Received 3 September 2021; revised 6 October 2021; accepted 13 October 2021.

Date of publication 26 October 2021; date of current version 15 November 2021.

Digital Object Identifier 10.1109/OJCOMS.2021.3122286

Path Loss Prediction Model Development in a

Mountainous Forest Environment
BILGUUNMAA MYAGMARDULAM 1 , NAKAYAMA TADACHIKA1 , KAZUYOSHI TAKAHASHI 2 ,
RYU MIURA3 , FUMIE ONO4 (Member, IEEE), TOSHINORI KAGAWA5 , LIN SHAN 3 (Senior Member, IEEE),
AND FUMIHIDE KOJIMA 6 (Member, IEEE)
1 Department of Materials Science, Nagaoka University of Technology, Nagaoka 940-2188, Japan

2 Department of Civil and Environmental Engineering, Nagaoka University of Technology, Nagaoka 940-2188, Japan

3 Wireless Network Research Center, National Institute of Information and Communications Technology, Yokosuka 239-0847, Japan

4 Global Strategy Bureau, Ministry of Internal Affairs and Communications, Tokyo 100-8926, Japan

5 System Engineering Research Laboratory, Central Research Institute of Electric Power Industry, Yokosuka 240-0196, Japan

6 Wireless Communication Center, National Institute of Information and Communications Technology, Yokosuka 239-0847, Japan

CORRESPONDING AUTHOR: B. MYAGMARDULAM (e-mail: [email protected])

This work was supported by the SCOPE Project of the Ministry of Internal Affairs and Communications Japan.

ABSTRACT We consider a method for developing a radio-wave propagation prediction model in a

mountainous forested area. A new path loss development approach uses a free-space path loss (FSPL)
model and an empirical path loss model. To improve the prediction accuracy, the transmission path
distance, free space area, and forest area were calculated separately. We obtained the transmission path
distance for free space and forest areas from the digital surface model (DSM), which represents surface
elevation information, including vegetation and object height. In this study, the results showed that by
combining the empirical model with FSPL for free space area, the accuracy for all the empirical models
was improved. We confirmed that the transmission distance calculation of the free space area and forest
area with a combination of the empirical models showed a better performance than the model with
physical distance. The predicted model results were validated using the actual radio wave propagation in
the 920 MHz band measurement data. The overall path loss prediction accuracy was improved for the
empirical models average of 8.05 dB on the experimental data.

INDEX TERMS Digital surface model, LoRa, path loss prediction, modified empirical model, drone
mapper.

I. INTRODUCTION radio wave propagation characteristics in forest environ-

R ADIO-WAVE propagation modeling is a fundamen-

tal study in wireless communication technology. Since
the first radio transmission of Guglielmo Marconi in 1901,
ments. Traditionally, path loss propagation models in forested
environments can be divided into two categories: analytical
radio propagation prediction and empirical radio propaga-
radio wave propagation characteristics have been studied tion prediction. Analytical models were developed based
extensively in the development of wireless communica- on the physical characteristics of radio wave propagation.
tion around the world [1]. Radio propagation environments Physics-based analytical models require knowledge of the
have been widely divided into rural, suburban, and urban physical laws of the radio wave, the ability to use math-
areas [2]. Depending on the transmission environment, the ematical calculations, and heavy computational resources.
obstacles cause the radio wave to be reflected, refracted, Empirical models can be developed easily based on the
diffracted, scattered, or blocked. The path-loss model rep- measurement of attenuation data. The main advantage of
resents the signal power loss when it propagates from the empirical model is its simplicity and applicability
the transmitter to the receiver. Since the 1960s, numer- for any experimental area compared to the analytical
ous studies have been conducted to investigate the actual model [3].
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

2494 VOLUME 2, 2021

FIGURE 1. Experimental setup: (a) Map of Japan; (b) Experimental area; (c) receiver antenna located at Mt. Takatouge; (d) transmitter antenna mounted on the roof of vehicle.

