TMCnet News
Sensor Drift and Predicted Calibration Intervals of Handheld Temperature and Relative Humidity Meters Under Residential Field-Use Conditions [Journal of Environmental Health](Journal of Environmental Health Via Acquire Media NewsEdge) Abstract Handheld temperature and relative humidity (T/RH) meters are commonly used in residential indoor air surveys. Although popular, T/RH meters are prone to sensor drift and consequent loss of accuracy, and thus instrument manufacturers often recommend annual calibration and adjustment. Field-use conditions, however, have been shown to accelerate electronic sensor drift in outdoor applications, resulting in out-of-tolerance measurements in less than one year. In the study described in this article, sensor drift was evaluated under residential field use for 30 handheld T/RH meters to predict needed calibration intervals based on hierarchical linear modeling. Instruments were used in 43 home visits over a 93-day period and were calibrated (without adjustment) 49 times over the study period with a laboratory standard. Analysis of covariance showed significant drift among temperature sensors for all three instrument types (p < .0001) and among humidity sensors in two instruments. The authors' study suggests calibration frequency should he based on instrument performance under specific sampling conditions rather than on predetermined time intervals. Introduction Temperature and relative humidity (T/RH) have become increasingly important considerations in residential health surveys. Cold housing, for instance, is associated with increased mortality, circulatory diseases, respiratory problems, and mental ill health (Butler, Williams, Tukuitonga, & Paterson, 2003; Marmot Review Team, 2011), and high indoor temperature is associated with increased mortality, particularly among the elderly (Centers for Disease Control and Prevention [CDC], 2003; Semenza et al., 1996; Yip et al., 2008). Indoor RH above 50%-55% can support house dust mite (HDM) growth, which is associated with the development and exacerbation of asthma (Arlian, Confer, Rapp, Vyszenski-Moher, & Chang, 1998; Ellingson, LeDoux, Vedanthlan, & Weber, 1995; Korsgaard, 1982). Likewise, many fungi produce allergenic spores and can grow on building surfaces above 75% RH (Arundel, Sterling, Biggin, & Sterling, 1986; Block, 1953; Grant, Hunter, Flannigan, & Bravery, 1989). In addition to supporting HDM and mold growth, damp indoor environments are associated with upper respiratory infections, wheezing, and coughing (Arundel et al., 1986; Gunnbjörnsdóttir étal., 2006; Koskinen, Husman, Meklin, & Nevalainen, 1999). Indoor dampness is also linked to mental illness and depression, independent of visible mold growth (Hopton & Hunt, 1996; Hyndman, 1990). Thus, accurate assessment of indoor environmental conditions is essential prior to making recommendations to home or building owners, residents, or housing authorities; or prior to interpreting research findings related to residential health. Handheld electronic T/RH meters are common instruments for measuring indoor environmental conditions, and continue to be used as the primary instrument for measuring air temperature and RH in many environmental studies (Cho et al., 2006; Grimsley et al., 2012; Martin & Coffey, 2007; Perry et al., 2008; Rabito et al., 2007; Ren, Jankun, Belanger, Bracken, Leaderer, 2001). T/RH meters most often use negative temperature coefficient (NTC) thermistors to measure air temperature and thin-film capacitance-based sensors to measure RH. One significant disadvantage to using T/RH meters, however, is that NTC and capacitance-based sensors are prone to drift and decalibration (Childs, Greenwood, & Long, 2000; Visscher & Körnet, 1994). To ensure accurate measurements, manufacturers often recommend annual sensor calibration. Studies show, however, that annual calibra- tion may not be sufficient under outdoor atmospheric monitoring applications, particularly for capacitance-based RH sensors (Freitag et al., 1994; Lake, Noor, Freitag, & McPhaden, 2003; Visscher & Körnet, 1994). Given the propensity for T/RH meter sensors to drift, it is possible that annual calibration may be too infrequent for residential field applications as well. The likelihood of sensor drift and the difficulty of conducting inhouse T/RH meter calibration may lead field practitioners to continue using instruments after they have drifted outside of manufacturers' specified tolerances. Although sensor drift is a recognized concern, little is known about how field-use conditions affect T/RH meter accuracy. Wight (1994) suggested sensor drift may be accelerated