Next Article in Journal
Radiocarbon Investigation of the Historic African Baobabs of Omusati, Namibia
Next Article in Special Issue
Design of Game-Based Virtual Forests for Psychological Stress Therapy
Previous Article in Journal
The Separation Mechanism of Bamboo Bundles at Cellular Level
Previous Article in Special Issue
Psychological Effects of Green Experiences in a Virtual Environment: A Systematic Review
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Forests
Volume 13
Issue 11
10.3390/f13111895
forests-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

A Model Combining Forest Environment Images and Online Microclimate Data Instead of On-Site Measurements to Predict Phytoncide Emissions

by
Yeji Choi
,
Sujin Park
,
Soojin Kim
,
Eunsoo Kim
and
Geonwoo Kim
*
Forest Human Service Division, Future Forest Strategy Department, National Institute of Forest Science, Seoul 02455, Korea
*
Author to whom correspondence should be addressed.
Forests 2022, 13(11), 1895; https://doi.org/10.3390/f13111895
Submission received: 11 October 2022 / Revised: 9 November 2022 / Accepted: 10 November 2022 / Published: 11 November 2022
(This article belongs to the Special Issue The Healing Power of Forests)
Browse Figures
Versions Notes

Abstract

:
In the existing phytoncide-prediction process, solar radiation and photosynthetically active radiation (PAR) are difficult microclimate factors to measure on site. We derived a phytoncide-prediction technique that did not require field measurements. Visual indicators extracted from forest images and statistical analysis were used to determine appropriate positioning for forest environment photography to improve the accuracy of the new phytoncide-prediction formula without using field measurements. Indicators were selected from the Automatic Mountain Meteorology Observation System (AMOS) of the Korea Forest Service to replace on-site measured climate data and the phytoncide-prediction equation was derived using them. Based on regression analyses, we found that forest density, leaf area, and light volume above the horizon could replace solar radiation and PAR. In addition, AMOS data obtained at 2 m altitudes yielded suitable variables to replace microclimate data measured on site. The accuracy of the new equation was highest when the surface area in the image accounted for 25% of the total. The new equation was found to have a higher prediction accuracy (71.1%) compared to that of the previous phytoncide-prediction equation (69.1%), which required direct field measurements. Our results allow the public to calculate and predict phytoncide emissions more easily in the future.

1. Introduction

Phytoncide, a natural volatile organic compound (NVOC), is emitted by plants and such emissions occur in large quantities in forest reserves. Its main components are terpenes, with monoterpene forming the majority portion. Phytoncide has been found to have a positive effect on human health such as anti-inflammatory, antioxidant, and antidepressant effects [1,2,3]. Specifically, alpha-pinene, a monoterpene that makes up the highest proportion of phytoncides, has been found to have anti-inflammatory, antioxidant, sedative, pain relieving, and sleep improving effects, and beta-pinene, which accounts for the next highest proportion, has anti-inflammatory, antioxidant, and antidepressant effects [3,4,5]. Other monoterpenes that make up phytoncides include limonene, camphene, and camphor are examples of monoterpenes, which also have anti-inflammatory, antibacterial, and metabolic-promotion effects [3,6,7]. As monoterpene has a positive effect on human health, there is increased public interest in phytoncide emissions and its uses [8,9]. Accordingly, the need for research to measure and predict phytoncide emissions is increasing.
Due to its health promotion effects, various countries in Asia, America and Europe are steadily conducting research on the diverse characteristics of NVOCs such as its diurnal pattern and correlation to microclimate factors [10,11,12,13,14,15]. These studies are being conducted by directly measuring the quantity of emission of NVOCs of the survey site through professional equipment for a long duration. However, direct measurement of NVOCs in the field is challenging because a combination of several environmental factors affects phytoncide generation. Therefore, rather than direct measurement in the field, the emission quantity is often predicted using the phytoncide-prediction equation derived through regression analyses of previous studies, and these indirect methods are mainly conducted in countries such as Korea, Japan, and China. These equations were derived based on microclimate factors such as temperature, humidity, wind speed, photosynthetically active radiation (PAR), solar radiation, soil temperature, and soil humidity, which mainly act as independent variables [16,17,18,19,20]. Temperature, humidity, and wind speed are relatively easy to measure in the field and it is convenient to substitute their values with external data such as that obtained from the Korea Meteorological Administration and the Automatic Mountain Meteorology Observation System (AMOS) of the Korea Forest Service. However, solar radiation and PAR must be measured directly using specialized equipment. This makes it very difficult for citizens, unlike experts and researchers, to directly measure environmental variables used in the phytoncide-prediction formula in the field and use it to calculate phytoncide emissions. Therefore, it is urgent to derive a phytoncide-prediction algorithm that can be easily accessed by the general public.
It is necessary to figure out other related indicators that can be measured by non-experts to replace solar radiation and PAR, which are largely related to the forest light environment. The easiest forms of data that citizens can extract in the forest are images and videos, which consist of light, color, subject, and composition [21,22]. The values of different indicators can be obtained through the analysis of photographs taken of forest environments. Data related to the forest light environment can be obtained in relation to light, those related to leaf area and lower vegetation can be obtained in relation to color, and those related to tree species and forest density can be obtained in relation to the subject. In addition, the values of factors related to the adjustment of the surface-area ratio in the picture can be obtained in relation to the composition.
This study was conducted with the aim of deriving a novel phytoncide-prediction algorithm that can be applied without field surveys by linking the various visual indicators that can be extracted from forest images, such as those mentioned above, with AMOS data. In this study, we tried to answer the following research questions with the aim of deriving a new phytoncide-prediction equation that does not require actual measurements in the field: (1) what visual indicators are suitable for replacing solar radiation and PAR?; (2) how should we photograph the forest environment to increase the explanatory power of the newly developed phytoncide-prediction Equation?; (3) which AMOS indicators are suitable to replace microclimate data measured at the site?; (4) is the newly developed phytoncide-prediction equation based on the selected visual and AMOS indicators significant? In the face of growing interest in phytoncides, this study developed a model that can predict phytoncide using only a single forest image, which would provide an opportunity for the general public to calculate and predict phytoncide emissions more easily.

