Cost-efficient approaches to measure carbon dioxide (CO2) under different environmental factors such as temperature and humidity using mini loggers

(1)

1

Cost-efficient approaches to measure

carbon dioxide (CO

2

) under different

environmental factors such as temperature

and humidity using mini loggers

Jasmine Lander

Degree project for Bachelor of Science in Chemical Analysis, engineering, at the department of Thematic Studies—Environmental Change,

(2)

2

Abstract

Fluxes of carbon dioxide (CO2), being a major greenhouse gas, are of great importance to

understand and monitor. Findings have shown that while CO2 emissions enhanced by humans

triggered the greenhouse gas effect, several significant CO2 fluxes in nature that are climate

sensitive may still be poorly constrained, especially those from inland waters and its surrounding soils and sediments. This including different processes such as decomposers degrading organic material. At present, direct measurements of CO2from soils, waters or CO2

concentrations in surface water, are typically labour intensive or require costly equipment. Therefore, small inexpensive CO2 mini loggers, originally made for indoor air quality

monitoring, are for this project being developed further for field use, as a convenient equipment to measure CO2 emissions. However, a requirement is that the mini loggers are

stable and robust against interference by other air components, including water vapour, and physical factors such as temperature. Therefore, the mini loggers were for this project studied further under different environmental conditions such as temperature and humidity in a controlled environment. The results were analysed using multiple regression analysis where the CO2 concentration (CO2), measured by the LGR instrument, versus the logger IR signal

(IR), temperature (T) and relative humidity (RH) were studied. Unlike some previous studies, this project studied a large CO2 concentration interval (400 – 10 000 ppm).

The results show that there was a strong regression for IR versus CO2. The regression for both

RH and the T was on the other hand very weak. However, there were factors that could have affected the mini logger. This since it was noticed that the mini loggers’ ability to calculate the CO2 concentration was worsened when the humidity was increased during the

experiments. It is believed that the cause was condensed water, gathering in the inner parts of the mini logger. This because the sensor chamber had a lower temperature than the hot air entering the sensor chamber from the humidity bottle, together with the CO2, leading to

condensation. Hence, prevention of condensation inside the measurement cell is important. However, the results from the analysis shows that the factors; RH and T do not need to be taken into account when studying the CO2 concentration over a larger interval, as long as the

conditions are not condensing.

(4)

4

1. Introduction

Fluxes of carbon dioxide (CO2), being a major greenhouse gas with a fundamental

importance to the biosphere and the Earth’s climate [1] are of great importance to understand and monitor. It is known that the natural turnover of CO2 has been systematically affected by

humans through combustion of fossil fuels and changed land use, e.g. cultivation of land and logging etc, resulting in an increase of CO2 in the atmosphere. At present, CO2 is increasing

rapidly by 0.5 percent each year. Because CO2 absorbs infrared radiation (IR) from the Earth

in narrow bands of wavelengths [1], the Earth’s atmosphere is now exceeding a global warming, resulting in serious consequences for ecosystems [3].

Findings have shown that while CO2 emissions enhanced by humans triggered the greenhouse

gas effect, several significant CO2 fluxes in nature that are climate sensitive may still be

poorly constrained, especially those from inland waters and its surrounding soils and

sediments and including different processes such as decomposers degrading organic material. [4] In general, input of carbon compounds flushed to fresh waters by rain, lake mixing in autumns and other seasonal factors influence CO2 emissions in inland waters. [5] In general,

the processes that drive CO2 fluxes in inland waters are complex since they include several

biological and physical factors such as turbulence in the water, depth, bottom roughness as well as the temperature of the water. Therefore, the magnitude and mechanisms of CO2 fluxes

from fresh waters are still not fully understood and therefore not adequately quantified, leading to a high uncertainty in upscaling approaches. [6] Due to the importance of CO2 emissions, concentrations and the need to cover temporal variability, a number of automated techniques have been developed, and an increasing number of commercial systems have become available. Direct measurements of CO2 fluxes across soil-atmosphere and water-atmosphere interfaces often rely on flux chamber (FC) measurements. For this technique, the system in focus is covered by a chamber. The change in CO2 over time, in the chamber headspace, is then used to calculate the CO2 fluxes. These measurements of CO2 fluxes from soils or water, or CO2 concentrations in surface water, usually requires expensive equipment to measure and log carbon dioxide levels, or are labour intensive, and therefore time

consuming. This if gas concentrations in the chamber headspace depend on manual sampling and analyses. This can be problematic since a high cost of the measuring equipment restricts the study to a few measurement units, which leads to the information of spatial variability being lost. Therefore, there is a severe limitation of estimating CO2 exchange across land- or water surfaces and the atmosphere. [4] Recent studies may solve this problem by equipping these floating chambers with low-cost CO2 loggers (mini loggers) to quantify CO2fluxes and

waterside CO2 partial pressure [7]. These small, inexpensive mini loggers, originally made for indoor air quality monitoring, are therefore for this project being developed further for field use, as a convenient equipment to measure CO2 emissions. The mini loggers do not typically

have the same high performance and sensitivity as the present commercial instruments such as the Greenhouse Gas Analyzer (LGR instrument), but these mini loggers would be a cost- and labor efficient alternative for direct measurements and monitoring of CO2 in terrestrial and aquatic environments and may therefore be highly beneficial. [4] The mini loggers use non-dispersive infrared (NDIR) spectroscopy [19], and has an excellent durability, being one of the most reliable low-cost sensors for atmospheric CO2 concentration measurement. [8]

Previous studies have shown that these sensors may be good enough for environmental application, where if the lower cost would allow simultaneous deployment of a large number of measurement units [4]. However, a requirement is that the mini loggers are stable and robust against interference by other air components, including water vapour, and physical factors such as temperature.

