Spatiotemporal Dynamics of the Cropland Area and Its Response to Increasing Regional Extreme Weather Events in the Farming-Pastoral Ecotone of Northern China during 1992–2020

Abstract

The cropland area is closely related to food production. Previously, more focuses were paid on impacts of extreme events on food production, but less on cropland dynamics. This study used the piecewise linear regression, the correlation analysis, and the ridge regression to explore the spatiotemporal dynamics of cropland and its drivers in three time periods (1992–2020, 1992–2010, and 2010–2020) at the Farming-Pastoral Ecotone of Northern China (FPEN). Specifically, 13 driving factors were considered from the perspectives of extreme events, environmental conditions, socioeconomic development, urban sprawl, and ecological construction. Results showed that the cropland area increased significantly at a rate of 333.5 km²/a during 1992–2020 and was spatially clustered in the eastern FPEN. The impact extent and size of each driving factor on the cropland trend presented large spatiotemporal differences, but ecological construction had, overall, the greatest impact on cropland area changes, followed by urban sprawl. In comparison, extreme low temperature had the smallest. Since the 2010s, areas with urban sprawl being the dominant factor in cropland dynamics have increased by 41.9%, but it is still less than the ecological construction impact. Furthermore, this study found that extreme event effects on the cropland area trend evidently increased. Particularly, extreme high temperature displayed the highest increase (~99.4%). Cropland area changes dominated by extreme temperature events in 2010–2020 increased by nearly six times compared to those in 1992–2010. These results suggest that policy and planning makers should caution increasing impacts of regional extreme weather events on cropland area changes.

1. Introduction

In the context of the climatic warming and growing global population, global agriculture development sustainability and food security are facing large challenges [1]. As an irreplaceable physical carrier of food production, cropland bears a great responsibility for the survival and development of humanity and is a constant issue in the process of human development [2,3,4]. In recent decades, the amount and spatial distribution of cropland has changed dramatically under the influence of human activities and climate change [5,6]. Globally, the area of cropland has increased by 9% during 2003–2019, and the rate of cropland expansion has almost doubled [7]. Nevertheless, regionally, the expansion of cropland has not occurred uniformly [8]. Disparities in the spatial alteration of cropland stem from variations in the pace of economic growth, stages of industrialization and urbanization, the execution of ecological restoration initiatives, and diverse climatic conditions [9]. These factors collectively influence regional food security and the delicate equilibrium between supply and demand [10]. Hence, it is imperative to elucidate the distinctive traits of spatial shifts in the cropland extent and their underlying drivers. This is essential to establish a theoretical foundation for crafting well-informed policies aimed at safeguarding croplands and upholding food security.

The driving mechanism of the cropland area trend is a complex issue that is forced by multiple drivers of natural and human activities, such as climate, environmental conditions, socioeconomic development, urban sprawl, and ecological construction [11,12,13]. Urban sprawl is generally regarded as the dominant driver of a cropland area decrease [8,14]. In the past decades, half of decreased cropland was used for urban sprawl [9]. Ecological construction represents expansion of ecological land, such as forests and grasslands in a region, and is another key factor affecting the cropland area [15]. The major reasons for a cropland decrease in fragile ecological areas have been proven [9,15]. Some studies have explored the relationships between the cropland area trend and the environmental conditions and socioeconomic development in recent years [16,17]. For example, by analyzing relationships between the cropland area and environmental conditions, Zhong et al. [18] found that the cropland area decreased along with an elevation gradient increase during 1999–2006. Wang et al. [19] selected fifteen drivers, collecting from the perspectives of cropland status, environmental conditions, and socioeconomic development, to assess impacts of cropland expansion, and they found that all of these drivers had large impacts on the cropland area trend at different spatial and temporal scales.

Climate change and extreme events are also among the drivers of the cropland trend [5,10,20]. Their impacts on cropland change are indirect, mainly through changes in climatic conditions (e.g., light, heat, and water) that affect crop growth, development, and yield [21,22]. Significant regional differences exist in the impact of climate change on cropland [5]. Numerous studies have indicated that the warming climate has facilitated the northward expansion of cropland in China, leading to an augmentation in the potential suitability of northern regions for cropland cultivation [23,24]. For example, Shi et al. [25] reported that climatic warming trends facilitated the cropland area increase in arid and semi-arid regions. In contrast, extreme events typically exert greater impacts on agriculture than events of average magnitude [26,27]. They can result in substantial harm to croplands, frequently culminating in direct reductions in crop yields and, in severe instances, exacerbating land degradation [28]. Corey et al. [29] found that at the national level, a drought and extreme heat reduced cereal yields significantly by 9~10%. For cropland, yield reductions can lead to two contradictory outcomes. On the one hand, under pressure from food demand and economic returns, farmers will increase yields or minimize losses by expanding their cropland [5]. As reported by Zaveri et al. [13], repeated dry anomalies increase cropland expansion to compensate for lower yields. On the other hand, extreme events may lead to the conversion of cropland to more economically valuable types of land by reducing agricultural productivity, and further lead to a reduction in croplands [30]. Additionally, as the frequency, intensity, and duration of extreme events increase, they may lead to degradation of croplands that are no longer suitable for farming.