2. Materials and Methods

2.1. Study Sites

This study was conducted in the Hannam and Seogwipo experimental forests at Jeju island, Republic of Korea (Figure 1). The Hannam experimental forest is approximately 1200 ha in size, is located 300–750 m above sea level on the southeast slope of the Hallasan mountain, and has a mixture of different broad-leaved deciduous tree species. It is characterized by temperate, warm, and subtropical climate characteristics, and an annual precipitation as high as 3000 mm. In this study, research data from this forest were collected from the Cryptomeria japonica (L.f.) D. Don. reforested area, where the trees were between 30 to 80 years old. The geographical location of the study site is 33°19′29.704″ N, 126°39′10.612″ E. The Seogwipo experimental forest ranges from 400 to 1000 m above sea level in the middle mountainous area at the foot of the Hallasan mountain and has an area of approximately 1400 ha. It contains a mix of orchid green broad-leaved and deciduous broad-leaved trees and research data in this forest were collected from the Chamaecyparis obtusa (Siebold & Zucc.) Endl. reforested area. The geographical location of the study site is 33°18′28.202″ N, 126°32′50.751″ E.

2.2. Data Collection Methods

In this study, NVOC emissions and microclimate data were measured, and 360° photographs of the forest environment were obtained directly on the survey sites from December 2021 to March 2022. AMOS data between the same period were extracted from their service website to determine whether data measured directly in the field can be replaced with indirectly measured data [23]. For NVOCs and microclimate data, experiments were conducted at three different plots within a radius of 10 m of each survey site to reduce measurement errors. Details of the NVOC, microclimate environment, forest image, and AMOS data indicators that were measured for the study are shown in Table 1.

2.2.1. NVOCs

NVOCs were collected every 2 h from morning (09:00) to sunset (17:00) once a month from December 2021 to March 2022 in consideration of the peak forest visitation times. At each of the two study sites, NVOC measurements were conducted in three distinct circular plots positioned within 5 m of the central tree based on geographical features. Per plot, three pumps were put considering the vegetational characteristics. Samples were collected using the adsorption tube method. Adsorption was conducted in tubes containing 150 mg of Markes International Inc.’s Tenax TA (Sacramento, CA, USA). The NVOCs were measured using a metric of μg/m3, where m3 indicates the volume of the environment at the measurement sites. The collected air volume of 9 L during an hour was converted to m3 according to the detailed information below.
The sample capture equipment consisted of a mini pump (MP-30KN; Sibata Scientific Technology Ltd., Saitama, Japan), which was calibrated at an adsorption error measurement prior to using the flow meter. At a flow rate of 150 mL/min, nine liters of NVOC were collected. A previous study, which was conducted to increase the accuracy and efficiency of NVOC measurements in the forest, focused on the sampling quantity and found that 9 L produced the most accurate results when compared to 1, 3, 6, 9, 12, 24, and 48 L [24]; consequently, the present study was also conducted with a 9 L sampling volume. The sample equipment was positioned on a tripod 1.5 m above the ground, and the average value was determined by repeating the procedure at each location. During the experiment, disposable polyethylene gloves and antimicrobial masks were utilized to prevent accidental contact with the tube. After collection, the sample tubes were stored at a temperature below 4 °C for 48 h prior to analysis (Table 2). To minimize the possibility of error, the value of tubes obtained without Tenax TA was also considered.
Using a gas chromatography–mass spectrometer (7890N-5975; Agilent, Santa Clara, CA, USA) with a thermal desorption device (GC-MSD; Gerstel TDS, Gerstel, Germany), the samples were evaluated qualitatively and quantitatively. The adsorption tube chemicals are concentrated in a low-temperature cryofocusing system that draws 1 mL/min of helium gas of high purity from a thermal desorption device. The system desorbed the gas for three minutes at 210 °C while maintaining a temperature of −30 °C. The compounds were then heated at 220 °C for three minutes prior to injection into a GC spectrometer and detected through an MSD.
Few measures can be employed to validate the analysis equipment and the procedures. Twenty species of standard compounds, including α-pinene and β-pinene, were used to generate a calibration curve. While using the calibration curve to determine each element’s mass number and the square of its rate of diluting standard materials, it was discovered that the linearity of most of the materials exceeds 0.997%. Examples are α-pinene (R2 = 0.997), β-pinene (R2 = 0.998), and d-limonene (R2 = 0.999). Experiments with these materials have a high linear correlation coefficient reproducibility, making them appropriate for scientific research.

2.2.2. Microclimate Environments

At each study site, temperature, humidity, wind speed, solar radiation, and PAR were recorded to collect data on microclimate environment variables. In addition, the direction and slope of the site were measured based on its locational setting. At five-minute intervals, the physical characteristics of the site environment were monitored using a portable multifunction meter (HOBO-U23 V2; Onset Computer Corp., Bourne, MA, USA). Sensors for solar radiation (S-LIB-M003, Onset) and photosynthetically active radiation (PAR, S-LIA-M003, Onset) were installed in the same area and monitored during the experiment.
In consideration of the geological features, a wind monitoring sensor (Wind Monitor O5103-45; R. M. Young Company, Traverse City, MI, USA) was put at each targeted site to collect wind velocity data. The meter was installed at a height of 1.5 m on a tripod 5 m away from a mini-pump, and the digitalized measurement records were saved and converted for the study. Using the HOBO-ware Pro software (Onset), the results were analyzed. In order to minimize the possibility of measurement errors, any data saved 5 min before and after each measurement were omitted from the study.

