Establishing Farm Spatial Characteristics using Landsat and Local Data for North-West Romania for Initialisation of an Agent-Based Land-Use Change Model

摘要

Understanding and predicting land-use and land-cover change (LUCC) remains a significant challenge due to the complex interaction of social-economic and biophysical drivers that affect farmer (and other land owners) decision-making. This paper reports upon the development of an agent-based model for assessing LUCC to 2050 under various climate change scenarios. Agent-based modelling (ABM) is a computer simulation methodology for understanding complex coupled social, economic, and environmental systems. This research focuses on a rural case study area in north-west Romania, which encompasses the two village communes of Poieni and Sacuieu (approx 100 km~(2)). A social survey of 124 farmers in the region found that the majority of land uses in the region are grazing pastures, hay meadows, potato fields and forest; and that land-use and land ownership have been largely static over recent years. An initial difference analysis of Landsat TM and ETM+ images from 1992 and 2002 respectively using Normalised Difference Vegetation Index difference (NDVI) confirms this, showing significant change only adjacent to the rivers, probably due to changing water-levels. Paper cadastral maps of the area exist dating from the 1860s (with records of subsequent changes in land ownership). The cadastre currently lacks georeferencing information except place names but shows boundaries, streams and dwellings, many of which still exist. The cadastre has been scanned and will be geo-located to aerial photographs and the Landsat images. The combined data will be used to guide the interpretation of the Landsat data to identify and delineate farm boundaries. Landsat TM and ETM+ images from 1980, 1992 and 2002 will be classified to determine the land-use of each field at those dates. The classified images and social surveys provide a baseline of land holdings and land-use for the agent-based model. The model incorporates human decision-making processes through farmer agents who can change the land-use of their fields and trade land-parcels. Projections can then be tested using hindcasting against the satellite-derived land-cover data.