The main methods used to quantify the contribution of drivers to cropland change include a partial correlation analysis, a regression analysis, logistic regression models, a principal component analysis, a geographical detector, and geographically weighted regression [17,19,31,32,33]. Each of these methods has its own advantages and disadvantages and is applicable to different situations. Among them, a partial correlation analysis is able to analyze the relationship between variables and dependent variables while excluding the influence of other factors, but it is limited to a small number of variables [34]. The remaining methods can quantify the effects of several variables on changes in area, but do not exclude the influence of other factors on the results. There are interactions between the factors involved in this study, which may lead to less accurate results when using the above methods. The ridge regression model proposed by Hoerl and Kennard [35] is well able to overcome the problem of imprecise calculation results due to the interaction of independent variables (presence of multicollinearity). For this reason, this method was used in this study to quantify the contribution of the drivers to the change of croplands.

China, as a populous and agriculturally significant nation, holds approximately 21% of the world’s population and 9% of global cropland [36]. According to the latest data from the Food and Agriculture Organization (FAO, Rome, Italy), China’s per capita cropland area stands at ~0.09 hectares [37]. It is significantly lower than the global average per capita cropland area [11]. Consequently, the issue of food security in China has garnered considerable attentions [38,39]. Since the launch of reform and opening-up policy, China’s rapid economic growth has led to an accelerated process of urbanization and industrialization, resulting in substantial cropland loss [10]. Despite efforts to curtail the reduction in the cropland area, the Chinese government has implemented a series of cropland policies aimed at safeguarding croplands since 1978, with the goal of regulating changes in both the quantity and quality of cropland at a macro level [10]. However, existing research findings indicate that in China, a substantial amount of lower-quality cropland in the northern regions is displacing higher-quality cropland in the southern regions. Despite being characterized by low crop productivity, severe land degradation, and a fragile ecological environment, the lower-quality cropland in the northern areas is emerging as the primary source of supplementary cropland resources [16,17,40,41]. The Farming-Pastoral Ecotone of Northern China (FPEN) is ecologically fragile and particularly vulnerable to human activities and climate change [42,43]. In recent years, the quantity and spatial distribution of cropland have changed dramatically under the influence of climate change and human activities [44]. Due to its coverage of typical geographical features such as the Northeast Plain, the Inner Mongolia Plateau, and the Loess Plateau, as well as substantial regional variations in policies and climate, this region provides a natural experimental setting for studying the spatiotemporal characteristics of cropland area changes and the driving factors.

Existing studies have shown that a cropland area exhibits non-linear changes under the combined effect of several factors; in other words, it must have one or more turning points [45]. In order to better understand the causes of cropland area change in more depth, this study hypothesizes that the non-linear changes in a cropland area result from variations in the contributions of drivers before and after a turning point in the time series. In summary, the aims of this study, using the FPEN as an example, are (1) to analyze the spatiotemporal dynamics of the cropland area in the FPEN and the turning point, (2) to quantify the relative contributions of extreme events, environment conditions, socioeconomic development, urban sprawl, and ecological construction on change in the cropland area, and (3) to clarify the differences in contribution of driving factors before and after the turning point, and the dominant driver of the cropland area trend at the county scale.

2. Data and Methods

2.1. Study Area

The FPEN’s fragile ecological environment makes its boundaries highly vulnerable to changes in climate and human activities [46,47]. To facilitate the quantification of the impacts of drivers on the cropland area at the county level, FPEN (34.8°–48.5° N, 100.9°–124.6° E) covers 245 counties of 10 provinces (Figure 1). Croplands and grasslands are the main land use types, accounting for ~73.9% of FPEN, of which the cropland area was ~2.02 × 10⁵ km² (Figure 1a). Its terrain is complex and diverse, which is low in the northeast and high in the southwest. FPEN has a temperate continental monsoon climate, with rainfall considerably affected by the summer monsoon, predominantly in summer, with an average annual precipitation of 250~600 mm (Figure 1d). The average annual temperature is 6.8 °C (Figure 1c). Considering the large differences in the climate, environment, and policies in the study area, FPEN was subdivided into three regions. Specific information is given in

Table 1. Basic information on the three regions.

Region	Descriptions	Average Annual Temperature (°C)	Average Annual Precipitation (mm)
Eastern	Located in the east of the study area, it belongs to the Northeast China Plain. It has abundant black soil resources, and flat terrain, providing good conditions for agricultural activities and mechanized production. It is also China’s main commercial grain production base and a key implementation region for cropland protection policies [42,43].	5.8	468.2
Central	Located in the center of the study area, connecting the Inner Mongolia Plateau. And the terrain is complex and diverse, and areas of cropland and grassland are the largest. Due to the prominent problems of land and environmental degradation, it has become a key ecological reserve in China [47].	7.5	551.0
Western	Located in the west of the study area. It has low vegetation cover, severe soil erosion, and a particularly fragile ecological environment. And it is a key implementation area for ecological projects such as the “Grain for Green” project [48].	7.4	600.8

.

2.2. Data and Processing

The data used in this study included land use/land cover data, meteorological data, drought index data, volumetric soil water data, net primary productivity data, population data, and night-time light index data.