2.2.3. The 360° Forest Images

The 360° images of the forest environment were collected to find visual indicators that do not require on-site measurement and can replace solar radiation and PAR, which are indicators related to the forest light environment in the existing phytoncide-prediction method. The forest environment image was collected through a 360° imaging device (ONE X2; Insta360, Irvine, CA, USA) within a distance of 1 m from the tree at the center of each survey site. It was filmed at a height of 2.5 m facing north, and was filmed four times a day in the same way as for NVOC, every 2 h from 09:00 to 17:00 h, for connection with NVOC data.
These collected images were reprocessed to find the best way to photograph forest environments for predicting phytoncide emissions. A total of six two-dimensional images were extracted from one 360° file based on the area of the surface in the image. They include five photos processed such that the surface accounts for 0%, 25%, 50%, 75%, and 100% of the entire image, and one fisheye image (Figure 2). This image classification through surface area calculation was performed using Matlab R2022a based on the total number of pixels of the image.
A total of five forest light environment-related visual elements were extracted from each pre-processed image using Matlab, which are forest density, leaf area, lower vegetation ratio, and quantity of light above and below the horizon (Figure 3). In the case of leaf area and lower vegetation, the ratio was calculated by extracting the green color from the original image and dividing it based on the horizon. In the case of forest tree density, it was calculated based on brown pixels extracted from the original image. Finally, in the case of the quantity of light above and below the horizon, white and blue pixels extracted from the original image were calculated.

2.2.4. AMOS

Instead of measuring directly at the survey site, climate data for a particular site may be easily accessed online through various secondary platforms. Representative institutions that provide weather data in Korea include the Korea Meteorological Administration and K-weather, but it is difficult to apply their data to forest-related research because climate in the forests has very different characteristics from that in cities [25,26,27]. Therefore, in this study, data from AMOS that measures and collects forest microclimate in real time was used through the climate measurement tower installed in the forest.
AMOS data provided by the Korea Forest Service were collected to select climate indicators that can replace on-site measured microclimate environments such as temperature, humidity, and wind speed. AMOS data can be accessed through an online system provided by the National Institute of Forest Science under the Korea Forest Service [23,28]. In this study, mountain climate data on the day of phytoncide measurement between December 2021 and March 2022 were extracted. The extracted mountain climate data includes measurements of temperature, humidity, and wind speed at altitudes of 2 m and 10 m. When there was no AMOS measurement tower at the survey site, data from the closest measurement tower were used.

2.3. Data Analysis Methods

In this study, the analysis was performed based on a total of 418 sets of data, with phytoncide, microclimate data at the site, 360° images, and AMOS data. The analysis was performed using R 4.2.1, Python 3.10.7, and Matlab R2022a, and the detailed analysis process is shown in Figure 4. The tree species, forest density, leaf area, lower vegetation rate, and light intensity data were extracted from the six forest images included in each set, and the extracted visual indicators were linked to microclimate data measured directly at the survey site. Based on these linked data, visual indicators that can replace solar radiation and PAR were selected through multiple regression analyses using R, and a forest photography method that was most suitable for phytoncide prediction was derived.
Next, multiple regression analysis was performed using R by linking the visual indicators selected in the above process with AMOS data. AMOS indicators that can replace the microclimate indicators measured in the field were selected based on results of the analysis. Through these two analyses, a significant phytoncide-prediction equation that does not require field measurement was derived based on visual and AMOS indicators.

3. Results

Research results for each question are described in the following subsections.

3.1. Selection of Significant Visual Indicators That Can Be Extracted from Forest Images

In this study, a total of five visual elements were extracted from each image of the forest such as forest density, leaf area, lower vegetation ratio, quantity of light above the horizon, and quantity of light below the horizon. We used multiple regression analysis to examine whether there were any visual indicators that could replace solar radiation and PAR, which require direct field measurements. Two regression analyses were performed. First, NVOCs’ predictive power was examined using microclimate (temperature, humidity, and wind speed), solar radiation, and PAR values measured in the field as independent variables (Model 1 in Table 3). Second, the predictive power of the NVOCs was examined again using microclimate data measured in the field and the five visual elements described above as independent variables (Model 2 in Table 3). In the case of Model 1 (Table 3), based on on-site measured data, the predictive power of the phytoncide-prediction equation was significant at 69.1%. In the case of Model 2, based on visual indicators that did not include solar radiation and PAR, the predictive power of the phytoncide-prediction equation was found to be significant at 71.1%. This implies that the predictive power of Model 2 was slightly higher than that of Model 1, and a meaningful phytoncide-predictive equation can be derived even if visual indicators are used as phytoncide predictors instead of solar radiation and PAR. Looking at the significance values for each visual indicator of Model 2, it can be seen that forest density, leaf area, and light quantity above the horizon have a significant influence on the results. Through this, it can be seen that the visual indicators above the horizon, such as forest density, leaf area, and light quantity above the horizon, have a more significant influence than the visual indicators below the horizon such as lower vegetation ratio and light quantity below the horizon. Therefore, in deriving a new phytoncide-prediction formula, it is desirable to use forest density, leaf area, and light quantity above the horizon as indicators to replace solar radiation and PAR.

3.2. Selection of Forest Photography Techniques to Increase Prediction Power

When deriving the phytoncide-prediction equation based on the visual indicators extracted from the images of the forest environment, the angle and method of the forest environment being photographed can be of significant influence. Therefore, in this study we figured out the most desirable methods for photographing the forest with regard to the predictive power of the phytoncide-prediction equation. A total of six different images were extracted from the 360° forest image (Figure 2), and the most suitable type of image with significant phytoncide-predictive power was detected through multiple regression analysis. As shown in Table 4, the predictive power of the phytoncide-prediction equation differed depending upon the method of forest photography. The predictive power of the phytoncide-prediction equation was the highest at 87.0% when the surface area in the image accounted for 25% of the total area, followed by 81.9% in the fisheye photograph, and 75.1% when the surface area was 50%. Therefore, to derive a meaningful phytoncide-prediction formula based on visual indicators, it is desirable to photograph in a manner wherein the surface area in the image occupies about 25% of the total.

