TIMBER HARVEST AREA PLANNING IN MALAYSIA USING A NETWORK ALGORITHM

Abstract TIMBER HARVEST AREA PLANNING IN MALAYSIA USING A NETWORK ALGORITHM Norizah Kamarudin 1, Woodam Chung 2 In Peninsular Malaysia, the most common logging systems are cable skidder and log fisher. A log fisher is an excavator based hydraulic timber harvest system that can winch logs up to 300 meters. While cable skidders can be used only on gentle slopes, log fishers can be applied regardless of slope conditions. Since both harvesting methods are applicable for gentle ground, a decision on harvesting system has to be made depending on existing and proposed forest road network, landing locations, and harvesting costs. Usually, harvesting cost of log fisher is higher than cable skidder, but does not require skid trail construction, which is mandated for cable skidder operations in Malaysia. We developed conceptual network models to represent this complex forest operations planning problem where decisions need to be made simultaneously on harvest system, landing location, and forest road location. We applied the models to a 102 ha timber harvest area in Malaysia and solved the problem using a heuristic network algorithm to obtain the least cost harvesting option while considering both fixed (i.e., road and trail construction costs) and variable costs (i.e., harvesting costs). In this paper, the problem formulation approach and case study solutions are presented. Keywords: Timber harvesting, road planning, Peninsular Malaysia, extraction system, cost efficiency INTRODUCTION Determining the optimal logging systems for a timber harvest area is a difficult task because many considerations need to be taken into account including timber volume and distribution, terrain and environmental conditions, costs and productivity, and the existing road infrastructures. The most common logging systems in Peninsular Malaysia include 1) groundbased system using a cable skidder and 2) cable winching system called log fisher. Log fisher is an excavator based hydraulic cable winching system that can winch logs up to 300 meters. While cable skidders are normally used only on gentle ground, log fishers can be applied regardless of slope conditions. Saharudin et al. (2004) and Norizah et al. (2012) found that log fishers have a higher productivity than cable skidders where ground slope exceeds 20 degrees. However, log fishers have higher machine and operating costs than cable skidders. Another unique difference between the two logging systems in Malaysia is that skid trail construction is mandated by the government for skidding operations, whereas log fishers do not rely on skid trails. In addition, according to the forest regulations in Malaysia, the total length of skid trails from any log landing should not exceed 300m. Because of these differences in costs, productivity, and regulations, Norizah et al. (2012) suggested that the two logging systems could be complementary to each other, and applying a combination of the systems might reduce the overall costs of timber harvesting in a given harvest unit. In this study, we have attempted to 1 Dr, Universiti Putra Malaysia, 43400 UPM, Serdang, Selangor, Malaysia. norizah_k@upm.edu.my. 2 Associate Professor, University of Montana, Missoula, MT 59812, USA, woodam.chung@umontana.edu. 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 1

model and solve this harvest area planning problem to minimize total costs of timber extraction while considering the two aforementioned logging systems and designing required skid trails and forest roads. Although there were several past studies that attempted to solve different harvest area and road network layout problems (Chung et al. 2003, Anderson and Nelson 2004, Epstein et al. 2006, Najafi et al. 2008, Chung et al. 2008), those studies did not simultaneously consider logging systems, skid trails, and road constructions. Also, the two logging systems, as well as skid trail construction requirement and its limitation on maximum length, make the harvest area planning problem in Malaysia unique, which has not been modeled and solved mathematically. Our attempt to model and solve this unique problem is still in progress. In this paper, we present different modeling approaches to tackle such unique and challenging problems based on graph theory and compare preliminary solutions of the approaches. METHODS Network representation of timber harvesting operations Once a harvest area is rasterized, timber flows from individual tree locations to landing and then to the final destination through road network can be represented by series of links developed between grid cells (i.e., nodes). In our approach, some grid cells represent timber volume and location (i.e., timber nodes), others may represent locations of skid trails, landings, or forest roads. Links between grid cells may represent skidding activities, log fisher operations, or truck transportation. Each link is attributed with variable or fixed costs or both depending on the activity that the link represents. In the case of skidding activities, links are attributed with both variable and fixed costs representing skidding and skid trail construction costs, respectively. For log fisher operations, only variable cost will be assigned to the link. For timber transportation, both variable and fixed costs are assigned to the links representing new roads, while only variable costs (i.e., haul cost) exist on the links representing the existing roads. The same physical links between grid cells may often represent multiple activities. For example, a link between two adjacent grid cells can represent skidding activity, log fisher operation, or timber transportation. Since no duplicate links are allowed in network formulation, we used different labels of from- and to-nodes along with different cost attributes to differentiate activities that occur in the same physical locations. Costs associated with each activity are estimated based on a productivity study conducted by Norizah (2013). However, there has not been any productivity study on timber transportation so far in Peninsular Malaysia. Thus, hauling cost was estimated for this study according to expenses allotted for log transportation by forest concessionaire. Road construction cost was estimated using the past study by Mohd Hasmadi et al. (2008). All the cost data used in this study are summarized in Table 1. 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 2