The land use/land cover data for 1992–2020 were obtained from the European Space Agency (ESA, Paris, France), which has an overall accuracy of about 75.4%, with a high accuracy of 92% for cropland (code 10). The main data used in this paper are arable land data; therefore, this dataset can be used in this study (details can be found under the Land Cover CCI Product Validation and Intercomparison Report (PVIR) v2, https://www.esa-landcover-cci.org/?q=webfm_send/138, accessed on 13 January 2022). Using the reclassification function of ArcGIS, the land use/land cover data were grouped into six categories, croplands, forests, grasslands, water body, built-up land, and bare land, according to the classification system of the International Geosphere-Biosphere Programme (IGBP) and the Chinese Academy of Science (detailed information is provided in

Table A1. International Geosphere-Biosphere Programme (IGBP) and Chinese Academy of Science classification system.

Chinese Academy of Science Classification System	Primary Land Use Category	Secondary Land Use Category	Descriptions	IGBP Classification System	Note
1	Cropland	Cropland, rainfed	Cropland (>60%)	10	Cropland, rainfed
				11
				12
				20	Cropland, irrigated or post-flooding
		Mosaic cropland/natural vegetation	Mosaic cropland (>50%)/natural vegetation (tree, shrub, herbaceous cover) (<50%)	30
2	Forests	Needle-leaved evergreen forests	Dominated by needle-leaved evergreen (canopy > 2 m)	70	Tree cover, needle-leaved, evergreen, closed to open (>15%)
				71	Tree cover, needle-leaved, evergreen, closed (>40%)
				72	Tree cover, needle-leaved, evergreen, open (15–40%)
		Broad-leaved evergreen forests	Dominated by broad-leaved evergreen and palm (canopy > 2 m)	50	Tree cover, broadleaved, evergreen, closed to open (>15%)
		Deciduous needle-leaved forests	Dominated by deciduous needle-leaved (canopy > 2 m)	80	Tree cover, needle-leaved, deciduous, closed to open (>15%)
				81	Tree cover, needle-leaved, deciduous, closed (>40%)
				82	Tree cover, needle-leaved, deciduous, open (15–40%)
		Deciduous broadleaved forests	Dominated by deciduous broad-leaved (canopy > 2 m) Forest cover (>60%)	60	Tree cover, broadleaved, deciduous, closed to open (>15%)
				61	Tree cover, broadleaved, deciduous, closed (>40%)
				62	Tree cover, broadleaved, deciduous, open (15–40%)
		Mixed leaf type	Tree cover, mixed leaf type (broad-leaved and needle-leaved). Forest cover (>60%)	90	Tree cover, mixed leaf type (broadleaved and needle-leaved)
		Shrubland	Dominated by woody perennials (1~2 m tall)	120	Shrubland
				121	Evergreen shrubland
				122	Deciduous shrubland
		Mosaic tree and shrub/herbaceous cover	Mosaic tree and shrub (>50%)/herbaceous cover (<50%)	100
		Mosaic herbaceous cover/tree and shrub	Mosaic herbaceous cover (>50%)/tree and shrub	110
				40	Mosaic natural vegetation (tree, shrub, herbaceous cover) (>50%)/cropland (<50%)
3	Grasslands	Grasslands	Dominated by annual herbaceous plants (<2 m)	140	Lichens and mosses
				130	Grasslands
4	Water body	Permanent wetlands	Permanent wetland cover between 30 and 60% and vegetation cover (>10%)	160	Tree cover; flooded, fresh, or brackish water
				170	Tree cover; flooded, saline water
				180	Shrub or herbaceous cover; flooded, fresh/saline/brackish water
		Water body	Permanent water body cover (>60%)	210	Water body
5	Built-up	Urban areas	Urban areas	190
6	Bare land	Sparse vegetation	Sparse vegetation	150	Sparse vegetation (tree, shrub, herbaceous cover) (<15%)
				151	Sparse tree (15%)
				152	Sparse shrub (15%)
				153	Sparse herbaceous cover (15%)
		Permanent snow and ice	More than 60% of the area is covered by snow and ice for at least 10 months of the year	220	Permanent snow and ice
		Bare land	At least 60% of the area is barren with no vegetation (sand, rock, soil) and <10% vegetation	200	Bare areas
				201	Consolidated bare areas
				202	Unconsolidated bare areas
				170	Unconsolidated bare areas
				180	Shrub or herbaceous cover; flooded, fresh/saline/brackish water
255	No data		Land use categories that could not be identified due to missing inputs	0	No data

). To analyze spatial changes of the land use area, the sum area of croplands, sum area of forests and grasslands, and urban area within each 9 km × 9 km grid cell were counted. Specially, for the mosaic croplands (>50%) in the IGBP classification system, only half the area for croplands was counted. In addition, the sum of forests and grasslands area is considered in our research as the ecological land area, and the area of built-up land as the construction land area.

Meteorological data and volumetric soil water data were all derived from the European Centre for Medium Range Weather Forecasts (ECMWF) Reanalysis ERA5 dataset. Here, daily maximum and minimum temperature, and annual average temperature were calculated using the hourly maximum and minimum 2-m temperature data. Daily precipitation and annual precipitation were computed using the hourly precipitation data. Subsequently, the daily maximum and minimum temperature and precipitation were used to estimate extreme climate indices. Yearly volumetric soil water and potential evaporation were calculated using the hourly volumetric soil water data of soil layer 1 (0~7 cm) and potential evaporation data. Drought index data, the daily evapotranspiration deficit index (DEDI), were obtained from a study of Zhang et al. [49]. The DEDI was calculated based on the daily ERA5 product, and defined as the standardized deficit of difference between the actual and potential evapotranspiration. DEDI not only considered the aggravating effect of warming on the severity of a drought but also reflected the degree of the impact of a drought on vegetation and soil moisture in time. More importantly, it was extremely sensitive to an agricultural drought [50]. Thus, compared with other commonly used drought indices, such as the Standardized Precipitation Index (SPI) and Standardized Precipitation Evapotranspiration Index (SPEI), DEDI is more suitable for analyzing the impact of a drought on the cropland area trend under climate warming [51]. The yearly net primary productivity data were derived from the United States Geological Survey (USGS) MOD17A3HGF dataset.