(5)

5

Therefore, the mini loggers were for this project further studied under different temperatures and humidity’s in a controlled environment. The results were studied using multiple

regression analysis where the regressions of the CO2 concentration (CO2), measured by a

LGR instrument [10], versus the IR signal (IR), temperature (T) and relative humidity (RH), measured by the mini logger, were studied. Unlike some previous studies where the

performance of the mini loggers has been studied around 400 ppm [9], the regressions were for this project studied for a larger interval of CO2 concentration (400 – 10 000 ppm). In order

to study these interferences, an optimal experimental setup and injection method for the CO2

(6)

6

2. Instrumentation and methods

2.1. Instrumentation and experimental setup

The experimental setup is described in the sections below. The setup consisted of an LGR instrument (section 2.1.3.), a sensor chamber (section 2.1.1.) which contained a mini logger (section 2.1.2), and a system used to change the humidity inside of the sensor chamber (section 2.2.2). The setup for the humidity changes was placed outside of the sensor chamber (Figure 7).

2.1.1 Sensor chamber

The sensor chamber consisted of a 40 cm long and 16 cm wide transparent plastic tube, and two caps which were attached to each end of the plastic tube (Figure 1). On one of the caps, five 2.0 cm wide (OD) holes were drilled, and on the other cap one 2.0 cm wide (OD) hole was drilled (Figure 2). Plastic cylinders containing cleats, with four holes in each cleat, were put in each of these holes except for one. This since a tube as wide as the hole was pulled through this hole (Figure 2). Therefore, no cleat was needed for this hole in order to isolate the sensor chamber from the surrounding environment. The tube which was pulled though this hole was a 2.0 cm wide (OD) and 1.0 m long plastic tube (PVC hose reinforced). This was the gas inlet for the sensor chamber. Through the other holes on the same cap, a USB-cable for the mini logger and USB-cables connecting the mini logger to the power supply was pulled. On the other cap, an approximately 4.0 m long plastic tube (5 mm OD and 3 mm ID), made of polyurethane plastic, was pulled through the hole. This was the gas outlet for the sensor chamber. A small fan was connected to the end of this plastic tube, inside of the sensor chamber. In the remaining empty holes, plastic plugs were put. To further seal the chamber, and thereby reducing the risk of gas leakage, plastic caps were screwed on top of the plastic cylinders, containing the different cables and tubes (Figure 2). The mini logger was screwed onto a 30 cm long and 9 cm wide metal plate (Figure 3). This plate was placed inside of the sensor chamber. The power cables were connected to the sensor chamber and to a power supply, placed outside of the chamber. The USB-cable was connected to the mini logger and the computer, via an extension cord.

Figure 1, The sensor chamber in which the experiments took place. The chamber contained a mini logger and a small pump. To the mini logger, an external power supply was connected. During the experiment, the carbon dioxide would enter through the gasinlet (the plastic tube to the left). The concentration of the carbon dioxide would then be measured by the mini logger before exiting the sensor chamber through the gas outlet (the plastic tube to the right).

(7)

7

Figure 2, One of the two caps of the sensor chamber. This cap contained five 2.0 cm wide (OD) holes. These holes contained plastic cylinders, in which a plastic cleat with four holes was put. Cables connecting the mini logger to an external power supply and a USB-cable was pulled through the cleats. The plastic cylinder containing the 2.0 cm wide (OD) and 1.0 m long plastic tube (gasinlet) did not contain a cleat. Plastic plugs were put in the holes of the plastic cleats which did not contain cables or tubes. This in order to isolate the sensor chamber from the surrounding environment.

(8)

8

Figure 3, The metal plate, placed inside of the sensor chamber, containing the mini logger. A USB-cable and a cable, connecting the mini logger to an external power supply, was connected to the mini logger.

2.1.2. Mini logger

The mini loggers (CO2 sensors) used was the K33 ELG sensor made by Senseair

(www.senseair.com). These mini loggers are suitable for low power applications and can be put into sleep-mode between measurements. The K33 ELG sensors measure and stores records of relative humidity (RH) and temperature and have the ability to detect carbon dioxide (CO2) up to 10 000 ppm, using non-dispersive infrared (NDIR) spectroscopy with an optical filter. The logging of CO2, temperature and relative humidity is simultaneous. The operating temperature range is 0-50°C, with temperature compensated CO2 values, and the sensors are fully functionally at high humidity, from 0-99 %, given non-condensing conditions. [11]

The mini logger was bought from Senseair [11] and prepared according to section “2.2

Sensor adaption for field use and initial calibration” in [4]. However, the mini logger used for this project was not varnished because the mini logger was not exposed to an outside

environment but was placed in a sensor chamber where the environment was controlled.