3.3. Selection of Significant AMOS Indicators to Replace On-Site Measured Microclimate Data

Of the five major independent variables used in the existing phytoncide-prediction formula, this study confirmed that solar radiation and PAR could be replaced by visual indicators extracted from forest images. The remaining three main independent variables are temperature, humidity, and wind speed and, in the currently used phytoncide-prediction Equation, values measured directly at the survey site are used. However, this study investigated whether these microclimate indicators can be replaced with online accessible AMOS data, as it aims to derive a new phytoncide-prediction equation that does not require field surveys. AMOS data includes temperature, humidity, and wind speed data measured at 2 m and 10 m altitude points in the measurement tower installed in the forest. In this study, multiple regression analysis was performed with data measured in the field using measuring equipment, AMOS data at 2 m and 10 m altitudes, and AMOS data from both 2 m and 10 m altitudes. As shown in Table 5, the predictive power of the phytoncide-prediction equation that was derived from the temperature, humidity, and wind speed data measured directly on site was 87.0%. In the case of the phytoncide-prediction equation derived from the temperature, humidity, and wind speed data from AMOS, data measured at an altitude of 2 m were found to have the highest explanatory power of 84.5%. Next, the predictive power of the phytoncide-predictive equation based on data including both 2 m and 10 m altitudes was 74.4%, and the explanatory power of data at 10 m altitude was 72.8%. Although all three types of AMOS data were found to have significant explanatory powers, it was found out that it is most desirable to use AMOS data measured at an altitude of 2 m when substituting for on-site measured microclimate data.

3.4. Derivation of a New Phytoncide-Prediction Equation Based on Selected Indicators

Taken together, our results show that to derive a new phytoncide-predictive expression that does not require field measurement, the forest environment should be photographed such that the surface area accounts for about 25% of the total image. In addition, forest density, leaf area, light quantity above the horizon extracted from the image obtained using the photography method detailed in Section 3.1, and temperature, humidity, and wind speed measured at an altitude of 2 m extracted from AMOS should be included as independent variables. Therefore, in this subsection, based on the above-mentioned independent variables selected through this study, the predictive power of the newly derived phytoncide-prediction equation was examined through multiple regression analysis (Table 6). In Model 3 (Table 6), it was found that all independent variables had significant explanatory powers, and the overall phytoncide-predictive equation also had a significant explanatory power of 84.5%. Therefore, it can be said that the new phytoncide-prediction equation that does not require on-site measurement suggested through this study was successfully derived with a high explanatory power. The newly derived phytoncide-prediction equation is as follows:
NVOCs = −4.565 + 1.341 × TEMP2m + 0.001 × HUM2m − 1.578 × WIND2m − 0.732 × DENSITY + 6.881 × LA − 0.865 × LIGHTABOVE
where, TEMP2m: AMOS 2 m altitude temperature; HUM2m: AMOS 2 m altitude humidity; WIND2m: AMOS 2 m altitude wind speed; DENSITY: forest density; LA: leaf area; and LIGHTABOVE: quantity of light above the horizon.

4. Discussion

Phytoncide is one of the forest healing elements that the general public pays significant attention to when visiting forests [16,29,30]. With the increase in the knowledge about the health-promotion effects of NVOCs through various media, there has been an increase in the attempts to directly measure the quantity of phytoncide generated in visiting forests [3,7,16,17,18,19,20,31]. However, to utilize the phytoncide-prediction formula developed so far, the general public has to measure its independent variables for an extended duration using specialized measuring equipment in the field, which is an impractical situation. Therefore, in this study, we attempted to predict phytoncide emissions using a combination of AMOS data that is accessible online and forest images that can be taken directly by forest visitors without resorting to field measurements.

4.1. Selection of Significant Visual Indicators That Can Be Extracted from Forest Images

This study conducted an analysis based on a total of four research questions. The first research question was to find visual indicators extracted from forest images that were suitable for replacing solar radiation and PAR, which are indicators that require field measurements with professional measuring equipment among the independent variables of the existing phytoncide-prediction formula. Therefore, it is essential to replace these indicators with others to predict phytoncide emissions without actual measurements. Solar radiation and PAR are indicators related to the forest light environment, and to find their replacements, five visual indicators related to the light environment were extracted from the image of the forest. These are forest density, leaf area, lower vegetation, light quantity above the horizon, and light quantity below the horizon. Among them, indicators showing significant phytoncide-predictive power were selected to substitute for solar radiation and PAR. The study found that the visual indicators from the upper part of the horizon, such as the forest density, leaf area, and light quantity above the horizon, act as significant independent variables compared with the visual indicators from the bottom part of the horizon such as lower vegetation ratio and light quantity below the horizon. The phytoncide-prediction equation that was derived based on forest density, leaf area, and light quantity above the horizon instead of solar radiation and PAR, shows a slightly improved predictive power compared with that of the existing formula, indicating that these indicators are highly likely to be utilized. Several previous studies have shown that forest density, leaf area, and forest light quantity affect NVOC generation and divergence characteristics, but no previous study has analyzed these indicators by extracting them from forest photographs [18,20,32,33,34]. Therefore, the results of this study are of significance as they prove the possibility of phytoncide prediction through visual data and fill this research gap.

4.2. Selection of Forest Photography Techniques to Increase Prediction Power

As this study found that visual indicators extracted from forest images have a significant effect on phytoncide prediction, the second question then arises: how to take forest images to further improve phytoncide prediction? This study aimed to derive a phytoncide-prediction method that would not require on-site measurements and could be easily used by the general public. Therefore, instead of specialized equipment, analysis was performed using images taken by easily accessible cameras, such as those on smartphones. The images of the forest environment were captured at a total of six different angles, which would be easy for the general public to understand and to take pictures accordingly. Based on the results of this study, when the surface area accounts for about 25% of the total picture, phytoncide-predictive power is the highest. As previous studies have shown that visual elements at the top of the surface have more influence on phytoncide prediction than those at the bottom, it may be because the phytoncide-predictive power is improved when the indicators at the top of the surface are reflected at a higher level. Therefore, when developing a service for the general public for predicting phytoncide emissions based on forest images in the future, a guiding line must be displayed at the quarter level of the camera screen to allow easy adjustment of the surface area in the image to 25%.