The population data were collected from the LandScan, which is the available and fine-resolution global population data. The annual night-time light index data were retrieved from a study of Li et al. [52]. This annual data were generated by inter-correcting night-time light index data from different satellites using a sigmoid function and derived relationship. It was verified that the corrected night-time light index data showed a good consistency and were superior to other night-time light index data in terms of the temporal consistency. In addition, the data also showed good agreements with the temporal trends of socioeconomic activities.

In order to maintain spatial consistency, all the data in this research, except the population, were resampled to a spatial resolution of 9 km using a bilinear resampling algorithm. For the population data, a population of 9 km × 9 km was counted. Details of the data used in this study can be found in

Table A2. Research usage data details and sources.

Data Types	Period	Spatial Resolution	Temporal Resolution	Data Source
Land use/land cover	1992–2020	300 m	yearly	ESA/CCI viewer (ucl.ac.be, accessed on 13 January 2022)
Maximum and minimum 2 m temperature	1992–2020	0.25°	hourly	ERA5 hourly data on single levels from 1979 to present (copernicus.eu, accessed on 13 January 2022)
Two meter temperature	1992–2020	0.25°	monthly	ERA5 monthly averaged data on single levels from 1979 to present (copernicus.eu, accessed on 13 January 2022)
Precipitation	1992–2020	0.25°	hourly	ERA5 hourly data on single levels from 1979 to present (copernicus.eu, accessed on 13 January 2022)
Potential evaporation	1992–2020	0.1°	hourly	ERA5-Land monthly averaged data from 1950 to present (copernicus.eu, accessed on 13 January 2022)
Volumetric soil water layer 1	1992–2020	0.25°	hourly	ERA5 hourly data on single levels from 1979 to present (copernicus.eu, accessed on 13 January 2022)
Night-time light	1992–2020	30″	yearly	Harmonization of DMSP and VIIRS night-time light data from 1992 to 2020 at the global scale (figshare.com, accessed on 13 January 2022)
Population	2000–2020	30″	yearly	LandScan Datasets|LandScan™ (ornl.gov, accessed on 13 January 2022)
Net primary productivity	2000–2020	500 m	yearly	LP DAAC—MOD17A3HGF (usgs.gov, accessed on 13 January 2022)
Daily evapotranspiration deficit index	1992–2020	0.25°	daily	https://doi.org//10.11922/sciencedb.00906, accessed on 13 January 2022

. In addition, for some vacant time series, replace them with the most recent year; for example, the population data in 1999 were replaced by those in 2000. All these data processing and calculations are based on ArcGIS 10.7 and MATLAB 2022b software tools.

2.3. Methods

2.3.1. Main Research Framework

In this study, impacts of driving factors on the cropland area trend at the county scale were quantified, and the differences in the contribution of driving factors before and after the turning point were compared. The main research framework is illustrated in Figure 2, which incorporated four steps:

(1) Spatial–temporal change analysis. The spatiotemporal dynamics of the cropland area from 1992 to 2020 were estimated by using the trend analysis approach and its time series turning point was detected with the piecewise regression approach.

(2) Driving factor determination. The correlation between the cropland area and drivers using a Pearson correlation analysis was quantified at the county scale, and whether to put initial driving factors into the ridge regression model was judged by considering the correlation coefficient R values and the ratio of the significant area to total area.

(3) Relative contribution of drivers. The filtered factors and cropland area were put into the ridge regression model; the accuracy of the model was tested based on the RMSE, R², and RMSE/mean (Appendix A: Equations (A2)–(A4); Figure A5); and finally the relative contribution was obtained.

(4) Differences in contribution of driving factors before and after the turning point. The turning point obtained in step (1) was taken as a node, and the relative contribution value of the driving factors before and after the turning point and their relative changes were calculated, respectively. Finally, the spatial distribution of the dominant and biggest change driver was visualized at the county scale.

2.3.2. Initial Driver Factor Selection and Driver Factor Determination

Previous studies have shown that the potential driving factors that could affect the cropland area change were grouped into five categories: extreme events, environmental conditions, socioeconomic development, urban sprawl, and ecological construction (

Table A3. The categories of driving factors for cropland area trend.

Categories	Driving Factors
Environmental conditions	Annual average temperature
	Annual precipitation
	Potential evapotranspiration
	Volumetric soil water
	Net primary productivity
Extreme events	Annual frequency of extreme high temperature
	Annual frequency of extreme low temperature
	Annual frequency of drought
	Annual frequency of extreme precipitation
Socioeconomic development	Sum of population
	Annual average night-time light
Urban sprawl	Construction land area
Ecological construction	Ecological land area

) [16,17,19,20].