2.1.3. Greenhouse Gas Analyzer (LGR instrument)

The Greenhouse Gas Analyzer (LGR instrument) used was the Greenhouse Gas Analyzer (CH4, CO2, H2O) model GGA-24EP, manufactured by Los Gatos Research

(9)

9

measurements of carbon dioxide (CO2), methane (CH4) and water vapor (H2O). The LGR instrument is easy to use, and its ability to simultaneously measure these three gases at high speed and over a wide range of mole fractions makes it a good choice for chamber flux measurements. The LGR instrument uses Off-axis ICOS technology, a fourth-generation cavity enhanced laser absorption technique. Off-axis ICOS has many advantages such as a short measurement time, providing measurements over a wide dynamic range and being rugged and alignment insensitive. The LGR instrument also have an internal computer which can store data on its hard disk drive. [10]

2.1.4. Experimental setup

The experimental setup consisted of an LGR instrument, a sensor chamber in which the mini logger was located (See section 2.1.1), a system used for changing the humidity (See section 2.2.2) and plastic tubes, which connected every part of the system to each other (Figure 4). Hence, the system (the inner parts of the experimental setup) was isolated from the

surrounding environment while the experiments were running. To the LGR instrument, an external pump was connected. The pump used was a S2000 vacuum diaphragm pump (Max. pump speed: 9.6 L/min, Max. vacuum: 10 mbar) made by Picarro (www.picarro.com). The workflow of the system, and the different parts of the experimental setup, will now be explained. The pump created a flow of air inside of the different parts of the system and the plastic tubes, connecting every part of the system to each other. Hence, the air, containing the CO2, inside of the closed system travelled from the LGR instrument, through the system and back to the LGR. The plastic tubes (5 mm OD and 3 mm ID) were made of polyurethane plastic. The length of these tubes depended on the distance between the different parts of the system, varying from 2-4 m. The CO2 was injected at the gas inlet (Figure 4) and mixed using a mixing syringe (See section 2.2.1). In order to seal the system, and isolate it from the

surrounding environment, 3-way luer-lock valves were placed between the gas inlet/mixing syringe and the polyurethane plastic tube, connecting the polyurethane plastic tube to the gas inlet. The external pump, connected to the LGR instrument, would then transport the gas mixture through the humidity system (See section 2.2.2) and into the sensor chamber. A USB-cable was used to connect the mini logger inside of the sensor chamber to a computer via an extension cord. This allowed the CO2 values etc, measured by the mini logger, to be displayed on the computer. The gas mixture would then travel out of the sensor chamber, through a water trap (to stop liquid water condensing in the tubes), into the LGR instrument.

(10)

10

Figure 4, A schematic diagram showing the workflow of the system (experimental setup). The CO2 was injected into the

polyurethane plastic tubes with the gas stream (background air) through the gas inlet. The pump, connected to the LGR instrument, transported the CO2 through the plastic tube, the humidity bottle, the sensor chamber and the water trap before

reaching the LGR instrument. The mini logger, inside of the sensor chamber, would carry out five short measurements of the CO2 concentration and calculate the mean of these measurements before displaying this mean on the computer, connected to

the mini logger via a USB-cable. The LGR instrument also did measurements of the CO2, which values were displayed on

the LGR instrument itself.

2.2. Methods

2.2.1. Injection method

It was tried to inject the CO2 (100 mL) using an injection needle. However, this was not an optimal method due to a very small increase in the CO2 concentration for each injection. Therefore, the CO2 was injected simply by breathing into the sensor chamber via the gas inlet (Figure 4). For this injection method, a 3-way luer lock valve was used to open and close the tube, which was connected to the sensor chamber via the humidity bottle (see section 2.2.2.). When injecting the CO2, the valve was turned to create a free passage for the CO2 to pass from the lungs of the experimenter and into the plastic tube (Figure 5). The experimenter would then blow quite hard into the gas inlet for ten seconds. This due to the small, 5 mm (ID) wide opening of the gas inlet. The CO2 inside of the plastic tube was mixed ten times immediately after the injection, using a mixing syringe with a volume of 100 mL, connected to the same tube via the valve. This by first turning the valve, creating a free passage between the mixing syringe and the plastic tube while closing the passage between the gas inlet and the plastic tube, and then drawing out the CO2, filling the mixing needle, and pushing it back into the system. This might not have been a necessary step of the injection since it did not seem to contribute to a faster or more precise measurement of the CO2. Despite this, it was decided to mix the gas since it was a quick and easy step of the injection method. The CO2 was injected ten times for each humidity level (Figure 6). This to be able to reach the maximum concentration of 10 000 ppm CO2, starting at approximately 400 ppm.