4.3. Selection of Significant AMOS Indicators to Replace On-Site Measured Microclimate Data

The existing phytoncide-prediction equation includes microclimate data on temperature, humidity, and wind speed as major independent variables besides solar radiation and PAR. Thus, the third research question arises with regard to suitable climate indicators that would be accessible online and could replace the need for on-site measurement of microclimate data. Although there are various sources of climate data that can be accessed online, most of them are weather environment data obtained from city centers. Unfortunately, as several previous studies have shown, the microclimate in urban and forest environments has significantly different characteristics [24,25,26]. Therefore, this study attempted to replace on-site measurements of microclimate indicators using AMOS data provided by the Korea Forest Service. AMOS provides forest microclimate data measured at altitudes of 2 m and 10 m. According to the results of this study, the highest phytoncide-predictive power was based on temperature, humidity, and wind speed data measured at an altitude of 2 m. Its predictive power also does not differ much from the explanatory power of the currently used phytoncide-prediction equation derived from on-site measured microclimate data. As shown previously, the forest microclimate environment changes significantly with altitude [35,36,37]. Therefore, considering that previous studies related to phytoncide prediction measured the forest microclimate environment at an altitude of 1.5 m, it is expected that AMOS microclimate data measured in a similar environment were significant.

4.4. Derivation of a New Phytoncide-Prediction Equation Based on Selected Indicators

Through the analysis process, it was found that linking forest density, leaf area, and light quantity above the horizon with the temperature, humidity, and wind speed values collected at an altitude of 2 m through AMOS had a positive effect on improving phytoncide-predictive power. Therefore, as the last step in this study, we examined whether the explanatory power of the phytoncide-prediction Equation based on the six new indicators selected through this study was significant. As a result of deriving the phytoncide-prediction equation through multiple regression analysis, it was possible to derive the prediction equation with a high explanatory power of 84.5%. In addition, all six indicators that were set as independent variables were found to have significant influence. Therefore, this study successfully derived a new phytoncide-prediction equation that does not require specialized equipment or field measurements.

4.5. Limitations and Future Research

There are several limitations to this study. Here, visual indicators were extracted and analyzed from images of the forest environment. Therefore, visual indicators such as height and diameter at breast height were not included in the picture, and forest environments such as age of stand and slope could not be examined. This is a limitation that can be easily supplemented using images taken with drones or professional cameras at various angles and elevations. However, in this study, which pursued the convenience of the general public, only visual indicators that can be captured using general smartphone cameras were extracted. In addition, since solar radiation and PAR, the main independent variables of the existing phytoncide-prediction formula, are indicators related to the light environment, there is a limitation in examining only the visual indicators related to light in an image of the forest environment. In future studies, it is necessary to find out whether additional visual indicators such as tree species that can be extracted by techniques such as deep learning can be used as meaningful independent variables. Several open-source libraries are being developed that make it easier for researchers to utilize deep and machine learning techniques. Several previous studies have been conducted to predict seasonal variables and grain yields using these methods, and it is expected that it will be a significant technical development if it is applied to future phytoncide-prediction studies [38,39,40,41]. In addition, this study has a limitation that the study site is limited to Jeju island in the Republic of Korea. As both survey sites are composed of single species forest trees, diversity in the forest environment may be insufficient. Therefore, caution is needed when applying the results of this study to other regions and forests, and it is necessary to examine the accuracy and applicability of the new phytoncide-prediction equation developed in this study through follow-up studies.
Although this study has several limitations, it is still relevant as it is the first attempt to predict phytoncide emissions using images of the forest environment and microclimate data collected online without a field survey. In addition, the new phytoncide-prediction technique and prediction formula developed through this study are expected to provide an opportunity for the general public to calculate and predict phytoncide generation more easily in the future. Therefore, it is expected that the national forest healing service for the general public can be advanced by providing the prediction guideline of phytoncide emissions based on the prediction model and algorithms developed in this study.

5. Conclusions

In this study, we aimed to derive a phytoncide-prediction technique that did not require field measurements. Several visual indicators extracted from forest images were used to select suitable indicators to replace solar radiation and PAR, and statistical analysis was conducted to determine the most accurate way to photograph forest environments to improve the accuracy of the new phytoncide-prediction formula without the need for field measurements. In addition, indicators were selected from AMOS that could replace on-site measured climate data and, finally, the new phytoncide-prediction equation was derived based on the selected indicators. According to the results of this study, forest density, leaf area, and light volume above the horizon were the main variables that could replace solar radiation and PAR. In addition, AMOS data from 2 m altitude were suitable as variables to replace actual climate data measured on site. In addition, the accuracy of the new phytoncide-prediction equation was found to be the highest when the surface area in the image accounted for 25% of the total. Therefore, in this study, a significant phytoncide-prediction equation was derived by linking visual data extracted from images taken with a surface area of 25% and AMOS data from 2 m altitude. The results of this study, which developed a model that can predict phytoncide emissions using only a single forest image and online microclimate data without on-site measurements, are expected to advance the national forest healing service. Furthermore, this study is expected to provide an opportunity for the general public to predict phytoncide emissions more easily in the future through the process of simply putting the values calculated from the model derived from this study into the newly developed prediction-equation.

Author Contributions

Conceptualization, Y.C., G.K. and S.P.; methodology, Y.C.; software, Y.C.; validation, E.K. and S.K.; formal analysis, Y.C.; investigation, E.K. and S.K.; resources, G.K. and S.P.; data curation, G.K.; writing—original draft preparation, Y.C. and E.K.; writing—review and editing, G.K. and S.P.; visualization, Y.C.; supervision, G.K. and S.P.; project administration, G.K. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data presented in this study are available on request from the corresponding author. Data are not publicly available as they are collected for internal research purposes by the National Institute of Forest Science, Korea.

Acknowledgments

This study was carried out with the support of the ‘R&D Program for Forest Science Technology (Project No. 2021394C10-2223-0103)’ provided by the Korea Forest Service (Korea Forestry Promotion Institute).

