Use of LIDAR-based digital terrain model and single tree segmentation data for optimal forest skid trail network

Abstract

The implementation of rational forest management that takes into consideration the requirements of sustainable forest development requires many decisions to be made, ranging from simple rules to extremely complex procedures. This is also true of logging operations, where the applied techniques and technologies should relate not only to economic aspects, but also be maximally adapted to the specific forest environment. One of the most important determinants of environmentally safe and effective logging work is forest accessibility through an appropriately planned road and skid trail network. This paper presents the possibilities of using the Airborne Laser Scanner (ALS) and Geographic Information System (GIS) to determine the optimal or near-optimal locations of forest skid trails. Choosing skid trails depends on the adopted logging method, existing road network, forest stand development phase, and terrain conditions. The process of optimization takes into account existing stand gaps to reduce the number of trees that must be removed for the network. Segments representing single trees and the Digital Terrain Model (DTM) served as inputs for the GIS analysis. The research was carried out in Scots pine stands. The obtained results show that the total length of skid trails, with 40 m distance between them, after optimization for traditional harvesting method decreased by 2%. For fully mechanized harvesting method, the skid trail network was decidedly denser (20 m) and an original length was reduced only by 0.06%. The results obtained confirmed the practical usefulness of the adopted procedures.

Introduction

The efficiency of logging operations in forest stands are largely dependent on the availability of an adequate access through a transportation network, in terms of roads adjacent to wood storage areas and operational trails allowing machines to easily move inside the stand.

Secondary forest roads, known as strip roads or skid trails, are constructed infrastructure occasionally used to execute the tasks of a management plan. They are primarily intended for skidding or forwarding operations ([22]). The density of skid trails and their arrangement depend on many factors ([19]): timber harvesting method and system, configuration, terrain rockiness, tree density and dimensions, and the position of the main forest truck road, to which the timber is extracted.

Optimizing the system of roads and trails is a complex process. Optimal trail spacing can be influenced by many factors, such as logging method, price of products, taxation policies, landing costs, overhead costs, equipment opportunity costs, road width and size of landing, skidding pattern, profit of logging contractor, slope, topography and soil disturbance ([4]).

The first forest road-spacing scheme, along with the theoretical mean skidding distance, was presented by Matthews ([13]). Models of optimal road networks for forest areas have been published many-fold in the literature. Some papers proposed an evaluation of alternative trail routes based on economic considerations only ([7], [20], [6]), while others were based on specific information obtained by interpreting aerial photographs, field surveys and integrated DTM data ([26], [2], [3], [8]).

The problem of accessing a forest through a network of secondary forest traffic infrastructure is particularly important for thinning, and for those mature stands where selective harvesting is performed. Even though machinery can move freely over the entire area in a uniform clear-cut, it is always recommended fixed trails to be used.

Almost all the negative effects of the use of machines in the forest reflect on trails ([16]), mainly soil compaction and damages to residual trees after forest operations. Skidders, tractors, and other heavy machinery used in forest practices may compact soils remarkably, depending on the type of equipment used, soil characteristics, etc. Compaction can take many years to recover, depending on the level of damage and many other environmental conditions. Limiting the area where compaction occurs by reducing machine drives to appropriately designed skid trails is one of best way to alleviate such problem ([24]). Skid trail planning before felling operations can also reduce skidding damages. Machines moving only on trails cause damages which are located mostly near their working route ([18], [30]).

In recent years, new opportunities have been developed to optimize roads and skid trails. This is mainly due to the use of information from the Airborne Laser Scanner (ALS) and Geographic Information System (GIS). Laser scanning provides precise data layers representing terrain surfaces in the form of Digital Terrain Models (DTM). DTM layers have attracted great interest to support forestry operations since they provide thorough and detailed information about terrain topography, which in turn is used to choose the best skidding system in complex forest areas ([15], [12], [32]). Krč & Košir ([10]) have also used Digital Elevation Models (DEM) to develop a model for terrain classification based on the best predicted skidding direction on a steep terrain. High resolution DTMs are also used to define various soil moisture indexes, providing information about the path to be taken by machines on a given terrain ([17]).

Apart from DTM, ALS provides very precise and complex information about vegetation. Many different stand and tree characteristics can be determined with the use of ALS point cloud or Canopy Height Model (CHM), such as height, structure, biomass volume, growth and density ([21], [23], [28], [29]). Once acquired ALS data can be utilized not just in forest inventory or monitoring, but also in many other management activities, such as optimization of the forest skid trail network.