(11)

11

Figure 5, The carbon dioxide (CO2) was injected through the gas inlet into the polyurethane plastic tube. The CO2 was then

mixed ten times using the mixing syringe. This by first turning the valve (3-way luer lock valve), creating a free passage between the mixing needle and the plastic tube while closing the passage between the gas inlet and the plastic tube, and then drawing out the CO2, filling the mixing needle, and pushing it back into the system. After this, the valve was turned, isolating

the system from the surrounding environment after the injection of the CO2. 2.2.2. Humidity setup - Controlling the humidity

For the experimental setup, a system of so-called humidity bottles was used to control the humidity inside of the sensor chamber (Figure 1). Three bottles were used for each of the humidity levels; a low, a high and one in between. The low humidity level was reached by placing the humidity bottle in an ice-bath and/or was filled with 1-2 dl of silica [21]. This depending on the decrease in humidity. The high was reached by heating up water in the humidity bottle using a hot plate, and the mid-level was reached by using a room temperature bottle (Figure 7). The room temperature bottle was either cooled down or heated up

depending on the temperature of the climate room, where the experiments took place. This in order to reach a good mid-level for the humidity. The humidity bottles used for the mid- and highest humidity levels were filled with different amounts of water. For an experiment, the humidity level in question was kept during the injection of CO2 (400 – 10 000 ppm). When the highest concentration of CO2 had been reached, the sensor chamber was opened. This in order to ventilate the chamber. After this, the humidity was increased for the next humidity level, and CO2 was injected once again. Hence, the experiment started at the lowest humidity level and finished at the highest humidity level (Figure 6).

(12)

12

Figure 6, A schematic diagram showing the changes in humidity over time during an experiment. The experiment starts at the lowest humidity level. The carbon dioxide (CO2) is injected ten times in order to systematically reach a concentration of

10 000 ppm, starting at a concentration of 400 ppm. Once the highest concentration of CO2 has been reached, the sensor

chamber is opened and ventilated. Then the sensor chamber is closed once again, where after the humidity is increased to the mid-level. After this, CO2 is injected ten times, and thereafter the sensor chamber is opened and ventilated once the

highest CO2 concentration has been reached once again. The sensor chamber is then closed before increasing the humidity

one last time. When the highest humidity level has been reached, CO2 is injected ten times. During the injections, the

humidity level is kept the same.

The humidity level was changed by moving the cap, connected to one of the three humidity bottles and to the sensor chamber via a 2.0 cm wide (OD) and 0.8 m long plastic tube, to the bottle containing a higher humidity than the one previously used (Figure 7). The different humidity levels were kept by changing the temperature of the environment inside of the bottles using ice, a hot plate or by placing the humidity bottle on the floor of the climate room.

Figure 7, A schematic diagram of the experimental setup used for changing the humidity inside of the sensor chamber. Three humidity levels were used; a low, a high and one in between. The low humidity level was reached by placing the humidity

(13)

13

bottle in an ice-bath and/or was filled with 1-2 dl of silica, and the high humidity level was reached by placing the bottle on a hot plate. The mid-level was reached using a room temperature bottle, which was either cooled down or heated up depending on the low- and the high humidity levels. The humidity bottles used for the mid- and the high humidity levels were filled with different amounts of water. The humidity level was changed by moving the cap, connected to one of the three humidity bottles and to the sensor chamber, to the humidity bottle containing a higher humidity than the one previously used.

The experiments were carried out in a climate room, in which all of the instrumentation (see section 2.1.4), except the LGR instrument, was put. The different changes in the experimental setup, and the result of these alterations, can be studied in Table 1 below. The changes were being made due to difficulties in achieving a rapid change in humidity and increase of carbon dioxide, which was desired. The final experimental setup is described in section 2.1.4. above.

Table 1, A table listing the different alterations in the experimental setup, used to control the humidity inside of the sensor chamber. The results of these changes are also displayed in the table. The resulted is deemed satisfying or not satisfying.

Experimental setup/changes in the setup Result

The humidity bottle used for the lowest humidity level was an empty 1 L bottle, which was placed in an ice-bath. For the mid-level, a 1 L bottle filled with 2 dl of room temperature water was used. For the highest humidity level, a 1 L bottle, containing 2 dl of boiling water, was used. A hot plate was used to keep the water boiling. A volume of 100 mL CO2 was injected 10-15 times (until a concentration of 10 000 ppm CO2 had been reached) for each humidity level. The CO2 was not mixed. Before changing to the next humidity level, the sensor chamber was opened in order to vent the chamber from CO2.

Not satisfying results. The increase in CO2

concentration was very slow. Despite this, the rate of the increase was accepted at this time. The increase in humidity between the different humidity levels was also very slow. The desired humidity for the lowest humidity level could not be reached.

An approximately 1.0 m long (5 mm OD and 3 mm ID) plastic tube made of polyurethane, placed between the gas inlet and the humidity bottle (Figure 4), was replaced by a copper tube (5 mm OD and 2 mm ID).

Satisfying results. A blockage in the system

was detected and removed. This increased the rate of both the change in humidity and the CO2 concentration remarkably. The copper tube was kept since it seemed to slightly accelerate the rate of the increase in CO2 concentration. However, it was desired to increase the rate further since it was still not optimal. The filter, connected to the LGR, was removed

since the rate of the changes in humidity and CO2 concentration was still small. A small leak at the end of one of the plastic tubes, made of polyurethane, was detected and taken care of. This by tightening the plastic tubes using cable ties.

Satisfying results. By removing the filter, it

was eliminated as the cause to the small change in humidity and the CO2 concentration. The humidity and the CO2 concentration seemed to increase at a quicker rate, but the changes were still not optimal.

