Use of thermal signal for the investigation of near-surface turbulence

Abstract. Organised motion of air in the roughness sublayer of the atmosphere was investigated using novel temperature sensing and data science methods. Despite accuracy drawbacks, current fibre-optic distributed temperature sensing (DTS) and thermal imaging (TIR) instruments offer frequent, moderately precise and highly localised observations of thermal signal in a domain geometry suitable for micrometeorological applications near the surface. The goal of this study was to combine DTS and TIR for the investigation of temperature and wind field statistics. Horizontal and vertical cross-sections allowed a tomographic investigation of the spanwise and streamwise evolution of organised motion, opening avenues for analysis without assumptions on scale relationships. Events in the temperature signal on the order of seconds to minutes could be identified, localised, and classified using signal decomposition and machine learning techniques. However, small-scale turbulence patterns at the surface appeared difficult to resolve due to the heterogeneity of the thermal properties of the vegetation canopy, which are not immediately evident visually. The results highlight a need for physics-aware data science techniques that treat scale and shape of temperature structures in combination, rather than as separate features.


I thank the two reviewers for their helpful comments and suggestions. Detailed information on the changes made to the manuscript can be found below. As some sections were moved and restructured, also the figure order changed accordingly. In addition, I found errors in Table 1 that are now corrected.
I look forward to learn how we can proceed with the revised manuscript.
Kind regards,

General Comment
The author presents a measurement technique which combines a thermal imaging instrument and distributed temperature sensing system in order to monitor spatial and temporal fluctuations of the temperature. Those measurements are supplemented by point observations of the temporal wind fluctuations, which are used in order to provide a more thorough insight of the local ambient conditions. The article is in general well written and the measuring techniques and the data analysis are presented in detail.
However, I think that the results presented in the submitted manuscript are lacking an assessment of how accurate and precise the estimation of the measured temperature fluctuations are. This is discussed briefly in the text and a qualitive comparison could be performed visually from the results in Figure 3. However, it will contribute to the assessment of the measuring capability of this setup if a direct comparison with the reference sensors is performed.
Response: Assessments of measuring capabilities were the subject of earlier studies and I chose not the repeat those here to avoid redundance. As the reviewer points out, the reader may be inclined to use Figure 3 for a qualitative comparison of temperature observations, but the presented detail is not well-suited for a performance assessment. I agree that the addition of a comparison between the temperature methods would be helpful. I further suggest to add text to the interpretation of the Alan variance results ( Figure A7), which had not been included in previous studies and reveals insights for possible improvement of the calibration of DTS data. Figure R1: Timeline of the temperature comparison between Ta (aspirated), Tc (DTS cable) and Ts (sonic anemometer).

