczykulska, 17 Feb 2022
We thank the reviewer for the comments. As a general note we would like to stress that this work was primarily motivated by an observation of a spurious long-term trend in the derived volumetric water content (VWC) at some of the wetter sites in the COSMOS-UK network, for instance Glensaugh. It was identified that this trend was correlated with the Jungfraujoch counts used in the background neutron intensity correction, then of the form I(t)/Iref (so G = 1 in eqn. 8). We found that after applying other reported approaches (Hawdon et al., 2014), the long-term trend would still remain. So we took a data-centric approach and directly calibrated the normalised fluctuations of Jungfraujoch counts with the CRNS counts at the specific site of interest to find G for that specific site. We also took caution to not fit through soil moisture variations by applying a lower quartile regression when deriving our site specific Gs. Our method significantly reduced the observed long-term trend.
We of course appreciate the reviewer’s request for clarity, providing comparison of approaches and more scientific understanding of how G varies with respect to different physical processes. We will address these points in the new version of the manuscript. We also apologise for the misstatements in terms of what is revised: the form of the correction given in eqn. 8 appears in the literature (although with almost no explanation where it originates from) and it is the method for obtaining the site specific G that is revised. We will clarify these aspects in our manuscript.
Below are our responses to the individual comments.
1. “The authors claim to derive a "revised correction factor" using an amplitude scaling technique that has already been proposed in the literature before, while suggesting the use of median instead of reference count rates. In the same study, the authors find that the proposed median count rate often is insignicifantly different from the reference count rate, that it turns the whole approach to be highly dependent on the individual measurement period, and that it is less generalizable.”
We agree that the general form of the correction in eqn. 8 is not novel and has been proposed in the literature. The value of the paper comes not in the general form of the correction, but in the way we estimate G – using local cosmic ray neutron sensor data combined with a lower quartile regression fitting to fit through wetter points associated with more constant soil moisture in the UK. We apologise for the misstatements in the manuscript, and we will revise it to clarify this point.
To address the point of poorer generalizability due to the individual measurement period dependence, the following are two reasons why such dependence arises:
- G values are derived by comparing neutron count fluctuations from the reference neutron monitor (In) and local neutron fluctuations from the Cosmic Ray Neutron Sensor (Nn) and therefore there is a necessity to have sufficient variations in In to establish a reliable relationship between In and Nn. This is linked with the solar activity more than the time-period – it is already explained in the manuscript in L 141 and L 142, but we will clarify this further. Small variations in In indicate low solar activity in which case the intensity correction is of less significance. So while the method requires sufficient variations in the incoming background neutrons, it is also mostly not required when this is not the case.
- The CRNS data carry soil moisture signal and therefore the measurement period needs to be sufficiently long so that it is possible to fit through relatively constant soil moisture points (in the case of the UK, these are the points associated with wet periods). So it is rightly noted that for a short time series G will carry more uncertainty, but since the method is aimed at correcting long-term trends, the users requiring this method will most likely have a long time series available.
2. “Good agreement has been found by comparing the overall average at tens of sites with findings from other studies, but the reader is still left with no clear idea about the added value provided by this study.”
The original motivation for this study was an observation of a spurious long-term drying trend present at wetter COSMOS-UK sites, for instance Glensaugh, when G = 1. Using G = 1.22 based on cut-off rigidities, as given by Hawdon et al., 2014, also does not remove this trend completely. The added value of our study is that we calculate G for each individual COSMOS-UK site using directly neutron data from the local site of interest. While these neutrons are subject to soil moisture variability, we use a lower quartile regression fitting to put more weight on data points associated with more constant soil moisture. This approach corrects the observed unphysical drying trend present at some of the COSMOS-UK sites.
In the study conducted by Howat et al., 2018, a value of gamma = 1.19 is given for a specific location in Greenland and the authors use a neutron monitor (THUL) also located in Greenland. Gamma given in that work is therefore for that specific case study and is not expected to be the same at different locations. While we note in the manuscript that our COSMOS-UK site average G value coincides closely with the value of gamma obtained by Howat et al., 2018, we do not use it in the background neutron intensity correction. We always use the site specific value which can vary significantly from site to site (between G = 1 and G = 1.5).
We will improve the manuscript to stress more clearly the novelty of our approach and its implications on the derived soil moisture data. We will also add to the scientific understanding of how G varies with different physical parameters.
3. “what is the expected sensitvity of CRNS to these corrections (cite e.g. Baroni et al. 2018 (JoH))”
We will report on the sensitivity of CRNS to the background neutron intensity correction factor based on the literature.
4. “provide fundamental derivation of your approach and discuss differences to previous approaches, and”
If the reviewer is asking about how we derive G, then yes, we will clarify this further in the manuscript. We will also discuss the differences to other approaches.
5. “statistically sound evaluation (using all your individual site data, proper statstical measures, and uncertainty analysis)”
We will provide uncertainty estimates for individual site Gs.
Regarding the aspect of “using all your site data”, we did use all site data for the site specific G estimation. If what is meant is spatial analysis, then we will analyse our data in the context of understanding why G varies from site to site.
6. “The fundamental explanation of the involved processes and physics have been marginally addressed. The authors seem to treat the correction approach as yet-another function without discussing the meaning of the analytical form or the physical processes involved.”
We took a data-centric approach without complex modelling of the physical processes involved, but we will improve the manuscript by discussing the physical processes.
7. “Why should detected low-energy neutrons in any way behave proportional to incoming high-energy neutrons? What fraction of the detected signal is direct and indirect radiation and how does this change with site conditions? The authors you are citing have answers to that. This would help readers to actually understand your approach and this may have direct implications on the interpretation of the performance of your results at individual sites.”
These of course are very valid questions, which we had originally decided to defer to later work, but we now see that this work is somewhat incomplete without at least discussion of the relevant physical processes and we agree that these would help to inform the interpretation of the different G values across all 48 sites.
We will discuss the energy spectral response functions for the CRNS versus the RNMS with respect to the question of proportionality between the incoming low and high-energy neutrons, and the relevance of site conditions for the correction. As above, will discuss properly the literature on this topic.
