Proc., BIOENERGY '96 - The Seventh National Bioenergy Conference: Partnerships to Develop and Apply Biomass Technologies, September 15-20, 1996, Nashville, Tennessee.
The farmgate cost and available supply of biomass often exhibit considerable variation within a State. This variation, combined with the relatively high cost of transporting bulky biomass material, produces a wide range of expected delivered feedstock costs across a State. As a consequence, both production and transportation costs must be well-modeled when analyzing potential locations for conversion facilities. The Regional Integrated Biomass Assessment system consists of two phases. The descriptive phase characterizes a farmgate cost and supply "surface" for switchgrass production over a given State. These results are passed to the analytical phase, where a transportation model is used to compute the marginal cost of supplying an ethanol plant at a prescribed level of demand. The model generates a marginal cost "surface" that illustrates the most promising areas for locating an ethanol plant. Next, a sequential location model simulates the commercial development of ethanol production facilities. This model considers every road network node as a potential site and generates a sequence of likely plant locations. Results from the RIBA analysis demonstrate that the cost of switchgrass can increase dramatically from one location to another. This variation will seriously effect the economics of conversion in the proper sizing and locating of ethanol plant facilities.
Keywords: transportation, site location, biomass
For any agricultural resource, considerable variation in the farmgate cost, as well as the available supply, will be observed from one location to another across a State or region. The degree of variation in the spatial pattern is far greater than suggested by mere State or regional averages. Consider the case of Alabama, where the State's average market value of agricultural products sold averaged $62,503 per farm in 1991 (U.S. Census Bureau, 1992). However, in the best county, the market value of agricultural products sold per farm was nearly twice the State average, and 17 times the per farm average of the worst county. The variability of supply is similarly illustrated considering the 1991 corn crop in Alabama. Yields per unit of land area were more than three times greater in the best county compared to the worst county, yet total supply (in bushels) from the best county was 232 times the total supply from the worst county. Clearly, in areas with higher yields per acre, many more acres are devoted to the particular crop resulting in much greater variation in supply from one location to another than one would otherwise expect. Commercial production of potential biomass crops like switchgrass (Panicum virgatum) is likely to exhibit at least as much variation in farmgate costs and supplies as a conventional crop like corn. The variable pattern of farmgate costs and supplies across a State is the product of a complex interaction of features of the natural environment—such as climate and soils, combined with elements of a State's economy—such as demand for and prices of crops, the transportation infrastructure, and agricultural technology. Determining the prices and supplies of any potential biomass crop, such as switchgrass, requires combining features of the natural environment with information about a State's economy. Descriptive data characterizing the elements of the natural environment are combined with information about a State's economy within the context of a Geographic Information System (GIS) and estimates of farmgate cost and supply of switchgrass are developed for each square kilometer in the entire State. The descriptive model of farmgate price and supplies is used as a basic input to both parts of the analytical phase of the Regional Integrated Biomass Assessment (RIBA) system. The descriptive model is more fully explained by Graham, et al. (1996).
In the first part of the analytical phase of the RIBA system (the surface model), the model of farmgate prices and supplies is combined with a transportation modeling algorithm to determine the marginal cost of delivering a prescribed amount of switchgrass to any specific location. We consider a pixel to be representative of one square kilometer. Since every pixel is considered as a potential conversion facility site and has an associated marginal price, a map that characterizes the marginal cost "surface" of supplying switchgrass for that prescribed demand level across the State can be generated. The resulting map provides a graphic representation of the spatial variability of feedstock costs for supplying switchgrass to any location across a State. Although analysis based on State averages alone may indicate that commercial production of switchgrass may or may not be economically feasible, the marginal cost surface maps effectively identify niche areas where conditions may vary significantly from the "average" State condition. In these niche areas, commercial production of switchgrass for a potential conversion facility may be a viable alternative for the local agricultural economy.
