Performance Evaluation of an Infrared Thermocouple

The measurement of the leaf temperature of forests or agricultural plants is an important technique for the monitoring of the physiological state of crops. The infrared thermometer is a convenient device due to its fast response and nondestructive measurement technique. Nowadays, a novel infrared thermocouple, developed with the same measurement principle of the infrared thermometer but using a different detector, has been commercialized for non-contact temperature measurement. The performances of two-kinds of infrared thermocouples were evaluated in this study. The standard temperature was maintained by a temperature calibrator and a special black cavity device. The results indicated that both types of infrared thermocouples had good precision. The error distribution ranged from −1.8 °C to 18 °C as the reading values served as the true values. Within the range from 13 °C to 37 °C, the adequate calibration equations were the high-order polynomial equations. Within the narrower range from 20 °C to 35 °C, the adequate equation was a linear equation for one sensor and a two-order polynomial equation for the other sensor. The accuracy of the two kinds of infrared thermocouple was improved by nearly 0.4 °C with the calibration equations. These devices could serve as mobile monitoring tools for in situ and real time routine estimation of leaf temperatures.


Introduction
Monitoring of the physiological state of forests or agricultural plants has become a basic technique to control the crop environment and to modulate irrigation and fertilization [1]. Remote and non-destructive monitoring techniques are more convenient to use to collect the necessary physiological information, including important factors such as plant weights, stem diameters, sap flow rates, fruit diameters and leaf temperatures.
Jones [2] introduced an infrared thermometry method to detect stomatal closure conditions. Artificial wet and dry reference surfaces were used to provide the base indexes. Kumar et al. [3] applied a Tele temp AG-42 infrared thermometer to measure the canopy temperature and used these measurement values as an indicator of plant-water status. Gontia and Tiwari [4] selected an infrared thermometer to measure the leaf temperature of wheat crops. The difference in the canopy and air temperature and the vapour pressure deficit was used to calculate the crop water stress index.
The performance evaluation of the infrared thermometer is a key step to ensure the accuracy of the measured values. Amiro et al. [5] described the design of leaf chambers and proposed a nonlinear calibration equation. The accuracy of measurement values of their infrared thermometer was limited to within 0.2 C. Rapier and Michael [6] introduced a calibration technique for Everest 4000A Radiometers. A temperature control chamber was constructed for the calibration equipment and a polynomial calibration equation was established. Their results revealed that the root-mean-square differences were less than 0.6 C.
Kalma and Alksnis [7] described a calibration system for an infrared surface temperature transducer.
The observed significant measurement errors were due to the assumption of a linear relationship between the output voltage and standard values detected by a thermistor thermometer. Bugbee et al. [8] evaluated two types of infrared transducers and found that the need for a longer response period to reach the equilibium state induced significant errors. The best calibrations for these infrared thermometers were second-order polynomial equations. Baker et al. [9] tested the accuracy of an provides an opportunity to apply the technique for monitoring routine temperatures for forest management and agricultural production. The objectives of this study were: (1) to evaluate the performance of infrared thermocouples using a standard temperature calibrator and (2) to determine the adequate calibration equations to improve the accuracy of these sensors.

Infrared Thermocouple Devices
Two kinds of infrared thermocouples were used in this study, a Sentron SI-10AL (Sentron Eng. CO, Ltd, Taiwan) and a Trotec BP-20 (Trotec GmbH & CO, Heinsberg, Germany). The respective manufacturer's specifications are listed in Table 1.

Standard Temperature
The standard temperature of the blackbody source was maintained by a TC 2000 temperature calibrator (Instutek AS, Skreppestad Naringspak, Norway). The operating temperature ranged from −40 C to 150 C. The temperature of the standard environment was measured by a RTD (resistive temperature detector) thermometer calibrated by the U.S. NIST (National Institute of Standards and Technology). The uncertainty of this equipment from the calibration certificate was 0.03 C. An aluminum cylinder was installed into the oil bath of this calibrator. The size of this cylinder corresponded to the requirements of the blackbody source [15].

Test Procedures
The target temperature for calibration was maintained at 13, 17, 21, 25, 29, 33 and 37 C for the Sentron SI-10A and at 2 C intervals for the Trotec BP-20 within the same measuring range. The test environment was maintained at 25 C and the variation of the set room environment temperature was controlled within ±1.5 C. Several replicate measurements were made for each standard temperature. As one measurement has finished, the infrared thermocouple was taken out from the blackbody cavity for five minutes and then was put into the cavity for further measurement. The signals of the two kinds of infrared thermocouples were indicated in their LCD screens. The error of the data acquisition device was insignificant, according to the manufacturer's specifications. The sensor signals reached an equilibration state within one second and this was recorded by the visual method.