under field-use conditions due to harsh handling, extreme environments, and multiple users, and therefore calibration intervals should be based on experience with a given instrument under specific sampling conditions rather than on predetermined time intervals. No studies, however, have looked specifically at T/RH meter sensor drift under residential field-use conditions. Understanding the effect of field use on sensor drift is essential to determining whether the convention of annual calibration is appropriate for T/ RH meters. The purpose of our study, therefore, was to assess temperature and RH sensor drift over time in handheld T/RH meters used in home visits, and to use regression models to predict required calibration intervals. Methods Instrument Selection Ten each of three T/RH meter models (Extech 445580, Fluke 971, and General Tools PTH8707) were evaluated in our study (Table 1). U.S. National Institute of Standards and Technology (NIST)-traceable calibration was performed on all Extech and Fluke instruments by the manufacturer prior to data collection. NIST-traceable calibration was not offered by General Tools, and instruments were used as purchased. A Vaisala HMP 110 T/RH instrument was used as a standard from which to compare T/ RH meters. This instrument underwent NISTtraceable calibration prior to data collection. The Vaisala HMP 110 was chosen based on cost, temperature range (-40°C to 80°C), temperature accuracy (±0.2°C between 0°C and 40°C), RH range (0%-100%), and RH accuracy (±1.7% between 0%-90% RH). The Vaisala was theorized to be a suitable comparison instrument due to its use of a platinum resistance temperature detector and long-term RH sensor stability. The Vaisala was postcalibrated with a NIST-traceable standard in July 2012 following data collection. The humidity sensor was found to be slightly out of tolerance (-2.09%) at 75% RH, but was within tolerance at 0%, 12%, and 33% RH. The Vaisala was therefore considered an acceptable standard given that RH never exceeded 36% in our study. The temperature sensor was within tolerance at postcalibration. Field Visits To test the effect of field use on instrument performance, nine northern Utah homes were selected to represent homes where residential environmental surveys might be conducted. Homes were chosen from volunteers associated with the Cache County Center of the National Children's Study (NCS). Similar to environmental data collection performed in homes as part of the NCS, field use was simulated by having study personnel transport T/RH meters in a field transport bag to and from each home by personal vehicle. Study personnel were provided with instrument-specific training for each T/RH meter model. Study personnel were also trained in a standard operating procedure for collecting temperature and humidity in homes. Home visits were performed 43 times over a 93-day period (March-June 2012). Our study was approved by Utah State University's institutional review board. It was hypothesized that physical contact between instruments and other hard objects during transport could cause trauma to temperature and humidity sensors, which may accelerate instrument drift. To test this hypothesis, instruments were randomly assigned identification numbers between 1 and 10 within each model. Instruments that received an even number were assigned to the enhanced packaging group. Instruments in the enhanced packaging group were wrapped in bubble wrap for transport. Instruments that received an odd number were assigned to the standard packaging group. Instruments in the standard packaging group were placed in field bags with no attempt to limit contact with other hard objects during transport. Because only nine homes were in the study, one instrument of each model remained in the lab during each field visit. All 30 instruments were rotated through each of the nine homes to ensure uniformity and consistency in data collection. T/RH Meter Comparison to Vaisala HMP110 At the completion of home visits, instruments were unpacked in the study center laboratory by two technicians who conducted all packaging and instrument comparisons. Instruments were powered on and placed on a laboratory bench adjacent to the Vaisala and allowed to equilibrate for five minutes. After five minutes, temperature and RH readings were taken and differences between T/ RH meters and the Vaisala were calculated and recorded on a standardized data collection form. Data from home visits and laboratory comparisons were then manually put into a spreadsheet. Because only nine homes were used, one instrument of each model remained in the laboratory on any given