In the second part of the analytical phase (site modeling), the same farmgate price and supply data described above are combined with a plant siting algorithm that sequentially selects conversion facility sites in a least cost fashion. In choosing a site to locate an agricultural conversion facility, the variations observed in farmgate cost and supply are obvious factors irrespective of the agricultural resource being converted is corn for corn syrup, or switchgrass for ethanol. However, in comparison to conventional crops such as corn, supplying switchgrass to a conversion facility will require transporting a bulk item with relatively lower value per unit of volume. Thus, transportation will be a large com- ponent of the cost to supply a switchgrass conversion facility. The transportation cost, irrespective of who actually pays it, is always borne by the buyer. Assuming that the capital outlay for facility construction and the labor costs for operating a switchgrass conversion facility are consistent across a State, then the delivered feedstock cost is the main determinant in the variable cost of producing a unit of ethanol. This variable cost may fluctuate considerably across a State. As a consequence, production and transportation costs must be well-modeled when analyzing potential locations for conversion facilities.
For a given State, the descriptive phase of the RIBA system consists of land use informa- tion obtained from a national database developed from Advanced Very High Resolution Radar (AVHRR) satellite imagery. Since AVHRR imagery has a 1-km resolution, the land area of a State is composed of pixels of 1 km2 (100 ha). The RIBA system descriptive phase calculates the farmgate cost of switchgrass bales in $U.S. and the supply in dry tonnes per year for every pixel in the State. A digital file consisting of a unique pixel identification number, the farmgate price, and the supply quantity for every pixel in a State is created.
A 1:100,000 scale digitized road network is obtained for each State. The digital road network consists of three classes of roads encoded as links and nodes. Nodes are points to which the links are connected and usually represent intersections. The links represent the actual roads. Class I roads are interState and other four-lane divided highways. Class II roads are major four- or two-lane highways. Class III roads are paved two-lane secondary roads. Each road class is categorized as either fast or slow. Class I roads are categorized as fast with an assumed travel speed of 80 km/h (50 mph). Class II and Class III roads are categorized as slow with an assumed travel speed of 48 km/h (30 mph). The transportation model of RIBA requires a sufficient density of road network nodes. As a consequence, each State's digital road network must be manually edited to some extent. The objectives of the road network editing are to filter out excessive network nodes in congested urban areas and add nodes in sparsely populated rural regions. In addition, the network must have complete connectivity; i.e., it must be possible to travel from any network node to any other network node in the State. For example, if there is a road on an island that can only be reached via a ferry, the road must be either deleted, or the ferry route must be digitized as a Class III road (allowing for the relative slow speed of the ferry and the loading and unloading delays). The final edited version of the digital road network is not a precise planimetrically correct digital road map, but rather a reasonably accurate digital representation of a State's automotive transportation infrastructure. This final version of the digital road network is processed using an "all to all" shortest path algorithm to determine the shortest route (based on travel time) between a given node and every other node in the network.
Once the price and supply data for all pixels in a State are completed and the road network has been fully edited, each pixel is assigned to its nearest road network node. This process is accomplished through the use of a "nearing" algorithm within the ARC/INFO GIS software package, in combination with the digital land use information and the digital road network. This procedure produces an output file containing the unique pixel identification number, the road network node identification number, and the "crow's flight" distance between the pixel and its nearest node. This distance represents a virtual link between an individual pixel and a network node and is assigned the slow travel speed associated with Class II and Class III roads.
Before the marginal price of delivering a bale of switchgrass to the processing facility can be determined, one must first calculate the transportation cost of moving the bale from the origin pixel that supplies the bale to the destination pixel where the bale is received at the conversion facility. The transportation cost for moving a bale of switchgrass is a function of the total distance and time for hauling, and the time for loading/unloading. A cost parameter is developed for each of these elements.
The total distance incurred in a pixel-to-pixel haul consists of (1) the crow's flight distance from the origin pixel to its nearest network node; (2) the shortest time route distance across the network from the node nearest the origin pixel to the node nearest the destination pixel; and (3) the crow's flight distance from the destination pixel to its nearest network node. The crow's flight distances between every pixel and its nearest node are already determined by use of the nearing algorithm within the GIS, but the shortest time route over the network must be determined independently. For every link in the road network a travel time is computed by dividing the link distance by the associated link speed; i.e., fast link distances divided by fast speeds, slow link distances divided by slow speeds. Using these link travel times as a basis, the shortest time path along the road network can be algorithmically determined for each node to every other node. After the shortest time path along the road network is calculated, the total distance is calculated by adding the three components together. The total distance is multiplied by the distance cost parameter of approximately 5 cents/km. The value of the distance cost parameter is derived from empirical data on truck costs such as fuel, repairs, lubrication, maintenance, and tires, and trailer costs including repairs and tires. In addition, the truck and trailer are assumed to operate 2000 h/yr, there are three trailers per truck, and 12.7 dry tonnes can be hauled per trailer load. Thus, for a 40-km haul, the total distance cost function (including the truck's empty back haul) would be $2.00/dry tonne.
Once the total distance of a haul is known, the time cost can be calculated. The total dis- tance is measured in kilometers, but is composed of both fast links and slow links. The slow link travel time is estimated by summing all the slow link distances and dividing by the slow link travel speed. Similarly the fast link speed is estimated by summing the fast link distances and dividing by the fast link travel speed. The total travel time is then sim- ply calculated by adding the slow link travel time and the fast link travel time. The crow's flight distance from the origin pixel to the nearest road network node, and the crow's flight distance from the destination pixel to the nearest road network node are always assumed to be slow links, and therefore use the slow speed. The total travel time is multiplied by the time cost parameter of approximately $4.00/h/dry tonne. The value of the time cost parameter is derived from empirical data related to time-dependent cost variables such as the driver labor costs, truck and trailer time costs including depreciation, interest, insurance, and fees. The same assumptions about operating hours per year, number of trailers per truck, and haul capacity used for the distance cost apply. In addition, a real interest rate is assumed to be 6%. A 40-km, 45-min one-way haul would result in a time cost of roughly $3.00, depending on the labor costs of the region.
The transportation costing must also include loading and securing the bales onto a trailer and unsecuring the load and unloading the trailer at the conversion facility. It seems reas- onable to assume the loading and transportation activities would be conducted either by a contract hauler or by the conversion facility. It does not make good economic sense for each farm to purchase expensive transportation equipment that will be used very few hours per year. In our model, the tasks of loading and transporting are assumed to be conducted by two individuals. One individual (the driver), is responsible for spotting empty trailers at the farm and moving loaded trailers to the conversion facility. At the farm, the driver must unhitch an empty trailer from his truck and hitch a loaded trailer to his truck. At the conversion facility, he must unsecure the load and wait for the trailer to be unloaded by plant personnel and equipment. The second individual (the loader) oper- ates a forklift at the farm and is responsible for loading and securing bales onto the trailer. The activities of the loader are performed concurrently with the hauling. Since there must be an empty trailer available for loading while the trailer is being hauled, there must be at least one extra trailer that is not being moved by a truck at any given time. We assume 1.5 trailers are assigned to a truck to help accommodate variation in loading time, hauling time, and scheduling. One loader is estimated to be able to keep two trucks loaded and moving. A truck is estimated to be able to haul two to five loads per day, depending on the travel time between the farm and the conversion facility. The loading/ unloading cost is composed of trailer loading costs such as the loader's labor, the forklift equipment cost (depreciation, interest, insurance, and fuel), truck and trailer waiting cost during hitching, unhitching, and unloading, and driver labor cost during hitching, unhitching, and unloading. The forklift is assumed to operate 2000 h/y. Based on empirical data, the loading/unloading cost is calculated as approximately $3.00/dry tonne depending on the regional labor rate.
The delivered cost of receiving switchgrass at its destination pixel from a supplying pixel is computed by adding the pixel's farmgate price to the total calculated transportation cost (the sum of hauling and loading costs). In analyzing a destination pixel, a delivered cost is computed corresponding to every other pixel, and the delivered costs are sorted in a least cost fashion. The destination pixel is then supplied sequentially from the sorted list of origin pixels until the prescribed conversion facility demand level is achieved. The delivered cost of switchgrass from the pixel whose supply is needed to just meet the prescribed demand level is then assigned as the marginal cost of switchgrass to supply a conversion facility at the destination pixel. The procedure repeats by considering the next pixel as a destination, and continues until every pixel has been considered as a potential conversion facility site. The resulting set of marginal prices represent a spatially continuous cost "surface" for delivered switchgrass.
The surface model is developed for a specified processing-facility demand level and assumes that only one processing facility will be located in the State. For processing convenience and computational efficiency, multiple demand levels can be specified in a single RIBA analysis, but each cost surface that is generated essentially represents an individual analysis for each demand level specified, and in each case assumes only one processing facility will be located in the State. However, while only one facility may be located in the State, every pixel is considered as a potential site for that facility. The consideration of every pixel as a potential site is the essential component that allows for the generation of a cost surface. Since every pixel is considered, we compute a marginal cost for each. The total cost for delivering the last bale of switchgrass needed to meet the specified demand level of the processing facility is the marginal cost assigned to that pixel. Once the marginal cost is calculated and assigned for each individual pixel, a spatially continuous marginal cost exists for every location in the State. The spatially continuous nature of the assigned marginal cost values represents the cost "surface" for the State. When these assigned values are mapped, they graphically display the spatial variability of delivering switchgrass to a processing facility of the specified demand level at any location in the State.
The site modeling procedure utilizes the same data inputs as used previously with the surface model. In fact, the transport costs are calculated in much the same manner and with the same assumptions as in the surface modeling. Rather than compute marginal costs for all pixels, marginal costs to supply switchgrass are calculated for each road network node. The transport cost is calculated with the same hauling, and loading/ unloading cost parameters as described above. However, since only road network nodes are considered as potential sites, the haul distance only consists of the distance from a supply pixel to the nearest road network node, and the shortest time path distance from the origin pixel's nearest node to the destination node. The road network node that has the lowest marginal cost for switchgrass at the prescribed conversion facility demand level is selected as the first conversion facility site. Once the first site is selected, all pixels used to supply that location are removed from the available supply. Marginal costs are then recalculated and re-sorted for all remaining road network nodes. The node with the lowest marginal cost from among those remaining network nodes is selected as the next conversion facility site, and the pixels used to supply it are removed from the available supply. This process is repeated and continues until the available supply is exhausted, or there is insufficient leftover supply to meet the demand level for the next conversion facility. The results are presented in graphic form as a conversion facility map. This map reveals the number and locations of conversion facilities that can be located in a State at the prescribed demand level. It also displays the variable cost of delivered feedstock costs across the State.
The RIBA system is used to analyze the cost of supplying switchgrass to conversion facilities of demand levels of 100,000 and 600,000 tonnes/yr. At both levels of demand, the surface maps reveal considerable spatial variation in the feedstock costs of supplying switchgrass. The southeastern area of Alabama offers the most promising potential, with a pronounced extension of the low-cost surface areas into the famed "Black Belt" region of grassland soils in south central Alabama. Not surprisingly, at the 600,000 tonnes/yr demand level, a conversion facility must pay significantly higher prices for feedstock. However, the best areas at the 600,000-tonne demand level are much less expensive than some areas at the 100,000-tonne demand level. Clearly, selecting a conversion facility location will have a major impact on the cost of producing ethanol fuel. The conversion facility map for 100,000- tonne facility indicates over 60 facilities could be located in Alabama, mostly concentrated in the southeastern and central portions of the State. The marginal costs of feedstock would range from less than $35 in the best locations, upward to more than $60 in the worst areas. At the 600,000-tonne demand level, only a dozen sites can be located, and feedstock costs range from $45 to $80.
If the capital outlay for construction and the labor costs for operation of a switchgrass processing facility are consistent across Alabama, and accounted for half the cost of producing a unit of ethanol, then the best areas in the State could produce a unit of ethanol for almost 25% less than the worst areas, at either a 100,000- or a 600,000- tonne demand level.