Environmental conditions are represented from the perspective of meteorology, soil, and the grain production situation using annual average temperature, annual precipitation, potential evapotranspiration (PET), volumetric soil water, and net primary productivity (NPP), respectively [24,43,53]. These factors are closely related to grain growth and development, which in turn affect the change in a cropland area [19,20]. In addition, extreme events could be largely responsible for altering a cropland area due to a dramatic increase in the frequency and intensity of extreme events under climate change [13,23,26,30,54]. Therefore, the climate extremes index (covering an extreme high temperature, an extreme low temperature, extreme precipitation, and a drought) was also contained in our research.

Due to the unavailability of continuous socioeconomic data at the grid scale, predominantly, two types of socioeconomic factors were considered, namely population and satellite-based night-time light index (NTL) data [52]. The areas of construction land and ecological land represent urban sprawl and ecological construction in this study, respectively [15].

To further confirm whether the driving factors in Table A3 can reasonably explain the cropland area trend, before estimating the relative contribution of driving factors on the cropland area, the driving factors that had stronger relative value (R > 0.4) were screened out at first. Meanwhile, the driving factors that had a ratio of the significant area to total area that was less than 20% were also erased, due to the spatial heterogeneity. Their significance was judged with the help of the F-statistic test. The Pearson correlation analysis formula is as follows:

$${\mathrm{R}}_{cp,X_i}=\frac{cov\left(cp,X_i\right)}{{\sigma }_{cp}{\sigma }_{X_i}}, 0<X_i<245$$

(1)

where i is the county of FPEN; cp is the cropland area of i county; X is the driving factors of i county; $cov\left(cp,X_i\right)$ is the covariance of the cropland area and driving factors at the county scale; ${\sigma }_{cp}$ and ${\sigma }_{X_i}$ are the standard deviations of the cropland area and driving factors, respectively; and ${\mathrm{R}}_{cp,X_i}$ is the correlation coefficient of the cropland area and driving factors at the county scale.

2.3.3. Extreme Climate Index Calculation

Four indices were selected, including an extreme high temperature, an extreme low temperature, an extreme drought, and extreme precipitation, to describe the extreme climate as they were closely related to cropland area change/agricultural production in FPEN. These indices were calculated by referring to the definition of climate extremes by the Expert Team on Climate Change Detection and Indices (ETCCDI) [55,56]. In addition, to keep the unit of drought indices consistent with others, the annual frequency of a drought (D20p) was computed based on DEDI using the same method as ETCCDI. The detailed calculation method is shown in

Table A4. Definition of extreme climate and drought indices based on daily data in this study.

Indices	Attributes	Definition	Units
TH90p	Extreme high temperature	Count of days per each year where THij > Tmax90p. THij is the daily maximum temperature on day i in year j. Tmax90p is the 90th percentile centered on i in a five-day window of daily maximum temperature during 1992–2020.	Days
TL10p	Extreme low temperature	Count of days per each year where TLij > Tmin10p. TLij is the daily minimum temperature on day i in year j. Tmin10p is the 10th percentile centered on i in a five-day window of daily minimum temperature during 1992–2020.	Days
R95p	Extreme precipitation	Count of days per each year where Rij > R95p. Rij is the daily precipitation (Rij ≥ 1 mm) on day i in year j. R95p is the 95th percentile during 1992–2020.	Days
D20p	Drought (DEDI)	Count of days per each year where Dij > D20p. Dij is the daily DEDI on day i in year j. D20p is the 20th percentile centered on i in a five-day window of daily DEDI during 1992–2020.	Days

.

2.3.4. Trend Analysis

Linear regression was used to estimate the spatial–temporal trends of annual driving factors and the cropland area [57]. Their significance was judged with the F-statistic test in this study. The trend formula is as follows:

$${\mathrm{s}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}}_i=\frac{n{\sum }_{t=1}^{n}t{x}_i-\left({\sum }_{t=1}^{n}t\right)\left({\sum }_{t=1}^n{x}_i\right)}{n{\sum }_{t=1}^n{t}^2-{\left({\sum }_{t=1}^{n}t\right)}^2},0<\mathrm{t}\le 29$$

(2)

where n is the study coverage time; $x_i$ represents the value of the driving factors/cropland area in year t; and ${\mathrm{s}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}}_i$ is the slope value of this factor.

2.3.5. Turing Point Detection

A piecewise linear regression model was used to identify the turning point of the cropland area from 1992 to 2020 [58], and its significance was judged with the F-statistic test. The piecewise linear regression model is as follows:

$$\mathrm{y}=\left\{\begin{array}{ll}{\beta }_0+{\beta }_{1}t+\epsilon ,& t\le t_0\\ {\beta }_0+{\beta }_{1}t+{\beta }_2\left(t-t_0\right)+\epsilon ,& t>t_0\end{array}\right.$$

(3)

where y is the cropland area; t is the year; $t_0$ is the identified turning point; ${\beta }_0$ is the intercept; ${\beta }_1$ and ${{\beta }_1+\beta }_2$ are the values of trends before and after the turning point, respectively; and ε is the residual errors. The least squares method is used to compute coefficients, such as ${\beta }_0$, ${\beta }_1$, and ${\beta }_2$. In this study, the turning point was determined with multiple iterations [59]. Specifically, firstly, points were limited to between 1996 and 2016, to ensure that the time series before and after the point is greater than or equal to 5 years. Then, multiple fitted linear iterations were used to obtain the point where it had minimized the residual value square sum of both sides. Finally, the F-statistic test was used to judge whether the trend of time series on both sides of the point is significant. If it passes the significance test (p < 0.05), this point is regarded as the turning point of cropland area changes.