day of data collection. All instruments were calibrated against the Vaisala following field visits, however, including those that remained in the laboratory. Statistical Methods A mixed models repeated measures analysis of covariance (ANCOVA) was used to compare the reliability of T/RH meters. This analysis was used to determine if a change in mean deviation occurred from the Vaisala from beginning to end of data collection. The means of the first and last five days of data collection were compared to determine whether significant drift occurred from the Vaisala from beginning to end of data collection. Following this initial analysis, stage was replaced with day as a continuous variable and a hierarchical linear mixed model (HLM) was performed to quantify the rate of sensor drift over time. This model included an interaction between instrument type and day in order to estimate separate linear slopes for each instrument. Slopes were used to predict the time in days for instrument sensors to drift outside of manufacturers' reported levels of accuracy. All analyses were performed using the mixed procedure in SAS version 9.3. Results All 30 T/RH meters completed the study, which included 43 home visits and 49 laboratory comparisons against the Vaisala. T/ RH meters traveled an average distance of 1,081.5 km (672 miles) over the course of the study (25.1 km [15.6 miles] round trip per home visit). All home visits and laboratory comparisons were completed over a 93-day period from March to June 2012. Temperature A total of 490 laboratory comparisons for temperature were conducted for each instrument type. Ambient lab temperature, as measured by the Vaisala, ranged from 20.8°C to 24.6°C with a mean of 21.8°C. Across all 30 instruments the mean difference from the standard (instrument reading-Vaisala reading) was 0.08°C, with a range of -4.4°C to 4.9°C. Means, standard deviations (SD), 95% confidence intervals (Cl), and range of temperature deviations for each instrument type are provided in Table 2. Temperature ANCOVA None of the interactions or the main effect term for packaging were statistically significant in the full model (a = .05), thus these terms were removed using backward elimination. The parsimonious model contained only the terms for stage (beginning vs. end), instrument type, and baseline temperature. The parameter estimate for baseline temperature was -0.3570 (p < .0001). Thus, within the range of temperature encountered in the lab, as the ambient temperature increased, the instrument readings trended closer to the Vaisala reading. Table 2 provides the final ANCOVA results for the parsimonious model. The nonsignificant instrument type by stage interaction in the full model (p = .63) indicates that no significant difference occurred between the instruments in the change from beginning to ending. Thus all instruments were grouped together and the change from beginning to end was measured by the mean deviation of all instruments. At the beginning of data collection, all instruments were on average 0.23°C higher than the lab standard. By the end of data collection, the mean deviation was -0.12°C lower than the lab standard. The difference in mean deviation from the beginning of data collection was 0.34°C (SE-0.07; p< .0001). HLM Results for Temperature HLM was used to test linear decline in instrument accuracy over time. The test for the homogeneity of slopes was not significant (F = 1.02; p = .38). Consistent with the ANCOVA results, HLM showed no difference in the slopes of the three instrument types. Thus, the interaction term was removed from the model. Figure 1 shows the regression lines from the HLM model. If we assume constant drift over time from the HLM model, we can calculate the time in days until an instrument requires recalibration. The common slope for the three instruments had a value of 0.000538. As the manufacturers for the instruments we used gave accuracy ranges of 0.50°C (Fluke) and 1.0°C (Extech and General Tools), we used these parameters to estimate the days to calibration. We determined that temperature accuracy would fall out of tolerance with a ±0.50°C range in 929 days or approximately three years and out of compliance with a ±1°C range in 1,859 days or approximately five years. Relative Humidity As with temperature, RH comparisons were made against the Vaisala on 49 separate days. One or more of the General Tools instruments failed to display a humidity reading on 15 days during the sampling period, however. This problem occurred on days when ambient RH dropped below the limit of detection (20%). For the General Tools meters, only 417 valid measurements