The purpose of this paper is to present an analytical procedure that allows the optimal determination of skid trails by using data from the Airborne Laser Scanner and digital forest maps. Single tree segmentation results in such decision support designing system are involved for the first time. Specific aims of the paper can be defined as follows:

* creation of a procedure for forest skid trail network optimization, fully based on GIS and ALS data;

* evaluation of pre- and post-procedure results on network length and the biomass volume planned to be cut;

* analysis of the results with respect to the implementation of the method.

Material and Methods

Study site

The study site was a 1000 ha forest area owned by the Warsaw University of Life Sciences - SGGW, located at the experimental station in Gluchów, central Poland (51° 45′ 19″ N, 20° 06′ 30″ E). Different forest types exist in the area, ranging from single-layer stands of Scots pine (Pinus silvestris L.) and Common oak (Quercus robur L.) to multilayer, mixed stands of Birch (Betula pendula Roth), Alder (Alnus glutinosa L.), European beech (Fagus sylvatica L.) and Hornbeam (Carpinus betulus L.). Stand age varied between 30 and 120 years. The study area has a generally flat relief, with a mean elevation of about 185 m a.s.l. In order to verify the theoretical model presented hereinafter, Scots pine stands with different tree stocking and dimensions were studied, and skid trail network was optimized for each stand selected.

The model presented below was tested over 170 stands, where Scots pine was the dominant species (cover > 70%). The main characteristics of the studied stands are reported in Tab. 1.

Tab. 1 - General characteristics of selected Scots pine stands (no. stands = 170).

Parameter	Minimum	Maximum	Sum	Mean	Standard deviation
Area (ha)	0.38	24.91	481.33	2.38	3.11
Height (m)	5	28	-	21	4
Age (years)	21	80	-	52	17
Stocking	0.3	1	-	0.73	0.72
Dbh (cm)	5	38	-	23	6
Volume (m3/ha)	0	526	-	231	93

Airborne Laser Scanning - data and processing

The Falcon II airborne laser scanner system from Topographische Systemdaten GmbH (TopoSys, Biberach, Germany) was used for LIDAR small footprint data acquisition. The TopoSys System is based on two separate glass fiber arrays of 127 fibers each. Its specific design produces a push-broom measurement pattern of the ground. Details about the flight mission were described by Sterenczak & Kozak ([27]).

Based on data obtained, a 1-m digital terrain model was generated by the TSFFCVJC^® software (FELIS, Germany). For more details about software description and the algorithms implemented therein, see Weinacker et al. ([31]).

Optimization procedure

The model represented in Fig. 1 simulates the most common harvesting methods used nowadays in Poland: (i) the traditional method, based on timber felling by chainsaws and extraction by skidders (TR); (ii) the fully mechanized method, using harvesters and forwarders (FM). The selected algorithm takes into consideration three types of inputs: Canopy Height Model (CHM), Digital Terrain Model (DTM) and layers from Forest Digital Maps (road network and forest compartments). The main steps for the implementation of the optimization algorithm are described in following paragraphs.

Fig. 1 - Flow chart of the optimization algorithm.

(1) A single-tree segmentation was carried out based on the CHM from Airborne Laser Scanning (ALS) data ([14]). Segments representing single trees (polygon layer) were used in the next step as a point layer (Fig. 2). Each tree is represented by the centroid of a polygon.

Fig. 2 - Single tree segmentation results.

(2) The Digital Terrain Model was first processed. The slope layer was generated using the SPATIAL ANALYSIS^® extension of the ARC GIS^® 10.1 package (ESRI, Redwoods, CA, USA). To take into account the limits of traditional skidding by using skidders or agricultural tractors (maximum slope 40%. According to Antonczyk ([1]), the maximal slope limits for skidders are 33% uphill and 50% downhill, while is 18% (uphill and downhill) for agricultural tractors not equipped with Front Wheel Drive (FWD). In Poland, agricultural tractors without FWD are mostly used for thinning operations. For that reason, in the case of the traditional timber harvesting method, we assumed the maximum slope limit as 18%. In other cases or countries, when tractors are equipped with FWD, the maximum limit is higher, especially in downhill direction. Areas smaller than 10 m² with steep slopes were omitted from the analysis, being either local errors in the DTM or not important from a technological point of view. These data were used later for skid trail optimization.