8. “Can we expect this approach to be different for different regions on Earth? What has this to do with latitude or cut-off rigidity? The answers may have implications on the performance of your distributed sensors and should be discussed using the empirically obtained values for G.”
We will show the dependency that our empirical G values have on cut-off rigidity.
9. “What could be the physical reason for the necessity of an amplitude scaling in the form of G*(I/I_m-1)? This has implications on the choice of G and its meaning, and could answer whether situations could occur with G<1, too.”
We will discuss the form of the correction and try to elaborate why this form can be adopted.
10. “Also put the whole idea into context and note that it is about correcting for solar activity fluctuations and mention that the scale of those variations can be from years (solar cycle) to days (FDs, GLEs). This has direct implications on the conclusions that you could and could not draw from your analysis, e.g. by choosing a "monthly average", with which one would have no chance to resolve any short-term variations.”
In our study, we are concerned with the long-term trend (the solar cycle) that was observed at some of the wetter sites in the COSMOS-UK network, for instance Glensaugh. So we choose monthly averages in figures 3 and 4 to show the impact of the correction on long-term trends.
11. “The authors determine gamma empirically, while there are high chances of fitting through soil moisture variations. A more "safe" approach would be to calibrate gamma on quite periods (low solar activity) and at sites with constant soil moisture (desert, concrete, or lakes). Some authors (which you have cited) have already traveled along this path and it could be helpful to report on advantages and disadvantages of the various approaches.”
To prevent from fitting through wide soil moisture variations, we use a lower quartile regression when comparing the normalised neutron count fluctuations from the cosmic-ray neutron sensor (Nn) and the reference neutron monitor (In) to obtain G. This ensures that we can use all the data, but the fitting is biased towards wet periods which have more constant soil moisture in the UK. We will stress it more clearly in the manuscript.
We will emphasise more clearly that we can only get good G estimates, over a sufficient change in RNMS counts e.g. with significant change in solar activity.
Some of the co-authors have also travelled this path to a degree with COSMOS-Rover experiments. However, given that cut-off rigidity plays a role, we do not have convenient constant soil moisture sites close to the SMS locations. There are a few exceptions, but even translating results from concrete or 100% water (lake) presents issues. We will discuss these approaches along with other onsite reference measurement approaches.
12. “The authors spend two pages to derive their concept, and end up with an equation which is equal to the one from literature which has been presented already in the introduction. It is claimed that the new equation is now "revised", for two reasons:
- it uses median count rates instead of a reference count rate (although this option has already been suggested by Zreda et al. 2012 and others). Interestingly, in your own study you tell that the difference between the use of I_m or I_ref often is insignificant...
- the scaling factor G now is empirical (although this has been done already by other authors which are cited elsewhere in this very study), while they are hardly compared to values from existing approaches. In addition to that, even the relationships to the cut-off rigidity, as proposed by Hawdon 2014, for instance, are empirical. So instead of empirically building on their theory, you reduce the complexity of finding G even more by applying a purely empirical fit. This comes with the risk of low transferability and generalizability of the results.”
The revision comes in the way we calculate G using the site specific CRNS data. Some previous authors used neutron monitor data (different detector and not site specific to where the cosmic-ray neutron sensor is located) to relate gamma (or G) to other physical parameters (Hawdon et al., 2014 relates gamma to cut-off rigidity and Howat et al., 2018 relates gamma to elevation, but with no formula given in the latter case).
While both approaches are empirical, our approach directly finds G for the site of interest using that same site data and without making assumptions of what physical parameters are involved – we directly use the CRNS data from the site of interest and account for soil moisture variability. Using the CRNS data has also the advantage of calibrating G for detector differences that exist between the CRNS and the reference neutron monitor, even if it is only a first order approximation.
We discuss the generalizability aspect in point 1 of this response and of course parameterisation of G (if all relevant factors are included) would increase the generalizability of the results and improve scientific understanding of the processes involved. We will address this point by providing a discussion on the physical influences on G. We will also attempt to build on the Hawdon et al., 2014 approach for G to explain our results at the COSMOS-UK sites.
We would also like to note that the work presented here has a very pragmatic focus: to bias correct long-term UK soil moisture time series data, for use by the hydrological and land surface climate model communities. We welcome others to test the applicability to other regions, to bias correct their historical data. The results will be specific to the make and model of the CRNS, and to a degree the installation conditions, as well as the surrounding environment.
13. “Under these circumstances, I strongly suggest to properly discuss previous literature and not to sell the current development as a "revised correction approach". This study for me looks rather like a case study, where some of the existing suggestions have been tested on the UK network data. It is surely important to do that, but it would require a different storyline.”
We apologise for the unintentional mis-selling – and we agree that we present an empirical application of the earlier published correction. We do consider that we present important details of the exact application, to allow an empirical fit using the on-site CRNS data (without underlying assumption of the causes) and to weight our fits to wet data to find G. We will strengthen the paper, as you suggest to provide more analysis of the results and more insight, as a UK case study.
14. “L143: If the results are dependent on the variation of the incoming radiation, then it should be easy to evaluate this by testing the approach during high and low solar activity. Something that could be done at the COSMOS-UK Network.”
We perhaps should have made it clearer: since G is a slope of a linear regression (lower quartile regression) between x (In) and y (Nn) there needs to be sufficient variations in x to find a reliable estimate of the slope. Figure 5c shows how the T values of G estimates for the COSMOS-UK sites increase when the variation in Jungfraujoch counts increase.
15. “L147: "Preliminary studies show no simple relationship with ... latitude, altitude, or rigidity". In the last sentence before the conclusions you are mentioning the most important aspect, in my view, which should be central to this study. Consider elaborating on building a theoretical basis for your approach, which involves confrontation with the mentioned quantities. This would help to develop this approach further towards a generalized and transferable method.”
Agreed, as in earlier reviewer responses, we will improve this work by properly discussing the physical processes that may contribute to the fitted values.