were obtained. Over the course of our study, RH in the lab ranged from 13.19% to 35.87%, with a mean of 23.84%. For the 30 instruments, the mean difference in RH readings as compared to the standard was 0.59%, with a range of -7.01% to 11.22%. Means, standard deviations, 95% Cl, and range of RH deviations for each instrument type are provided in Table 3. Relative Humidity ANCOVA The main effect for packaging (p = .90) and both interaction terms containing packaging (p = .38 and p = .34) were nonsignificant in the full model. Therefore, these terms were removed using backward elimination. The parameter estimate for baseline humidity was -0.04119, indicating that within the range of RH encountered in the lab, as the relative humidity increased the instrument readings trended closer to the Vaisala readings. This finding was significant (p = .024). The final AN CO VA model included main effect terms for baseline RH, instrument type, stage, and an instrument type by stage interaction. Table 3 also provides the AN CO VA results for the final model. The interaction was significant (p < .0001), indicating a significant difference among the three instruments from beginning to end. For the Extech instruments, the mean difference from Vaisala was 1.59% in the beginning cluster and 2.18% in the ending cluster, a difference of 0.6%. This change was not significant (p = .11). Among the Fluke instruments, the mean difference from Vaisala in the beginning and end clusters were 0.04% and -0.99%, respectively. The difference in deviation from beginning to end was 1.03% (p = .008). Similarly, for the General Tools instruments, the deviations were 2.62% and -0.34% in the beginning and end clusters, respectively, with a change of 2.96% over time (p < .0001). Using a Bonferroni adjusted alpha for the three tests (a = 0.05/3 = 0.0167), we concluded that a significant change occurred in measurement reliability over the course of our study for the Fluke and General Tools instruments. Relative Humidity HLM The test for the homogeneity of slopes in the HLM model was statistically significant (F = 29.71; p < .0001), consistent with our ANCOVA results. Thus, three different slopes were fitted. The HLM slopes for each instrument are shown in Figure 2. Although Fluke and General Tools reported accuracies of ±2.5% and ±3.0% RH, respectively, an accuracy of ±5% of the instrument's reading was used for the purposes of our study and was assumed to be acceptable for most residential indoor air quality assessments. The highest RH observed over the course of the study was approximately 36%. Thus, assuming 36% RH, and using ±5% of the instrument's reading as being out of tolerance (1.8% RH), we estimated the slope and the number of days until calibration would be required for each type of instrument. These results are presented in Table 4. Assuming constant drift over time, we predicted that the Extech instruments would remain in tolerance beyond the conventional annual recalibration interval. The Fluke instruments would remain within ±5% accuracy for approximately four months, and the General Tools instruments would require recalibration at just 47 days. Discussion Our study supports previous work showing that field-use may accelerate capacitancebased RH sensor drift (Freitag et al., 1994; Lake et al., 2003; Visscher & Körnet, 1994). Furthermore, capacitance-based RH sensor drift was significant enough for some instruments to warrant calibration intervals of less than one year. Based on these study conditions, calibration intervals of 2, 4, and >12 months are appropriate for General Tools, Fluke, and Extech instruments' RH sensors, respectively. This finding supports Wight's (1994) recommendation to tailor calibration intervals based on instrument performance rather than on predetermined time schedules. Although calibration intervals tailored to specific instruments and sampling conditions may be ideal, limitations imposed by available equipment and procedures may make frequent calibrations difficult to perform in house for some organizations. Optimally, equipment and procedures for field calibration of T/RH meters should be affordable, easy to perform in a relatively short time period, and designed such that field technicians can adjust instruments to match calibration standards. One method of in-house calibration is to compare instrument temperature readings to a liquid-in-glass precision thermometer. Based on this study, temperature sensors were predicted to stay within manufacturers' specified tolerances for at least three years. In the absence of a method for in-house adjustment, NTC sensor drift could be monitored