(3) Forest Digital Maps were used to select the stands to be harvested and the roads to be used for timber transportation. Selected stands were 21 to 80 years old (Fig. 3). Based on the selected roads and compartments (forest units), a skid trail network was then created. If a forest road was already present, the longest side of the compartment was always used as the basis for planning skid trails. In cases of irregular-shaped forest units, trails were plotted in such a way to optimize the access (for example, along the longest axis of the unit), thus minimizing the sum of their lengths. Theoretical skid trails were designed with orthogonal direction to the base road and spacing of 40 m (traditional logging) or 20 m (fully mechanized logging).

Fig. 3 - Selected stands and slope classes.

Results

Based on the procedure described above, skid trails were traced and the trail network expanded to its maximum possible extent by taking advantage of the existing gaps in the study site.

Total length of skid trails for the traditional harvesting method was 109.17 km before optimization (TRBO_L), whereas it decreased by 2.14 km (2%) to 107.03 km after optimization (TRAO_L). For fully mechanized harvesting, the skid trail network was remarkably denser and longer, with an original length of 227.27 km (FMBO_L) that was reduced by 142 m (0.06%) to 227.13 km after optimization (FMAO_L).

A comparison of the average length of skid trails per ha before and after the optimization step is reported in Fig. 4 for the different harvesting methods considered. The large difference between trail lengths obtained for traditional and mechanized logging methods (over 60%) was mainly due to the different distances between adjacent trails, which was initially set to 20 and 40 m, respectively. Optimization by taking into account stand gaps did not significantly change the overall trail length. The observed difference in the average length of skid trails per ha using traditional methods was 2.01%, while with fully mechanized methods was only 0.12%. As shown in Tab. 2, such differences were not statistically significant.

Fig. 4 - Average skid trail length (m ha-1) in the study area before and after optimization.(TRBO_L): Traditional logging method before optimization; (TRAO_L): traditional logging method after optimization; (FMBO_L): fully-mechanized logging method before optimization; (FMAO_L): fully-mechanized logging method after optimization.

Tab. 2 - Results of the Student’s t-test for the comparison of the trail length per ha and the number of trees to be felled before and after the optimization process. Labels are the same reported in Fig. 4 and Fig. 5.

Variable	Comparison	Mean 1	Mean 2	t	df	p	N 1 and N 2	SD 1	SD 2	F ratio	p
Av. length of skid trails ha-1	TRBO_L vs. TRAO_L	399.11	391.09	0.1239	338	0.9013	170	60.76	53.58	1.0275	0.8601
	FMBO_L vs. FMAO_L	641.02	640.28	0.0092	338	0.9926	170	59.92	71.21	1.0064	0.9669
N. trees to be felled	TRBO_N vs. TRAO_N	59.59	58.06	0.1687	338	0.8660	170	24.11	21.29	1.0371	0.8127
	FMBO_N vs. FMAO_N	113.22	112.35	0.2259	338	0.8213	170	31.88	35.23	1.0374	0.8115

Also, the single tree segmentation carried out allowed to determine the total number of trees to be felled for the construction of the skid trails in the study site. Before optimization, such figure was 16 785 using the traditional harvesting method (TRBO_N), whereas after optimization (TRAO_N) it decreased by 465 (2.8%) to 16 320. Using the fully mechanized harvesting method, the number of trees was originally 55 995 (FMBO_N), while after optimization (FMAO_L) this number was reduced by 308 (0.55%) to 55 687.

Fig. 5 shows the number of trees (per ha) standing on the surface of the designed skid trails in the study area. The trends obtained were similar to those observed for the trail length. However, differences between traditional and mechanized logging methods were higher, increasing to about 90%. Also in this case, the optimization process adopted did not significantly reduce the number of trees to be felled. Differences in the numbers of trees before and after optimization were not statistically significant (Tab. 2), amounting to 2.57% and 0.77% for the traditional and the mechanized methods, respectively.

Fig. 5 - Number of trees to be removed before and after the optimization process. (TRBO_N): Traditional logging method before optimization; (TRAO_N): traditional logging method after optimization; (FMBO_N): fully-mechanized logging method before optimization; (FMAO_N): fully-mechanized logging method after optimization.

Discussion

Accessing forest stands using an appropriately planned skid trail network allows not only logging and skidding operations to be carried out, but also other works related to silviculture and forest protection.