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Li, Q.; Kobayashi, M.; Wakayama, Y.; Inagaki, H.; Katsumata, M.; Hirata, Y.; Hirata, K.; Shimizu, T.; Kawada, T.; Park, B.J.; et al. Effect of phytoncide from trees on human natural killer cell function. Int. J. Immunopathol. Pharmacol. 2009, 22, 951–959. [Google Scholar] [CrossRef] [PubMed]
  2. Bach Pagès, A.; Peñuelas, J.; Clarà, J.; Llusià, J.; Campillo i López, F.; Maneja, R. How should forests be characterized in regard to human health? Evidence from Existing Literature. Int. J. Environ. Res. Public Health 2020, 17, 1027. [Google Scholar] [CrossRef] [Green Version]
  3. Antonelli, M.; Donelli, D.; Barbieri, G.; Valussi, M.; Maggini, V.; Firenzuoli, F. Forest volatile organic compounds and their effects on human health: A state-of-the-art review. Int. J. Environ. Res. Public Health 2020, 17, 6506. [Google Scholar] [CrossRef]
  4. Kim, S.E.; Memon, A.; Kim, B.Y.; Jeon, H.; Lee, W.K.; Kang, S.C. Gastroprotective effect of phytoncide extract from Pinus koraiensis pinecone in Helicobacter pylori infection. Sci. Rep. 2020, 10, 9547. [Google Scholar] [CrossRef]
  5. Woo, J.; Lee, C.J. Sleep-enhancing effects of phytoncide via behavioral, electrophysiological, and molecular modeling approaches. Exp. Neurobiol. 2020, 29, 120–129. [Google Scholar] [CrossRef]
  6. Dehsheikh, A.B.; Sourestani, M.M.; Dehsheikh, P.B.; Mottaghipisheh, J.; Vitalini, S.; Iriti, M. Monoterpenes: Essential oil components with valuable features. Mini Rev. Med. Chem. 2020, 20, 958–974. [Google Scholar] [CrossRef]
  7. Bach, A.; Yáñez-Serrano, A.M.; Llusià, J.; Filella, I.; Maneja, R.; Penuelas, J. Human breathable air in a Mediterranean forest: Characterization of monoterpene concentrations under the canopy. Int. J. Environ. Res. Public Health 2020, 17, 4391. [Google Scholar] [CrossRef]
  8. Ha, K.; Shin, W. Changes of the forest therapy paradigm in the post-corona era: Focusing on analysis of news search words related to forest therapy and COVID-19. J. Tour. Manag. Res. 2021, 25, 611–637. [Google Scholar]
  9. Roviello, V.; Roviello, G.N. Less COVID-19 deaths in southern and insular Italy explained by forest bathing, Mediterranean environment, and antiviral plant volatile organic compounds. Environ. Chem. Lett. 2022, 20, 7–17. [Google Scholar] [CrossRef]
  10. Miyama, T.; Tobita, H.; Uchiyama, K.; Yazaki, K.; Ueno, S.; Saito, T.; Matsumoto, A.; Kitao, M.; Izuta, T. Differences in monoterpene emission characteristics after ozone exposure between three clones representing major gene pools of Cryptomeria japonica. J. Agric. Meteorol. 2018, 74, 102–108. [Google Scholar] [CrossRef] [Green Version]
  11. Xu, Y.; Li, S.-J.; Yuan, X.-Y.; Feng, Z.-Z. Emission characteristics of biogenic volatile compounds (BVOCs) from common greening tree species in Northern China and their correlations with photosynthetic parameters. Huanjing Kexue 2020, 41, 3518–3526. [Google Scholar] [CrossRef] [PubMed]
  12. Jiang, Y.; Veromann-Jürgenson, L.; Ye, J.; Niinemets, Ü. Oak gall wasp infections of Quercus robur leaves lead to profound modifications in foliage photosynthetic and volatile emission characteristics. Plant Cell Environ. 2018, 41, 160–175. [Google Scholar] [CrossRef] [PubMed]
  13. Nunes, T.; Pio, C. Emission of volatile organic compounds from portuguese Eucalyptus Forests. Chemosphere-Glob. Change Sci. 2001, 3, 239–248. [Google Scholar] [CrossRef]
  14. Laffineur, Q.; Aubinet, M.; Schoon, N.; Amelynck, C.; Müller, J.-F.; Dewulf, J.; Van Langenhove, H.; Steppe, K.; Šimpraga, M.; Heinesch, B. Isoprene and monoterpene emissions from a mixed temperate forest. Atmos. Environ. 2011, 45, 3157–3168. [Google Scholar] [CrossRef] [Green Version]
  15. Jing, X.; Lun, X.; Fan, C.; Ma, W. Emission patterns of biogenic volatile organic compounds from dominant forest species in Beijing, China. J. Environ. Sci. 2020, 95, 73–81. [Google Scholar] [CrossRef]
  16. Choi, Y.; Kim, G.; Park, S.; Kim, E.; Kim, S. Prediction of natural volatile organic compounds emitted by bamboo Groves in urban forests. Forests 2021, 12, 543. [Google Scholar] [CrossRef]
  17. Kim, G.; Park, B.; Koga, S. Development of a prediction model for NVOC concentration with changing microclimate in Camellia japonica temple forest. J. Facult. Agric. Kyushu Univ. 2021, 66, 105–113. [Google Scholar] [CrossRef]
  18. Kim, G.; Park, S.; Kwak, D. Is it possible to predict the concentration of natural volatile organic compounds in forest atmosphere? Int. J. Environ. Res. Public Health 2020, 17, 7875. [Google Scholar] [CrossRef]
  19. Jo, Y.; Park, S. Prediction equations of phytoncide concentration in Korean pine (Pinus koraiensis) forest. Environ. Epidemiol. 2019, 3, 303–304. [Google Scholar] [CrossRef]
  20. Choi, Y.; Kim, G.; Park, S.; Lee, S.; Kim, S.; Kim, E. Statistical evidence for managing forest density in consideration of natural volatile organic compounds. Atmosphere 2021, 12, 1113. [Google Scholar] [CrossRef]
  21. Luo, Y.; Tang, X. Photo and video quality evaluation: Focusing on the subject. In Lecture Notes in Computer Science; Springer: Berlin/Heidelberg, Germany, 2008; pp. 386–399. [Google Scholar] [CrossRef]