The filter was reconnected to the LGR. The 1 L bottle used for the mid-humidity level was filled entirely with water. This to be able to increase the humidity at a quicker rate. A water trap was placed between the sensor chamber and the inlet to the LGR.

Satisfying results. The humidity was increasing

at a quicker rate. The water trap prevented condensed water from the sensor chamber to enter the LGR.

A small fan was connected to the gas outlet of the sensor chamber and the power supply inside of the chamber. This to circulate the air around the mini logger in order to receive more accurate measurements of the CO2

Satisfying results. There was no visual change

of the measurements displayed on the computer when installing the fan. The silica seemed to decrease the humidity further and at a quicker rate when studied visually.

(14)

14

concentration, IR signal etc. To the bottle used for the lowest humidity level, 2 dl of silica was put in order to decrease the humidity further and at a quicker rate.

Since the pump inside of the LGR stopped working due to dirt gathering inside of the pump, an external pump was installed. This pump was connected to the outlet, of the sensor chamber, on the LGR.

Satisfying results. The external pump

transported the air around the closed system via the plastic tubes.

The copper tube, between the gas inlet and the humidity bottle (Figure 4), was removed since the humidity did not appear to be affected by the metal. The copper tube was replaced by the 1.0 cm wide plastic tube, which was used before replacing this with the copper tube.

Not satisfying results. There was no visual

difference in the humidity when removing the copper tube from the system.

The 1 L bottles used for controlling the

humidity was switched to 2 L bottles, except the one for the lowest humidity level. This 1 L bottle still contained 2 dl of silica.

Satisfying results. A visual change in the rate

of the change in humidity was detected.

However, this change was small and could have been affected by other unknown factors. In order to fix the problem of the temperature

increasing slightly inside of the sensor chamber during the experiment (for the highest humidity level), one more bottle was installed. This bottle was filled with water and placed between the humidity bottle and the sensor chamber.

Not satisfying results. The temperature inside

of the sensor chamber was still affected. However, the temperature did not seem to increase by more than three degrees Celsius. Therefore, the change in temperature could easily be tracked using a thermometer if needed. Furthermore, this bottle caused condensed water entering the sensor chamber during the

experiments. Since the mini logger’s ability to measure the CO2 concentration are affected when contaminated with condensed water, this bottle had to be removed from the experimental setup.

The bottle described above was removed from the system.

Satisfying results. The problem of condensed

water entering the sensor chamber was

removed. The temperature inside of the sensor chamber was not negatively affected by

removing this bottle. This since the temperature of the sensor chamber did not increase by more than three degrees Celsius and therefore easily could be tracked using a thermometer. 2.3. Result evaluation - Multiple regression analysis

The data measured by the mini logger and the LGR instrument was studied using multiple regression analysis, where the regression was performed with log-transformed sensor IR-signal data. Multiple regression is an extension of the simple linear regression, used in the analysis of relationship between dependent and multiple independent variables [19]. Multiple regression formula is used to predict the value of a variable based on the value of two or more other variables. For this project, the CO2 concentration measured by the LGR instrument is the variable we want to predict. Therefore, the CO2 concentration is the dependent variable (yi). The variables used to predict the value of the dependent variable are called the

independent variables (xi). Therefore, the independent variables are; the IR-signal (IR) produced by the mini logger and the temperature (T) and the relative humidity (RH), measured by the mini logger. [20]

(15)

15

In the general multiple regression model, the dependent variable (yi) is a linear combination of the independent variables (xi). For modelling n data points, there is therefore independent variables and parameters (𝛽_𝑖). The formula of the multiple regression analysis is represented by the following equation yi [19],

𝑦𝑖 = 𝛽0+ 𝛽1𝑥𝑖 + 𝛽2𝑥𝑖2+ 𝜀𝑖 , 𝑖 = 1, … , 𝑛 𝛽0, 𝛽1, 𝛽2 = 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠

𝜀_𝑖 = error term

Given a random sample from the population, the population parameters can be estimated, and the sample linear regression model can be obtained,

𝑦̂_𝑖 = 𝛽̂ + 𝛽₀ ̂ 𝑥₁ _𝑖

The residual (𝜀𝑖) is the difference between the value of the dependent variable predicted by the model 𝑦̂𝑖 and the true value of the dependent variable 𝑦𝑖. A method of estimation is the ordinary least squares, where the method obtains parameter estimators that minimize the sum of squared residuals (SSR). If this function is minimized, it will result in a set of normal equations; a set of simultaneous linear equations in the parameters. These are solved to yield the parameter estimators (𝛽̂ , 𝛽₀ ̂ ). ₁

The multiple regression analysis was made using excel – data analysis – regression. The level of the confidence interval was 95 %. The results were analysed by studying r-square and standard error for each of the equations (See 3. Results, table 2). The r-square (coefficient of determination) value is the proportion of variance in the dependent variable that can be explained by the independent variables. That is, the r-square value is the proportion of variation accounted for by the regression model. Unlike the p-value, which states if there is a regression, the value of r-square and standard error shows how strong this regression is, where a high value of r-square and a low value of standard error shows a strong regression for the variables. [20] This because as r-square increases and standard error decreases, the data point moves closer to the regression line [9]. Therefore, a low value of r-square and a high value of standard error shows a weak regression. The standard error of the regression is in the units of the dependent value and provides the absolute measure of the typical distance that the data points fall from the regression line. R-square provides the relative measure of the fraction of the dependent variable variance that the model explains. Hence, r-square can range from 0 to 1. [20]