16. “Common and vague language is used quite often, rather that clear and factual language. E.g., L17 "perhabs less obvious", L85 "gives a gradient G, not of 1, but 1.5", L122: "change is very much greater".
We will improve the quality of the language used.
17. “L33: the Desilets equation is semi-empirical, the paramters have been derived from observations only. In general, this paragraph can be much shorter. The choice of the conversion function and vegetatiion or organic carbon corrections are of lesser importance here.”
According to Desilets et al., 2010, the calibration curve was obtained by “fitting ground-level neutron fluxes simulated in MCNPX [Pelowitz, 2005] to the shape-defining function”, so we are not sure why the reviewer suggests that “the parameters have been derived from observations only”. And yes, we can shorten this paragraph.
18. “Fig 3: The comparison of time serieses with different G is not insightful. Please provide reference data, e.g. from the neutron monitor (to demonstrate a changing correlation to their signals) or soil moisture data (to demonstrate better match).”
We will revise the plot and add neutron monitor data to it.
19. “Fig 4: Now you show TDT data, but only with G=1, so there is no chance to compare the approaches and to evaluate the performance of your idea. Moreover, consider showing a weighted average using different depths, instead of only 10 cm data.”
Yes, we will add a plot with the revised G to figure 4. Unfortunately, we only have TDT data for a depth of 10 cm, so we will not be able to show a weighted average.
Citation: https://doi.org/10.5194/hess-2021-564-AC3
- The HESS manuscript type "Technical notes" requires to "report new developments, significant advances, and novel aspects of experimental and theoretical methods and techniques which are relevant for scientific investigations within the journal scope." The addressed problem and topic are relevant for studies published in HESS, but it is not clear how to identify the significant advancement and novel aspect of this study. While the manuscript "should be short (a few pages only)", the authors elaborate unnecessarily on less related aspects, such as the importance of soil moisture for hydrology or drying and wetting characteristics of individual sites. I would suggest to provide a more concise document with a strong focus only on the *technical* aspect of incoming neutron correction:
Interactive discussion
Status: closed
-
CC1: 'Comment on hess-2021-564', Heye Bogena, 24 Nov 2021
Based on data from the UK Cosmic-ray Soil Moisture Observing System network the authors found that the correction of incoming neutrons variations caused spurious trends in the soil moisture estimates, especially for a site in Scotland with high organic carbon content.
This is a timely topic, as more and more Cosmic-ray neutron sensors (CRNS) are installed world-wide. Correction for incoming neutron variability is essential for the application of CRNP since in addition to the short-term fluctuations, a strong positive trend in incoming neutrons was observed in 2016 and 2017 due to the solar cycle. Therefore, without appropriate correction, the CRNS data will show spurious trends in soil moisture.
For this purpose, data from neutron monitors are widely used in the CRNS community. However, neutron monitors show differences in the variations of incoming neutrons, mainly due to different cut-off rigidities. It is clear that these differences also propagate in soil moisture estimates from CRNS signals, which requires the necessity of trend adjustment of the neutron monitor data. Up to now, a dedicated study on this topic is missing. Therefore, this paper is of great interest for the growing community of CRNS users.
I have the following general comments:
The authors argue that the spurious trend becomes more obvious at site with high soil moisture values due the non-linear calibration function. But the COSMOS-UK site Sourhope (https://cosmos.ceh.ac.uk/sites/SOURH) only ~160 km away from Gensaugh also features high soil moisture values up to ~70 Vol.% due to high soil porosity (soil density is 0.65 g/cm³), but there is no spurious trend like Glensaugh.
Therefore, I think there is a need to test this method at several sites to exclude the possibility that the spurious trend is actually due to a local hydrological change at the Glensaugh site triggered by the extreme drought in 2018. Such decreasing trends after this drought can be observed also at other sites in Europe.
So far, the authors used only one COMOS-UK site (Glensaugh) for developing their method and another site (Bunny Park) to test it. The COMOS-UK network as well as the recent COSMOS-Europe data paper (Bogena et al., 2021) provide a perfect basis for a better testing of the method.
The proposed method for correcting influences of incoming neutron variations on CRNS readings has the following shortcomings, which should be discussed in more detail:
1) Although it uses standardised differences using the median, the correction factor derived from the relationship is still influenced by soil moisture dynamics (i.e. it would only perfectly work for a permanently wet site or a lake). I suggest excluding dry periods from the analysis to avoid dry bias.
2) The correction factor depends on the measurement period. It is unclear, how many years are need for achieving a reliable correction factor.
Citation: https://doi.org/10.5194/hess-2021-564-CC1 -
AC1: 'Reply on CC1', Magdalena Szczykulska, 06 Dec 2021
We thank the reviewer for appreciation of our work and for the comments. Below are our responses to the general and specific comments.
1. “The authors argue that the spurious trend becomes more obvious at site with high soil moisture values due the non-linear calibration function. But the COSMOS-UK site Sourhope (https://cosmos.ceh.ac.uk/sites/SOURH) only ~160 km away from Glensaugh also features high soil moisture values up to ~70 Vol.% due to high soil porosity (soil density is 0.65 g/cm³), but there is no spurious trend like Glensaugh.”
We did not state a reason why the trend seems to be more obvious at wetter sites. We simply reported our observation that within the COSMOS-UK network, wetter sites show more obvious trend. COSMOS-UK website already utilises the improved intensity correction and hence VWCs from Sourhope shown there will have any trend removed or reduced.
2. “Therefore, I think there is a need to test this method at several sites to exclude the possibility that the spurious trend is actually due to a local hydrological change at the Glensaugh site triggered by the extreme drought in 2018. Such decreasing trends after this drought can be observed also at other sites in Europe.
So far, the authors used only one COSMOS-UK site (Glensaugh) for developing their method and another site (Bunny Park) to test it. The COSMOS-UK network as well as the recent COSMOS-Europe data paper (Bogena et al., 2021) provide a perfect basis for a better testing of the method.”
