1. Introduction
Cropland abandonment, an extreme manifestation of land marginalization [
1], is an ongoing trend in both developed and developing countries worldwide [
2], attracting great attention at a global level [
2,
3,
4,
5]. Due to rapid economic growth, urbanization and industrialization have been vigorously promoted in recent years, accelerating the transfer of the large rural labor force to urban areas [
6]. In 2000, the number of rural migrant workers in Chinese cities reached 150 million; this figure increased to 269 million in 2013, accounting for 60–70% of the total rural labor population [
7]. Due to the combined influences of rural-to-urban migrations and other natural socio-economic factors, cropland abandonment in China has intensified, making it an important issue.
Briefly, cropland abandonment is described as the cessation of cultivation in cropland for a certain period. However, the defined period of ceased cultivation remains the subject of debate [
8]. For example, the Food and Agriculture Organization (FAO) defined abandoned cropland as cropland that has not been utilized for agricultural production or other agricultural purposes for at least five years [
9], while the International Symposium on Land Consolidation and Land Reserve in 2011 defined cropland abandonment as two or more years [
10]. In this study, the latter definition, i.e., ceased cultivation of cropland for at least two successive years, was adopted. Here, two years of successive cessation was emphasized to distinguish abandonment from fallowed land. In China, land fallowing is a new phenomenon that has emerged in recent years, although it remains unpopular due to the huge pressure for food security. Most fallowed land is uncultivated for one year [
11,
12].
The forms of cropland abandonment are spontaneous and induced [
13]. Spontaneous cropland abandonment is an initiative of farmers based on their own planting decisions, while induced cropland abandonment is a result of policy guidance, such as the Common Agricultural Policy (CAP) in Europe and the Grain-for-Green Policy in China. The former policy encouraged the withdrawal of agricultural land from production [
14,
15,
16], while the latter encouraged farmers to convert sloping cropland to forestry or grassland [
17,
18].
In general, two types of approach have been adopted to estimate the scale of cropland abandonment in previous research: household surveys and remote sensing monitoring. The highest merit of the former approach is ease of use: There are few technical difficulties in using household surveys to assess the quantity of abandoned cropland. Therefore, many scholars have adopted the survey approach; for example, Li et al. [
19] conducted a large-scale survey to estimate the extent of cropland abandonment in mountainous China and found that the average cropland abandonment rate was 14.32%. In spatial, the abandoned cropland in China was found mostly distributed in southern China based on a meta-analysis [
20]. A field survey from Nepal showed that nearly 46.6% of land has been abandoned due to the growth of market towns and urban centers, and the opening of the country to the international job market [
21]. In the U.S., a farmers’ perceptions survey of cropland abandonment showed that 28% of farm land owners considered some of their land abandoned [
22]. While household surveys can easily acquire the quantity of abandoned cropland, their weakness is in mapping the spatial information of abandoned cropland. In addition, a typical survey cannot usually provide a full view of the whole cropland abandonment in the study area. In spite of the household survey and satellite image datum, the cadastral datum is a defendable datum to help mapping abandoned cropland [
23]. However, in some regions such as China, there are no public cadastral databases, limiting the utilization of the cadastral datum.
Freely accessible satellite images have provided many datum sources to map abandoned cropland. Based on the spatial resolution of satellite images, the Moderate Resolution Imaging Spectroradiometer (MODIS), Landsat, and high-resolution satellite data, such as Quick-Bird, and aerial photographs are generally adopted to identify abandoned cropland. The advantage of the MODIS data is the high temporal resolution. However, the spatial resolution is low, 250 m, which means that fragmented abandoned cropland usually cannot be identified. When MODIS datum has been used to extract abandoned cropland, the Normalized Difference Vegetation Index (NDVI) time series covering the whole growing season of crops was employed to detect the vegetation change [
8]. The spatial resolution of Landsat data is about 30 m, which is much higher than that of MODIS, and has the advantage of identifying fragmented abandoned cropland. However, due to the low temporal resolution (about 16 d), it is usually difficult to find enough cloud-free images to establish long time series NDVI over a large region. Therefore, direct land use classification has been performed to capture land use conversions and identify abandoned cropland. The Random Forest model [
24,
25,
26] and Support Vector Machine [
27,
28] are the two most popular methods adopted to classify land use. In addition, other global or regional land use and land cover databases, such as the History Database of the Global Environment [
29] and National Land Cover Database of the United States [
30], have been used to assist in mapping abandoned cropland.