Data Analysis
The performance of these infrared thermocouples was assessed by their accuracy and precision. The accuracy is expressed as the closeness with which a measurement value approached the standard temperature, i.e.: where e i is the error, T r is the readout value of infrared thermocouple, and T s is the standard temperature. The smaller the e i value, the better the accuracy.
The precision P is expressed as the repeatability of these measurement values in the same standard environment. The precision P is exactly the statistical standard deviation computed from the available set of temperature measurements: where T ave is the average value of reading values and n is the number of data. A smaller P value indicates better precision performance.

The Calibration Equation
The calibration equations were established using the regression analysis technique. The criteria for selecting of the best equation are the coefficient of determination R 2 , the standard deviations of estimator, s, for the calibration equation, the t-test of each parameter for the equation, and residual plots [16]. The R 2 value is calculated as follows: where y' is the predicted value of the regression equation, y ave is the average value of the dependent value and y i is the the dependent value.
where n is the number of data and p is the number of parameters of equation.
The t value of each parameter is calculated as: where b i is the parameter values and se(b i ) is the standard estimation values of b i .
The residual plot is a qualitative criterion to evaluate the adequateness of calibration equations. As the data distribution of residual plots indicated a uniform pattern, the model could be recognized as an adequate model. If the residual plots revealed a clearly systematic pattern, a model cannot be accepted.
There are two types of calibration equations. For the classical calibration equation, the readout values of infrared thermocouples were assumed as dependent variables and the standard values of the calibrator were selected as independent variables. For the inverse calibration equation, the standard values maintained by the temperature calibrator were recognized as dependent variables. Because the inverse calibration equation has better predictive ability and is easy to apply [12,13], the inverse calibration equation was adopted in this study: A polynomial equation could be used to fit the calibration data. However, as higher order parameters were selected, the predicted errors could be increased significantly [16], so the selection of an adequate

Performance of the Sentron SI-10A Sensor
The relationship between the reading values of the Sentron SI-10A infrared thermocouple versus standard values maintained by TC-2000 calibrator is shown in Figure 1. A nonlinear distribution of the data scattering was found. The error distribution of the Sentron SI-10A is shown in Figure 2.   Below the measurement temperature of 25 C, the standard deviations were less than 0.2 C. In the higher temperature range, the standard deviations were less than 0.1 C. From the viewpoint of practical applications, this sensor had good precision performance. The estimated parameters and statistics of the calibration equations for this Sentron SI-10A infrared thermocouple within the range from 13 C to 37 C are listed in Table 2. A clear pattern or uniform distribution of residual plots could serve as a qualitative criterion to evaluate the calibration equation. If the data distribution of residual plots indicated a uniform distribution, the model could be recognized as adequate.
A clean systematic pattern of residual plots was found for linear and polynomial equations. However, a random distribution of residual plots was found for the three-order polynomial equation, indicating that the three-order polynomial equation was the adequate calibration equation for this sensor within the range from 13 C to 37 C. The standard deviation of estimators for the three calibration equations was 0.5541, 0.5410 and 0.3924 C, respectively. The nonlinear curves could be viewed as linear lines by narrowing the measurement range. As the measurement data within the narrower range were reevaluated, the calibration equations and statics of the Sentron SI-10A infrared thermocouple for the measurement data ranging from 20 C to 35 C are presented in Table 3. This range is the most common and useful for routine temperature estimation.

Performance of the TROTEC BP-20 Sensor
The distribution between readout values of the Trotec BP-20 infrared thermocouple and the standard temperature are shown in Figure 1. A typical nonlinear data distribution was found. The error distribution is presented in Figure 2. The errors distribution showed a clear nonlinear pattern. The standard deviation of readout values at each standard temperature is presented in Figure 3. The good repeatability was confirmed for this sensor. The numeric values of standard deviations were less than 0.18 C. The calibration equations and criteria are listed in Table 4.  Table 5.