The method has been applied to all sites in the COSMOS-UK network. This is stated and discussed further in section 3 of the technical note. As mentioned in the response above, the COSMOS-UK website already utilises the revised background intensity correction. We very much hope that the methodology described in the technical note is explored by other ‘cosmos’ networks and agree that the COSMOS-Europe data set will provide an interesting opportunity for further analyses. We note that some of the sites in Figure A.3 of Bogena et al. 2021 (https://doi.org/10.5194/essd-2021-325) may display a similar trend to those seen within the COSMOS-UK network (e.g. Harrild, Rollesbroich and Ruraue).
3. “Although it uses standardised differences using the median, the correction factor derived from the relationship is still influenced by soil moisture dynamics (i.e. it would only perfectly work for a permanently wet site or a lake). I suggest excluding dry periods from the analysis to avoid dry bias.”
We agree that the wet periods contribute points with more stable soil moisture. In the attempt to avoid the dry bias, we applied a lower quartile regression which puts more weight towards counts corresponding to the wetter periods while still utilising all the available data. We feel this is more appropriate than attempting to identify wet and dry periods given the temperate maritime climate of the UK. This is discussed in section 2.1. In practice, the reviewer’s suggested approach and the lower quartile regression yield similar results for the COSMOS-UK network sites.
4. “The correction factor depends on the measurement period. It is unclear, how many years are needed for achieving a reliable correction factor.”
We discuss this aspect in section 3. Since the regression is performed between COSMOS-UK and Jungfraujoch (JUNG) neutron monitor counts, it is not so much the record length but the range of available JUNG counts which very much depends on the solar cycle. For instance, in the period between 2018 and 2021, there was not much variation in the JUNG counts. As a result, sites with data for this period only would not have a reliable estimate of G.
Citation: https://doi.org/10.5194/hess-2021-564-AC1
-
AC1: 'Reply on CC1', Magdalena Szczykulska, 06 Dec 2021
-
RC1: 'Comment on hess-2021-564', Anonymous Referee #1, 27 Dec 2021
The comment was uploaded in the form of a supplement: https://hess.copernicus.org/preprints/hess-2021-564/hess-2021-564-RC1-supplement.pdf
-
AC2: 'Reply on RC1', Magdalena Szczykulska, 17 Feb 2022
We thank the reviewer for the comments. Below are our responses.
- “The Authors present a revised incoming correction factor that could be used for improving soil moisture estimated with cosmic-ray neutron sensing (CRNS). The topic is interesting and timing and probably the study fits a technical manuscript in HESSD. The analysis is based on the extensive datasets collected within the COSMOS-UK sites and the manuscript is generally also well structured and written.”
We thank the reviewer for appreciation of our work and for the positive general comment.
- “[1] the study proposes a revised factor (eq.8) and the results are compared to the state-of-the-art (eq. 3). However, as far as I have understood, it is also well known an additional method (eq. 4) that so far it seems to be not widely adopted by the “CRNS community” but only used in some studies e.g., (Howat et al., 2018). So why do the Authors did not test this “not so common” revised method before proposing something new?”
These approaches were tested, but we did not report this in the paper. We will add a comparison of the different correction approaches and how they affect the corrected counts and volumetric water content (VWC) at the COSMOS-UK sites. In particular, we will provide a comparison between our approach and the approaches given in Hawdon et al., 2014 (i.e. gamma based on cut-off rigidity approach and a nearest neutron monitor based approach).
To add for clarity, Howat et al., 2018 give gamma for a particular site in Greenland using a neutron monitor (THUL) also located in Greenland, so their result is for that particular case study.
- “[2] as far as I have understood, the revised factor (eq.8) converges to what was currently available but not widely implemented (eq.4) not only in the form but, even more important, to the actual parameter i.e., G = 1.2 vs. gamma = 1.19 (See (Howat et al., 2018)). So, did I misunderstand or should the Authors be already satisfied by using eq. 4 without the need to propose a revised factor?"
Howat et al., 2018 obtain the value of gamma = 1.19 (or beta as stated in the paper) for a particular site in Greenland without explanation beyond ‘based on regressions to the global neutron monitor dataset’. Our study details how to derive G values at specific COSMOS-UK sites (there are 48 of them), so site specific incoming neutron intensity corrections. These site specific G values vary between 1 and 1.5 for the COSMOS-UK network. While the mean value across sites seem to converge to 1.2, it is not the case at individual sites.
We will clarify this aspect in the manuscript.
- “for the development of the new revised factor (eq.8) the Authors compare incoming neutrons from RNMS (e.g., Jungfraujoch) to CRNS neutrons locally collected at a soil moisture site (SMS) during period where it is expected low variability due to soil moisture changes. Namely, removing local influences due to soil moisture, variability in the neutron counts should then be related to incoming fluctuations. The Authors then compare these local fluctuations to the RNMS. It is well discussed that, if these fluctuations are not the same, on a longer term, should be due to different cutoff rigidity and altitude between the RNMS and SMS. But since also eq.4 was developed to account for these factors, from my understanding it should be not a surprise that this revised method converge to eq. 4. So, overall, it seems to me that the Authors simply analyzed some time series and found empirically what is already know and addressed in literature. I might be wrong but, if this is the case, I encourage the Authors to clarify and improve the manuscript to better convey the novelty of the study.”
The local fluctuations at SMS are different from RNMS due to the differences in cut-off rigidity and altitude, but also likely due to other differences, such as detector differences and site conditions. Whilst the CRNS detectors each have nominally the same response function as each other (but different to the RNMS), it is the interplay between the CRNS response function and the local SMS neutron environment (neutron energy spectrum) which can lead to site specific values of G (or gamma), even for sites at the same cut-off rigidity. For example Weimar et al., 2020 note that “The relative thermal contribution of the signal of the standard CRNP is in particular large for moist soil.” It is also worth noting that none of the detectors deployed for COSMOS-UK have any thermal shield, so will be susceptible to thermal neutron leakage. In the revised manuscript, we will properly discuss these and other effects (e.g. wet soil generated neutrons) which may affect G, and in turn suggest possible physical interpretations of G.