Table 1. Cost data used in this study for timber harvesting activities. (i) Stump-to-landing Skidding cost with cable skidder MYR 0.06/m 3 /m Skid trail preparation MYR 0.5/m Winching cost with log fisher MYR 0.13/m 3 /m (ii) Timber transportation Hauling cost MYR 0.096/m 3 /km Road construction cost MYR 18/m * MYR 1.0 is approximately equivalent to USD 0.31 Log fisher operations Links representing potential log fisher operations were built from each timber node to multiple neighboring nodes that are within 300m proximity of the timber node (Figure 1). Links are attributed with variable cost only because no skid trail is required. If one of the links is selected, to-node of the link becomes a log landing, and then is connected to the forest road system for truck transportation. Figure 1. A link pattern representing log fisher operation from a timber entry node to neighboring nodes within a 300m distance. Cable skidder operations Timber node Neighbor node Possible link with costs assigned Due to the unique regulations on skidding operations in Malaysia (i.e., skid-trail construction requirement and maximum skidding distance limit of 300m), formulating the harvest area planning problems becomes challenging. We have not yet identified a precise approach to formulate and solve the unique problem mathematically, but our attempt is still in progress. In this paper, we present three different proxy models we developed to represent the unique problem while relaxing one or more regulations. In the first model (i.e., Model A), we assumed that skidding operation can be done in a similar way to log fisher operations (Figure 1) where the log can directly skidded to a log landing within a 300 m from the stump location. However, unlike log fisher operations, skid trails need to be built for skidding operations. Therefore, each link representing skidding operation includes both fixed and variable costs. On the contrary, the second model (i.e. Model B) does not assign fixed cost to each link, but instead skid trail construction cost of the maximum distance (i.e., 300m) is added at the log landing, assuming that skid trail construction cost does not vary with skidding distance. In Model B, the links (i.e., skid trails) that can be used by each timber node are labeled uniquely using a combination of grid cell location and relative direction from timber node (Figures 2a through 2e). This labeling method allows tracking the total length of skid trails from 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 3

individual timber nodes and thus limiting the maximum skidding distance. The same skid trail may be labeled differently when it gets used by multiple timber nodes. In this case, an imaginary link with no cost is built between the two different labels representing the same grid cell so that the same skid trail can be shared by multiple timber nodes. When a skid trail ends at a landing location, the pre-determined skid trail construction cost is added to the total costs (Figure 2f). Figure 2. Node labels representing skidding activities from a timber entry node to multiple neighboring nodes based on grid cell locations and relative directions from timber nodes (Figures 2a through 2e). Model C assumes there is no maximum skidding distance limit. Therefore, each node is linked to its eight adjacent nodes through skidding activities. Both fixed and variable costs are assigned to each link for skid trail construction and skidding operations (Figure 3). Figure 3. Links representing skidding activities from a timber entry node to its eight adjacent nodes. Each link was attributed with both variable and fixed costs. Road construction and timber transportation Timber node Neighbor node Possible link with costs assigned Because every to-node of the links representing skidding or log fisher operations can potentially serve as a log landing, the to-nodes need to be connected to the existing roads through another set of network representation of road building and truck hauling operations. In order to do this, an imaginary link is developed to connect each grid cell used as to-nodes of timber harvesting operation to itself that represents forest road (e.g., S13 -> R13 in Figure 4). The node is then further connected to its eight neighbors by links that represent forest roads. These links are attributed with both fixed costs (road construction costs) and variable costs (truck hauling costs). Any grid cells representing the existing forest roads serve as the final destination of timber. 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 4

S1 L1 S3 L3 S5 L5 S11 S15 S21 S13 Imaginary link L13 L11 L15 L21 S23 L23 S25 R13 L25 (a) (b) Figure 4. Grid cells that represent three different activities in the same physical location with three different labels (a), and an example of link created for a timber entry node 13 representing skidding activities (labeled with S), log fisher operation (labeled with L), and road activities (labeled with R) (b). Solution method NETWORK 2000 equipped with a heuristic network algorithm (Chung and Session, 2003) was applied to find the least cost routes from each timber node to the final destination. With the ability to solve fixed and variable cost network problems, the tool provides near-optimal solutions that determine the least-cost extraction methods (i.e., skidder vs. log fisher), skid-trail locations, landing locations, new forest road locations, and transportation routes for timber harvesting operations in a given harvest unit. APPLICATION R1 R3 R5 R11 R15 R21 R23 R25 We applied the methods developed in this study to a 102 ha harvest unit located in Ulu Jelai Forest Reserve, Pahang of Peninsular Malaysia (Figure 5). The study area was rasterized with a 30m resolution, and a total of 550 grid cells were attributed with timber volumes of 6,632m 3 in total, which serve as entry nodes in the network problem. There are 50 grid cells representing the existing roads. Any of these road grid cells serve as the final destination in our network problem. Figure 5. Study area located in Ulu Jelai Forest Reserve, Pahang of Peninsular Malaysia, showing distribution of timber nodes on gentle slopes (<20 degrees) and slopes of 20 degrees and more. 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 5

