Analysis of spatial pattern of farmland and its impacting factors in coastal zone of Circum Bohai

Abstract: In this paper, coastal zone of Circum Bohai Sea Region which covers an area of approximately 170, 000 km2 was selected as the study area. The spatial distribution characteristics of farmland of this study area were analyzed and the relationship between farmland distribution and natural, social or economic impacting factors was explored. Based on Landsat TM images acquired in 2009/2010, farmland distribution map was created through visual interpretation with auxiliary data in ArcGIS 9.3. Then farmland distribution map was overlaid with a lattice map to statistic area of farmland in each 5 km × 5 km lattice. Impacting factors of farmland consisted of elevation, slope, distance to nearest coastline, distance to nearest railway, distance to nearest road, distance to nearest residential area, distance to nearest river, average yearly precipitation, average yearly temperature and population density, which were compiled into raster format data with a spatial resolution of 5 km × 5 km and normalized between 0 and 1 in ArcGIS 9.3. As conventional statistical methods assumed that the data to be analyzed was statistically independent, it was inappropriate to use traditional statistical method to analyze spatial land use data which had a tendency to be dependent. In this study, ordinary least square linear regression model (OLS), spatial error model (SEM), spatial lag model (SLM) and geographically weighted regression model (GWR) were established from global and local perspectives. Several evaluation indexes were selected to assess the performance of those models. The results showed that: 1) Farmland was the main land use type, which occupied 53% of the whole study area. Positive spatial autocorrelation that decreased gradually with distance was detected in both farmland distribution and impacting factors; 2) Spatial autoregressive models and GWR had a better goodness-of-fit than conventional linear regression model. As to spatial autoregressive models, SEM performed better than SLM in this study, as was indicated by higher preudo R2 value and maximum likelihood logarithm (LIK) value, and lower Akaike information criterion (AIC) value, Schwartz criterion (SC) value and residuals for the former model; 3) GWR could be used to explore spatial variation in the relations between cultivated land distribution and different impacts factors, providing more detailed information, while SEM could only explore the relations from a global view; 4) The SEM showed a positive correlation between farmland and elevation, slope, distance to the nearest roads, as well as a negative correlation between farmland and distance to nearest shoreline, distance to nearest railroad, distance to nearest settlements, average yearly temperature, average yearly precipitation from a global perspective; and 5)The GWR revealed both positive and negative correlations between farmland and impacting factors (expect for average yearly precipitation). Among the most sensitive factors affecting farmland distribution, average yearly temperature and average yearly precipitation were the main positive factors, while elevation, slope and distance to nearest residential area were the main negative factors.