Eqn. 4 accounts for some of the differences between SMS and RNMS by normalising neutron intensities at time of interest to a reference value (I/Iref), but also via gamma. Hawdon et al., 2014 relate gamma to cut-off rigidities of locations of interest. This does not take into consideration the other above mentioned differences which do not necessarily cancel out via I/Iref. In Howat et al 2018, both the cosmic ray neutron sensor and the reference neutron monitor are located in Greenland. The authors are primarily concerned with elevation differences and find gamma = 1.19 for that specific case study.
COSMOS-UK sites and the Jungfraujoch neutron monitor have very different location characteristics when compared to the study conducted by Howat et al., 2018. We also do not try to say that G has a single value for the whole COSMOS-UK network, but rather we investigate each individual site and find that G varies between 1 and 1.5, and we only report in L 144 that the average value of G is 1.2. We will clarify this point.
The novelty is that we calculate G for each individual COSMOS-UK site using directly neutron count data from the local site of interest (so we do not have to make assumptions about what causes the differences). We take a data-centric approach, and find the relationship directly from the data. While these neutrons are subject to soil moisture variability, we use a lower quartile regression fitting to put more weight on the wetter points (lower count rates) which have more constant soil moisture in the case of the UK. This method successfully reduces the unphysical drying trend in the CRNS soil moisture observations at wetter COSMOS-UK sites, for instance Glensaugh, when G=1 and provides better performance in this case when compared to G based on cut-off rigidity as given by Hawdon et al., 2014. This is an indicator of other influences which, as above, we will discuss further in our revised manuscript.
We will improve the manuscript to stress where the novelty of this study lies.
- “[4] my last comment is related to the general assumption that incoming neutron counts from a RNMS adequately represents the relevant incoming neutron flux at the SMS and the revised factor accounts for some additional differences (L61-63). Based on that, the Authors conclude and suggest (L165-171) several research activities that could be performed for further improvements. Indeed I agree that using incoming fluctuation from RNMS is a first order correction that has to be considered also for CRNS applications. This assumption has however two shortcomings that should be considered. First, time series at RNMS need also several corrections that are still under investigations and the focus of current research activities and improvements. Thus, these time series are not error-free. Second, some local incoming fluctuations at SMS are not detected by RNMS. Thus, these time series could not well inform local incoming fluctuations even in the case they were error-free. For these reasons, personally I do not see a good suggestion to push much effort in improving a method that is based on input (i.e., the time series at RNMS) that has these drawbacks. In contrast, I have seen that the use of alternative detectors installed directly at the SMS for the detection of incoming fluctuations has been suggested in literature using e.g., muons detectors (Stevanato et al., 2019; Stowell et al., 2021) or neutron spectrometers (Cirillo et al., 2021; Fersch et al., 2020). Personally, I believe that improving and working with these approaches could be much more valuable suggestions for further studies and developments instead of improving the manipulation of no error-free and non-representative time series from RNMS. Alternative, a discussion of the added value of the present revised factor in comparison to abovementioned approaches should be reported.”
We thank the reviewer for this helpful comment – yes we agree that (as above) there will be neutron energies (epithermal and thermal) which the CRNS detects, but not the RNMS. Indeed, for future work, the promising development of local reference detectors should be considered as a significant advancement, where affordable. We will add a brief discussion of these approaches.
However, given that for COSMOS-UK alone, we now have more than 5 years of data for 48 sites, with no in situ reference measurements, it is of great value to improve the long-term bias correction of the derived soil water content of these historical datasets, using the RNMS approach. Furthermore, our analysis of G across the whole of the COSMOS-UK network is valuable to highlight what improvements may be required (such as the deployment of reference detectors), and their relative cost-benefit.
Citation: https://doi.org/10.5194/hess-2021-564-AC2
-
AC2: 'Reply on RC1', Magdalena Szczykulska, 17 Feb 2022
-
RC2: 'Comment on hess-2021-564', Anonymous Referee #2, 20 Jan 2022
This manuscript "Technical note: A revised incoming neutron intensity correction factor for soil moisture monitoring using cosmic-ray neutron sensors" deals with the problem of incoming neutron radiation correction for cosmic-ray soil moisture sensors.
The authors claim to derive a "revised correction factor" using an amplitude scaling technique that has already been proposed in the literature before, while suggesting the use of median instead of reference count rates. In the same study, the authors find that the proposed median count rate often is insignicifantly different from the reference count rate, that it turns the whole approach to be highly dependent on the individual measurement period, and that it is less generalizable. Good agreement has been found by comparing the overall average at tens of sites with findings from other studies, but the reader is still left with no clear idea about the added value provided by this study.
While the problem statement is relevant for the CRNS community, the novelty of the approach as well as the quality of the explanations are low. However, there is a potential to accept the manuscript with major revisions, particularly by further elaborating on the scientific foundations, on the statistical quality, and on the demonstration of measurement results.
# Major concerns
- The HESS manuscript type "Technical notes" requires to "report new developments, significant advances, and novel aspects of experimental and theoretical methods and techniques which are relevant for scientific investigations within the journal scope." The addressed problem and topic are relevant for studies published in HESS, but it is not clear how to identify the significant advancement and novel aspect of this study. While the manuscript "should be short (a few pages only)", the authors elaborate unnecessarily on less related aspects, such as the importance of soil moisture for hydrology or drying and wetting characteristics of individual sites. I would suggest to provide a more concise document with a strong focus only on the *technical* aspect of incoming neutron correction:
- what is the expected sensitvity of CRNS to these corrections (cite e.g. Baroni et al. 2018 (JoH)),
- provide fundamental derivation of your approach and discuss differences to previous approaches, and
- statistically sound evaluation (using all your individual site data, proper statstical measures, and uncertainty analysis).
