Comment on “Capabilities and limitations of tracing spatial temperature patterns by fiber‐optic distributed temperature sensing” by Liliana Rose et al.

The need of precise thermal measurements at high spatial and temporal resolutions for scientific discovery has pushed the development of fiber-optic distributed temperature sensing (FO-DTS) in geosciences. Since its first appearance in the hydrologic community [Selker et al., 2006a, 2006b], FO-DTS has been used in numerous environmental and engineering applications [Lutz et al., 2012; Striegl and Loheide, 2012; Krause and Blume, 2013; Suárez et al., 2014] and apparently its use will continue to grow in the future. Researchers have realized that one of the main challenges for successful FO-DTS data collection is to achieve the best calibration setup [Tyler et al., 2009; Suárez et al., 2011; Hausner et al., 2011; van de Giesen et al., 2012]. On this subject, the work of Hausner et al. [2011] offers different experimental design and calibration methodologies that allows a quantitative assessment of accuracy, precision and resolution. These metrics, combined with the instruments' spatial and temporal resolution are probably the most important factors for correct interpretation of FO-DTS results. The work of Rose et al. [2013] discusses potential uncertainties related to the interpretation of FO-DTS results, showing how a temperature signal of variable size and intensity is revealed in an FO-DTS trace (the thermal profile along the cable obtained from the DTS instrument). Although the FO-DTS limitations discussed by Rose et al. [2013] can be inferred if the temperature and spatial resolutions of the instrument are known (recall that FO-DTS returns the average temperature measured along a spatial and temporal extent), Rose et al. [2013] provides a valuable clarification that should be useful for future investigations. However, their data show a consistent bias and inconsistencies at almost all the spatial scales that may be due to calibration issues. In this comment, I would like to emphasize that the physical installation (also known as deployment) combined with calibration methodologies are crucial for successful FO-DTS data collection. Rose et al. [2013] used a duplex single-ended configuration [Hausner et al., 2011] and a constant temperature ice bath for calibration purposes. The calibration section passed through this bath at positions 20–40 m and 240–260 m, and they matched the temperatures of those sections at a fixed value that was independently measured with a thermistor. The main issue with this physical setup is that the reference sections have the same (or very similar) temperature. As discussed by Hausner et al. [2011], to estimate the calibration parameters of a FO-DTS trace at least two different temperatures are needed in the reference sections [see Hausner et al., 2011, Figure 1]; otherwise, the system of equations that governs the physics of backscattering inside a fiber-optic cable may not have a unique solution, which could yield erroneous temperature estimations. To illustrate what happens when reference sections at the same temperature are used to calibrate FO-DTS traces, I applied the calibration algorithms presented by Hausner et al. [2011] in a previously validated FO-DTS data set (Figure 1). All the calibration algorithms are based on the same raw anti-Stokes and Stokes signals returned by the DTS instrument (Figure 1a). This data set was collected using a 2 m spatial resolution Sensornet Sentinel DTS instrument (Sensornet Ltd., Hertfordshire, UK), and a duplex single-ended configuration that had two reference sections at different temperatures. This physical configuration resulted in four reference sections along the fiber-optic cable (see Suárez et al. [2011] for complete details about the experimental setup). Figure 1b shows three calibrated temperature traces. The “validated data” were calibrated using three reference sections (two at the beginning and one at the end of the cable). The fourth reference section, located farther away from the source of light, was only used to estimate calibration metrics such as root mean square error (RMSE) and bias. Based on 575 traces collected during 48 h, these data have a RMSE of 0.131°C and a bias of −0.011°C in the fourth reference section (complete details of this data set and its metrics are presented in Hausner et al. [2011]). Thus, temperature fluctuations larger than ∼0.2°C in any location of the fiber are the result of real temperature changes that are greater than the “noise of the FO-DTS installation.” In addition, independent thermal measurements at different locations along the fiber and outside the reference sections were consistent with these calibration metrics [Suárez et al., 2011]. The “ambient bath-based calibration” and the “heated bath-based calibration” data (Figure 1b) were calibrated using the following algorithm (based on the discussion provided by Hausner et al. [2011]): if a single-ended calibration is performed using two reference sections (or points) at the same temperature, the differential attenuation (Δα) has to be assumed constant and can be interpolated from a length of fiber at uniform temperature using Beers' Law. The other calibration parameters (C and γ) must be determined through optimization by minimizing an objective function, e.g., the mean bias, and constraining the solution to satisfy the criteria of having the same temperature in the reference sections at a known value. This could be one of the many algorithms that would satisfy a “match sections” calibration. If the calibration parameters are based on a single data set taken over some initial integration period, the calibration is typically called “static” or “fixed,” and if the calibration parameters are recalculated during each time step of the data acquisition process, the calibration will be “dynamic.” From the description of Rose et al. [2013], it is