Abstract: A lot of open space including farmland is being invaded and occupied because of economic development and city expansion. Sustainable use of farmland is increasingly concerned for the ecological functions, and non-market value of farmland has been recognized and accepted widely. The key issue of rational use of farmland is characterizing the spatial distribution pattern of farmland conversion and identifying the locations of hotspots and cold spots of farmland conversion. In this paper, farmland conversion data were extracted by using spatial statistics and spatial analyst functions in ArcGIS based on the land use data between 1990 and 2011. To explore the characteristics of spatial distribution and moving route of spatial agglomeration districts of farmland conversion, spatial autocorrelation analysis is applied to different periods of farmland conversion in the Wuhan City. Based on the analysis, the main conclusions are as follows: Firstly, the global Moran's I index indicated that the farmland conversion in the Wuhan City showed significant global spatial autocorrelation during the all four periods from 1990 to 2011, and the aggregation intensity displayed an increasing-decreasing tendency. The aggregation intensity trend reflected that the farmland conversion in the Wuhan City presented spread and out of order in space in the period from 2005 to 2011. Secondly, Empirical Bayes methods could increase the stationarity of variables while the local Moran's I statistic is computed for rates or proportions. So we used the local indicators of spatial association cluster map with the Empirical Bayes methods to show the location of cluster areas of farmland conversion in the Wuhan City and the spatial distribution of different types of aggregation (e.g., low-low, low-high and high-high). The areas presenting high-high type clustering are exactly the hotspots of farmland conversion. The conversion between the types of aggregations indicates that low-high type clustering is a potential high-high type clustering, and low-high type clustering will have a greater probability to convert into high-high type clustering. So low-high type clustering present the priority regulation areas in the perspective of the protection of farmland resources. The number of accumulation areas continued rising in the Wuhan City, and the leading role of farmland conversion has changed from low level of farmland conversion areas to high level ones. Both these description means farmland protection is becoming more difficult than before. Thirdly, the location of cluster areas of farmland conversion in the next period can be pre-estimated through drawing the moving route of different types of aggregation. The centers of gravity of low-high type clustering and high-high type clustering are expected to appear to the southeast of centroid in Wuhan, and that of low-low type clustering is expected to appear to the western. Recognizing the spatial characteristics of farmland conversion by spatial autocorrelation analysis is the foundation to protect farmland effectively and use efficiently. The research results and analysis methods can provide a theoretical reference and basis for the rational allocation and regulation of land resources.

Key words: spatial autocorrelation, farmland conversion, hot spots, Wuhan city