JIA-2019-11

2629 ZHANG Xi-wang et al. Journal of Integrative Agriculture 2019, 18(11): 2628–2643 1. Introduction The key to global food security is sustainable agriculture, and improved agricultural practices require current, spatially explicit information on crops (Conrad et al. 2014; Zhang and Zhang 2016). Timely information on crop acreage at regional and national scales is essential for accurately predicting yields, optimizing spatial patterns, and agricultural planning (Wu and Li 2012; Zhang et al. 2019). Wheat is one of the most important food crops in the world. The largest share of global wheat production is concentrated in China, which contributes 17.3% of the global production (FAO 2014). Furthermore, Henan Province accounts for a quarter of China’s total wheat production (Zhou T et al. 2017). Thus, it is vital to conduct research on winter wheat in Henan for reasons related to both food security and environmental sustainability (Qiu et al. 2017). Remote sensing has proven to be an effective and widely used tool in the agricultural field, for applications such as mapping and monitoring, for many decades (Oetter et al. 2001; Bolton and Friedl 2013; Potgieter et al. 2013), by providing positive solutions to pressing agricultural problems (Murakami et al. 2001; Gallego and Bamps 2008; Xiao et al. 2014). By using medium or high spatial resolution remote sensing data, spectral feature-based methods have been frequently used to estimate crop acreage and distributions (MacDonald and Hall 1980; Dawbin and Evans 1988; Löw et al. 2013; Zhou Q et al. 2017). However, the reflectance similarity of different crops may appear in agricultural areas with many crop types. In these conditions, multi-temporal images can be used to identify crop types based on the spectral differences at different times in the growing season (Liu et al. 2003; Carrao et al. 2008; Cai et al. 2018). The main problem is that the availability of such frequent ideal optical remote sensing images is seriously limited in many areas by cloud cover, adverse atmospheric conditions and other factors (Baumann et al. 2011; Schmidt et al. 2015). When ideal images cannot be acquired during a specific period, the identification accuracy will be affected due to the seasonal variability of different vegetation types and land surface characteristics. Some studies have shown that the images obtained on the optimal date can outperform the worst images by 13% in land cover classification (Nitze et al. 2015). The frequent remote sensing data with coarse resolution, such as NOAA/AVHRR (1.1 km), SPOT/VEGETATION (1.15 km), and TERRA/MODIS (250–1000 m), provide a feasible tool to delineate vegetation phenology and monitor crop types (Pan et al. 2015). The rapid availability of these data is necessary for the accurate and operational crop acreage estimations and mapping over large regions (Xiao et al. 2005; Wardlow and Egbert 2008; Geerken 2009; Upadhyay et al. 2016). However, these types of images do not provide sufficiently detailed spatial change characteristics. When using hard classification techniques, errors will be introduced in estimating crop acreage and mapping crop types, particularly for areas with a planting structure comprised of many crop types (Zheng et al. 2012; Estel et al. 2015). Thus, several methods have been developed to extract sub-pixel information (i.e., acreage ratios of different crop types within a pixel) based on such time series data (Lobell and Asner 2004; Busetto et al. 2008; Ozdogan 2010). The major challenge lies in selecting pure endmembers in coarse resolution data, even with the aid of finer resolution data (Pan et al. 2012). This issue is very relevant in assessing the crop acreage in China because of the small fields and common practice of mixing different crops (Wu and Li 2012). In addition, these methods output only coarse resolution abundance and cannot account for the spatial distribution of sub-pixels. Spatial and temporal resolutions of remote sensing products are the two main features to consider when extracting information (Poggio and Gimona 2013; Li et al. 2017). In practical applications, there is generally a trade- off between temporal and spatial resolutions, as current sensors rarely have both high spatial resolution and high temporal resolution (Amorós-López et al. 2013; Zhu et al. 2016; Zhang et al. 2018). This trade-off generally leads to non-ideal results. Therefore, optimal results are obtained by combining the spatial features of finer spatial resolution remote sensing images and the temporal frequency of coarser spatial resolution images (Gao et al. 2006; Hilker et al. 2009; Zhu et al. 2010). This combination can provide a feasible and economical solution (Ling et al. 2012; Gao et al. 2017), especially when ideal finer spatial resolution images cannot be obtained. In the normalized difference vegetation index (NDVI) can be improved by integrating spectral and temporal information. Since the identification of winter wheat is performed in the space corresponding to each MODIS pixel, the influence of differences of environmental conditions is greatly reduced. This advantage allows the proposed method to be effectively applied in other places. Keywords: temporal change characteristics, membership, abundance, winter wheat, multi-resolution remote sensing