Figure R2: Comparison between Ta and Tc (left) and Ta and Ts (right).
A comparison between temperature measurements by the different systems was made, using sensors that were closely located in space. For the cross-comparison, sensors at 3 m height were selected nearest to the Ta observations (aspirated sensor profile) and a period with approximately equal number of daytime and nighttime hours (12:00 19 Jul 2016 to 04:00 20 Jul 2016). The results show that both Tc and Ts are linearly correlated with Ta, but did not exactly match the aspirated temperature in absolute value and high frequency variance ( Figure R1 and R2). The Ta sensors exhibit a slower thermal constant due to the size of the metal sensor rod, leading to a dampened high-frequency response compared to Ts and Tc. Furthermore, Ts was cross-calibrated among the EC sensors prior to this experiment, but not calibrated to a temperature reference scale. This shows a need to include further calibration or Ts in future experiments. However, differences between aspirated and non-aspirated sensing methods are inevitable and explain part of the deviation between sensing systems seen here.
A comparison between Tc and Ts shows linearity between the signals, which appeared to improve in absence of direct sunlight during the night ( Figure R3). However, the computed coherence between Tc and Ts deteriorates at high frequency (f > 0.1 Hz, dt < 10 s; Figure R4). This can partly be explained by noise in one or both of the sensors, as well as small differences between the location and path length of the sensors being compared.  Moreover, I think that it is not explained clearly the reason for selecting the specific shape and size for the experimental setup. This information is going to be useful for understanding and interpreting the results of this study.
Response: Thanks for the suggestion.
In a nutshell, the design was intended to support studies on coherent structures, advection processes and conditional sampling methodology. Both as empirical study and in combination with fluid dynamics models. Placement of the wind sensors on the corners and at the center was intended for the determination of a representative wind vector at the walls of the box. It was thought that placing tripods at the center of a wall would result in more uncertainty. The dimensions and shape were primarily limited by the maximum range of the DTS instrument, which can support up to 1.8 km of optical fibre per channel. The DTS profile height was extended to reach sufficiently above the sonic anemometers at 3 m. In an initial design, the guyed mast would be taller and placed in the center. A compromise had to be made during deployment and the mast was moved to a corner. This also had obvious implications for the field of view of the TIR system.
The setup was not a stand-alone experiment. Other experiments were conceptualized to use the setup, as part of cooperative research during the experimental campaign (ScaleX; https://scalex.imk-ifu.kit.edu/; line 70-76). To name a few: • Adjacent to the setup, a horizontal gradient of Tc was measured over a distance across the shallow valley, and in parallel to a transect of sonic anemometers during a different period (Mauder and Zeeman 2018) • The combined DTS setup was included in an area of TIR mapping by UAV (Brenner et al. 2018) to evaluate an approach similar to the aquatic study cited in the review comments below (Dzara et al). Response: Thank you for the comment. The following change was made: 'Organised motions of air in the roughness sub-layer of the atmosphere …'.
Line 6. What is meant with the term "Variance Events"?
Response: Thanks for pointing out the use of jargon. The term variance event is used to distinguish signal in the time series without (obvious) periodic pattern, but representing significant excursions from a mean or trend. I suggest to rephrase 'variance events' to help the reader.
'Events in the temperature signal...' Line 9. I suggest replacing the "with the naked eye" with visually.
Response: Thanks for the comment, I agree. The change was made as suggested.
Line 13 -14. The author states that "the available methods to determine energy and scalar fluxes from terrestrial land surface are relatively imprecise due to a multiscale of irregularities in the land surface and the turbulent transport mechanisms". I think that this is statement is not very clear. This imprecision originates from limitations in the precision of the measuring methodologies or does the imprecision refer to the need for spatially distributed measurements?
Response: The line refers to limitations in the measuring methodologies, which in part relate to spatiotemporal irregularities in surface-atmosphere interactions.
The methods are precise, just not applicable everywhere and all the time. As a consequence, there is a pattern, perhaps bias, as to where and when energy and scalar fluxes are observed today using micrometeorological techniques. In general, complex terrain is avoided, and periods with stable atmospheric conditions (e.g., night) or hydrometeorological events (e.g., rain, fog) are excluded. This introduces a systematic uncertainty.
Many researchers have asked the question: 'what [are we] missing outside the applicable range of the methodology and assumptions?'. Spatial distributed measurements may not be the final answer, but I think they could help with the assessment of processes and the development of empirical methods.
Line 24. An abbreviation should be added after the "roughness sublayer". Later in the document (line 26) is referred to as RSL.
Response: I agree, thank you for noting the omission. The definition of the abbreviation was made as suggested by the reviewer.
Line 45. What it is meant by the following statement: "the quantification of Tb outside a controlled laboratory environment is a challenge, in and of itself"?
Response: There are many possible sources of interference in thermal imaging, which in a controlled laboratory environment can be observed and corrected for. In field studies this is much more difficult.