2.3.6. Ridge Regression Model

The ridge regression model was designed to solve the problem of multi-collinearity between independent variables [35,60,61]. This study involved nine variables, and the results of the inflation factor (VIF) test showed that there was multi-collinearity (VIF ≥ 10) between driving factors (Figure A1). The ridge regression model is expressed as follows:

$$\widehat{\beta }\left(k\right)={\left(X^{T}X+kI\right)}^{-1}{X}^T{A}_{Cropland}$$

(4)

where $A_{Cropland}$ is the cropland area that is spatial–temporal standardized, $\widehat{\beta }\left(k\right)$ is the regression coefficient, and X is a two-dimensional matrix of spatial–temporal standardized driving factors. The detailed calculation method that is spatial–temporal standardized is shown in the Methods section of Appendix A (Equation (A1)). K is the ridge parameter. The principle of selecting the value of k is as follows: if k = k₀, the $\widehat{\beta }\left(k\right)$ of driving factors tends to be stable, and k₀ will be the ridge parameter. In our research, k is 1.

To account for the long-term trend in the effects of drivers on the cropland area, the ridge regression coefficients were multiplied by the trend of the corresponding drivers in order to characterize the contribution of drivers for cropland area trends. The contribution of the driving factor to the cropland area is expressed as follows:

$${\eta }_{ij}=\widehat{\beta }{X}_{ij\_trend}$$

(5)

The relative contribution is

$${R\eta }_{ij}=\frac{{\eta }_{ij}}{{\sum }_{j=1}^9\left|{\eta }_{ij}\right|}$$

(6)

where $X_{ij\_trend}$ is the spatial–temporal standardization of driving factors’ trends; ${R\eta }_{ij}$ represents relative contribution of driving factors to trends of the cropland area; and the positive and negative of ${\eta }_{ij}$ indicate the positive and negative effects of ${\eta }_{ij}$. ${R\eta }_{ij}$ the larger its value, the greater its impact.

The accuracy of the simulated results was evaluated using three metrics (root-mean-square error (RMSE), coefficient of determination (R²), and RMSE/mean). The detailed calculation method is shown in the Methods section of Appendix A (Equations (A2)–(A4)).

3. Results

3.1. Spatial–Temporal Dynamics of Cropland Area

The cropland area significantly increased at a rate of ~333.5 km²/a in FPEN during 1992–2020 (p < 0.05). The area increased from 19.5 × 10⁴ km² in 1992 to 20.2 × 10⁴ km² in 2020 (Figure 3c). Regionally, the cropland area of the eastern region continued to increase in the entire period with a growth rate of ~357.3 km²/a (p < 0.01, Figure 3d). The cropland area in the central region showed a decreasing trend during 1992−2020, but it was not significant (Figure 3e). The cropland area of the western region showed a significantly decreasing trend (p < 0.05). The entire cropland loss area was 875.2 km² from 1992 to 2020 (Figure 3f). Spatially, the increased cropland area of the entire FPEN focused on the eastern region, and the significantly increased area was 14.13 × 10⁴ km² (p < 0.05), particularly in the eastern part of Inner Mongolia. The decreased cropland area was primarily in the central and western regions, and its significantly increased and decreased trend areas were 16.08 × 10⁴ km² and 22.38 × 10⁴ km², respectively (Figure 3a).

In FPEN, ~38.90% of cropland had a significant turning point (p < 0.01, Figure 3b). The turning point of the cropland area mainly occurred after 2010, accounting for ~24.85% of FPEN (63.88% of entire turning point regions). The pixels of croplands having a turning point before 2000 only accounted for ~1.55% of FPEN. Through an overlay analysis of the cropland area trend and turning point, it was found that a turning point was more prone to occur after 2010, if the cropland area exhibited a significant decreasing trend in the pixel scale. Therefore, it was speculated that there might be large differences in the impact of driving factors on the cropland area trend around 2010.

3.2. Selecting Driving Factors Based on the Ridge Regression Model

This study used the Pearson analysis to quantify the correlations between driving factors and the cropland area in FPEN (Figure 4a–m). The results showed that the mean absolute of the correlation coefficient between all driving factors with the cropland area was greater than 0.4. Among them, the mean absolute of the correlation coefficients of EC and urban sprawl was much higher than other factors (R values of ~0.9 and 0.8, respectively). At the significance level of 0.05, the areas where the cropland area was significantly correlated with socioeconomic development, urban sprawl, and ecological construction all exceeded 50% of FPEN. On the contrary, except for NPP, the areas of significant correlation between environmental conditions and extreme events with the cropland area were all less than 50% of FPEN, of which the annual average temperature, annual precipitation, volumetric soil water, and annual frequency of extreme precipitation were significantly correlated with the cropland area in less than 20% of FPEN (Figure 4c,e,f,h). Due to spatial heterogeneity, the small-range significant correlation cannot clearly explain the reason for the cropland area trend in the entire FPEN. Thus, the annual average temperature, annual precipitation, annual frequency of extreme precipitation, and volumetric soil water were excluded from the follow-up study.