The most popular empirical path loss model in The experiment was conducted in Kagoshima University
forests was proposed by Weissberger [4] in 1982. The Forest, which is the 5th largest university forest. There are
empirical models were developed based on the experimen- 27 universities with a university forest in Japan. The univer-
tal data COST235 [5], ITU-R [6], FITU-R [7], and lateral sity forest is a large classroom for students [22]. Focusing
ITU-R [8]. They were applied to fit the data based on the on Japan’s forestry, most of the forests are in the mountains.
experiments where each parameter had different values of There are many places where radio waves do not reach due
the determined parameters [4]–[8]. Empirical models based to the influence of forest and terrain [23].
on regression methods are commonly used to estimate propa- To establish a reliable communication network for forest
gation based on the type of vegetation environment [9]–[11]. ICT and inside of the forest there are no signal for the
Radio-wave propagation characteristics in forest environ- cellphone therefore safety of the workers and students in
ments can be divided into direct, reflected, and lateral the forest, we need a precise path loss prediction model that
waves [12]. The direct and reflected waves propagate through is suitable for this laurel forest environment.
the forest, and the lateral waves propagate through the tree
crowns over the top of the vegetation. The models were II. MATERIAL AND METHODS
developed for the situation where antennas are both located A. EXPERIMENTAL SETUP
inside the forest near the grove of trees and the signal The field experiment was conducted in February 2020 at
propagates mainly through the trees. Therefore, existing Takakuma Experimental Forest (31.5◦ N, 130.7◦ E)
empirical models in forest areas have specific advantages of Kagoshima University, Tarumizu City, Kagoshima
and disadvantages and do not apply to every propagation Prefecture, southern Japan. The dominant tree species were
environment. the Japanese cedar and Japanese cypress. The details of the
However, many empirical models have been developed measurement setup are shown in Figure 1. A map of Japan
in the measured environment, but they fail when applied to is shown in Figure 1(a), where the dark blue colored area is
different measurement environments and experimental setups the Kagoshima prefecture; in Figure 1(b), the red rectangle
for which it was developed [13]. is the location of the experimental area in this study. In this
Modified empirical models in forest environments were study we used low-cost broadcast-based location information
evaluated and developed in [18], [19], and [20]. sharing system called Drone Mapper using 920 MHz.
The reliability of the network access depends on the accu- Drone Mapper system is based on simple device-to-device
racy of the propagation model performance. Hence, the need (D2D) broadcasting protocol can establish the communica-
for a simple and realistic prediction of the empirical model’s tion network that not requiring the infrastructure or access
improvement in the prediction accuracy is maintained. point [25]. The purpose of this system is location information
The main objective of this research is to develop a simple sharing between unmanned aircrafts (UA) and their operators
and accurate model based on the combination of an empiri- or between the flying vehicle to ensure their safe operation.
cal path loss model and a free-space path loss (FSPL) model Drone mapper system collect the information name of the
to improve the overall prediction accuracy in a forested envi- vehicle, time, latitude, longitude, altitude, speed, direction,
ronment. We proposed a model in which the total signal path and the signal strength [24], [25], and [26].
loss between the transmitter and receiver is divided into two The two Drone Mapper units (the receiver and the trans-
parts: free space loss and forested space loss. In addition, the mitter) were deployed in this experiment. The experiment
proposed model was evaluated using field experiment data. was conducted by installing a receiver antenna (Rx) at

VOLUME 2, 2021 2495

MYAGMARDULAM et al.: PATH LOSS PREDICTION MODEL DEVELOPMENT IN MOUNTAINOUS FOREST ENVIRONMENT

TABLE 1. Experimental setup. TABLE 2. Parameter values for empirical model.

This model considers only the frequency and the

propagation distance and does not consider radio wave
propagation effects, such as reflection, refraction, diffrac-
tion, and absorption. Path loss expresses the loss in signal
strength as a function of the distance.
The FSPL model can be calculated as follows:

Lfspl (db) = 20 log 10(4π d/λ), (1)