The sentence was modified:
'The quantification of \Tb\ outside a controlled laboratory environment is a challenge, in and of itself; there are many possible sources of interference in thermal imaging, which in a controlled laboratory environment can be anticipated and corrected for.' Line 57. I recommend that this statement about the goal of this study is also mentioned in the abstract. It will give a clearer idea to a reader about the objective of this study. Please note that I am not neither the author or any of the co-authors of the aforementioned study.
Response: Thank you, I was not aware of the study. The suggested study used the two techniques in an aquatic environment. Please note that TIR in the study by Dzara et al was used to generate thermal maps (one per season), which were then matched and compared to statistics of a DTS time series. In contrast, this manuscript presents an approach where DTS and TIR are both used as high-resolution spatial time series.
Lines 65: What does ICOS stand for?
Response: Thanks, the definition is indeed missing. The acronym was added to the text: 'The study was conducted at the DE-Fen station, Fendt--Peissenberg, Germany, which is a TERerstrial ENvironment Observatories (TERENO) and Integrated Carbon Observation System (ICOS) core site …' Lines 68 -69: In these lines the author gives details about elements of the landscape surrounding the experimental area. It is not clear how this information is relevant to the study. I suggest that the author explain briefly the impact of the landscape to the experiment presented in the manuscript or remove that part.
Response: I fully agree with the reviewer's suggestion to remove this part.
Line 70: What are the ScaleX campaigns?
Response: I agree that this concept is not properly introduced and suggest the following.
'During intensive field campaigns at DE-Fen, additional experiments were conducted for the investigation of scale interactions between the atmospheric boundary layer and the surface, as well as validation of measurement techniques (ScaleX; …).
Line 74: Why is the period between 18 -22 Jul 2016 considered as a reference period?
Response: This is indeed not a necessary qualification. The word 'reference' was removed from the sentence.
Line 75: What was the purpose of the UAV use? And how could they have an impact on this study?
Response: UAVs were used for mapping surface brightness temperature at a larger spatial scale and for in situ measurement of wind field and air quality properties. Those studies were also part of ScaleX and referenced in the paragraph text. In some instances, horizontal transects were flown above and upwind of the setup. Particularly the heavy airborne platforms, e.g., carrying a gas analyzer, generated a downwash jet that could be sensed at some distance. Also, some UAV operations required teams of people moving in and out of the field, e.g., for hourly off-site charging of batteries. The flight tracks were recorded in detail, the movement of people and vehicles not.
Added to the text: …'perturbing the air by down wash and frequent transit through the area.' Line 78: What does EC stand for?
Response: The definition used is 'Ultrasonic anemometer (EC)' (line 77). EC refers to the Eddy Covariance technique in which these instruments are used. No changes were made to the text.
Line 78: I think that it would be very helpful for a reader if the author specify that is the figure 1c and table 1.
Response: I fully agree. The references were updated as suggested.
Line: 82: What was the reasoning for the number of sonic anemometers used, the selection of the locations of the tripods and the heights of the sonic anemometers?
Response: This was limited by available hardware at the time. Ideally, only 3-axis sonic anemometers would be used. A compromise had to be made for the number of 3-axis sonic anemometers deployed here and elsewhere during the campaign. The height of the 3-axis type instruments on the tripods was chosen to be similar to the ICOS station and other studies using the Eddy Covariance technique on permanent grassland (including DE-Fen; see also the study by Mauder and Zeeman cited on line 80). There are sensitivity limitations for working with current model ultrasonic anemometers close to the surface. A level of 2-axis sonic anemometers closer to the ground (0.25 m) was planned but the appropriate mounting hardware could not be arranged in time for deployment.

Details were added to section '2.5 Design considerations'.
Also, from the Figures 1 and 2 it is visible that the sonic anemometers were located between the supporting poles of the DTS mast. Could there be any interference to the sonic anemometers measurements acquired during the period selected in this study from wakes generated from the supporting poles?
Small-scale interference in the wake is possible. The distance from the DTS masts to the EC profiles (on tripods) was 3 m. DTS masts had a diameter of 0.1 m. Increasing the distance would have required for the suspension cable to be mounted higher and with larger tension force to keep the steel cable straight. This could not (safely) be realized during the deployment.

Details were added to section '2.5 Design considerations'.
Line 97: Where was the TIR system pointed to?
Response: The TIR system was pointed to the ground at a slanted angle to include as much surface within the DTS box as well as static objects for georeferencing. The guyed lattice mast was planned to be taller, but had to be kept below 10 m for safety of nearby glider planes. This limited our options for the camera viewpoint during the deployment.