(16)

16

3. Results

The regression between the dependent variable (yi); CO2 concentration (CO2), and the independent variables (xi); IR-signal (IR), Temperature (T) and relative humidity (RH) were studied using the values from the mini logger and the LGR instrument. In Table 2 below, the results from the multiple regression analysis of the variables can be studied. The mole fraction of carbon dioxide in air (in ppm), measured by the LGR instrument, is presented as CO2. The IR-signal (infrared light), produced by the mini logger, is presented as IR and the temperature and the relative humidity, measured by the mini logger, is presented as T and RH. The regressions of the different variables were studied for the temperatures; 5, 20 and 30 degrees Celsius, merged together in one analysis. In Table 2, the equations for the multiple regression analyses is presented and the values of r-square and standard error can also be studied. The standard error of the regression is in the unit of the dependent value; mole fraction of carbon dioxide in air in ppm and provides the absolute measure of the typical distance that the data points fall from the regression line. R-square provides the relative measure of the fraction of the dependent variable variance that the model explains. Hence, r-square can range from 0 to 1. A high value of r-r-square and a low value of standard error shows a strong regression (See section 2.5. Result evaluation - Multiple regression analysis).

Table 2, A table listing the different equations of the multiple regression analysis made, and the results of these analysis. The results are presented as r-square and standard error, where a high value of r-square and a low value of standard error demonstrates a strong regression. The standard error of the regression is in the unit of the dependent variable CO2; mole

fraction of carbon dioxide in air in ppm, and the r-square ranges from 0-1. IR, T and RH denote the sensor IR signal, temperature (C) and relative humidity (%), respectively.

Equation of the multiple regression analysis r-square Standard error (ppm)

CO2 = 205779.17 – 47112.24logIR – 30.58T – 6.76RH 0.951 615 CO2 = 204136.81 – 46851.57logIR – 21.44T 0.949 628 CO2 = 202029.39 – 46473.61logIR – 0.14RH 0.944 659 CO2 = 4589.52 + 23.23T + 9.14 RH 0.005 2780 CO2 = 202013.71 – 46471.51logIR 0.944 658 CO2 = 5314.26 + 11.12T 0.001 2783 CO2 = 5385.43 + 4.16RH 0.001 2783

Studying Table 2 above, the value of r-square for the independent variable IR versus the dependent variable CO2 is very high for each analysis; around 0.94-0.95. The value of

standard error is also much lower when the variable IR has been taken into account compered to when the variable IR is left out of the analysis, decreasing from approximately 2800 to 600. This shows a strong regression for IR versus CO2. Furthermore, it could during the experiments be visually studied that the CO2 correlates with IR; an increasing CO2

concentration gives a deceasing IR signal (Figure 8). However, this possible correlation was not studied further for this project (See section 4. Discussion).

(17)

17

Figure 8, A graph showing the correlation between the CO2 concentration and the IR signal (not log-transformed data),

where the CO2 concentration (ppm) is set against the IR-signal. It can be seen that an increasing CO2 concentration

correlates with a decreasing IR-signal.

It can be seen that leaving T or RH out of the equations does not seemingly affect the value of r-square or standard error. This since leaving these independent variables out of the equations gives a very small decrease of approximately 0.01 in r-square, and also a small increase in standard error. In addition, the value of r-square is very low at 0.001 for the multiple regression analysis of RH and T, and even lower when studying the equations for each of these variables alone. This compared to the analysis where IR has been taken into account. Therefor it can be assumed that the variables RH and T do not have any clear effect on the regression. The value of r-square and standard error barely differs between each other when studying the effect of T and RH. Hence, the effect of these variables is assumed to be negligible. 300 2300 4300 6300 8300 10300 12300 14000 16000 18000 20000 22000 24000 26000 C (C O 2 ) in ppm IR-signal

Visual representation of the correlation between the CO

2

(18)

18

4. Discussion

The regression for the CO2 concentration (CO2) versus the IR signal (IR) is strong; the r-square value from the multiple regression analysis is approximately 0.94-0.95, and the value of standard error is about 2000 units lower than the value of the analysis leaving the independent variable IR out of the equation. This corresponds with theory as the mini loggers use non-dispersive infrared (NDIR) spectroscopy to measure the concentration of CO2 inside of the sensor chamber [11]. The mini loggers contain small diode laser which sends out broad band light into the measurement cell and an optical filter makes only specific CO2 absorbing wavelengths hit the detector. When light is detected, the detector produces IR signals as a result of the wavelengths reaching the detector [15]. Hence, the IR signals is a direct response to the amount of CO2 detected and the optical filter was effective in excluding light wavelengths adsorbed by water vapour. [15] This explains the strong regression for the IR signal versus the CO2 value. Furthermore, the correlation between the CO2 concentration and the IR signal can also be visually studied; an increasing CO2 concentration gives a decrease of the IR signal.