where Lfspl is the free space path loss in dB, d is the distance
between the receiver and transmitter antenna in meters, and
λ is the wavelength in meters.
the summit of Mt. Takatouge (722 m high), as shown in Link budget: The link budget calculates the total received
Figure 1(c). The height of the receiver antenna is approxi- signal strength in dBm, accounting for all gains and losses
mately 2.5 m above ground. The transmitter car ran through in a transmission system. The link budget can be calculated
the inside of the experimental forest and transmit the sig- as follows:
nal collected at the receiver antenna. Figure 1 (d) shows
Pr (dbm) = Pt + Gr + Gt − L, (2)
a transmitter antenna (Tx) mounted on the roof of a vehi-
cle approximately 2 m above the ground with a 13 dBm where Pr is the received power expressed in dBm, Pt is the
transmitter power. It has an antenna gain of 0 dB. Long transmitter power in dBm, Gr is the receiver antenna gain
Range (LoRa) communication technology is used in this in dB, and Gt is the transmitter antenna gain in dB.
study. The specifications of the experimental setup are Empirical Path Loss Model in Forest: The exponential
listed in Table 1. Both the receiver (Rx) and the trans- decay approach to modeling path loss in vegetation, which
mitter (Tx) had a whip-type antenna (horizontal surface assumes that the loss increases exponentially with distance,
omnidirectional). The measured data include the received was first proposed by Weissberger [4] in 1982. The modi-
signal strength indicator (RSSI), altitude of the transmitter fied version of the model was included in the International
antenna, and geographical coordinates. Radio Consultative Committee Recommendations (CCIR) in
1986 [5].
In this study, five well-known empirical path loss models
B. OVERVIEW OF RADIO WAVE PROPAGATION MODEL for vegetation, namely, Weissberger’s [4] modified expo-
The empirical path loss model was developed based on the nential decay model, ITU Recommendation (ITU-R) [5],
experimental data, accounting for the effect on vegetation. COST235 [6], FITU-R [7], and lateral ITU-R [9] are
Each model uses a specific parameter for prediction with developed. The modified exponential decay (MED) path loss
good accuracy. The empirical path loss model provides a models A, B, and C parameter values can be empirically
method for modeling path loss data in each range of mea- determined. The A value is determined based on the foliage
sured data, but each path loss model has advantages and type, and the B and C values represent the frequency and
disadvantages. Each model has limitations that depend on distance dependencies, respectively. Models A, B, and C
the propagation environment and experimental setup, which are estimated from the experimental data, and depending on
are unable to model the path loss. To model path loss in a the experimental data, each parameter value of the fitted
mixed forest, Tx-Rx as distance is considered. We propose parameters is different. In this study, for all models, only
a path loss model combining the FSPL model using the dis- the in-leaf foliage models were used in our experimental
tance of a free space area and an empirical path loss model site. In Table 2, all empirical models used in this study are
for forest areas. summarized.
Free Space Path Loss model: The FSPL model is the The MED model can be calculated as:
theoretical data that predicts radio wave propagation path
loss in free space without obstructing objects over a distance. LMED (dB) = Af B dC , (3)

2496 VOLUME 2, 2021

 TABLE 3. Glossary of variables and parameters.

the signal. Our proposed model considers the transmission

path distance separately for free space and forest areas using
digital surface model (DSM) data.
Proposed path loss prediction model for this study can
FIGURE 2. Illustration of the direct path, the transmission path distance for the be calculated using the following equations from (4)-(9). In
forest area and for the free space area in the proposal methodology.
the model the total transmission path distance between the
transmitter and the receiver, D is divided into two parts, the
where the parameters A, B, and C are fitted values, f is the total transmission path distance for forest areas Dforest , and
frequency, and d is the distance between the transmitter and the total transmission path distance for free space areas Dfree .
the receiver.
Weissberger’s model was developed under the situation D = Dforest + Dfree , (4)
where both antennas were located in the ground with a where D is the total transmission path distance between
foliage depth of 400 m and a frequency range of 230 MHz to the transmitter and the receiver in m, Dforest is the total
95 GHz. The ITU-R model was developed at a frequency of transmission path distance for the forest area in m, and Dfree
200 MHz and 95 GHz, and both antennas were located near a is the transmission path distance for the free area.
tree grove. The COST 235 model was proposed at millimeter The Dforest and Dfree are calculated using the equa-
frequencies from 9.6 GHz to 57.6 GHz and the antennas were tion (5) and (6).
located near a tree grove. The Fitted ITU-R (FITU-R) model 
n
forest
was proposed at frequencies of 11.2 GHz and 20 GHz. The Dforest = di (5)
lateral ITU-R (LITU-R) model was proposed at frequencies i
of 240 MHz and 700 MHz, and the model considering the 
m
free
lateral wave effect. Dfree = dj (6)
j
C. PROPOSED PATH LOSS MODEL forest free
The di and
the dj
are the i-th and j-th transmission path
The proposed path loss model is obtained by combining the distance for a forest area and a free space area respectively.
FSPL path distance for the free area and empirical models
using distance for the forest area. In Fig. 2, the illustration Pfspl = Pt + Gr + Gt − Lfspl , (7)
of the proposed path loss model shows the transmission
Here, Pfspl is the received power expressed in dBm, Pt is the
distance from Tx to Rx (a), transmission path distance for
transmitter power in dBm, Gr is the receiver antenna gain
the forest (b), and free space area (c). The definitions of
in dB, Gt is the transmitter antenna gain in dB, and Lfspl is
variables are presented in Table 3.
the free space path loss in dB.
When transmitting the signal between the transmitter and
the receiver, the propagation path can be divided into direct PMED = Pt + Gr + Gt − LMED , (8)
line-of-sight paths and indirect non-line-of-sight paths. The
where PMED is the received power expressed in dBm, Pt is
direct line-of-sight path corresponds to the free space area
the transmitter power in dBm, Gr is the receiver antenna
between the transmitter (Tx) and receiver (Rx). The indirect
gain in dB, Gt is the transmitter antenna gain in dB, and
non-line-of-sight path corresponds to the forest area propa-
LMED is the path loss through vegetation in dB.
gation path by physical objects (such as buildings, terrain,    
or trees), and could affect the level of the signal or block Pproposal = Pfspr Dfree + PMED Dforest , (9)