- The fundamental explanation of the involved processes and physics have been marginally addressed. The authors seem to treat the correction approach as yet-another function without discussing the meaning of the analytical form or the physical processes involved. For exampe:
- Why should detected low-energy neutrons in any way behave proportional to incoming high-energy neutrons? What fraction of the detected signal is direct and indirect radiation and how does this change with site conditions? The authors you are citing have answers to that. This would help readers to actually understand your approach and this may have direct implications on the interpretation of the performance of your results at individual sites.
- Can we expect this approach to be different for different regions on Earth? What has this to do with latitude or cut-off rigidity? The answers may have implications on the performance of your distributed sensors and should be discussed using the empirically obtained values for G.
- What could be the physical reason for the necessity of an amplitude scaling in the form of G*(I/I_m-1)? This has implications on the choice of G and its meaing, and could answer whether situations could occur with G<1, too.
- Also put the whole idea into context and note that it is about correcting for solar activity fluctuations and mention that the scale of those variations can be from years (solar cycle) to days (FDs, GLEs). This has direct implications on the conclusions that you could and could not draw from your analysis, e.g. by choosing a "monthly average", with which one would have no chance to resolve any short-term variations.
- The authors determine gamma empirically, while there are high chances of fitting through soil moisture variations. A more "safe" approach would be to calibrate gamma on quite periods (low solar activity) and at sites with constant soil moisture (desert, concrete, or lakes). Some authors (which you have cited) have already traveled along this path and it could be helpful to report on advantages and disadvantages of the various approaches.
- The authors spend two pages to derive their concept, and end up with an equation which is equal to the one from literature which has been presented already in the introduction. It is claimed that the new equation is now "revised", for two reasons:
- it uses median count rates instead of a reference count rate (although this option has already been suggested by Zreda et al. 2012 and others). Interestingly, in your own study you tell that the difference between the use of I_m or I_ref often is insignificant...
- the scaling factor G now is empirical (although this has been done already by other authors which are cited elsewhere in this very study), while they are hardly compared to values from existing approaches. In addition to that, even the relationships to the cut-off rigidity, as proposed by Hawdon 2014, for instance, are empirical. So instead of empirically building on their theory, you reduce the complexity of finding G even more by applying a purely empirical fit. This comes with the risk of low transferability and generalizability of the results.
- Under these circumstances, I strongly suggest to properly discuss previous literature and not to sell the current development as a "revised correction approach". This study for me looks rather like a case study, where some of the existing suggestions have been tested on the UK network data. It is surely important to do that, but it would require a different storyline.
- L143: If the results are dependent on the variation of the incoming radiation, then it should be easy to evaluate this by testing the approach during high and low solar activity. Something that could be done at the COSMOS-UK Network.
- L147: "Preliminary studies show no simple relationship with ... latitude, altitude, or rigidity"
In the last sentence before the conclusions you are mentioning the most important aspect, in my view, which should be central to this study. Consider elaborating on building a theoretical basis for your approach, which involves confrontation with the mentioned quantities. This would help to develop this approach further towards a generalized and transferable method.
# Specific comments:
- Common and vague language is used quite often, rather that clear and factual language. E.g., L17 "perhabs less obvious", L85 "gives a gradient G, not of 1, but 1.5", L122: "change is very much greater".
- L33: the Desilets equation is semi-empirical, the paramters have been derived from observations only. In general, this paragraph can be much shorter. The choice of the conversion function and vegetatiion or organic carbon corrections are of lesser importance here.
- Fig 3: The comparison of time serieses with different G is not insightful. Please provide reference data, e.g. from the neutron monitor (to demonstrate a changing correlation to their signals) or soil moisture data (to demonstrate better match).
- Fig 4: Now you show TDT data, but only with G=1, so there is no chance to compare the approaches and to evaluate the performance of your idea. Moreover, consider showing a weighted average using different depths, instead of only 10 cm data.
Citation: https://doi.org/10.5194/hess-2021-564-RC2 -
AC3: 'Reply on RC2', Magdalena Szczykulska, 17 Feb 2022
We thank the reviewer for the comments. As a general note we would like to stress that this work was primarily motivated by an observation of a spurious long-term trend in the derived volumetric water content (VWC) at some of the wetter sites in the COSMOS-UK network, for instance Glensaugh. It was identified that this trend was correlated with the Jungfraujoch counts used in the background neutron intensity correction, then of the form I(t)/Iref (so G = 1 in eqn. 8). We found that after applying other reported approaches (Hawdon et al., 2014), the long-term trend would still remain. So we took a data-centric approach and directly calibrated the normalised fluctuations of Jungfraujoch counts with the CRNS counts at the specific site of interest to find G for that specific site. We also took caution to not fit through soil moisture variations by applying a lower quartile regression when deriving our site specific Gs. Our method significantly reduced the observed long-term trend.
We of course appreciate the reviewer’s request for clarity, providing comparison of approaches and more scientific understanding of how G varies with respect to different physical processes. We will address these points in the new version of the manuscript. We also apologise for the misstatements in terms of what is revised: the form of the correction given in eqn. 8 appears in the literature (although with almost no explanation where it originates from) and it is the method for obtaining the site specific G that is revised. We will clarify these aspects in our manuscript.
Below are our responses to the individual comments.
1. “The authors claim to derive a "revised correction factor" using an amplitude scaling technique that has already been proposed in the literature before, while suggesting the use of median instead of reference count rates. In the same study, the authors find that the proposed median count rate often is insignicifantly different from the reference count rate, that it turns the whole approach to be highly dependent on the individual measurement period, and that it is less generalizable.”
We agree that the general form of the correction in eqn. 8 is not novel and has been proposed in the literature. The value of the paper comes not in the general form of the correction, but in the way we estimate G – using local cosmic ray neutron sensor data combined with a lower quartile regression fitting to fit through wetter points associated with more constant soil moisture in the UK. We apologise for the misstatements in the manuscript, and we will revise it to clarify this point.