with a precision thermometer, and reference laboratory calibration intervals could be set based on instrument performance, which according to our findings may be needed only every 2-3 years. High quality liquid-in-glass precision thermometers used as reference instruments should only require one complete calibration in their lifetime, followed by periodic recalibrations in the user's laboratory of only a single temperature, usually 0.0°C (Wise, 1991). Salt solutions used for T/RH meter calibration may provide an affordable alternative to frequent reference lab calibration for humidity sensors. In cases where dedicated in-house laboratory calibration is available, traceable humidity chamber or other calibration systems may be preferred. If resources are limited, however, salt bath calibration may provide a method for monitoring drift over time to establish calibration intervals. Our study aimed to identify sources of instrument drift during field use. We hypothesized that trauma to T/RH meter sensors during transport might be a primary contributor. We found no significant difference between instruments packed in bubble wrap and those transported with no protection. Other possible explanations for accelerated drift may be extreme temperatures during vehicle transport or sensor poisoning from contaminants in participant homes. Future studies may consider these and other potential causes of drift. One limitation of our study was the time allowed for data collection. We found significant sensor drift during field use, but our observations were limited to 93 days. Whether the rate and direction of drift will change over longer periods of time is unknown. Findings are also limited to the instruments under consideration. The wide variation among the three types of instruments assessed in our study suggests that these findings are not generalizable to all T/ RH meters. Furthermore, the mean RH over the course of this study was 24%. Thus, these findings may not apply to sensor performance under high ambient RH and future research is needed to evaluate sensor performance over broader RH and temperature ranges. Conclusion Our study suggests that RH sensors in handheld T/RH meters may drift outside of manufacturers' specified tolerances in less than one year under field-use conditions. Calibration intervals should be based on instrument performance under a given set of sampling conditions, rather than on predetermined time frames. HÜ Acknowledgements: Support for this project was provided by the National Institutes of Health contract HHSN275200800027C/ N01-HD-80027. The authors of this manuscript did not receive any financial support from the commercial manufacturers of the instruments evaluated and believe no conflicts of interest exist. The authors would also like to thank the volunteers who opened their homes for data collection. References Arlian, L.G., Confer, P.D., Rapp, C.M., Vyszenski-Moher, D.L., & Chang, J.C.S. (1998). Population dynamics of the house dust mites Dermatophagoides farina, D. pteronyssinus, and Euroglyphus maynei (Arcari: Pyroglyphidea) at specific relative humidities. Journal of Medical Entomology, 35,46-53. Arundel, A.V., Sterling, E.M., Biggin, J.H., & Sterling, T.D. (1986). Indirect health effects of relative humidity in indoor environments. Environmental Health Perspectives, 65, 351-361. Block, S.S. (1953). Humidity requirements for mold growth. Applied Microbiology, 1(6), 287-293. Butler, S., Williams, M., Tukuitonga, C., & Paterson, J. (2003). Problems with damp and cold housing among Pacific families in New Zealand. The New Zealand Medical Journal, 116(1177), U494. Centers for Disease Control and Prevention. (2003). Heat-related deaths-Chicago, Illinois, 1996-2001, and United States, 19791999. Morbidity and Mortality Weekly Report, 52(26), 610-613. Childs, P.R.N., Greenwood, J.R., & Long, C.A. (2000). Review of temperature measurement. Review of Scientific Instruments, 71(8), 2959-2978. Cho, S.