I Introduction
Different natural landforms such as valleys, forests, and other areas have a significant impact on the propagation of radio waves[1]. The forest area is covered by dense trees, which makes the multipath fading and nonlineofsight channel very complex, and differs from the effects of urban buildings[2]. Some main factors affecting radio wave propagation, such as the distance between the transmitting and receiving antennas, the height of the antenna, and the type of ground objects, are reflected in the path loss formula as variable functions[3]. However, in different geographical environments, topographic relief, vegetation height and density, climate, and other factors have various degrees of influence on propagation[4]. Therefore, when these propagation models are applied in specific environments, the corresponding variable functions should be different, and it is necessary to identify a reasonable channel model. The accuracy of the channel model is very important for network deployment, because the inappropriate channel models will lead to significant reduction of network coverage[5].
Different typical propagation models have different characteristics, and they are applicable to different environment. Typical propagation models include Okumura Hata model[6], cost231 Hata model[7], SPM model[8], etc. These models are suitable for cities, suburbs and villages, but not for forest areas. In[9]
, the authors proposed the Erceg model which is based on extensive experimental data collected by AT&T Wireless Services across the United States in 95 existing macro cells at 1.9GHz. The terrains are classified in three categories. Category A is hilly terrain with moderatetoheavy tree density, category B is hilly terrain with light tree density or flat terrain with moderatetoheavy tree density and category C is mostly flat terrain with light tree density. Soon later, In
[10], Stanford University proposed Stanford University Interim (SUI) channel model is a set of 6 channel models representing three terrain types and a variety of Doppler spreads, delay spread and lineofsight/nonlineofsite conditions that are typical of the continental US. The terrain type A, B, C are same as those defined earlier for Erceg model. However, these models were proposed earlier and are mainly applicable to the North American environment.Other scholars have proposed channel models for specific forest environment, in addition to the typical channel models. In [11], the authors studied the oblique leaf path of roadside woodland, including three vegetation types, and obtained the attenuation loss results of oblique path at Cband yielding 0.9 dB overall improvement and up to 20 dB regional improvement in root mean square errors. In [12], the authors investigated the propagation behavior of 28GHz millimeter wave in coniferous forests and model its basic transmission loss, and proposed novel fully automated sitespecific models. The rootmeansquare deviations between model predictions and simulation results are 11.3 dB for an ITU woodland model and 6.8 dB for a sitespecific model published in this paper. In [13], the authors characterized the wireless channel for a shortrange, temperate, medium density forest environment at the 5GHz band. In [14], the authors presented measurement results and propose empirical models of ultrawideband signal propagation in forest environments after recording more than 22000 measurements at 165 sites in four different forest environments in Virginia and Maryland. However, these works only focus on the single scenario.
In this paper, we carry out propagation measurement campaigns in two different types of forest areas, and record the measured values of signal propagation loss in those areas. Owing to the large attenuation of the signal in the forest area, the propagation effect of the lower frequency band is better, and in order to study the communication status of the emergency frequency band, so we adopted the 605 MHz frequency band for measurement. Then we use three classical largescale path loss models and forest excess attenuation models to characterize the measured data, providing a comprehensive model comparison. Through the analysis of the results, we develop a new forestspecific path loss model, which has better performance than representative existing models.
Ii Description of measurements
In order to study the impact of different forest areas on signal propagation, we selected two different types of areas which are Jiaozi snow mountain and Puduriver dryhot valley where channel propagation measurement campaigns were conducted in March 2022.
Jiaozi snow mountain is located at the junction of Luquan County and Dongchuan District in Yunnan Province, China, with a maximum altitude of 4344.1 meters (m), a minimum altitude of 2300 m, and a relative height difference of more than 2000 m. Jiaozi snow mountain belongs to seasonal snow mountain, which is the lowest snow mountain in the northern hemisphere. There are 15000 mu of Abies lanceolata primary secondary forest and Rhododendron forest, which are typical dense forest scenes. Fig. 1 shows the geographical scenario of Jiaozi snow mountain.