To address the point of poorer generalizability due to the individual measurement period dependence, the following are two reasons why such dependence arises:
- G values are derived by comparing neutron count fluctuations from the reference neutron monitor (In) and local neutron fluctuations from the Cosmic Ray Neutron Sensor (Nn) and therefore there is a necessity to have sufficient variations in In to establish a reliable relationship between In and Nn. This is linked with the solar activity more than the time-period – it is already explained in the manuscript in L 141 and L 142, but we will clarify this further. Small variations in In indicate low solar activity in which case the intensity correction is of less significance. So while the method requires sufficient variations in the incoming background neutrons, it is also mostly not required when this is not the case.
- The CRNS data carry soil moisture signal and therefore the measurement period needs to be sufficiently long so that it is possible to fit through relatively constant soil moisture points (in the case of the UK, these are the points associated with wet periods). So it is rightly noted that for a short time series G will carry more uncertainty, but since the method is aimed at correcting long-term trends, the users requiring this method will most likely have a long time series available.
2. “Good agreement has been found by comparing the overall average at tens of sites with findings from other studies, but the reader is still left with no clear idea about the added value provided by this study.”
The original motivation for this study was an observation of a spurious long-term drying trend present at wetter COSMOS-UK sites, for instance Glensaugh, when G = 1. Using G = 1.22 based on cut-off rigidities, as given by Hawdon et al., 2014, also does not remove this trend completely. The added value of our study is that we calculate G for each individual COSMOS-UK site using directly neutron data from the local site of interest. While these neutrons are subject to soil moisture variability, we use a lower quartile regression fitting to put more weight on data points associated with more constant soil moisture. This approach corrects the observed unphysical drying trend present at some of the COSMOS-UK sites.
In the study conducted by Howat et al., 2018, a value of gamma = 1.19 is given for a specific location in Greenland and the authors use a neutron monitor (THUL) also located in Greenland. Gamma given in that work is therefore for that specific case study and is not expected to be the same at different locations. While we note in the manuscript that our COSMOS-UK site average G value coincides closely with the value of gamma obtained by Howat et al., 2018, we do not use it in the background neutron intensity correction. We always use the site specific value which can vary significantly from site to site (between G = 1 and G = 1.5).
We will improve the manuscript to stress more clearly the novelty of our approach and its implications on the derived soil moisture data. We will also add to the scientific understanding of how G varies with different physical parameters.
3. “what is the expected sensitvity of CRNS to these corrections (cite e.g. Baroni et al. 2018 (JoH))”
We will report on the sensitivity of CRNS to the background neutron intensity correction factor based on the literature.
4. “provide fundamental derivation of your approach and discuss differences to previous approaches, and”
If the reviewer is asking about how we derive G, then yes, we will clarify this further in the manuscript. We will also discuss the differences to other approaches.
5. “statistically sound evaluation (using all your individual site data, proper statstical measures, and uncertainty analysis)”
We will provide uncertainty estimates for individual site Gs.
Regarding the aspect of “using all your site data”, we did use all site data for the site specific G estimation. If what is meant is spatial analysis, then we will analyse our data in the context of understanding why G varies from site to site.
6. “The fundamental explanation of the involved processes and physics have been marginally addressed. The authors seem to treat the correction approach as yet-another function without discussing the meaning of the analytical form or the physical processes involved.”
We took a data-centric approach without complex modelling of the physical processes involved, but we will improve the manuscript by discussing the physical processes.
7. “Why should detected low-energy neutrons in any way behave proportional to incoming high-energy neutrons? What fraction of the detected signal is direct and indirect radiation and how does this change with site conditions? The authors you are citing have answers to that. This would help readers to actually understand your approach and this may have direct implications on the interpretation of the performance of your results at individual sites.”
These of course are very valid questions, which we had originally decided to defer to later work, but we now see that this work is somewhat incomplete without at least discussion of the relevant physical processes and we agree that these would help to inform the interpretation of the different G values across all 48 sites.
We will discuss the energy spectral response functions for the CRNS versus the RNMS with respect to the question of proportionality between the incoming low and high-energy neutrons, and the relevance of site conditions for the correction. As above, will discuss properly the literature on this topic.
8. “Can we expect this approach to be different for different regions on Earth? What has this to do with latitude or cut-off rigidity? The answers may have implications on the performance of your distributed sensors and should be discussed using the empirically obtained values for G.”
We will show the dependency that our empirical G values have on cut-off rigidity.
9. “What could be the physical reason for the necessity of an amplitude scaling in the form of G*(I/I_m-1)? This has implications on the choice of G and its meaning, and could answer whether situations could occur with G<1, too.”
We will discuss the form of the correction and try to elaborate why this form can be adopted.
10. “Also put the whole idea into context and note that it is about correcting for solar activity fluctuations and mention that the scale of those variations can be from years (solar cycle) to days (FDs, GLEs). This has direct implications on the conclusions that you could and could not draw from your analysis, e.g. by choosing a "monthly average", with which one would have no chance to resolve any short-term variations.”
In our study, we are concerned with the long-term trend (the solar cycle) that was observed at some of the wetter sites in the COSMOS-UK network, for instance Glensaugh. So we choose monthly averages in figures 3 and 4 to show the impact of the correction on long-term trends.
11. “The authors determine gamma empirically, while there are high chances of fitting through soil moisture variations. A more "safe" approach would be to calibrate gamma on quite periods (low solar activity) and at sites with constant soil moisture (desert, concrete, or lakes). Some authors (which you have cited) have already traveled along this path and it could be helpful to report on advantages and disadvantages of the various approaches.”
To prevent from fitting through wide soil moisture variations, we use a lower quartile regression when comparing the normalised neutron count fluctuations from the cosmic-ray neutron sensor (Nn) and the reference neutron monitor (In) to obtain G. This ensures that we can use all the data, but the fitting is biased towards wet periods which have more constant soil moisture in the UK. We will stress it more clearly in the manuscript.
We will emphasise more clearly that we can only get good G estimates, over a sufficient change in RNMS counts e.g. with significant change in solar activity.
Some of the co-authors have also travelled this path to a degree with COSMOS-Rover experiments. However, given that cut-off rigidity plays a role, we do not have convenient constant soil moisture sites close to the SMS locations. There are a few exceptions, but even translating results from concrete or 100% water (lake) presents issues. We will discuss these approaches along with other onsite reference measurement approaches.