-H., Reponen, T., LeMasters, G., Levin, L., Huang, J., Meklin, T., Ryan, P, Villareal, M., & Bernstein, D. (2006). Mold damage in homes and wheezing in infants. Annals Of Allergy, Asthma, & Immunology, 97(4), 539-545. Ellingson, A.R., LeDoux, R.A., Vedanthlan, P.K., & Weber, R.W. (1995). The prevalence of Dermatophagoides mite allergen in Colorado homes utilizing central evaporative coolers. Journal of Allergy and Clinical Immunology, 96(4), 473^-79. Freitag, H.P., Feng, Y., Mangum, L.J., McPhaden, M.J., Neander, J., & Stratton, L.D. (1994). Calibration procedures and instrumental accuracy estimates of TAO temperature, relative humidity and radiation measurements. U.S. Department of Commerce, National Oceanic and Atmospheric Administration, Pacific Marine Environmental Laboratory. Retrieved from http://docs.lib.noaa.gov/noaa_documents/OAR/PMEl7TM_ERL_PMEL/TM_ERL_PMEL_104.pdf Grant, C, Hunter, C.A., Flannigan, B., & Bravery, A.E (1989). The moisture requirements of moulds isolated from domestic dwellings. International Biodeterioration, 25(4), 259-284. Grimsley, L.E, Chulada, P.C., Kennedy, S., White, L., Wildfire, J., Cohn, R.D., Mitchell, H., Thornton, E., El-Dahr, J., Mvula, M.M., Sterling, Y., Martin, W.J., Stephens, K.U., & Lichtveld, M. (2012). Indoor environmental exposures for children with asthma enrolled in the HEAL study, post-Katrina New Orleans. Environmental Health Perspectives, 120(11), 1600-1606. Gunnbjörnsdottir, M.I., Franklin, K.A., Norbäck, D., Björnsson, E., Gislason, D., Lindberg, E., Svanes, C, Omenass, E., Norrman, E., Joqi, R., Jensen, E.J., Dahlman-Hoglund, A., & Janson, C. (2006). Prevalence and incidence of respiratory symptoms in relation to indoor dampness: The RHINE study. Thorax, 61(3), 221-225. Hopton, J.L., & Hunt, S.M. (1996). Housing conditions and mental health in a disadvantaged area in Scotland. Journal of Epidemiology and Community Health, 50(1), 56-61. Hyndman, S.J. (1990). Housing dampness and health amongst British Bengalis in East London. Social Science and Medicine, 30(1), 131-141. Korsgaard, J. (1982). Preventive measures in house-dust allergy. American Review of Respiratory Diseases, 125(1), 80-84. Koskinen, O.M., Husman, T.M., Meklin, T.M., & Nevalainen, A.I. (1999). The relationship between moisture or mould observations in houses and the state of health of their occupants. European Respiratory Journal, 14(6), 1363-1367. Lake, B.J., Noor, S.M., Freitag, H.P., & McPhaden, M.J. (2003). Calibration procedures and instrumental accuracy estimates of ATLAS air temperature and relative humidity measurements. U.S. Department of Commerce, National Oceanic and Atmospheric Administration, Oceanic and Atmospheric Research Laboratories, Pacific Marine Environmental Laboratory. Retrieved from http7/www. pmel.noaa.gov/pubs/PDF/lake2569/lake2569.pdf Marmot Review Team. (2011). The health impacts of cold homes and fuel poverty. Retrieved from http://www.instituteofhealthequity. org/projects/the-health-impacts-of-cold-homes-and-fuel-poverty Martin, S.B., & Coffey, C.C. (2007). An indoor environmental quality investigation of the Fayette County (Pennsylvania) courthouse. Indoor and Built Environment, 16(5), 456-464. Perry, T.T., Vargas, P.A., Bufford, J., Field, C, Flick, M., Simpson, P.M., Hamilton, R.G., & Jones, S.M. (2008). Classroom aeroallergen exposure in Arkansas head start centers. Annals Of Allergy, Asthma & Immunology, 100(A), 358-363. Rabito, EA., Iqbal, S., Holt, E., Grimsley, L.E, Islam, T.M., & Scott, S.K. (2007). Prevalence of indoor allergen exposures among New Orleans children with asthma. Journal of Urban Health, 84(6), 782-792. Ren, P, Jankun, T.M., Belanger, K., Bracken, M.B., & Leaderer, B.P. (2001). The relation between fungal propagules in indoor air and home characteristics. Allergy, 56(5), 419^-24. Semenza, J.C., Rubin, C.H., Falter, K.H., Selanikio, J.D., Flanders, W.D., Howe, H.L., & Wilhelm, J.L. (1996). Heat-related deaths during the July 1995 heat wave in Chicago. The New England Journal Of Medicine, 335(2), 84-90. Vtsscher, G.J.W., & Körnet, J.G. (1994). Long-term tests of capacitive humidity sensors. Measurement Science and Technology, 5(10), 1294-1302. Wight, G.D. (1994). Fundamentals of air sampling. Boca Raton, FL: Lewis Publishers. Wise,J.A. (1991 ). A procedure for the effective recalibration of liquidin-glass thermometers. U.S. Department of Commerce, National Institute of Standards and Technology, Process Measurements Division. Retrieved from http://www.nist.gov/calibrations/upload/ sp819.pdf Yip, EY., Flanders, W.D., Wolkin, A., Engelthaler, D., Humble, W, Neri A., Lewis, L., Backer, L., & Rubin, C. (2008). The impact of excess heat events in Maricopa County, Arizona: 2000-2005. International Journal Of Biometeorology, 52(8), 765-772. James D. Johnston Brianna M. Magnusson Department of Health Science Brigham Young University Dennis Eggett Department of Statistics Brigham Young University Kyle Mumford Center for Persons With Disabilities Utah State University Scott C. Colling wood Department of Pediatrics University of Utah Scott A. Bernhardt Department of Biology Utah State University Corresponding Author: Scott A. Bernhardt, Department of Biology, Utah State University, UMC5305 Old Main Hill, Logan, UT 84322. E-mail: [email protected]. (c) 2014 National Environmental Health Association |