JIA-2018-09

2105 WANG Di et al. Journal of Integrative Agriculture 2018, 17(9): 2096–2106 spatial autocorrelation level could be reduced almost to zero. Consequently, traditional sampling methods can be reasonably used to estimate crop area. Overall, from the perspective of the reasonable selection of sampling unit size and stratification criteria, this study provides an important basis for formulating an effective spatial sampling scheme for crop acreage estimation, based on the analysis of spatial autocorrelation of crop area contained in the sampling units. Geographic entities may be not mutually independent owing to the influence of spatial interactions and diffusion, which results in spatial autocorrelation between them. Spatial autocorrelation is an important property of geographic data and a common phenomenon existing in various spatial contexts. Since Tobler (1970) put forward the First Law of Geography, more and more attention has been paid to the spatial autocorrelation of geographic data. Using an index, such as Moran’s I or Geary’s C , many previous studies have demonstrated the existence of spatial autocorrelation among various geographic objects, and pointed out that spatial autocorrelation among variables in survey sampling units should be considered. The variables of interest in prior studies mainly focused on ecology, hydrology, soil, geology, forestry and environment. However, there is no reported research on the spatial autocorrelation among crop acreage included in the sampling units. Based on the various thematic maps derived from satellite-based remotely sensed images, this study explored the spatial autocorrelation characteristics of crop acreage contained in sampling units of different sizes, and then analyzed the influence of stratification criteria and sampling unit size on the spatial autocorrelation of crop acreage, arriving at a reasonable stratification design and the sampling unit size scope that can be used in the spatial sampling scheme for crop area estimation. The results of this study provide reference points for designing a reasonable spatial sampling survey scheme to estimate crop acreage at a regional scale. Crop planting structures and spatial distributions generally vary between regions. To improve the applicability of this study, more research is needed to verify the existence and explore the characteristics of spatial autocorrelation in different crop producing regions. This study analyzed the spatial autocorrelation of four variables (crop type, CPI, cultivated land fragmentation and average ground slope), but was limited by the availability of experimental data. Other factors exhibiting spatial autocorrelation and, therefore, potentially affecting the choice of sampling units for crop area estimation, should be explored, to strengthen and extend the findings in this study. In addition, when there is strong spatial autocorrelation in a variable among the sampling units, a study on how to take this into account in the calculation of sample size, the extrapolation of population parameters, and the estimation of sampling errors in the spatial sampling scheme for crop acreage estimation, will be a future research focus. 5. Conclusion This study shows that although the spatial autocorrelation level among maize and rice acreages within the sampling units generally decreased with increasing sampling unit size, there was still a significant spatial autocorrelation among the two crop acreages included in the sampling units (Moran’s I varied from 0.49 to 0.89), regardless of their sizes. The spatial autocorrelation of rice area within the sampling units was stronger than that of maize. When the sampling unit size was less than 3000 m×3000 m, the stratification design that used crop planting intensity (CPI) as the stratification criterion, with a stratum number of 5 and a stratum interval of 20% decreased the spatial autocorrelation level to almost zero for the maize and rice area included in sampling units within each stratum. Therefore, traditional sampling methods can be used to estimate the two crop acreages. Compared with CPI, there was still a strong spatial correlation among the two crop acreages included in the sampling units belonging to each stratum when cultivated land fragmentation and ground slope were used as stratification criteria. Acknowledgements This research was financially supported by the National Natural Science Foundation of China (41471365, 41531179). We thank the anonymous reviewers for their comments and suggestions. We also thank Dr. Leonie Seabrook from Liwen Bianji, Edanz Group China (http://www.liwenbianji.cn/ac ), for editing the English text of this manuscript. References Anselin L. 1988. Spatial Econometrics : Methods and Model s. Kluwer Academic Publishers, Dordrecht. pp. 86–95. Benedetti R, Bee M, Espa G, Piersimoni F. 2010. Agricultural Survey Methods . John Wiley and Sons, Chippenham. pp. 125–136. Betts M G, Diamond A W, Forbes G J, Villard M A, Gunn J S. 2006. The importance of spatial autocorrelation, extent and resolution in predicting forest bird occurrence. Ecological Modelling , 191 , 197–224. Buarque D C, Clarke R T, Mendes C A B. 2010. Spatial correlation in precipitation trends in the Brazilian Amazon. Jouranl of Geophysical Research , 115 , 1–14. Carfagna E, Gallego F J. 2005. Using remote sensing for agricultural statistics. International Statistical Review , 73 , 389–404.