It is recommended to establish trails during the first thinning associated with logging operations. Trails should be designated along the shortest routes to extraction roads; however, schematic, rectilinear routes should be avoided, as well as the less accessible portions of the stand, such as wetlands, ponds, areas with steep slopes etc. Existing gaps, trails, etc., should be taken into account. An example of the skid trail network obtained after optimization on part of the study area is displayed in Fig. 6.

Fig. 6 - Skid trail network for the chosen forest area.

Conclusions

The application of laser scanning technology and the use forest digital maps, along with the development of adequate algorithms, may allow the optimization of skid trail networks, as well as the reduction of number of trees to be felled in the network construction. Both the length of skid trails and the number of trees to be felled are mainly affected by the distance required between trails. This in turn depends on the logging and wood extraction technology to be used.

The use of existing stand gaps for trail delineation may contribute to the reduction of the number of trees to be felled in the construction of the skid trail network. However, our results did not show significant differences between the modeled network variants before and after the optimization procedure applied.

Acknowledgements

We are grateful to Krzysztof Bedkowski (Warsaw Universities of Life Sciences, Poland) for the permission of using ALS data obtained within the project of the Polish Ministry of Science and Higher Education: “Using airborne and terrestrial LIDAR scanning for analyzing forest spatial structure and forest function in landscape” (project no. 2 P06L 02229).

References

(1) Antonczyk SZ (1995). Physiographic factors and their impact on a timber extraction. In: “Proecological and productive functions of skid trails in the modern forest management”. Foundation “Development of SGGW”, Warsaw, Poland, pp. 7-13. [in Polish] Gscholar

(2) Chung W, Sessions J (2001). Transportation planning under multiple objectives. In: Proceedings of the “International Mountain Logging and 11th Northwest Skyline Symposium”. Seattle (WA, USA) 10-12 Dec 2001, pp. 194-200. Online | Gscholar

(3) Ghaffarian MR, Sobhani H (2007). Optimization of an existing forest road network using Network 2000. Croatian Journal of Forest Engineering 28 (2): 185-193. Online | Gscholar

(4) Heinimann HR (1997). A computer model to differentiate skidder and cable-yarder based road network concepts on steep slopes. Journal of Forest Research 3 (1): 1-9.

(5) Hempel G (1968). Allometrische studie an Pinus cembra spp. sibirica (Rupr.) Kryl. und Abies sibirica (Ledeb.) [Allometric study of Pinus cembra spp. sibirica (Rupr.) Kryl. and Abies sibirica (Ledeb.)]. Archiv für Forstwesen 17 (11): 1099-1115. [in German] Gscholar

(6) Jourgholami M, Abdi E, Chung W (2013). Decision making in forest road planning considering both skidding and road costs: a case study in the Hyrcanian Forest in Iran. iForest 6: 59-64.

(7) Kobayashi H (1984). Planning system for road route locations in mountainous forests. Journal of the Japanese Forestry Society 66 (8): 313-319. Gscholar

(8) Kooshki M, Hayati E, Rafatnia N, Ahmadi MT (2012). Using GIS to evaluate and design skid trails for forest products. Taiwan Journal of Forest Science 27 (1): 117-24. Online | Gscholar

(9) König W (1970). The influence of optimum road-spacing on opening up the state forest of North-Rhein Westphalia. University Regensburg, Germany, pp. 21-31. [in German] Gscholar

(10) Krč J, Košir B (2008). Predicting wood skidding direction on steep terrain by DEM and forest road network extension. Croatian Journal of Forest Engineering 29 (2): 177-188. Online | Gscholar

(11) Leckie D, Gougeon F, Hill D, Quinn R, Armstrong L, Shreenan R (2003). Combined high-density lidar and multispectral imagery for individual tree crown analysis. Canadian Journal of Remote Sensing 5: 633-649.

(12) Lubello D (2008). A rule-based SDSS for integrated forest harvesting planning. PhD thesis, Università degli Studi di Padova, Italy, pp. 213. Gscholar

(13) Matthews DM (1942). Cost control in the logging industry. Mc Graw-Hill, New York ,USA, pp. 374. Online | Gscholar

(14) Miscicki S, Sterenczak K (2013). A two-phase inventory method for calculating standing volume and tree-density of forest stands in central Poland based on airborne laser-scanning data. Forest Research Papers 74 (2): 127-136.