VOLUME 2, 2021 2497

MYAGMARDULAM et al.: PATH LOSS PREDICTION MODEL DEVELOPMENT IN MOUNTAINOUS FOREST ENVIRONMENT

Here, Pproposal is the received power expressed in dBm, Pfspl

is the path loss for the free space area, and PMED is the path
loss for the forest area in dB.

D. DSM BASED OBSTACLE DETECTION IN THE

TRANSMISSION PATH
In this study, we obtained the transmission distance for the
free space area and the forest area using a DSM. The DSM
represents surface elevation information, including vegeta-
tion and object height. Specifically, all trees and terrain
height between the transmission paths were represented in a
3-dimensional (3D) database (DSM), which allowed forest
area identification in the transmission path using a simple
spatial analysis method of the profile. A DSM data resolution
of 30 m can be freely downloaded from the JAXA ALOS
Global Digital Surface Model website [21].
The surface profiles were generated using the geographic
coordinates of the transmitter (Tx) and receiver (Rx) antenna
locations. If the profile height values are below the fitting
line, they are considered the transmission path for the free
area, and if they are above the fitting line values, they are
considered for the forest area. Significant signal attenuation
is expected above the fitting line of a DSM value in the
forest area. This was repeated for all positions along the
transmission path, and the transmission path distance was
calculated.

E. EVALUATION OF THE PROPOSED MODEL

FIGURE 3. (a) Observed RSSI on a map. (b) Observed RSSI plotted on Digital
In this study, the mean absolute difference (MAD) is calcu- Surface Map.
lated to evaluate the proposed model’s results. The difference
between the observed and predicted path losses is evaluated TABLE 4. MED Parameter value of A used for this study.

based on the MAD, which was calculated as:

1
n
MAD = |xi − μi | (10)
n
i=1
Here, xi and μi are the values of the estimated and measured
RSSI data, and n is the total amount of data.

III. RESULTS
In our experiment, we measured data from the transmitter car
to the receiver antenna as well as the RSSI. The experimental the parameter values of Weissberger, ITU-R, COST 235,
environment was a mountainous forested area. The commu- FITU-R, and lateral ITU-R values at this experimental site
nication distance was up to 2000 m. The received RSSI plots were determined to be 0.0022, 0.00014, 8.01, 0.0022, and
are shown in Figure 3. The DSM map of the experimental 0.0021, respectively. However, some parameter (A) values
area is shown in Figure 3 (b). The received RSSI values were not significantly different between the original MED
indicated by the colored dots and black dots represent the values and modified parameter values.
receiver antenna. The received RSSI ranged between −87 Figure 4 shows comparisons between the measurement
and −125 dBm at the receiver, as shown in Figure 3 (a). data and predicted path loss using MED models with opti-
Communication loss occurred from 1.8 to 2 km because the mized A parameters using physical distance. The blue
transmitter car was running behind the hilly mountain with rectangle represents FSPL, light blue triangles represent
a height of 532 m. COST235, yellow rectangles represent Weissberger, red rect-
Parameter determination for MED models: From param- angles represent ITU-R, purple rectangles represent lateral
eters A, B, and C of the MED models, only parameter ITU-R, green dots represent FITU-R, and grey triangles rep-
value “A” was empirically determined through the regres- resent experimental data. From Figure 4 in this experimental
sion technique based on our experimental data. In Table 4, case, the Weissberger (yellow) and ITU-R (red) models could