The Difference Performance between Sensors
To evaluate the reproducibility of the calibration equations, the error distribution of two Sentron SI-10A infrared thermocouples from the same manufacturer and the same production batch are shown in Figure 4. The data distribution showed a significant difference between the two sensors. The   The error distributions of two TROTEC BP-20 infrared thermocouples are presented in Figure 5.
When the measurement data of sensor No. 4 was calculated with the calibration equation that was established from sensor No. 3, the predictive errors were higher than 1.2 C. Figure 5. The errors distribution of two Trotec BP-20 infrared thermocouples from the same manufacturer and production batch.
From the above results, we can conclude that we must establish a special calibration equation for each infrared thermocouple to ensure the accuracy of its measurement performance.
The results of this study indicated that the accuracy of two-kinds of infrared thermocouples could be significantly improved using calibration equations. 4000A infrared thermometer and the standard temperature [9]. The coefficient of determination was close to unity and the mean absolute error was 0.028 C. Similar results were reported by Savage and Heilman [10]. After finishing the laboratory test of twenty-one infrared thermometers, their report indicated that all the coefficients of determination were higher than 0.9965 and the RMSE value ranged near to unity. However, the residual plots showed that some equations cannot be recognized as adequate calibration equations, indicating that the coefficient of determination cannot be applied as the sole criterion to evaluate the calibration equation.
Mahan and Yeater [14] defined the infrared thermometer as the industrial quality infrared thermometer (IRT) and the infrared thermocouple as the consumer quality IRT. In their study, a question -can a low-cost consumer quality IRT replace an expensive industrial quality IRT for the applications in agriculture?‖ was addressed. From the above discussions, the accuracy of the infrared thermometer ranged from 0.03 C to 1.1 C [6,[8][9][10]. The results of this study indicated that the errors of the readout values without using calibration equations for the two kinds of infrared thermocouple ranged from −1.8 C to 1.8 C. However, the accuracy of these infrared thermocouples could be maintained at nearly 0.4 C after improvement by appropriate calibrations.
In the study of the estimation of stomatal conductance, the difference between canopy and air temperature was below 0.8 C [2]. The difference between canopy temperature and air temperature (canopy-air temperature) could serve as crop water stress index. The canopy-air temperature range was from −2 C to 9 C according to the study of Gontia and Tiwari [4], from −3 C to 3 C in the study of Abraham et al. [18] and from −5 C to 12 C in the study of Al-Faraj et al. [19]. The accuracy of the infrared thermocouple in this study was nearly 0.4 C after using the calibration equations. The performance test of this study was executed in the laboratory. The effects of the influencing factors on the performance in the field test need to be further studied and the measurement uncertainty of this sensor needs to be evaluated under those conditions. Considering the required performance of temperature measurement, the infrared thermocouple could not be applied to calculate the crop water stress index in this stage. The use of an infrared thermometer for in situ and real time measurements of the leaf temperature has been recommended [20]. This infrared thermocouple could serve as a mobile and useful tool to measure the leaf temperature. In this study, the performance of two kinds of infrared thermocouples was evaluated. A standard operation procedure (SOP) for infrared thermocouple performance evaluation is proposed in the following section. This SOP could be applied for other infrared thermocouples.

Standard Temperature
The standard temperature of the black source is maintained by a temperature calibrator and a blackbody source. The uncertainty of the temperature calibrator is within 0.03 C. The emittity of the blackbody source is nearly 0.9999.

Test Environment
The test environment was maintained at 25 C and the variation of the setting room environment temperature was controlled within ± 1.5 C.

Test Procedure
(a) The target temperature for calibration was maintained at different range according the practical requirement. The interval of testing temperature is 2 C.
(b) Three or more replicates are made on each measurement point.
(c) As one measurement is finished, the infrared thermocouple was taken out from the blackbody cavity for five minutes and then was put back into the cavity for further measurements.

Data Analysis
The performance of these infrared thermocouples was assessed by their accuracy and precision. The accuracy is expressed as the closeness with which a measurement value approached to the standard value. The error e i is defined as the difference between the readout value of the infrared thermocouple and the standard temperature. The precision P is expressed as the repeatability of these measurement values in the same standard environment. The precision was calculated as the estimated deviation of e i values. A smaller P value revealed better precision performance.

The Calibration Equation
The calibration equations were established using the regression analysis technique. The criteria for selecting of the best equation are the coefficient of determination R 2 , the standard deviations of estimator s, t-test of each parameters and residual plots. The inverse calibration polynomial equation was adopted.

Conclusions
In this study, the performance of two-kinds of infrared thermocouples was evaluated. The temperature maintained by the temperature calibrator and a specific black cavity served as the standard temperature. As the reading values of these sensors were used directly as true values, the error distributions ranged from −1.8 C to 1.8 C. The accuracy of these sensors could be maintained at nearly 0.4 C by the calibration equations. The form of the adequate calibration equations was different due to the different manufacturers of infrared thermocouple and the measurement range. Compared with the industrial level infrared thermometer, the main advantage of infrared thermocouples is their cost. From the performance evaluation, these sensors could be applied to in situ and real time measurement of leaf temperatures for the management of forest or agricultural resources. Future studies will include field tests and the calculation of the measurement uncertainty.