3.3. Factor Attributions of Cropland Area Trends at the County Scale

3.3.1. Relative Contributions of Driving Factors to Cropland Area Trends

This study compared the simulated cropland and satellite-driven cropland area in the entire FPEN and each of its counties. It was found that the simulated cropland area agreed well with the satellite-driven cropland area. Generally, R² is 0.99 and RMSE is 204.3; of which, regions with R² ≥ 0.9 accounted for ~90% of FPEN (Figure A5), indicating that the model was robust and suitable for further investigating the cropland area trend and the driving factor’s contributions.

Table 2. The relative contribution absolute values of the driving factors across the FPEN and three sub-regions during 1992–2020.

Relative Contribution Absolute (%)	TH90p	TL10p	D20p	PET	NPP	Pop	NTL	US	EC
FPEN	1.5	1.3	2.1	2.0	9.1	6.8	3.8	39.3	40.3
Eastern	1.8	1.4	2.3	2.6	10.4	6.1	4.1	37.6	38.0
Central	1.5	1.2	2.1	1.5	8.2	7.3	3.6	40.0	40.6
Western	1.2	1.4	1.8	3.7	9.4	6.3	3.9	39.2	41.7

summarizes the relative contribution absolute of the driving factors across the FPEN and three sub-regions. Generally, EC was the greatest contributor to the cropland area trend from 1992 to 2020 in FPEN, and was followed by urban sprawl. Their relative contribution absolute values were 40.3% and 39.3%, respectively. In contrast, TL10p was the minimal contributor to the cropland area trend. Its relative contribution absolute was 1.3%. Regionally, the impact of TH90p, D20p, NPP, and NTL on the cropland area trend in the eastern region was slightly larger than that in the central and western regions. Their relative contribution absolute values were 1.8%, 2.3%, 10.4%, and 4.1%, respectively. The influences of population and urban sprawl on the cropland area trend in the central region were slightly higher than in the eastern and western regions. Their relative contribution absolute values were 7.3% and 40.0%. Furthermore, it was also discovered that PET and EC had relatively greater impacts on the cropland area trend in the western region. These findings illustrate that cropland area trends in different regions performed the spatial heterogeneity of driving factors.

Spatially, the impacts of TH90p, PET, NTL, and EC on the cropland area trend were mainly positive, and accounted for 39.5%, 58.8%, 42.3%, and 49.8% of FPEN, respectively (Figure 5a,d,g,i, p < 0.05). On the contrary, the impacts of TL10p, D20p, NPP, population, and urban sprawl on the cropland area trends were mainly negative, and accounted for 51.5%, 41.4%, 39.0%, 47.8%, and 73.9% of FPEN, respectively (Figure 5b,c,e,f,h, p < 0.05).

3.3.2. The Relative Contribution Changes of Driving Factors

According to Section 3.1, it was found that the turning point years of the cropland area in FPEN were mainly concentrated after 2010. To spatially calculate and compare the relative contribution absolute of driver factors, the year of 2010 was thus taken as the node. The entire period was divided into two parts, i.e., the period of 1992–2010 and 2010–2020 (Figure 6, Figure A7 and Figure A8). The relative contribution absolute values of TH90p, TL10p, D20p, PET, population, NTL, and urban sprawl from 2010–2020 were higher than those from 1992–2010. Among them, the contribution value of TH90p had the highest increase (~99.4%), mainly from the increased positive contribution (Figure A6). Contrastingly, the relative contribution absolute of NPP and EC was decreased by ~33.8% and ~10.3%, respectively, mainly due to the decreased positive contribution (Figure A6).

Spatially, the relative contribution absolute values of TH90p, TL10p, and urban sprawl evidently increased in more than half of the study area (Figure 6a,b,h). Among them, the increased contribution of TH90p could be found in the eastern region, while that of TL10p and urban sprawl was mainly in the eastern and central regions. On the contrary, the relative contribution absolute of D20p, PET, NPP, population, NTL, and EC decreased significantly in most regions from 1992–2010 to 2010–2020 (Figure 6c–g,i). Among them, NPP involved the largest decreased area, accounting for ~70.2%, and was followed by EC (62.0%).

3.3.3. Dominance Drivers and Corresponding Changes at the County Scale

The relative contribution absolute values were used to determine the dominant factor at the county scale. It was found that the dominance areas of PET, TH90p, TL10p, population, and urban sprawl were increasing over time, but other factors were decreasing (Figure 7a–c). In the whole study period (1992–2020), the dominance of environment conditions, socioeconomic development, urban sprawl, and ecological construction covered 5.4%, 2.3%, 39.5%, and 52.7% of FPEN, respectively (Figure 7a). In 1992–2010, the dominated area of ecological construction was the largest, and accounted for 63.0% of FPEN (Figure 7b). It was followed by urban sprawl, which accounted for 30.0% of FPEN (Figure 7b). Environmental and socioeconomic factors only accounted for 4.9% and 1.9% of FPEN, respectively (Figure 7b). The smallest was extreme event factors, only covering 0.2% of FPEN (Figure 7b). In 2010–2020, the dominated areas of ecological construction and environmental factors (mainly due to the decreased areas dominated by NPP) decreased by ~16.1% and ~88.2%, respectively (Figure 7c). On the contrary, the dominated areas of socioeconomic factors and urban sprawl increased by ~35.2% and 41.9%, respectively (Figure 7c). Furthermore, it was found that the area dominated by extreme event factors in 2010–2020 was about six times greater than the area dominating in 1992–2010, mainly owing to the increased areas dominated by TH90p and TL10p (Figure 7c).