2498 VOLUME 2, 2021

 FIGURE 5. Comparisons between measurement data and proposed path loss using
MED models with free space area. Blue rectangle represents FSPL, light blue triangle
FIGURE 4. Comparisons between measurement data and predicted path loss using is COST235, yellow rectangle is Weissberger, red rectangle is ITU-R, purple rectangle
MED models with physical distance. Blue rectangle represents FSPL, light blue is Lateral ITU-R, green dot is FITU-R, and grey triangle is the experimental data.
triangle is COST235, yellow rectangle is Weissberger, red rectangle is ITU-R, purple
rectangle is Lateral ITU-R, green dot is FITU-R, and grey triangle is the experimental TABLE 5. Evaluation of the developed empirical model.
data.

not predict the path loss distance of more than 1400 m. Thus,
these models are unable to predict path loss over long dis-
tances. This is because the Weissberger and ITU-R models
were developed based on a short (<400 m) forest depth
with VHF and UHF bands. Therefore, these two models
cannot be applied to large foliage depths. COST 235, lat-
eral ITU-R, and FITU-R models are predicted distances of
up to 2000 m. The COST 235 and FITU-R models showed shows the evaluation of the proposed path loss models of
similar predictions. This is because both models are derived Weissberger, ITU-R, COST235, FITU-R, and Lateral ITU-R
from plan terrain with frequencies for COST 235 (9.6 and of values were 9.83, 9.94, 7.12, 7.04, and 6.32, respectively.
57.6 GHz) and FITU-R (11.2 and 20 GHz). By examining Table 5, the best performing model in this
Comparisons of the proposed path loss model results and study is the Lateral ITU-R.
measurement data are shown in Figure 5. The result of
the proposed method Weissberger (yellow) and ITU-R (red) IV. DISCUSSION
models predicted the path loss distance of more than 1400 m; In this study, the developed path loss models were compared
in this case, these models were able to predict the path loss with the experimental results. Figure 5 compares the empir-
of up to 2000 m. COST 235, Lateral ITU-R, and FITU-R ical path loss model combined with FSPL models versus
models well predicted a path loss distance of up to 2000 m. the experimental results in a forest area in Kagoshima. Our
For the distance from 500 m to 900 m, the forest might be proposed path loss model provides all the empirical models
denser causing a decrease in received power measurement. with acceptable prediction accuracy for distances of up to
Based on Figure 5, we observed that such matching makes 2000 m.
the model more specific to a particular set of empirically In Figure 4, the Weissberger and ITU-R models are
gathered data and is less usable. unsuitable for predicting the radio wave path loss in this
Table 5 shows the evaluation of the prediction model forest environment when the receiver antenna is above, and
that used the original parameter with physical distance. the transmitter antenna is under the forest trees. This is
Weissberger and COST 235 values of parameters were sig- because these models were developed for different frequency
nificantly lower than those of the original parameter in bands than our experimental data. Furthermore, it might be
MAD (I) showing the evaluation of prediction path loss that the majority of the signal propagates over the tree-
models that used the optimized parameter with physical dis- top level; therefore, the lateral wave will contribute more,
tance. The MAD values of modified Weissberger, ITU-R, as reported in [16], and Li et al. [17], [18] propagation in
COST235, FITU-R, and Lateral ITU-R of values were forests. Therefore, the lateral ITU-R model matches well in
19.81, 20.23, 9.30, 9.04, and 6.72, respectively. MAD (II) both cases with our experimental data.