  22. Peterson, B. Learning to See Creatively: Design, Color, and Composition in Photography; Amphoto Books: New York, NY, USA, 2015. [Google Scholar]
  23. Automatic Mountain Meteorology Observation System. Available online: http://mw.nifos.go.kr (accessed on 12 May 2022).
  24. Kim, G.; Park, B.-J.; Joung, D.; Yeom, D.-G.; Koga, S. Primary concentration measurements of natural volatile organic compounds in atmosphere using the headspace solid–phase microextraction method within the forest. J. Facult. Agric. Kyushu Univ. 2015, 60, 471–476. [Google Scholar] [CrossRef]
  25. Mues, A.; Manders, A.; Schaap, M.; Van Ulft, L.H.; Van Meijgaard, E.; Builtjes, P. Differences in particulate matter concentrations between urban and rural regions under current and changing climate conditions. Atmos. Environ. 2013, 80, 232–247. [Google Scholar] [CrossRef]
  26. Xiong, Y.; Liu, J.; Kim, J. Understanding differences in thermal comfort between urban and rural residents in hot summer and cold winter climate. Build. Environ. 2019, 165, 106393. [Google Scholar] [CrossRef]
  27. Rajagopalan, P.; Lim, K.C.; Jamei, E. Urban heat Island and wind flow characteristics of a tropical city. Sol. Energy 2014, 107, 159–170. [Google Scholar] [CrossRef]
  28. Yoon, S.; Won, M.; Jang, K. A study on optimal site selection for automatic mountain meteorology observation system (AMOS): The case of Honam and Jeju areas. Korean J. Agric. Forest Meteorol. 2016, 18, 208–220. [Google Scholar] [CrossRef] [Green Version]
  29. Lee, J.; Park, B.J.; Tsunetsugu, Y.; Ohira, T.; Kagawa, T.; Miyazaki, Y. Effect of forest bathing on physiological and psychological responses in Young Japanese Male subjects. Public Health 2011, 125, 93–100. [Google Scholar] [CrossRef]
  30. Choi, K.M.; Shin, W.S.; Yeoun, P.S.; Cho, Y.M. The influence of forest walking exercise on human, stress and fatigue. J. Korean Inst. For. Recreat. 2011, 15, 61–66. [Google Scholar]
  31. Bach, A.; Maneja, R.; Zaldo-Aubanell, Q.; Romanillos, T.; Llusià, J.; Eustaquio, A.; Palacios, O.; Penuelas, J. Human Absorption of Monoterpenes after a 2-h Forest Exposure: A field experiment in a Mediterranean Holm Oak Forest. J. Pharm. Biomed. Anal. 2021, 200, 114080. [Google Scholar] [CrossRef]
  32. Wang, H.; Wu, Q.; Liu, H.; Wang, Y.; Cheng, H.; Wang, R.; Wang, L.; Xiao, H.; Yang, X. Sensitivity of biogenic volatile organic compound emissions to leaf area index and land cover in Beijing. Atmos. Chem. Phys. 2018, 18, 9583–9596. [Google Scholar] [CrossRef] [Green Version]
  33. Oderbolz, D.C.; Aksoyoglu, S.; Keller, J.; Barmpadimos, I.; Steinbrecher, R.; Skjøth, C.A.; Plaß-Dülmer, C.; Prévôt, A.S.H. A comprehensive emission inventory of biogenic volatile organic compounds in Europe: Improved seasonality and land-cover. Atmos. Chem. Phys. 2013, 13, 1689–1712. [Google Scholar] [CrossRef] [Green Version]
  34. Guenther, A. Seasonal and spatial variations in natural volatile organic compound emissions. Ecol. Appl. 1997, 7, 34–45. [Google Scholar] [CrossRef]
  35. von Arx, G.; Dobbertin, M.; Rebetez, M. Spatio-temporal effects of forest canopy on understory microclimate in a long-term experiment in Switzerland. Agric. Forest Meteorol. 2012, 166–167, 144–155. [Google Scholar] [CrossRef]
  36. Anderson, J.; Keppel, G.; Thomson, S.-M.; Randell, A.; Raituva, J.; Koroi, I.; Anisi, R.; Charlson, T.; Boehmer, H.J.; Kleindorfer, S. Changes in climate and vegetation with altitude on mount Batilamu, Viti levu, Fiji. J. Trop. Ecol. 2018, 34, 316–325. [Google Scholar] [CrossRef]
  37. Rapp, J.M.; Silman, M.R. Diurnal, seasonal, and altitudinal trends in microclimate across a tropical montane cloud forest. Clim. Res. 2012, 55, 17–32. [Google Scholar] [CrossRef]
  38. Chang, C.-C.; Lin, C.-J. LIBSVM: A Library for support vector machines. ACM Trans. Intell. Syst. Technol. 2011, 2, 1–27. [Google Scholar] [CrossRef]
  39. Guo, Y.; Fu, Y.; Hao, F.; Zhang, X.; Wu, W.; Jin, X.; Bryant, C.R.; Senthilnath, J. Integrated phenology and climate in rice yields prediction using machine learning methods. Ecol. Indic. 2021, 120, 106935. [Google Scholar] [CrossRef]
  40. Zhou, X.; Kono, Y.; Win, A.; Matsui, T.; Tanaka, T.S. Predicting within-field variability in grain yield and protein content of winter wheat using UAV-based multispectral imagery and machine learning approaches. Plant Prod. Sci. 2021, 24, 137–151. [Google Scholar] [CrossRef]
  41. Fan, R.-E.; Chang, K.-W.; Hsieh, C.-J.; Wang, X.-R.; Lin, C.-J. LIBLINEAR: A library for large linear classification. J. Mach. Learn. Res. 2008, 9, 1871–1874. [Google Scholar] [CrossRef]
Forests 13 01895 g001 550
Figure 1. Image of the phytoncide-emission study sites on Jeju island, Korea.
Figure 1. Image of the phytoncide-emission study sites on Jeju island, Korea.
Forests 13 01895 g001
Forests 13 01895 g002 550
Figure 2. Reprocessed 360° images of the forest environment.
Figure 2. Reprocessed 360° images of the forest environment.
Forests 13 01895 g002
Forests 13 01895 g003 550
Figure 3. Example of color-extraction process.
Figure 3. Example of color-extraction process.
Forests 13 01895 g003
Forests 13 01895 g004 550
Figure 4. Flow-chart depicting the data analysis process.
Figure 4. Flow-chart depicting the data analysis process.