inferred that they utilized a “fixed match sections” calibration algorithm with two reference sections at known and constant temperature, which is similar to that used in the “ambient bath-based calibration” and in the “heated bath-based calibration” data. The main conclusion that can be drawn from the results presented in Figure 1b is that when two reference sections at the same temperature are used to calibrate the FO-DTS temperatures (whatever the choice of the reference temperature), large errors (>5–10°C) could occur in many locations along the fiber, highlighting the importance of a well calibrated data set. Figure 1b also shows that better temperature estimations occur at temperatures near the temperature of the reference section used to calibrate the FO-DTS trace. (a) Raw Raman spectra data (Stokes and anti-Stokes) used to calibrate temperatures along a fiber. The raw data were collected using a duplex single-ended configuration that had two reference sections at different temperatures. This physical configuration resulted in four reference sections along the fiber-optic cable. (b) Calibrated temperatures obtained from the raw FO-DTS data and using three different algorithms for a duplex single-ended configuration. When using calibrated temperatures based on two reference sections at the same temperature, better estimations occur at temperatures near the temperature of the reference section used to calibrate the FO-DTS trace but important errors occurs elsewhere. The shaded areas correspond to approximate locations where the cable is exposed to ambient temperature. In these locations, temperature estimations based on the ambient bath-based calibration agree fairly well with the validated data. Reviewing Rose et al. [2013, Figure 2A], it can be seen that for a signal size of 25 times the sampling interval of the DTS instrument the results appear coherent, where temperature changes in the warm bath occurs approximately within the spatial resolution reported by the instrument manufacturer, and with a temperature difference between the FO-DTS system and thermistor of 0.22°C. However, for signal sizes of 7.5 and 5 times the sampling interval of the DTS instrument, the FO-DTS temperatures diverge in more than 3°C from the temperatures measured by the thermistor. For these signal sizes, which are larger than the spatial resolution of the DTS instrument, and excluding edge effects, i.e., neglecting the locations in which the reported temperature is an average between the water and ambient temperature, one would expect a match between the peak temperature measured by the DTS instrument (at approximately the middle of the test section) and the temperature measured by the thermistor—this behavior would have been consistent with the data of the “25 times the sampling interval” experiment. Although this behavior is not occurring in the warm bath, it is actually happening in the cold bath for all the sampling experiments in which the signal size is greater than the FO-DTS spatial resolution [Rose et al., 2013, Figures 1B and 2A and Table 1]. Remarkably, the data reported by Rose et al. [2013] behave in a similar fashion than the data calibrated with two reference sections at the same temperature presented in Figure 1b. For instance, the data based on the ambient temperature bath presented in Figure 1b yields better estimations when temperatures near the reference temperature (ambient temperature) are being measured along the cable and worse estimations for warmer conditions. As opposed to the data presented by Rose et al. [2013], however, the data based on the ambient temperature bath overestimate the temperatures in the heated bath (Figure 1b). Therefore, claims like “a critical deterioration of quantitative accuracy of FO-DTS measured temperatures with reducing signal size” cannot be fully supported unless a more detailed description of the calibration and validation procedures is presented. Although this discussion emphasizes the relevance of the experimental design (physical installation, configuration of the reference sections and calibration algorithm), it is important to note that the “deterioration of quantitative accuracy” claimed by Rose et al. [2013] could be due to many other factors such as drift of the instrument, disconnecting and reconnecting connectors, having nonuniform differential attenuation along the fiber, nonuniform temperature within the reference sections, or other time-dependent factors [Tyler et al., 2009; Hausner et al., 2011; van de Giesen et al., 2012]. For instance, if Rose et al. [2013] used the manufacturer's calibration, it would be useful to understand how that calibration is carried out. Typically, FO-DTS systems have an internal coil that may be used for calibration purposes. In this case, the fiber in the internal coil not necessarily will have the same differential attenuation than the cable used to measure temperatures in the field and, consequently, it may result in an incorrect calibration. Therefore, it would be very valuable to confirm that the data presented by Rose et al. [2013] are well calibrated by validating the temperature traces in a location of the cable other than the reference baths or the cold/warm bath, and by presenting and analyzing the raw Stokes and anti-Stokes data of their experiments. This will assure that the calibration procedure will deliver a unique solution for the thermal profile along the fiber-optic cable (as opposed to what was shown in this comment) and will validate or invalidate their conclusions. This type of analysis will be very valuable for researchers that are learning and understanding the underlying physics of FO-DTS, which is crucial for data interpretation. The author acknowledges funding from the Comisión Nacional de Investigación Científica y Tecnológica de Chile (CONICYT, Fondecyt/Iniciación/11121208), and from the Centro de Desarrollo Urbano Sustentable (CEDEUS-CONICYT/FONDAP/15110020).