Puduriver dryhot valley is located in the Pudu river area in Yunnan Province as well. There are seven vegetation types, 11 vegetation subtypes, 17 formation groups, and 28 formations in the reserve, including dryhot valley hard leaf evergreen oak forest, semihumid evergreen broadleaved forest, mountaintop bryophyte dwarf forest, cold temperate shrub, and cold temperate meadow. Fig. 2 shows the geographical scenario of Puduriver dryhot valley.
Base station information  Parameters or information 

Transmitting antenna height  1.5 m 
Coordinate of Jiaozi snow mountain  (102.848226, 26.0845327) 
Coordinate of Puduriver dryhot valley  (102.7342136, 26.02112129) 
Base station transmission power  43 dBm 
Base station antenna type  Omnidirectional antennas 
Transmit antenna gain  5 dBi 
Receiving antenna gain  0 dBi 
Transmission band  605 MHz 
Cell reference signal power  15.2 dBm 
In the propagation measurement campaigns, we used a backpack base station as the signal transmitter (Tx) in a fixed position, equipped with an omnidirectional antenna, while a handheld device was used as the receiver (Rx), with an omnidirectional receiving antenna inside. Tx and Rx devices are shown in Fig. 3. The transmit power of the base station is 43 dBm, the transmit antenna gain is 2.5 dBi, the receive antenna gain is 0 dBi, and the carrier frequency is 605 MHz. The relevant information of the measurements is provided in Table I. We recorded the longitude and latitude of every transmission and reception position, adopted the continuous wave as the signal source to transmit the signal, and carried out the onboard test on the preset routes.
We collected and recorded the pilot signal received power at the onboard test cell phone. Since the test was aimed to obtain the path loss data in the actual network, the test data truly reflects the propagation of broadband signals in the local wireless environment. As there is no need to set up a base station by itself, this test scheme is simple and convenient. Fig. 4 illustrates the measurement trajectory, and different colors of trajectory indicate different reference signal received power where green represents the minimum and red represents the maximum. The measurement data in Jiaozi snow mountain and Puduriver dryhot valley will be shown in Fig. 5.
Iii Experimental results
After obtaining the data set, we consider three path loss models as the baseline generic model which are the alphabeta–gamma (ABG) model[15][16], the closein freespace reference distance (CI) model[15][16][17], and the freespace path loss (FSPL) model. The expressions of CI, ABG, and FSPL are as follows:
(1) 
(2) 
(3) 
where denotes the path loss exponent (PLE), is the closein freespace reference distance and is set to 1 m[15], is the 3D TR separation distance in meters, and are coefficients showing the dependence of path loss on distance and frequency, respectively, is an optimized offset value for path loss in decibels, is the carrier frequency in GHz, is the speed of light.
Note that the CI model has a very similar form compared with the ABG model, but has fewer model parameters and more solid physical basis[15][16]. Since additional attenuation is caused by the occlusion of vegetation in the forest area, we use the ITU horizontal forest model[19] as the excess path loss model.The expressions of ITU horizontal forest model is as follows:
(4) 
where denotes the specific attenuation for very short vegetative paths (dB/m) and denotes the maximum attenuation for one terminal within a specific type and depth of vegetation (dB). Next, we combine FSPL model and ITU horizontal forest model to fit the measured data[20]. The expression of FSPLH is given by:
(5) 
For shortdistance specific forest scenes, we build a simple but powerful scenespecific model by more carefully characterizing forestspecific propagation loss, which can simplify the expression and parameters compared with directly combining two types of models presented above. We name the model Beijing University of Posts and Telecommunications horizontal forest model (BHF). The expression of BHF is as follows:
(6) 
where is a coefficient showing the dependence of path loss on the conventional logscaled distance, is an optimized offset value for path loss in decibels, is a coefficient characterizing the path loss caused by vegetation attenuation.
