|  |

| |

Enabling better global research outcomes in soil, plant & environmental monitoring.

Differences in Quality Between Old and New HRM Instruments

Cleiton Breder Eller1, Caio Guilherme Pereira1, Rafael Silva Oliveira1, Alec Downey2,3, Stephen Burgess3
1Departamento de Ecologia, IB, Universidade Estadual de Campinas, Campinas, SP, Brazil
2ICT International, Armidale, NSW 2350 Australia
3School of Plant Biology, University of Western Australia, 35 Stirling Highway, Crawley WA 6009 Australia



Measuring the flux of water inside the xylem of plants (sap flow) is extremely useful for ecological, hydrological and agronomic studies. Sap flow data can be used to answer a number of important ecological and physiological questions about plant water use (Burgess et al 1998, Dawson et al 2007), whole plant carbon assimilation (Hu et al 2010) and large scale water fluxes relevant to land and water resource management (Whitley et al 2008; Kagawa et al. 2009; Pfautsch et al 2010; Brooksbank et al. 2011; Doody and Benyon 2011). The Heat-Ratio Method (HRM; Burgess 2001a,b) is becoming a widely used technique to monitor the sap flow in plants. The capacity of HRM to measure slow and reverse sap flow (Burgess et al 2000) and its capacity to accurately determine zero flow reference conditions are some of the reasons for the popularity of this method. Despite all the advantages, recently some concerns have arisen over improving the accuracy of this method at higher sap flow rates (Bleby et al 2008).

Considering that the fundamental basis for the HRM is examining the ratio between small temperature changes recorded by probes upstream and downstream of a heater element, it’s reasonable to expect that different hardware and experimental techniques may produce temperature responses and sap flow calculations of different accuracy. In our investigation we sought to compare the results obtained using a dedicated Sap Flow Meter (SFM1) specifically designed for HRM measurements, with results obtained using a generic multichannel datalogging platform. This latter approach uses traditional thermocouple-based temperature probes, which have been used quite widely, while the new SFM1 uses thermistors and custom designed signal-conditioning circuitry. We hypothesised that the SFM1 specifically designed for HRM would hold improvements for capturing high quality signals. We used the linearity and dispersion of the data to infer the data quality from the sensors.


We used a calibration system as described by Burgess et al. 2012 (in prep) which uses a high pressure pump to force a known amount of water through the xylem vessels of sample wood tissue. In this instance we used wood from Ochroma pyramidale (Balsa tree). We chose this species because its sapwood has a very homogeneous distribution of the xylem vessels, minimising any uncertainties and errors that might arise from non-homogeneous flow around the implanted sensor. O. pyramidale wood is also a very soft wood with very large vessels making it easy to shave the cut surface with a razor blade and observe whether the cut vessel ends are free from blockages which might also interfere with the homogeneity of flow around sensors.

We installed one SFM1 (manufactured by ICT International Pty Ltd Australia) and one traditional thermocouple-based sensor (also manufactured by ICT International Pty Ltd Australia) close together in the same piece of sapwood. Both sensors were inserted with an axial probe spacing of 7.5 mm. The traditional sensor system was programmed to emit heat pulses using a 15V power source for 3 seconds. The SFM1 was configured to emit a 40J heat pulse by controlling output from its own internal 12V source. We used raw temperature data collected from the temperature sensors to calculate the heat pulse velocity and then from the heat pulse velocity we calculated the sap velocity and finally the flow rate. All calculations and corrections were made according to Burgess (2001a,b).

As we intended to evaluate only the linearity and dispersion of the data, we chose to homogenise the values given by the sensors expressing them as percentage of the maximum value reached by each sensor during the experiment. We used a Pearson product-moment correlation test and a linear regression analysis to evaluate the relationship between the sensor values and the real values, which were calculated with the high-pressure pump data. Then we used a Student’s t-test to evaluate the difference in the standard errors between the sensors. The standard error data was normalised prior to analysis with a natural logarithmic transformation.

To investigate the mechanism which could lead to a difference in the performance between the sensors, we also analysed the raw heat pulse data from each sensor. We compared the noise levels of the heat pulse of each sensor by using a log normal regression to produce an equation which fitted the observed data from each sensor. Then we compared the differences between the error (standard error predicted – standard error observed (∇SE)) and the mean squared error (arithmetical average (observed – predicted)2 (MSE)) for each sensor. All the charts and analysis were made using the software STATISTICA, SigmaPlot and Excel 2010.



The log normal model fitted very well the raw heat pulse data from all the sensors (R2>0.95 in every case). The observed data from the SFM1 had a smaller standard error to the predicted data than the observed data from the traditional sensor system (SFM1 ∇SE=-1.32e-04(upstream); -9.76e-05(downstream); traditional ∇SE= 4.94e-04(upstream); 3.22e-04(downstream)). The SFM1 sensor also had a smaller MSE than the traditional sensor system. These results suggest that the SFM1 provided raw heat pulse data with less noise than the traditional sensor system (Figure 1).



Figure 1. Raw heat pulse data from the SFM1 and the traditional system. The black line is the temperature change from the upstream probe and the grey line is the temperature change from the downstream probe. The x-axis is the time in seconds after the beginning of the heat pulse.