PRELIMINARY RESULTS The least cost solutions found by NETWORK 2000 for the three problem formulation models indicate that cable skidder is a preferred harvesting system in most gentle areas regardless of the models (Figures 6a, 6b and 6c, for Models A, B, and C, respectively). Skid trail locations from Models A and B are not shown in Figure 6 because no specific skid trail routes can be identified with Model A and skid trails need to be manually tracked down for displaying the Model B results. For comparison, we also presented a timber harvest area plan manually developed by the Malaysian Forestry Department (Figure 6d). Table 2 compares the three solutions and the manual plan in terms of total forest road length, number of landings, and selected logging systems. The Model B result shows more log fishing operations selected compared to the other two Models. This is because Model B overestimates skid trail construction costs by considering construction of the maximum skid trail distance (i.e., 300m) regardless of skidding distance, which made log fisher more competitive in some areas. In addition, it appears that fewer number of log landings were selected (Table 2) because skid trail construction costs are added at the landing in Model B, and thus the more landings would cause the higher overall costs. This fewer number of log landings further reduced the total length of forest roads required. Forest roads proposed by the Forestry Department are located throughout the entire harvest unit. The Model A shows the similar pattern of road locations, whereas the results from Models B and C show different road patterns. Figure 6. Forest road locations resulted from Models A, B and C (a, b, and c), and road locations proposed by the Malaysian Forestry Department (d). 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 6

Table 2. Summary of harvest area plans resulted from Models A, B and C, and the Malaysian Forestry Department. Model A B C Forestry Department Road length (m) 4836.88 4646.48 4977.09 5001.32 Skid trail length (m) - - 4816.48 10892.12 (max: 247.84m) (max: 312m) Total 57 60 62 13 No of Cable skidder 51 19 59 - landing Log fisher 24 54 16 - No of timbers with cable skidder 482 328 486 - No of timbers with log fisher 68 222 64 - CONCLUDING REMARKS Timber harvest area planning problems in Malaysia are unique because of the government regulations on skid trail construction and maximum skidding distance limit. Our attempt to exactly formulate and analyze the unique problem using a network algorithm is still in progress. Here we presented three problem formulation models that simplified the problem by relaxing one or more regulations. Each model has its own limitations and advantages that will be further investigated in our on-going research. Developing a precise problem formulation and solution method would benefit forest practitioners in Malaysia in finding the least-cost options for their timber harvesting operations. References Anderson, A.E. and Nelson, J. 2004. Projecting vector-based road networks with shortest path algorithm. Canadian Journal Forestry Research, 34 (2004): 1444-1457. Chung, W., Stuckelberger, J., Aruga, K. and Cundy, T.W. 2008. Forest road network design using a trdeoff analysis between skidding and road construction costs. Canadian Journal Forestry Research, 38(2008):439-448. Chung, W. and Sessions, J. 2003. NETWORK 2000: a program for optimizing large fixed and variable cost transportation problems. In Proceedings of the 8th Symposium on Systems Analysis in Forest Resources, Snowmass Village, 27-30 September 2000, Colorado. Edited by G.J. Arthaud and T.M. Barrett. Kluwer Academic Publishers, Dordrecht, the Netherlands. Pp. 109-120. Epstein R., Weintraub A., Sessions J.B., Sessions J., Sapunar P., Neto E., 2006. A combinatorial heuristic approach for solving real size machinery location and road design problems in forestry planning. Operations Research, 54: 1017 1027. Mohd. Hasmadi, I, Kamaruzaman, J and Muhamad Azizon, J. 2008. Forest road assessment in Ulu Muda Forest Reserve of Peninsular Malaysia. Journal of Modern Applied Science 2(3): 101-108. Najafi, A., Sobhani, H., Saeed, A., Makhdom, M. and Mohajer, M.M. 2008. Planning and assessment of alternative forest road and skidding networks. Croatian Journal of Forest Engineering 29 (1):63-73. Norizah, K. 2013. Advanced forest harvest operation and road location planning to reduce negative environmental impacts in Peninsular Malaysia. Ph.D Thesis, Universiti Putra Malaysia, Malaysia. Norizah, K., Mohd Hasmadi, I., Kamaruzaman, J. and Alias, M.S. 2012. Operational Efficiency of Rimbaka Timber Harvester in Hilly Tropical Forest. Journal of Tropical Forest Science, 24(3): 379-389. Saharudin, A., Azmi, H. and Mohd-Zamri, R. 2004. Kajian mengenai kos dan produktiviti sistem pengusahasilan hutan bukit di Semenanjung Malaysia, Jabatan Perhutanan Semenanjung Malaysia: Kuala Lumpur. 37th Council on Forest Engineering Annual Meeting. 2014. Moline, Illinois 7