The regression for both relative humidity (RH) and temperature (T) versus the CO2 value (CO2) was very low, giving a r-square value of 0.001-0.005 depending on which of the independent variables; T and RH were included in the equation for the multiple regression analysis. This shows that these variables have a low, almost non-existing, effect on the dependent variable CO2 when studying these effects over a larger CO2 concentration interval (400 – 10 000 ppm). However, there could be a small effect on the mini loggers for a higher humidity due to condensed water inside of the sensor chamber. The possible reason for this will be explained in the following text. During the experiments and the gathering of data it was noticed that a higher humidity (approximately 70-90 % relative humidity) had a negative effect on the mini logger; its ability to calculate the CO2 concentration was worsened. The IR signal is produced by the mini logger as a response to how much IR light that has been absorbed by the CO2 molecules, and therefore the mini loggers’ ability to produce a correct IR signal was also worsened. This defect, caused by a higher humidity inside of the sensor chamber, was visually expressed in either more random IR signals/CO2 values or no signals/values being displayed at all. It is believed that the cause of this was small, condensed water particles, gathering inside of the mini logger measurement cell. This since a hot plate was used to increase the humidity for the highest humidity level, resulting in the air from humidity bottle having a higher temperature than the air inside of the sensor chamber. When the air, containing a higher humidity level, entered the sensor chamber the air was therefore cooled down [16]. This is the possible reason for the condensed water seen in the sensor chamber. This condensed water had most likely also gathered in the inner parts of the mini loggers, affecting the calculations of the CO2 concentration. This is the reason for why the temperature itself also could have a small effect on the mini loggers’ ability to calculate the correct CO2 concentration. However, when the mini logger calculates the CO2 concentration, the small changes in temperature is taken into account [13]. Therefore, it is unknown how much the

temperature could affect the mini logger or if the temperature would have a negative effect on them at all. Despite this, the results from the multiple regression analysis shows that the small amounts of water gathering inside of the mini logger does not have a remarkable impact on the regression when studying the variables for a larger CO2 concentration interval (400- 10 000 ppm). Hence, the optical filter, which filters out wavelengths which otherwise would have been absorbed by water molecules, seems to work good under environmental conditions such as interference by air components such as water vapour and physical factors such as temperature. Therefore, it is assumed that the factors; RH and T must not be taken into account when studying the CO2 concentration over a larger interval. This if the temperature differences between the humidity bottle and the sensor chamber is avoided, or if the humidity is kept low enough to avoid water from condensing inside of the sensor chamber.

(19)

19

5. Conclusions and future work

It was concluded that the regression for the IR signal (IR) versus the CO2 concentration

(CO2), measured by the LGR instrument, was very strong. This since the value of r-square for IR versus CO2 was very high for each analysis; approximately 0.94-0.95. The value of

standard error was also much lower when the variable IR had been taken into account compared to when the variable IR had been left out of the analysis, decreasing from approximately 2800 to 600. This is consistent with theory; the mini loggers use

non-dispersive infrared (NDIR) spectroscopy to measure the concentration of CO2 inside of the sensor chamber (See section 4. Discussion). In addition, the correlation between CO2 and IR can be visually studied; an increasing CO2 concentration gives a decrease of the IR signal. It was noticed that leaving T or RH out of the equations did not seemingly affect the value of r-square or standard error. Therefor it can be assumed that the variables RH and T do not have any clear effect on the regression.

In order to improve future studies, the mini logger should be isolated from water gathering in the inner parts of the mini logger. This if the conditions are condensing. An easier solution would be to avoid condensation by keeping an even temperature in every part of the

experimental setup. In addition, it would be preferable to keep a closed system, and therefore to not expose the inside of the system to the surrounding environment when changing the humidity level during an experiment (in order to vent the sensor chamber). This since opening the closed system affected the climate inside of the closed system, resulting in an increased or decreased humidity which then had to be adjusted. In order to spare some time, plastic tubes, used to connect the different parts of the experimental setup to each other, should be rearranged using 3-way luer-lock valves for example. If possible, some type of ventilator could be placed after the LGR instrument in order to slowly ventilate the sensor chamber after reaching the maximum CO2 concentration of 10 000 ppm, resulting in a decreasing CO2 concentration for each humidity level. In addition, this would enable studies of the decrease of the CO2 concentration at each humidity level (See section 2.2.2.

Controlling the humidity). This because the decrease of CO2 was time consuming using the chosen experimental setup. Therefore, only the increase of the CO2 concentration was studied due to a time limit. At present, the mini loggers cannot measure CO2 concentrations above 10 000 ppm, but it would be interesting to see if these mini loggers could still be used to study environments with higher concentrations of CO2. This because of the correlation between the CO2 value and the IR signal; an increasing CO2 concentration correlates with a deceasing IR signal. Therefore, it could be possible to measure higher CO2 concentrations than 10 000 ppm, using the mini loggers, by studying the IR signal. Due to the time limit, this was not studied further.

(20)

20

Acknowledgments

I would like to thank Thanh Duc Nguyen, research engineer at the department of Thematic Studies—Environmental Change (Tema Miljöförändring), Linköping University, Sweden, for engaged discussions and valuable assistance regarding the equipment. I am also grateful to thank David Bastviken, professor at Linköping university, for the advices and support during this project. This research was supported by grants from the European Research Council (ERC; grant 725546), Swedish Research Council (grant 2016-04829) and FORMAS (grant 2018-01794).