Details were added to section '2.5 Design considerations'.
Line 101: What is meant that the location was determined in post-processing?
Response: Thank you for the comment. This means that it required a georeferencing step as described in the text below the line and Appendix A. Change the text now make this clear.
'Each EC, DTS and TIR record was stored with an accurate time stamp and locations were georeferenced in post-processing. The calibration and georeference details are provided in Appendix A.' Line 116. The air temperature (Ta) is mentioned here, but it is only discussed how it is measured in Appendix A3.3. I would suggest a brief statement about those measurements also in section 2.4.
Response: I agree. This is an issue and I agree with the suggested solution.
Added sentence: 'Reference air temperature measurements were made using resistance temperature devices in fan-aspirated enclosures (Table 1; Appendix A)' Additionally, regarding Figure 3. What is the sampling frequency of the time series presented in Figure 3?
Response: Those are 1 min averages in all panels. I suggest to change the caption accordingly to better inform the reader.
Changes were made to include the frequency of the time series in the captions of Fig 3 and Fig 4. How is the Tc estimated at the presented heights? Is it the average over all the four sides of the box?
Response: Yes, it is the average over all Tc profiles.
Which sonic anemometer's data is being used in the Figure 3 c?
Response: Only the 3-axis sonic anemometer models provide a measurement for Tv, hence these are data for 3.0, 6.0 and 9.0 m height.
Line 127. The author states that "some turbulence statistics were rarely acceptable … " What is it meant by the words "some" and "acceptable"?
Response: For the application of the eddy covariance technique, it is currently recommended to perform a number of (self-)validation tests, e.g., based on stability, stationarity and friction velocity. The test results can be simplified as a quality classification for the averaging period. During the nights it was rare to find the quality of averaging period results classified as 'acceptable'.
I suggest to rephrase this to 'eddy-covariance flux computations rarely produced acceptable results…' Line 129-130. It is not clear how what are the assessment criteria used here to assess the quality of the flux computations.
Response: The assessment criteria may be the same, the computation to derive a stability classification is different.
Line 132. How were the temperature gradients calculated?
Response: Thank for noting the omission. This is indeed not mentioned clearly. The definition for temperature gradient was added to the list in Appendix B.
Line 136. Figures 3a-c allow a visual comparison of the time series. However, there is a lack of a statistical comparison of the different methods (e.g. correlation, mean absolute error). I suggest that the author elaborate more on this part.
Response: I agree. Please see the response to the General Comments.
Line 140. In Figures 4b-c time series of the normalized by the Obukhov length scale height and the friction velocity are presented. Measurements from which sonic anemometer were used for those calculations. Which criterion has by used to assess the atmospheric stability is stable or unstable?
Response: Only results from the 3-axis sonic anemometers can be used for the computation.
Details about height were added to the caption of the figure 4.
Line 145. Is it the air temperature or the cable temperature presented in Figure 6?
Response: These panels are derived from cable temperature. I think this is sufficiently clear from the caption and legend.
Line 189. The DTS measurement set-up has a rectangular shape. What is meant here the mean wind was mostly aligned to the set-up?
Response: Thanks for the comment. I now recognize the wording can lead to confusion for the reader and this aspect should be rephrased. What was meant is that for a period of time the mean wind was either perpendicular or parallel to the walls of the box.
The sentence was modified for clarity: 'During the daytime animated period, the mean wind is mostly aligned with the DTS measurement setup, i.e., either perpendicular or parallel to the walls of the boxshaped array, with mean flow from a SSE direction. Response: Thanks for pointing this out.
The surface was not homogenous in terms of TIR signature. Some signal in the TIR image time series were revealed when and where the background (the surface) and foreground (air that had interacted with the surface upstream) show a different heat signature. Therefore, motion was revealed from hot air advecting away from relatively hot areas in the plant canopy, against a background of cooler surfaces. I agree with the suggestion to rephrase line 196.
The following line was added to the paragraph: 'The surface was not homogeneous in terms of TIR signature. Some signal in the TIR image time series were revealed when and where the background (the canopy surface) and foreground (air that had interacted with surface upstream) show a different heat signature.' Line 213. Why is there a sudden jump in the TKE in Figure 11 between 00:00 and 12:00 in 21 jul 2016?
Response: The jump is correlated to the passing of a short storm with brief precipitation (See panels Figure 3f and Figure 4a, and the text at line 120). Response: Thank you for the comment. A short recommendation can be added here (see the response to the General Comments).
Line 285. How did the author recognize the period with winds from the north?
Response: Both the sonic anemometer (EC) network and the wind observations from the DE-Fen station indicate wind direction. This wind sector is frequently observed. The situation is maintained for several half-hour periods during the day, due to the proximity of the Alps to the south. Assumed was that wind from this sector would have limited wake effects on any of the sonic anemometers by mast structures or topography.
Lines 369 -380. Why is this paragraph in the appendix? Isn't this part of the results?
Response: Thanks for pointing this out. The paragraphs were moved to the methods section.
What is the physical meaning of the grouping of the clusters presented in Figure A7? Also, what is the impact of variations of atmospheric stability in the results presented in Figure A7?
Response: The original study on the TED method shows the shape of the temperature variance events for each cluster, as extracted from idealized data. Different shapes suggest a different physical meaning. Table 2. What is the reason for mentioning the different ways of parameterizing the atmospheric stability? How is this used in this study?
Response: The different parameterizations of atmospheric stability are used as background information for the reader. I am not sure at this point if or how any of the parameterizations can help improve the classification of turbulence events. Personally, I found the differences between a low (1.0 m) and high (3.0 m) location in the gradient intriguing and indicative, without exploring possible explanations in more detail here.