(15) Mohtashami S, Bergkvist I, Löfgren B, Berg S (2012). A GIS approach to analyzing off-road transportation: a case study in Sweden. Croatian Journal of Forest Engineering 33 (2): 275-284. Online | Gscholar

(16) Moskalik T, Sadowski J (2000). Forest accessibility for the fully mechanized timber harvesting. In: Proceedings of the “Communication infrastructure in multifunctional sustainable forestry”. Rogow (Poland) 14 Dec 2000. Warsaw University of Life Sciences, Warsaw, Poland, pp. 81-88. [in Polish] Gscholar

(17) Murphy PN, Ogilvie J, Meng FR, Arp P (2008). Stream network modeling using lidar and photogrammetric digital elevation models: a comparison and field verification. Hydrological Processes 22 (12): 1747-1754.

(18) Naghdi R, Rafatnia N, Bagheri I, Hemati V (2008). Evaluation of residual damage in felling gaps and extraction routes in single selection method (Siyakhal forest). Iranian Journal of Forest and Poplar Research 16: 87-98. Gscholar

(19) Nevečerel H, Pentek T, Pičman D, Stankič I (2007). Traffic load of forest roads as a criterion for their categorization - GIS analysis. Croatian Journal of Forest Engineering 28 (1): 27-38. Online | Gscholar

(20) Newnham RM (1995). ROADPLAN: a tool for designing forest road networks. Journal of Forest Engineering 6 (2): 17-26.

(21) Nilsson M (1996). Estimating tree heights and stand volume. Remote Sensing of Environment 56: 1-7.

(22) Pentek T, Nevečerel H, Poršinsky T, Pičman D, Lepoglavec K, Potočnik I (2008). Methodology for development of secondary forest traffic infrastructure cadastre. Croatian Journal of Forest Engineering 9 (1): 75-83. Online | Gscholar

(23) Popescu SC, Wynne RH, Nelson RF (2003). Measuring individual tree crown diameter with lidar and assessing its influence on estimating forest volume and biomass. Canadian Journal of Remote Sensing 29: 564-577.

(24) Schnepf CH (2008). Prevent forest soil compaction - designate skid trails. UI Extension Forestry Information Series, Forest Management No. 8, University of Idaho, Moscow, ID, USA, pp. 1. Online | Gscholar

(25) Sedlak O (1983). General principles of planning forest road nets. Food and Agriculture Organization publication, FAO, Rome, Italy, vol. 14, pp. 17-36. Gscholar

(26) Shiba M (1992). Optimization of road layout in opening up of forests. In: Proceedings of the “Computer Supported Planning of Roads and Harvesting Workshop”. Feldafing (Germany) 26-28 Aug 1992, pp. 1-12. Gscholar

(27) Sterenczak K, Kozak J (2011). Evaluation of digital terrain models generated in forest conditions from airborne laser scanning data acquired in two seasons. Scandinavian Journal of Forest Research 26 (4): 374-384.

(28) Sterenczak K, Miscicki S (2012). Crown delineation influence on standing volume calculations in protected area. ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. XXXIX-B8, pp. 441-445. Online | Gscholar

(29) Sterenczak K, Mielcarek M (2014). Assessing one year pine growth at stand level with single tree detection based on ALS data. In: Proceedings of the “6th Precision Forestry Symposium: The anchor of your value chain” (Ackerman P, Gleasure E, Ham H eds). Stellenbosch University (Stellenbosch, South Africa) 3-5 March 2014, pp. 16-18. Online | Gscholar

(30) Tavankar F, Majnounian B, Bonyad AE (2013). Felling and skidding damage to residual trees following selection cutting in Caspian forests of Iran. Journal of Forest Sciences 59 (5): 196-203. Online | Gscholar

(31) Weinacker H, Koch B, Heyder U, Weinacker R (2004). Development of filtering, segmentation and modeling modules for LIDAR and multispectral data as a fundament of an automatic forest inventory system. In: Proceedings of the “ISPRS Working Group on Laser-Scanners for Forest and Landscape Assessment” (Thies M, Koch B, Spiecker H, Weinacker H eds). Institute for Forest Growth, Institute for Remote Sensing and Landscape Information Systems, Freiburg, Germany, pp. 50-55. Online | Gscholar

(32) Vega-Nieva DJ, Murphy PN, Castonguay M, Ogilvie J, Arp PA (2009). A modular terrain model for daily variations in machine-specific forest soil trafficability. Canadian Journal of Soil Science 89 (1): 93-109.

(33) Zarzycki S (1995). Principles of skid trails and landings design. In: “Proecological and productive functions of skid trails in the modern forest management”. Foundation “Development of SGGW”, Warsaw, Poland, pp. 86-97. [in Polish] Gscholar