(21)

21

References

[1] Nationalencyklopedin, koldioxid. [cited 2019-10-09] Available from: http://www.ne.se/uppslagsverk/encyklopedi/lång/koldioxid

[2] Nationalencyklopedin, kol. [cited 2019-11-09] Available from: http://www.ne.se/uppslagsverk/encyklopedi/lång/kol

[3] Globalamålen, Mål 13: Bekämpa klimatförändringarna [cited 2019-11-10] Available from:

https://www.globalamalen.se/om-globala-malen/mal-13-bekampa-klimatforandringarna/

[4] D. Bastviken, I. Sundgren, S. Natchimuthu, H. Reyier, and M. Gålfalk.

2015. Technical Note: Cost-efficient approaches to measure carbon dioxide (CO2)

fluxes and concentrations in terrestrial and aquatic environments using mini loggers. Departm ent of Thematic Studies– Environmental Change, Linköping University, Linköping, Sweden.

[5] S. Natchimuthu, I. Sundgren, M. Gålfalk, L. Klemedtsson, D. Bastviken. 2017. Spatiotemporal variability of lake pCO2 and CO2 fluxes in a hemiboreal catchment.

Department of Thematic Studies—Environmental Change, Linköping University, Linköping, Sweden. Department of Earth Sciences, University of Gothenburg, Gothenburg, Sweden

[6] _{K. Attermeyer, P. Bodmer. 2016. Assessing CO2 Fluxes from European Running Waters.}

Uppsala University, Sweden. Leibniz-Institute of Freshwater Ecology and Inland Fisheries (IGB), Berlin, Germany.

[7] K. T. Martinsen, T. Kragh, and K. Sand-Jensen. 2018. Technical note: A simple and cost-efficient automated floating chamber for continuous measurements of carbon dioxide

gas flux on lakes. Freshwater Biological Laboratory, Biological Institute, University of Copenhagen, Copenhagen, Denmark

[8] T. Yasuda, S. Yonemura and A. Tani. 2012. Comparison of the Characteristics of Small Commercial NDIR CO2 Sensor Models and Development of a Portable CO2 Measurement Device. [cited 2019-11-22] Available from:

https://www.mdpi.com/1424-8220/12/3/3641/htm

[9] Bastviken D., professor at Tema Miljöförändring at Linköping university, and examiner for this project. 2019.

[10] LGR Los Gatos Research, Greenhouse Gas Analyzer (CH4, CO2, H2O) [cited

2019-10-28] Available from: http://www.lgrinc.com/analyzers/overview.php?prodid=23

[11] Sensair.com, K33 ELG [cited 2020-01-16]. Available from: https://senseair.com/products/flexibility-counts/k33-elg/

[12] VAISALA. 2013. Humidity conversion formulas, calculation formulas for humidity. Helsinki: Vaisala Oyj. Available from: https://www.vaisala.com/en/lp/make-your-job-easier-humidity-conversion-formulas

(22)

22

[13] Nguyen T. D., research engineer at Tema Miljöförändring at Linköping university, and supervisor for this project. 2019.

[14] ABBMeasurementExpert: Ultra-portable, laser-based gas analyzers for environmental and industrial applications (2013) [Video]. Available from:

https://www.youtube.com/watch?v=p3qiYVErCc8 (2019-10-26)

[15] CO2METER.COM. 2019. Hur fungerar en NDIR CO2 Sensor Arbete. [cited 2019-11-25]. Available from: https://www.co2meter.com/blogs/news/6010192-how-does-an-ndir-co2-sensor-work

[16] SMHI, 2019. Luftfuktighet. [cited 2019-11-29]. Available from: https://www.smhi.se/kunskapsbanken/meteorologi/luftfuktighet-1.3910

[17] SG X Sensortech. 2007. Infrared Sensor Application Note 1 A Background to Gas Sensing by Non-Dispersive Infrared (NDIR). [cited 2019-10-01]. Available from: https://www.sgxsensortech.com/content/uploads/2014/08/AN1-–-A-Background-to-Gas-Sensing-by-Non-Dispersive-Infrared-NDIR.pdf

[18] CO2METER.COM. 2019. Hur fungerar en NDIR CO2 Sensor Arbete. [cited 2019-11-25]. Available from: https://www.co2meter.com/blogs/news/6010192-how-does-an-ndir-co2-sensor-work

[19] WallStreetMojo. Multiple Regression Formula. [cited 2020-03-20]. Available from: https://www.wallstreetmojo.com/multiple-regression-formula/

[20] Lærd statistics. Multiple Regression Analysis using SPSS Statistics. [cited 2020-03-20]. Available from: https://statistics.laerd.com/spss-tutorials/multiple-regression-using-spss-statistics.php

[21] Avantor. Silica gel, granules, Chameleon C 2 - 6 mm drying agent in sachets. [cited 2019-10-28]. Available from: https://uk.vwr.com/store/product/4786498/silica-gel-granules-chameleon-c-2-6-mm-drying-agent-in-sachets

Cost-efficient approaches to measure carbon dioxide (CO2) under different environmental factors such as temperature and humidity using mini loggers