The data provided by both sensors were strongly correlated with the real sap flow values before any correction for wounding. The relatively high intercept value observed in the traditional sensor regression equation suggests misalignments in the probe. When the real flow was zero (i.e. when the high-pressure pump was turned off), the sensors values should be close to zero if the sensors were perfectly aligned. Probe misalignment however is quite common during sensor installation and should not affect the linearity and consistency of the data, which is what we were concerned with in our study. There is a possibility that rather than sensor misalignment, the thermocouple temperature sensors showed more bias compared to the thermistor sensors, but to demonstrate such errors further testing would be required. The standard error in the traditional sensor system was generally higher than the SFM1 (t-test: N=7; DF=12; t-value=-2.84; p<0.05) suggesting that the traditional sensor system produced less consistent data than the SFM1 (Figure 2).



Figure 2. Relationship between the real flow data (given by the high pressure pump) and the data from the sensors, before any correction for wounding. The black circles are the means from three measurements (except some data points in the traditional sensor system that were actually measured four times) and the bars represents the standard error. The black line across the data points is the linear regression line derived from the equation in black in the chart. The red line is derived from a 3rd degree polynomial regression (equation on red) and represents the model which better fits into the data. The data provided by both sensors were linearly correlated with the real flow value (Pearson product-moment correlation: p<0.05).


The purpose-built SFM1 provided more accurate and consistent sap flow data than the traditional thermocouple based sensors interfaced to a generic logging platform. The raw temperature traces provided by the SFM1 were also much smoother (Figure 1). The better quality heat pulse data in the SFM1 may be explained by a combination of purpose-built signal conditioning electronics and use of thermistor probes. Thermistors are known to be much more sensitive to temperature changes than thermocouples in terms of the amount of voltage they produce per °C of change. The SFM1 also had a very compact design and consequently shorter wiring between needles and electronics than the traditional sensor system, which means less chance of interference with the electrical signal.

The noise level in the raw heat pulse traces is probably the reason why the sap flow data calculated using the traditional sensor system data had a larger standard error than the SFM1 (Figure 2). The HRM relies on the ratio between upstream and downstream temperature and if the upstream and downstream temperatures vary a lot (as observed in the traditional sensor system) than the HRM will produce more variable sap flow data as well. However, the noise levels in the heat pulse data did not seem to interfere with the overall linearity of the data. In the present case some non-linearity was observed for both sensors, particularly at very high flow rates, but this is consistent with other methods such as the compensation heat pulse method and is routinely rectified by the application of wound correction functions (see Swanson and Whitfield 1981). In fact, our results, which are some of the first using a new calibration technique, will, along with numerical modelling exercises, help refine correction functions and improve the performance of the HRM over a wider range of conditions and flow rates. Our calibration methods themselves require some refinement (e.g. cross-checking of system performance under high pressures) and a more detailed journal publication on these topics will follow in the near future.


We acknowledge the financial support of ICT international and CAPES. We also acknowledge the help and support provided by the staff from the University of Western Australia.


Bleby TM, McElrone AJ, Burgess SSO. 2008. Limitations of the HRM: great at low flow rates, but no yet up to speed? 7thInternational ISHS Workshop on Sap Flow Abstracts, Sevilla, Spain

Brooksbank K, Veneklaas EJ, White DA, Carter JL (2011) Water availability determines hydrological impact of tree belts in dryland cropping systems. Agricultural Water Management 100:76-83. doi: 10.1016/j.agwat.2011.08.016

Burgess SSO,Adams MA,Turner NC,Ong CK.1998.The redistribution of soil water by tree root systems. Oecologia 115:306–311.

Burgess SSO, Adams MA, Bleby TM. 2000. Measurement of sap flow in roots of woody plants: a commentary. Tree Physiology 20: 909–913.

Burgess SSO, Adams MA, Turner NC, Beverly CR, Ong CK, Khan AAH, Bleby TM. 2001a. An improved heat pulse method to measure low and reverse rates of sap flow in woody plants. Tree Physiology. 21: 589-598

Burgess SSO, Adams MA, Turner NC, Ong CK, Khan AAH, Beverly CR, Bleby TM. 2001b. Correction: an improved heat pulse method to measure low and reverse rates of sap flow in woody plants. Tree Physiology 21:1157.

Burgess, SSO Bleby TM and others 2012. Calibration of plant sap flow gauges: past approaches and present options. Tree Physiology
Doody T, Benyon R. 2011.Quantifying water savings from willow removal in Australian streams. J. Environ. Manage. 92:926-935. doi: 10.1016/j.jenvman.2010.10.061

Hu J, Moore DJP, Riveros-Iregui DA, Burns SP, Monson RK. 2010. Modeling whole-tree carbon assimilation rate using observed transpiration rates and needle sugar carbon isotope ratios. New Phytologist 185:1000-1015.

Kagawa A, Sack L, Duarte K, James S. 2009. Hawaiian native forest conserves water relative to timber plantation: Species and stand traits influence water use. Ecological Applications 19:1429-1443. doi: 10.1890/08-1704.1

Pfautsch S, Bleby TM, Rennenberg H, Adams MA. 2010. Sap flow measurements reveal influence of temperature and stand structure on water use of Eucalyptus regnans forests. Forest Ecology and management 259: 1190-1199.

Swanson RH, Whitfield DWA. 1981. A numerical analysis of heat pulse velocity and theory. Journal of Experimental Botany. 32:221-239.

Whitley R, Zeppel M, Armstrong N, Macinnis-Ng C, Yunusa I, Eamus D. 2008.A modified Jarvis-Stewart model for predicting stand-scale transpiration of an Australian native forest. Plant and Soil 305:35-47.