Site  Model 





Jiaozi snow  CI  3.8      
mountain  ABG  2.9  31.8  2.0  
FSPLH  40.0  1.2    
BHF  1.6  1305.2  1407.0  
Puduriver  CI  4.0      
dryhot valley  ABG  1.9  57.7  2.0  
FSPLH  43.8  4.6    
BHF  0.8  48.3  64.2 
The BHF model contains the negative exponential characteristic attenuation caused by vegetation, where the changes are made on the basis of ITU horizontal model, and the additional attenuation is close to the hyperbolic tangent function. Compared with the ITU horizontal model in (4), the function about the additional attenuation of vegetation changes more gently in the BHF model. We fit the BHF model and three baseline models by using the leastsquare method to find the optimal model parameters. The optimized model parameters are shown in Table II. It can be seen from the Table II that the coefficients of the CI and FSPLH models are relatively stable, with little change across environments, due to the freespace reference term acting as an anchor point. In contrast, the coefficients of the ABG and BHF models are strongly influenced by the environment.
Fig. 6 shows the fitting results of the CI, ABG, FSPLH, and BHF models. It can be observed from Fig. 6 that the CI model and ABG model are two straight lines. Because the relationship between the parameter term and the 3D TR separation distance in the expression is multiplication, when other coefficients are given, the function between path loss and distance in the logarithmic scale is a linear function. Since the adaptation ability of straight lines is relatively limited, the fitting errors of CI and ABG models are large. The CI model has fewer parameter variables than the ABG model, rendering larger errors. The relationship between and parameters of the other two models are more complex, so the fitting effect is better. It can be seen that the BHF model has stronger capability of adapting with the trend of measured data in Fig. 6(b).
  Traditional  Proposed  
Model  CI  ABG  FSPLH  BHF 
Jiaozi snow mountain  4.6  4.1  3.6  3.0 
Puduriver dryhot valley  13.1  9.7  9.3  8.3 
Number of model parameters  1  2  2  3 
We use rootmeansquare error (RMSE) and the number of parameters to quantify the fitting effect of the models, which are given in Table III. Note that for single frequencies, in the ABG model is set to 2, thus there are actually two model parameters in the ABG model.
It can be seen from Table III that the fitting error for Jiaozi snow mountain is smaller than that for Puduriver dryhot valley in general, because the different types and densities of vegetation and the geographical environment have an impact on the transmission of signals. The terrain of Puduriver dryhot valley is steeper than that of Jiaozi snow mountain. Accoding to Table III, the RMSEs of the BHF model are 3.0 dB and 8.3 dB for Jiaozi snow mountain and Puduriver dryhot valley, respectively. This shows that the BHF model proposed in this paper has the best fitting effect overall and is more suitable for the forest environment.
Iv Conclusion
In this paper, we have provided results from realworld measurement campaigns to assess channel characteristics for the forest environment. The signal measurement data near Jiaozi snow mountain and Puduriver dryhot valley are used to compare the attenuation of vegetation with comprehensive largescale path loss models. Inspired by these results, we have developed a new sitespecific path loss model. Compared with typical traditional models, the model proposed in this paper yields significantly smaller fitting errors with acceptable computational complexity, and is thus more suitable for the forest environment.
References
 [1] I. T. U. ITU, “Attenuation in vegetation,” ITUR Recomm. 8336, 2017.
 [2] J. Hejselbæk, J. Ødum Nielsen, W. Fan, and G. F. Pedersen, “Empirical study of near ground propagation in forest terrain for internetofthings type devicetodevice communication,” IEEE Access, vol. 6, pp. 54 052–54 063, 2018.

[3]
A. Kavak, W. Yang, K. R. Dandekar, and G. Xu, “Effects of base station antenna height and mobile terminal movement on the vector propagation channels,” in
1999 IEEE 49th Vehicular Technology Conference (Cat. No. 99CH36363), vol. 1. IEEE, 1999, pp. 777–781.  [4] S. Rougerie, J. Israel, and T. Kan, “Validation of itur p. 8339 tree attenuation model for land mobile satellite propagation channel at ku/ka band,” in 2021 15th European Conference on Antennas and Propagation (EuCAP). IEEE, 2021, pp. 1–5.
