Tracking the spatio-temporal change of planting area of winter wheat-summer maize cropping system in the North China Plain during 2001–2018

Highlights

• An effective scheme to map long-term and large-scale distribution of winter wheat-summer maize was proposed.

• The impacts of different temporal resolutions of MODIS data and sample categories on classification were investigated.

• The spatio-temporal change of winter wheat-summer maize planting area during 2001–2018 in the North China Plain was analyzed.

• The study provided important information for agricultural water resources management.

Abstract

Discriminating winter wheat-summer maize distribution is fundamental for agricultural water resources management in the North China Plain (NCP) which confronts severe conflicts between food production and agricultural water scarcity. However, how the planting area changed since 2000 is still unclear. Therefore, the winter wheat-summer maize from 2001 to 2018 was identified by using the maximum likelihood algorithm of supervised classification based on the Moderate Resolution Imaging Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) product. Three schemes with different temporal resolutions of MODIS data and sample categories were compared to obtain the most reasonable identification results. The results reveal that using 16-day 0.0025° MODIS NDVI data and six sample categories including winter wheat-summer maize, winter wheat-rice, spring maize, cotton, other one cropping systems and other double cropping systems is an effective way for tracking the variation of winter wheat-summer maize distribution. The winter wheat-summer maize planting area showed an insignificant increasing trend at a rate of 34.2 km²/year (p = 0.955, CV = 7.49%) during 2001–2018, while it decreased at the rate of −5065.2 km²/year (p = 0.074) after 2012. Spatially, the planting area shrunk in northern NCP, the piedmont plain of the Taihang Mountains, the west of dry sub-humid zone and the southwest part of the NCP, accounting for 58.6% of the counties during 2001–2018; while 17.3% of the counties significantly expanded winter wheat-summer maize in the irrigation district along the Yellow River in Shandong province, the southeast of Hebei Plain and the central humid zone. Our results provide fundamental information for quantifying the change of crop water consumption and optimizing crop water resource allocation.

Introduction

The North China Plain (NCP) is one of the largest food production areas in China (Li et al., 2014, Tao et al., 2017a), where cropland takes up more than 80% of this region. Winter wheat and summer maize rotational cultivation is the typical cropping mode, and approximately 75% and 35% of China’s wheat and maize are produced in this region (Mo et al., 2017). However, crop water consumption (i.e., evapotranspiration (ET)) is usually higher than precipitation in the winter wheat growing season because of the high crop water consumption and uneven seasonal distribution of precipitation. The multi-year average seasonal ET during the wheat and maize seasons is around 238 mm and 221 mm, respectively, while the respective precipitation is only around 169 mm and 410 mm in the NCP (Fang et al., 2020). Thus, irrigation mainly from groundwater or rivers is necessary to favor crop production (Fang et al., 2010, Mo et al., 2009). However, continuous pumping from groundwater has caused serious overexploitation which led to a dramatic declining trend of groundwater table, especially in the Hebei Plain in the north part of the NCP over the past 40 years (Zhang et al., 2016, Zhang et al., 2018). To alleviate this problem, the government has implemented a series of policies to constrict groundwater pumping in recent years. For example, the project of reducing winter wheat fields to fallow fields in deep groundwater seriously over-exploited area was formulated in Hebei province in 2014 (Wang et al., 2016, Zhang et al., 2018). Under the implementation of these policies, accurately mapping the winter wheat-summer maize planting area since 2000 is fundamental for evaluating the changes in crop water consumption.

Compared with labor-intensive and time-consuming traditional crop identification methods (Jiang et al., 2019, Li et al., 2015, Tang et al., 2018), remote sensing technology has the advantages of providing timely, large coverage and repetitive information and low cost or even free, making it a useful tool to cropland mapping (Atzberger, 2013, Wardlow et al., 2007). However, the land use types and cropping systems in the NCP are complicated, manifested by (1) fragmented cropland resulted from highly intensive villages, small towns (Tan et al., 2006), and build-up areas with high industrial intensity (Liu et al., 2019); (2) various cropping systems and seasonal dynamic patterns caused by the household contract responsivity system (Wu et al., 2005) which enables peasants to autonomously choose crop species and cultivated time in their fields; (3) the inter-annual fluctuant cropping systems affected by market prices, policies and climate conditions. These complicated tempo-spatial characteristics make large-scale and long-term remote sensing mapping of winter wheat-summer maize remains a challenge.