Forests 13 01895 g004
Table
Table 1. Indicators for natural volatile organic compounds (NVOCs), microclimate environment, forest images, and Automatic Mountain Meteorology Observation System (AMOS) data that were measured for phytoncide-prediction equations.
Table 1. Indicators for natural volatile organic compounds (NVOCs), microclimate environment, forest images, and Automatic Mountain Meteorology Observation System (AMOS) data that were measured for phytoncide-prediction equations.
NVOCsMicroclimate
Environment		360° Forest
Image		AMOS
α-Pinene, β-Pinene, camphene, limonene, benzaldehyde, myrcene, phellandrene, sabinene, camphor, α-terpinene, γ-terpinene, terpinolene, 3-carene, terpineol, bornyl acetate, sabina ketone, cineole, longifolene, pinocarvone, sabinene hydrate, cymene, valencene, α-bisabolol, farnesene, caryophyllene, nerol, nerolidol, pulegone, borneol, menthol, geraniol, D-fenchoneTemperature,
humidity,
wind speed,
solar radiation, photosynthetically active radiation (PAR)		Forest density,
leaf area,
lower vegetation ratio, quantity of light above the horizon, quantity of light below the horizon		Temperature
humidity
and wind speed
at 2 m and 10 m altitudes
Table
Table 2. Conditions for the operating parameters for NVOC detection.
Table 2. Conditions for the operating parameters for NVOC detection.
ParametersConditions
ColumnHP-INNOWAX (60 m × 0.25 mm × 0.25 μm, film thickness)
Carrier gas flowHe at 1 mL/min
Injection modePulsed splitless
Injection port temp.210 °C
Transfer line temp.210 °C
Over temp. programInitialRateFinal
3 min40 °C8 °C/min220 °C3 min40 °C
Post run220 °C, 5 min
Table
Table 3. Results of regression analysis for selecting significant visual indicators.
Table 3. Results of regression analysis for selecting significant visual indicators.
Multiple Linear Regression of Model 1 and Model 2
IndicatorsBSEβTp2Adjusted R2Fp3
Model 1(Constant)−4.7656.1100.000−1.2630.011 *0.6914.5890.037 *
Temp.0.0670.0540.5781.5440.029 *
Humidity0.0510.0690.4840.5580.000 ***
Wind Speed5.3156.7600.5190.5830.001 **
Solar Rad.−0.0150.0400.2270.4670.047 *
PAR−0.0140.039−0.2720.5530.019 *
Model 2(Constant)−0.6770.2710.000−7.3040.001 **0.7118.4050.020 *
Temp.0.0360.0100.6594.4490.000 ***
Humidity0.0450.0380.4411.7090.032 *
Wind Speed−0.4610.072−0.508−4.4180.000 ***
Forest Density−0.0041.9350.856−0.0020.049 *
Leaf Area4.7583.3120.7711.4370.015 *
Lower Veg.19.71418.9870.2041.0380.299
Light Above−0.4172.127−0.588−0.1960.044 *
Light Below−3.8396.909−0.463−0.5560.057
* p < 0.05, ** p < 0.01, *** p < 0.001. p2 Calculated probabilities of each independent variable. p3 Calculated probabilities of each model. B: unstandardized coefficient; SE: standard error; β: standardized coefficient; Temp.: temperature; Solar Rad.: solar radiation; Lower Veg.: lower vegetation area.
Table
Table 4. Results of regression analysis for the selection of the forest photography method.
Table 4. Results of regression analysis for the selection of the forest photography method.
VariablesF-StatisticsAdjusted R2Rank
100% Surface Area4.0760.1045
75% Surface Area4.0100.1026
50% Surface Area3.1510.7513
25% Surface Area3.8450.8701
0% Surface Area4.2420.1554
Fisheye6.3610.8192
Table
Table 5. Results of regression analysis for selecting AMOS indicators.
Table 5. Results of regression analysis for selecting AMOS indicators.
VariablesF-StatisticsAdjusted R2Rank
On site (HOBO)3.8450.870-
AMOS (2 m)4.5820.8451
AMOS (10 m)3.4610.7283
AMOS (Total)4.0600.7442
Table
Table 6. Results of regression analysis based on selected visual and AMOS indicators.
Table 6. Results of regression analysis based on selected visual and AMOS indicators.
Multiple Linear Regression of Model 3
IndicatorsBSEβTp2Adjusted R2Fp3
Model 3(Constant)−4.5651.2830.000−3.5580.000 ***0.84514.580.000 ***
Temp.1.3410.1790.5387.5100.000 ***
Humidity0.0010.0180.4060.0620.016 *
Wind Speed−1.5780.476-0.377−3.3140.001 **
Forest Density−0.7321.702-0.214−0.4300.006 **
Leaf Area6.8812.8270.7352.4340.015 *
Light Above−0.8651.863-0.512−0.4640.043 *
* p < 0.05, ** p < 0.01, *** p < 0.001. p2 Calculated probabilities of each independent variable. p3 Calculated probabilities of each model. B: unstandardized coefficient; SE: standard error; β: standardized coefficient; Temp.: temperature.
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Choi, Y.; Park, S.; Kim, S.; Kim, E.; Kim, G. A Model Combining Forest Environment Images and Online Microclimate Data Instead of On-Site Measurements to Predict Phytoncide Emissions. Forests 2022, 13, 1895. https://doi.org/10.3390/f13111895

AMA Style

Choi Y, Park S, Kim S, Kim E, Kim G. A Model Combining Forest Environment Images and Online Microclimate Data Instead of On-Site Measurements to Predict Phytoncide Emissions. Forests. 2022; 13(11):1895. https://doi.org/10.3390/f13111895

Chicago/Turabian Style

Choi, Yeji, Sujin Park, Soojin Kim, Eunsoo Kim, and Geonwoo Kim. 2022. "A Model Combining Forest Environment Images and Online Microclimate Data Instead of On-Site Measurements to Predict Phytoncide Emissions" Forests 13, no. 11: 1895. https://doi.org/10.3390/f13111895

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop