Probabilistic Modelling of Cattle Farm Distribution in Australia I. V. Emelyanova & G. E. Donald & D. J. Miron & D. A. Henry & M. G. Garner Received: 2 August 2007 / Accepted: 21 January 2008 # Springer Science + Business Media B.V. 2008 Abstract A probabilistic Bayesian method called weights of evidence (WofE) was used to develop a synthetic dataset of cattle farm locations at a national scale across Australia. The synthetic dataset was required for the modelling of livestock movements with a view to assessing biosecurity implications. The WofE method is based on the analysis of spatial relationships between evidential patterns with respect to an event, such as the actual location of a farm. The evidential patterns of cattle farms were derived from maps of land use, land tenure, drainage systems, roads, settlements and long-term averaged rainfall. These eviden- tial patterns were used for delineating and ranking land areas suitable for cattle farming. For each evidential pattern statistics such as a positive weight,a negative weight and a contrast were calculated for estimating the degree of correlation between the evidential patterns and known farm locations. The integrated evidential patterns of known farms were then used for estimating posterior probabilities and splitting land into five different classes according to its suitability for farming. Keywords Bayesian modelling . Weights of evidence . Conditional independence . GIS . Farm location . Spatial distribution of farms . Lacunarity 1 Introduction Movements of animals and interaction between farming establishments are key drivers for the spread of livestock diseases, both endemic and exotic. Of particular concern for Australia are diseases like foot and mouth disease (FMD). This exotic disease is a highly contagious viral infection of cloven-hoofed domestic and wild animals. It is spread primarily via respiratory aerosols and via direct and indirect contact between animals. Understanding the movement of livestock both spatially and temporally is critical to understanding and managing the key risks and vulnerabilities in our livestock industries, for scenario planning in the event of an outbreak, for informing policy for surveillance and in providing decision support when responding to emergencies. Such movements can be between farms, feedlots, traders, ports, saleyards and abattoirs. The one underpinning requirement of such an analysis is knowledge of where animals are in the landscape. Unfor- tunately, at the national level, Australia does not have a complete database of the geographic location of all farms containing livestock. An agricultural census is conducted by the Australian Bureau of Statistics every 5 years, but due to confidentiality/privacy issues, data on individual farms is not publicly available. Individual state and territory Environ Model Assess DOI 10.1007/s10666-008-9140-z I. V. Emelyanova (*) Commonwealth Scientific and Industrial Research Organization (CSIRO), Livestock Industries (LI), Private Bag 5, Wembley, WA 6913, Australia e-mail: irina.emelyanova@csiro.au G. E. Donald : D. J. Miron CSIRO, LI, Locked Bag 1, Armidale, NSW 2350, Australia D. A. Henry CSIRO, LI, Australian Animal Health Laboratory, Private Bag 24, Geelong, VIC 3220, Australia M. G. Garner Office of the Chief Veterinary Officer, Department of Agriculture, Fisheries and Forestry, GPO Box 858, Canberra, ACT 2601, Australia