 [5] M. Rebato, J. Park, P. Popovski, E. De Carvalho, and M. Zorzi, “Stochastic geometric coverage analysis in mmwave cellular networks with a realistic channel model,” in GLOBECOM 20172017 IEEE Global Communications Conference. IEEE, 2017, pp. 1–6.
 [6] Y. Okumura, “Field strength and its variability in vhf and uhf landmobile radio service,” Rev. Electr. Commun. Lab., vol. 16, pp. 825–873, 1968.
 [7] E. Damosso, and L.M. Correia, Eds., Digital Mobile Radio Towards Future Generation Systems Communications. COST 231 Final Report, CEC, Brussels, Belgium, 1999
 [8] C. Rodriguez, O. M. Arias et al., “Propagation characteristics of managua city based on standard propagation model (spm) at 850 mhz for 3gwcdma systems,” in 2014 IEEE Central America and Panama Convention (CONCAPAN XXXIV). IEEE, 2014, pp. 1–6.
 [9] V. Erceg, L. J. Greenstein, S. Y. Tjandra, S. R. Parkoff, A. Gupta, B. Kulic, A. A. Julius, and R. Bianchi, “An empirically based path loss model for wireless channels in suburban environments,” IEEE Journal on selected areas in communications, vol. 17, no. 7, pp. 1205–1211, 1999.
 [10] K. Hari, D. Baum, T. Kaitz, and B. D. Trinkwon, “Project ieee 802.16 broadband wireless access working group¡ http://ieee802. org/16¿ title channel models for fixed wireless applications,” 2001.
 [11] F. Teschl, F. P. Fontan, M. Schonhuber, R. P. Cerdeira, and R. Teschl, “Attenuation of spruce, pine, and deciduous woodland at cband,” IEEE Antennas and Wireless Propagation Letters, vol. 11, pp. 109–112, 2012.
 [12] Y. Zhang, C. R. Anderson, N. Michelusi, D. J. Love, K. R. Baker, and J. V. Krogmeier, “Propagation modeling through foliage in a coniferous forest at 28 ghz,” IEEE Wireless Communications Letters, vol. 8, no. 3, pp. 901–904, 2019.
 [13] D. W. Matolak, F.C. Yang, and H. B. Riley, “Short range forest channel modeling in the 5 ghz band,” in 2012 6th European Conference on Antennas and Propagation (EUCAP). IEEE, 2012, pp. 3337–3341.
 [14] C. R. Anderson, H. I. Volos, and R. M. Buehrer, “Characterization of lowantenna ultrawideband propagation in a forest environment,” IEEE Transactions on Vehicular Technology, vol. 62, no. 7, pp. 2878–2895, 2013.
 [15] S. Sun, T. S. Rappaport, T. A. Thomas, A. Ghosh, H. C. Nguyen, I. Z. Kovacs, I. Rodriguez, O. Koymen, and A. Partyka, “Investigation of prediction accuracy, sensitivity, and parameter stability of largescale propagation path loss models for 5g wireless communications,” IEEE transactions on vehicular technology, vol. 65, no. 5, pp. 2843–2860, 2016.
 [16] 3GPP TR 38.901, “TR 138 901  V16.1.0  5G; Study on channel model for frequencies from 0.5 to 100 GHz (3GPP TR 38.901 version 16.1.0 Release 16),” 3Gpp, vol. 0, 2020.
 [17] T. S. Rappaport, “Wireless communications–principles and practice, (the book end),” Microwave Journal, vol. 45, no. 12, pp. 128–129, 2002.
 [18] Study on 3D Channel Model for LTE (Rel. 12), 3GPP TR 36.873, V12.1.0,Mar. 2015.
 [19] I. T. U. ITU, “Attenuation in vegetation,” ITUR Recomm. 8334, 2016.
 [20] Y. Zhang, C. R. Anderson, N. Michelusi, D. J. Love, K. R. Baker, and J. V. Krogmeier, “Propagation modeling through foliage in a coniferous forest at 28 ghz,” IEEE Wireless Communications Letters, vol. 8, no. 3, pp. 901–904, 2019.