The spatial resolution of satellite images can be mainly divided into four categories: very high (<10 m), high (10 m–30 m), coarse (>1 km), and moderate (250 m–500 m). Though satellite images with much high spatial resolution, such as the GeoEye-1(Caturegli et al., 2015), Landsat TM/ETM+ (Hansen and Loveland, 2012, Zhu and Woodcock, 2014), HJ-1A/AB CDD (Yu and Shang, 2017), Sentinel-1 and Sentinel-2 (Tian et al., 2019), performed best to capture the spatial distribution of crops in a small filed, they confront the limitations of low temporal resolution and poor data quality (Atzberger, 2013, Jiang et al., 2019, Wardlow et al., 2007, Wardlow and Egbert, 2008) which limits their application in mapping long-term continuous variations of large-scale crop distribution for the NCP. Take the Landsat data as an example, each Landsat image is collected every 16 days and the quality is often damaged by frequent clouds and aerosols contamination (Tian et al., 2019, Yan et al., 2019). It makes it almost impossible to acquire sufficient images of good quality covering a large area during a crop growing season, which destroys the integrity of the vegetation index (VI) curve and fails to accurately record the crop growth process. This limitation affects the identification of rotation systems more seriously due to their short phenological stage and rapidly changing VI curve in a year. Though this problem can be circumvented by using the threshold method based on single or several images, the Landsat data covering a large-scale requires to be obtained by mosaicking satellite images with different acquisition dates and locations (Ozdogan, 2010, Zhu and Woodcock, 2014). These differences cause that the growth period and VI values of crops in each Landsat image to be inconsistent, resulting in unreasonable threshold selection and large-scale mapping highly uncertain accordingly. On the other hand, the coarse spatial resolution images, such as 1 km SPOT and 8 km AVHRR data, could cover “mixed” different land use types (Tian et al., 2019, Wardlow et al., 2007). Taking the case of Quzhou County which is a typical agricultural county in the NCP for example (Fig. 1. (c)), the minimum mean cultivated field patch area of towns is about 700 m × 700 m (Sun et al., 2019) which is smaller than the 1 km footprint, indicating that the coarse resolution images are inappropriate for mapping agricultural systems in this region. As a compromise, moderate resolution images provide pertinent information to separate crops in areas with complex planting structures and fragmented agricultural landscapes (Vintrou et al., 2012). Up to now, a series of high quality Moderate Resolution Imaging Spectroradiometer (MODIS) VI products such as normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) datasets are available for crop mapping (Arvor et al., 2011, Chen et al., 2018, García-Mora et al., 2012, Zhang et al., 2011). They had sufficient spectral, spatial, temporal and radiometric resolutions to discriminate cropping patterns where field size was 500 m × 500 m or larger (Doraiswamy et al., 2004, Wardlow et al., 2007).

For the classification method, the supervised classification based on MODIS data is the most common method for crop classification (Chen et al., 2018, García-Mora et al., 2012, Lu and Weng, 2007, Yu and Shang, 2017). In this method, sample categories and the MODIS VI product are two critical factors influencing the classification results. First, the sample category determines the pattern of the eventual maps and should represent the main cropping systems (Lu and Weng, 2007). However, the training categories in previous studies differed greatly, which could introduce large uncertainties. For example, in the entire NCP, Zhang et al. (2008a) divided four cropping systems (winter wheat-summer maize, winter wheat-cotton, spring maize, and cotton), while Dong et al. (2008) developed eight categories (wheat, cotton, garlic, vegetables, orchard, conifer, urban and water). In the north part of the Yellow River in the NCP, although the number of categories in the three studies (Pan et al., 2015, Wang et al., 2017a, Zhang et al., 2019) were all four, the target categories differed from each other, varying from winter wheat-summer maize, other crops, forest and built-up land (Pan et al., 2015) to winter wheat-summer maize, summer maize, spring maize and cotton (Wang et al., 2017a) to winter wheat-summer maize, single spring maize, cotton and forest/fruit trees (Zhang et al., 2019). This situation is more complicated for identifying the double cropping system as it could be either pure winter wheat-summer maize pixels or another cropping system that also shows bimodal signals caused by mixed crops, but the latter one was usually not distinguished in previous studies. Second, the MODIS VI product with an appropriate temporal and spatial resolution is the priority for successful classification (Lu and Weng, 2007). For spatial resolution, 250 m is more accurate than 500 m in discriminating winter wheat-summer maize in the NCP. For temporal resolution, it determines the degree of detail in characterizing the crop growth process (García-Mora et al., 2012). However, in previous studies, the MODIS NDVI/EVI temporal resolution varied from 8-day (Skakun et al., 2017, Yan et al., 2019, Zhang et al., 2008a) to 16-day (Wu et al., 2019, Zhang et al., 2011, Zhang et al., 2019). None studies have evaluated the classification performance of MODIS data with different temporal resolutions in the NCP, and it leaves confusion that which temporal resolution of MODIS is more suitable to identify winter wheat-summer maize.

Moreover, previous studies focused on identifying crop types in a given year (Sun et al., 2012, Tao et al., 2017b, Wang et al., 2017a, Zhang et al., 2008b) or analyzing changes in some parts of the NCP (Pan et al., 2015, Wang et al., 2015a). To our knowledge, the long-time continuous spatio-temporal change of winter wheat-summer maize planting area in the entire NCP was not investigated. Thus, the objectives of this study are to (1) explore a suitable and effective method to map the long-term and large-scale distribution of winter wheat-summer maize in the NCP, based on the supervised classification and MODIS data; (2) analyze the continuous spatio-temporal changes of winter wheat-summer maize from 2001 to 2018.