VOLUME 2, 2021 2499

MYAGMARDULAM et al.: PATH LOSS PREDICTION MODEL DEVELOPMENT IN MOUNTAINOUS FOREST ENVIRONMENT

The MAD (I) values for the models (physical distance) using either the distance for the forest area or the physi-
calculated in comparison with the measured values are as cal distance, indicating slight to no difference between the
follows: Weissberger (physical distance) MAD = 19.81 dB, two. This is because, in this scenario, the lateral wave effect
ITU-R (physical distance) MAD = 20.23 dB, COST 235 might be dominant. The lateral ITU-R model was developed
(physical distance) MAD = 9.30 dB, FITU-R (physical for leafy dense forests and considers the lateral wave effect.
distance) MAD = 9.04 dB, and Lateral ITU-R (physical Based on the empirical results, our proposed method
distance) MAD = 6.72 dB. is well suited for improving the applicability of all path
In our proposed prediction of radio wave propagation loss models in a mountainous forested environment. Our
in a mixed forest model, considering the transmission dis- proposed method could be used to develop a model for
tance for free space and forest areas, the accuracy of the any reference distance, which may improve the practical
existing prediction model was successfully improved, mak- application of the model. The prediction accuracy of the
ing it applicable to other empirical models. As shown in proposed model can be improved based on the transmission
Table 5, large differences exist between the path loss pre- path calculation method and DSM data quality.
dicted by the empirical models combined with the FSPL
and the empirical results. The comparison between the
ACKNOWLEDGMENT
results of (FSPL + COST 235), (FSPL + FITU-R), and
(FSPL + Lateral ITU-R) and the empirical measurements This study is based on the results obtained from the Forest-
produced small MADs (7.12, 7.04, and 6.32, respectively). ICT project promoted by Qyushu Geospatial Information
(FSPL + lateral ITU-R) has better results than the other Practice Conference and Committee. We would like to
models in a mixed forest. This is because the lateral ITU-R acknowledge Prof. Yukio Teraoka and Prof. Tsuyoshi Kajisa,
model was developed considering the lateral wave effect. both Kagoshima University, for their leadership on the
Tamir noted that the lateral wave was dominant when the experiment and the arrangement of measurement field in
forest depth increased [18]. Therefore, the models do not the Kagoshima University Research Forest. We would
predict well in the forest environment because they do not like to acknowledge the part of the project “Research
consider the lateral wave effect [19], [20]. and development of highly reliable wireless transmission
However, the predicted path loss obtained from the models technology that contributes to the utilization of next-
showed the worst case for MAD with the empirical ITU-R generation UAVs,” the SCOPE project, the Aerospace, and
model of 20.23 dB, which decreased to the MAD of an Environmental Engineering Education Sky-infra in Mongolia
estimated path loss of 9.94 dB. Therefore, the proposed path project, and the Mongolian Japanese Engineering Education
loss models provide better results than the other models. Development Project (MJEED).
When combining the parameter-optimized empirical
model with FSPL and excluding the obstacle area from REFERENCES
the calculation applied to all the empirical models in this [1] A. I. Trivedi, “Propagation of radio waves over natural obstacles,”
study. Empirical models that exclude the forest from calcu- M.Phil. dissertation, Dept. Commun. Elect., Imperial College London,
lations could predict the path loss in such an environment. London, U.K., 1977.
[2] T. S. Rappaport, Wireless Communications: Principles and Practice.
The results of the combined empirical model showed better Philadelphia, PA, USA: Prentice Hall, 1995.
performance than the parameter-optimized models. [3] Y. S. Meng, Y. H. Lee, and B. C. Ng, “Study of propagation loss
prediction in forest environment,” Progr. Electromagn. Res. B, vol. 17,
pp. 117–133, 2009.
V. CONCLUSION [4] M. A. Weissberger, “An initial critical summary of models for
In this study, we presented a method for the development of predicting the attenuation of radio waves by trees,” Electromagn.
empirical path loss models based on the combination of an Compat. Anal. Center, Annapolis, MD, USA, Rep. ADA118343, 1982.
[5] COST 235, COST 235: Radiowave Propagation Effects on Next-
FSPL and an empirical path loss model, using the distance Generation Fixed-Services Terrestrial Telecommunications Systems,
for free area obtained from a 3D site-specific model (DSM) Luxembourg, Rep., 1996.
in Kagoshima, Japan. We confirmed that the transmission [6] “Influences of terrain irregularities and vegetation on troposphere
propagation,” CCIR Report, Geneva, Switzerland, Rep. 235-236, 1986.
distance calculation of the free space area and forest area [7] M. O. Al-Nuaimi and R. B. L. Stephens, “Measurements and
with a combination of the empirical models showed a bet- prediction model optimisation for signal attenuation in vegetation
ter performance than the model with physical distance. The media at centimetre wave frequencies,” IEE Proc. Microw. Antennas
Propag., vol. 145, no. 3, pp. 201–206, Jun. 1998.
predicted model results were validated using the actual radio
[8] Y. S. Meng, Y. H. Lee, and B. C. Ng, “Empirical near ground path loss
wave propagation in the 920 MHz band measurement data. modeling in a forest at VHF and UHF bands,” IEEE Trans. Antennas
The overall path loss prediction accuracy was improved for Propag., vol. 57, no. 5, pp. 1461–1468, May 2009.
the empirical models average of 8.05 dB on the experimental [9] G. A. J. van Dooren, H. J. F. G. Govaerts, and M. H. A. J. Herben,
“COST 235: Radiowave propagation effects on next-generation
data. The results of this study showed that the lateral ITU-R fixed-services terrestrial telecommunications systems,” Technische
path loss model is suitable for predicting the propagation loss Universiteit Eindhoven, Eindhoven, The Netherlands, Rep., 1997.
in this scenario with a receiver antenna height of 2.5 m and [10] A. Raheemah, N. Sabri, M. S. Salim, P. Ehkan, and R. B. Ahmad,
“New empirical path loss model for wireless sensor networks in
the transmitter antenna moving inside the forest. The lateral mango greenhouses,” Comput. Electron. Agric., vol. 127, pp. 553–560,
ITU-R model was shown to have similar MAD values when Sep. 2016.