As remote sensing technology has improved, the spatial resolution of satellite images can reach one meter or better, providing more opportunities to map abandoned cropland. For example, researchers utilized Quick-Bird images and existing land use/land cover map layer to identify abandoned cropland in the Murcia Region of Southeast Spain [
31]. In general, aerial photographs have higher spatial resolutions than those of satellite images. When identifying the abandoned cropland in topographically variable areas, such as gullies [
32] and terraces [
33,
34], aerial photographs have been adopted. For ultra-high resolution tasks, aerial photographs are very useful for identifying detailed information regarding abandoned cropland, such as the height of succession vegetation after land abandonment [
35] and the forest transition from deforestation to afforestation [
36]. The advantages of high-resolution satellite images and aerial photographs are clear. However, there are also some disadvantages hindering the application of these data. Most high-resolution satellite images and aerial photographs are not freely available, and one scene of image or photograph only covers a very small area. This results in very high costs for identifying abandoned cropland, even in a relatively large region using these data. In addition, unlike satellite images, aerial photographs usually exist in few areas, and periodic updates are limited.
Although some research has been conducted in identifying abandoned cropland using remote sensing technology, huge challenges remain, e.g., differentiating abandoned cropland from fallowed cropland and recultivated abandoned cropland [
2,
8,
24,
26,
34,
37,
38,
39,
40]. In prior work, abandoned cropland was usually identified by dividing the changes in vegetation or land use conversions using snapshots of land use data at the start and end years of a given research period. For example, Prishchepov et al. [
27] mapped cropland abandonment in post-Soviet European Russia from 1990 to 2000 using the interpreted land use map in 1990 and 2000. Malavasi et al. [
41] identified abandoned cropland in the Mediterranean mountain areas over a total span of 29 years using Landsat images in 1987, 2003, and 2016. This approach can either underestimate or overestimate the abandoned land. For the lack of continuous land use information, the timing of a certain land parcel’s removal from cultivation cannot be accurately recorded. If the abandoned cropland plot, which was identified at the end of the time span, was just removed from cultivation for rotation, it should be fallowed land. Moreover, some active cropland identified at the end of a time span could also have been abandoned and recultivated. These possibilities generate some uncertainties in identifying cropland abandonment. In addition, distinguishing spontaneous abandonment from induced abandonment is particularly important for countries having implemented agricultural policies to intervene in land abandonment, and who want to better understand land abandonment behavior and the impacts of policy. Unfortunately, most previous research has not succeeded in differentiating the different types of abandonment.
To address the described challenges in mapping abandoned cropland and provide a technique support for policy making about addressing cropland abandonment, this research developed an annual land-use trajectory method to differentiate abandoned cropland land from fallow land, and to identify spontaneous and induced abandonment. By evaluating the timing of abandoned cropland, this research also assessed the recultivation of abandoned cropland. To reduce the cost of identification and effects of fragmented land parcels on identification, and to increase the transferability of the method, this research adopted freely available Landsat and HJ-1B images to conduct this research. In brief, the specific objective of this research is to develop a new approach to map different kinds of cropland abandonment in mountainous areas using free and medium resolution satellite images.
6. Conclusions
This research developed an annual land-use trajectory approach to identify abandoned cropland in mountainous areas of China. By coupling a CART classification model and land-use trajectory tracing approach, the abandoned cropland in Zhongduo Town was mapped. To better describe the abandoned cropland, a detailed classification was proposed. Five types of cropland, spontaneously abandoned, induced abandoned, fallow, stable, and lost, were mapped. In addition, using the land-use trajectory trace approach, the timing and recultivation of abandoned cropland were mapped.
It was found that the average spontaneous abandonment rate in Zhongduo from 2012 to 2017 was 23.16%, which is in good agreement with published household survey data from nearby regions. Among the abandoned and fallowed croplands, spontaneously abandoned cropland accounted for the largest proportion, 23.16%, followed by fallowed cropland, 13.69%, and induced, 8.40%. Spontaneously abandoned cropland was abandoned for an average of 3.45 years. In addition, 50.54% of spontaneously abandoned cropland was recultivated in comparison to only 12.91% of induced cropland. The relatively good land quality and policy subsidy assisted in maintaining stable induced abandoned cropland. Fluctuations in agricultural production prices and the land transfer policy has resulted in recultivated spontaneously abandoned cropland.