12. “The authors spend two pages to derive their concept, and end up with an equation which is equal to the one from literature which has been presented already in the introduction. It is claimed that the new equation is now "revised", for two reasons:
- it uses median count rates instead of a reference count rate (although this option has already been suggested by Zreda et al. 2012 and others). Interestingly, in your own study you tell that the difference between the use of I_m or I_ref often is insignificant...
- the scaling factor G now is empirical (although this has been done already by other authors which are cited elsewhere in this very study), while they are hardly compared to values from existing approaches. In addition to that, even the relationships to the cut-off rigidity, as proposed by Hawdon 2014, for instance, are empirical. So instead of empirically building on their theory, you reduce the complexity of finding G even more by applying a purely empirical fit. This comes with the risk of low transferability and generalizability of the results.”
The revision comes in the way we calculate G using the site specific CRNS data. Some previous authors used neutron monitor data (different detector and not site specific to where the cosmic-ray neutron sensor is located) to relate gamma (or G) to other physical parameters (Hawdon et al., 2014 relates gamma to cut-off rigidity and Howat et al., 2018 relates gamma to elevation, but with no formula given in the latter case).
While both approaches are empirical, our approach directly finds G for the site of interest using that same site data and without making assumptions of what physical parameters are involved – we directly use the CRNS data from the site of interest and account for soil moisture variability. Using the CRNS data has also the advantage of calibrating G for detector differences that exist between the CRNS and the reference neutron monitor, even if it is only a first order approximation.
We discuss the generalizability aspect in point 1 of this response and of course parameterisation of G (if all relevant factors are included) would increase the generalizability of the results and improve scientific understanding of the processes involved. We will address this point by providing a discussion on the physical influences on G. We will also attempt to build on the Hawdon et al., 2014 approach for G to explain our results at the COSMOS-UK sites.
We would also like to note that the work presented here has a very pragmatic focus: to bias correct long-term UK soil moisture time series data, for use by the hydrological and land surface climate model communities. We welcome others to test the applicability to other regions, to bias correct their historical data. The results will be specific to the make and model of the CRNS, and to a degree the installation conditions, as well as the surrounding environment.
13. “Under these circumstances, I strongly suggest to properly discuss previous literature and not to sell the current development as a "revised correction approach". This study for me looks rather like a case study, where some of the existing suggestions have been tested on the UK network data. It is surely important to do that, but it would require a different storyline.”
We apologise for the unintentional mis-selling – and we agree that we present an empirical application of the earlier published correction. We do consider that we present important details of the exact application, to allow an empirical fit using the on-site CRNS data (without underlying assumption of the causes) and to weight our fits to wet data to find G. We will strengthen the paper, as you suggest to provide more analysis of the results and more insight, as a UK case study.
14. “L143: If the results are dependent on the variation of the incoming radiation, then it should be easy to evaluate this by testing the approach during high and low solar activity. Something that could be done at the COSMOS-UK Network.”
We perhaps should have made it clearer: since G is a slope of a linear regression (lower quartile regression) between x (In) and y (Nn) there needs to be sufficient variations in x to find a reliable estimate of the slope. Figure 5c shows how the T values of G estimates for the COSMOS-UK sites increase when the variation in Jungfraujoch counts increase.
15. “L147: "Preliminary studies show no simple relationship with ... latitude, altitude, or rigidity". In the last sentence before the conclusions you are mentioning the most important aspect, in my view, which should be central to this study. Consider elaborating on building a theoretical basis for your approach, which involves confrontation with the mentioned quantities. This would help to develop this approach further towards a generalized and transferable method.”
Agreed, as in earlier reviewer responses, we will improve this work by properly discussing the physical processes that may contribute to the fitted values.
16. “Common and vague language is used quite often, rather that clear and factual language. E.g., L17 "perhabs less obvious", L85 "gives a gradient G, not of 1, but 1.5", L122: "change is very much greater".
We will improve the quality of the language used.
17. “L33: the Desilets equation is semi-empirical, the paramters have been derived from observations only. In general, this paragraph can be much shorter. The choice of the conversion function and vegetatiion or organic carbon corrections are of lesser importance here.”
According to Desilets et al., 2010, the calibration curve was obtained by “fitting ground-level neutron fluxes simulated in MCNPX [Pelowitz, 2005] to the shape-defining function”, so we are not sure why the reviewer suggests that “the parameters have been derived from observations only”. And yes, we can shorten this paragraph.
18. “Fig 3: The comparison of time serieses with different G is not insightful. Please provide reference data, e.g. from the neutron monitor (to demonstrate a changing correlation to their signals) or soil moisture data (to demonstrate better match).”
We will revise the plot and add neutron monitor data to it.
19. “Fig 4: Now you show TDT data, but only with G=1, so there is no chance to compare the approaches and to evaluate the performance of your idea. Moreover, consider showing a weighted average using different depths, instead of only 10 cm data.”
Yes, we will add a plot with the revised G to figure 4. Unfortunately, we only have TDT data for a depth of 10 cm, so we will not be able to show a weighted average.
Citation: https://doi.org/10.5194/hess-2021-564-AC3
- The HESS manuscript type "Technical notes" requires to "report new developments, significant advances, and novel aspects of experimental and theoretical methods and techniques which are relevant for scientific investigations within the journal scope." The addressed problem and topic are relevant for studies published in HESS, but it is not clear how to identify the significant advancement and novel aspect of this study. While the manuscript "should be short (a few pages only)", the authors elaborate unnecessarily on less related aspects, such as the importance of soil moisture for hydrology or drying and wetting characteristics of individual sites. I would suggest to provide a more concise document with a strong focus only on the *technical* aspect of incoming neutron correction:
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 688 | 269 | 33 | 990 | 20 | 27 |
- HTML: 688
- PDF: 269
- XML: 33
- Total: 990
- BibTeX: 20
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