2500 VOLUME 2, 2021

[11] F. Granda, L. Azpilicueta, C. Vargas-Rosales, P. Lopez-Iturri, KAZUYOSHI TAKAHASHI received the B.E., M.E., and Ph.D. degrees in
E. Aguirre, and F. Falcone, “Integration of wireless sensor networks engineering from the Nagaoka University of Technology, Nagaoka, Japan,
in intelligent transportation systems within smart city context,” in in 1992, 1994, and 2005, respectively, where he is currently an Associate
Proc. IEEE Int. Symp. Antennas Propag. USNC/URSI Nat. Radio Sci. Professor.
Meeting, 2018, pp. 375–376.
[12] D. Cama-Pinto, M. Damas, J. A. Holgado-Terriza, F. Gómez-Mula,
and A. Cama-Pinto, “Path loss determination using linear and cubic
regression inside a classic tomato greenhouse,” Int. J. Environ. Res.
Public Health, vol. 16, no. 10, p. 1744, 2019. RYU MIURA received the B.E., M.E., and Ph.D. degrees in electri-
[13] T. Tamir, “On radio-wave propagation in forest environments,” IEEE cal engineering from Yokohama National University, Yokohama, Japan,
Trans. Antennas Propag., vol. AP-15, no. 6, pp. 806–817, Nov. 1967. in 1982, 1984, and 2000, respectively. He joined the Radio Research
[14] N. Faruk, Y. A. Adediran, and A. A. Ayeni, “Error bounds of empirical Laboratory, Tokyo, in 1984, which was reorganized to the National Institute
path loss models at VHF/UHF bands in Kwara State, Nigeria,” in Proc. of Information and Communications Technology (NICT) in 2004. He served
Eurocon, 2013, pp. 602–607. as an Executive Researcher with Wireless Networks Research Center, NICT
[15] C. Phillips, D. Sicker, and D. Grunwald, “A survey of wireless until March 2020. He is currently a Senior Researcher with Wireless System
path loss prediction and coverage mapping methods,” IEEE Commun. Laboratory, Wireless Networks Research Center. He has taken the leadership
Surveys Tuts., vol. 15, no. 1, pp. 255–270, 1st Quart., 2013. of several national joint projects for many years with academia and indus-
[16] T. Tamir, “Radio wave propagation along mixed paths in forest tries in the Research and Development of radio communication technologies
environments,” IEEE Trans. Antennas Propag., vol. AP-25, no. 4, for unmanned aircrafts.
pp. 471–477, Jul. 1977.
[17] L.-W. Li, J.-H. Koh, T.-S. Yeo, M.-S. Leong, and P.-S. Kooi,
“Analysis of radiowave propagation in a four-layered anisotropic for-
est environment,” IEEE Trans. Geosci. Remote Sens., vol. 37, no. 4,
pp. 1967–1979, Jul. 1999. FUMIE ONO (Member, IEEE) received the B.E., M.E., and Ph.D. degrees
[18] L.-W. Li, C.-K. Lee, T.-S. Yeo, and M.-S. Leong, “Wave mode and in electrical engineering from Ibaraki University in 1999, 2001, and 2004,
path characteristics in an inhomogeneous anisotropic forest environ- respectively. She was affiliated with the Faculty of Engineering, Tokyo
ment,” IEEE Trans. Antennas Propag., vol. 52, no. 9, pp. 2445–2455, University of Science as a Research Associate from 2004 to 2006. From
Sep. 2004. 2006 to 2011, she was an Assistant Professor with the Division of Electrical
[19] Y. S. Meng, Y. H. Lee, and B. C. Ng, “Empirical near ground path and Computer Engineering, Yokohama National University. She is currently
loss modeling in a forest at VHF and UHF bands,” in IEEE Trans. working with the Global Strategy Bureau, Ministry of Internal Affairs and
Antennas Propag., vol. 57, no. 5, pp. 1461–1468, May 2009. Communication, Tokyo, Japan. She received the Young Engineer Award
[20] J. Hejselbæk, J. Ø. Nielsen, W. Fan, and G. F. Pedersen, “Empirical from IEICE in 2006 and the YRP Award from the YRP Award Committee
study of near ground propagation in forest terrain for Internet-of- in 2007.
Things type device-to-device communication,” in IEEE Access, vol. 6,
pp. 54052–54063, 2018.
[21] “ALOS Global Digital Surface Model ‘ALOS World 3D - 30m
(AW3D30),”’ JAXA, Tokyo, Japan. Accessed: May 16, 2021. [Online].
Available: https://www.eorc.jaxa.jp/ALOS/en/aw3d30/index.htm TOSHINORI KAGAWA received the B.E. and M.E. degrees from Chubu
[22] “The Kagoshima University Research Forests.” University, Aichi, Japan, in 2008 and 2010, respectively, and the Ph.D.
[Online]. Available: https://ace1.agri.kagoshima- degree from the University of Electro Communications, Tokyo, Japan,
.ac.jp/ takakuma/menu/gaiyou/gaiyou.html (Accessed: Jun. 11, in 2013. In 2013, he joined the National Institute of Information and
2021). Communications Technology (NICT). He is currently working with the
[23] T. Kajisa, R. Miura, T. Yamazaki, and A. Shibuya, Originating Central Research Institute of the Electronic Power Industry. He is a member
From Kyushu! Advanced ICT Trial in Barren Mountain Forest Area of the Institute of Electronics, Information, and Communication Engineers.
(in Japanese), Association Bulletin FORN Standard 338, pp. 20–23,
Jan. 2021. [Online]. Available: https://reea.or.jp/report/2722/
[24] L. Shan, H.-B. Li, R. Miura, and F. Kojima, “Local information sharing
system with wireless device-to-device communications,” IEEE Access,
LIN SHAN (Senior Member, IEEE) received the M.E. and Ph.D. degrees
vol. 8, pp. 37453–37464, 2020.
from the Graduate School of Informatics, Kyoto University, in 2008 and
[25] L. Shan, R. Miura, T. Kagawa, F. Ono, H.-B. Li, and F. Kojima,
2012, respectively. He is a Researcher with the National Institute of
“Machine learning based field data analysis and modeling for drone
Information and Communications Technology (NICT), Japan. His research
communications,” IEEE Access, vol. 7, pp. 79127–79135, 2019.
interests include network coding, multiuser-MIMO scheduling, coopera-
[26] L. Shan et al., “Field tests on ’drone mapper’ location information and
tive relaying, and resource allocation in ad hoc and cellular networks. He
remote ID sharing network in the 920MHz band for drones,” Proc.
received the Kyoto University President Prize in 2010. He was a recipi-
21st Int. Symp. Wireless Pers. Multimedia Commun. (WPMC2018),
ent of the IEICE RCS Active Research Award and the IEEE VTS Japan
Bangkok, Thailand, Nov. 2018, pp. 1–4.
Young Researcher’s Encouragement Award in 2011, and the IEEE Kansai
Section Student Paper Award and the IEICE Best Paper Award in 2012 and
2013, respectively.

BILGUUNMAA MYAGMARDULAM received the B.E. and M.E. degrees

in engineering from the Mongolian University of Science and Technology, FUMIHIDE KOJIMA (Member, IEEE) received the B.E., M.E., and D.E.
Nagaoka University of Technology, Nagaoka, Japan, in 2014 and 2017, degrees in electrical communications engineering from Osaka University,
respectively, where she is currently pursing the Ph.D. degree with the Osaka, Japan, in 1996, 1997, and 1999, respectively. Since 1999, he has
Graduate School of Materials Science. Her research interests include remote been affiliated with the Communications Research Laboratory, Ministry
sensing and GIS, wireless communication systems, and the machine learning of Posts and Telecommunications, in 1999, where he has been involved
algorithm. in research on ITS, disaster radios, and wireless grids, including SUN and
smart grids. He is currently the Director of the Wireless Systems Laboratory,
National Institute of Information and Communications Technology. His
NAKAYAMA TADACHIKA received the B.E. and M.E. degree in engineer- current research interests include research, development, standardiza-
ing from Osaka Prefecture University in 1995 and 1997, respectively, and tion, and promotion of 5G/B5G radio communication technology, IoT
the Ph.D. degree from Osaka University, Japan, in 2000. He is currently a radio communication technology, and dependable radio communication
Professor with the Nagaoka University of Technology, Nagaoka, Japan. technology.

VOLUME 2, 2021 2501