To further examine changes in driver contribution, the relative changes of driver contribution in the two periods were used to identify the factors with the largest change at the county scale (Figure 7d). The results showed that counties firstly leading with extreme events, socioeconomics, urban sprawl, environmental conditions, and ecological construction covered 43.0%, 16.9%, 15.6%, 13.3%, and 10.7% of FPEN, respectively. Individually, the area of TL10p with the biggest change was higher than other factors, accounting for 22.2% of FPEN, mainly located in the eastern and central regions.

4. Discussion

4.1. Analyzing the Spatiotemporal Dynamics of Cropland

By using land use/land cover data, linear regression, and the turning point test, this study investigated the spatial and temporal dynamic characteristics of the cropland area in FPEN from 1992 to 2020. And it was found that the area of croplands in the FPEN increased significantly during the study period, which is consistent with previous studies [31]. About 44.8% of the pixels in the FPEN increased significantly, mainly in the east, while the negative changes were concentrated in the central and western regions (Figure 3a). This phenomenon is understandable. The east of the FPEN is situated within the Northeast China Plain, characterized by fertile soil conditions and flat terrain, providing the material foundation for the expansion of cropland [42]. Additionally, a series of policies such as agricultural subsidies and tax reductions are also important driving factors for the expansion of cultivated land [43]. The cropland area in the FPEN shows a sharp downward trend in 2018–2020 (Figure 3c,e,f), which could largely be attributed to the 2017 Guidance Opinions on Agricultural Structural Adjustment in the FPEN, issued by the Ministry of Agriculture and Rural Affairs of the People’s Republic of China [62]. The general idea of this policy is to “build ecological agricultural and pastoral area”, and a policy has been put in place for the development of artificial turf cultivation based on existing cropland. This study also calculated the annual cropland transfer matrix, which further confirms this reason (Figure 8). Based on piecewise linear regression tests, the turning point in the FPEN cropland area from 1992 to 2020 was examined. We found that the turning point in the cropland area occurred mainly around 2010, suggesting that changes in certain drivers around 2010 may have led to an abrupt change in cropland area change. After analyzing the changes in the contribution of the drivers between 1992–2010 and 2010–2020, we can conclude that changes in extreme temperatures (high and low), potential evapotranspiration, the night-time light index, and urban sprawl may be responsible for the abrupt changes in the cropland area (Figure 6 and Figure A6).

4.2. Understanding Impacts of Driving Factors on Cropland Trends

This study quantified the impacts of extreme events, environment conditions, socioeconomic development, urban sprawl, and ecological construction on a cropland area trend. Overall, our research clarifies that the cropland area trend was dominated by ecological construction expansion and urban sprawl. It is consistent with previous studies [9,14,63,64]. Consistent with the hypothesis of this study, there was a large difference in the contribution of the drivers to the change in the cropland area before and after the turning point. And it was found that the relative contribution of all factors, except NPP and EC, increased. Among them, the relative contribution of TH90p increased the most (~99.4%). However, interestingly, it was found that while the impact of EC on the cropland area trend decreased (principally due to decreased positive contribution), EC remained the dominant factor on the cropland area trend in most of the study area (dominant area exceeded 50% of FPEN). There are two possible reasons. First, ecological restoration programs implemented in FPEN are largely successful, making the most croplands converted to ecological land (forests and grasslands) during the past two decades [9,64,65]. Second, less economic development makes the amount of croplands converted to construction land lower than the ecological land in FPEN [9,65,66]. Other factors, such as the vulnerable ecosystem and water scarcity, may have hindered further expansion of cropland into forests and grasslands, leading to a decrease in the contribution of EC [67,68]. Additionally, rapid urbanization and industrialization resulting in more of an agricultural labor force far away from croplands may also be an important factor contributing to the decline in the influence of EC [8]. Furthermore, our results also showed that the dominant area of urban sprawl increased while EC decreased.

Although climate change and extreme events have been regarded as important drivers for cropland area changes under global warming, they are still rarely mentioned in previous studies, particularly in exploring the driving mechanism of cropland area change [5,15,69]. Therefore, the contribution of indices was estimated related to climatic change and extreme events on the cropland area. It was found that the cropland area is highly correlated to all the climate factors involved in this paper (R > 0.4, p < 0.05). However, areas that were significantly correlated with basic climate variables required by vegetation growth and development, such as temperature and precipitation, were very small (Figure 4e,f). In contrast, the area of extreme events was much larger (Figure 4a,b,d). This indicates that extreme events largely and broadly threaten agricultural activities in FPEN. Moreover, it was found that regardless of the positive or negative contributions of TH90p and TL10p on the cropland area trend, their values increased in most regions of FPEN (Figure 7a,b). This indicates that extreme temperature events are more closely related to the cropland area trend in the context of climatic warming [30]. Unlike Zaveri, Russ, and Damania [13], we found that a drought remains one of the vital factors that limits the cropland expansion. This situation illustrated that although the coverage and intensity of irrigation facilities have increased in the past few years, they had little effect on offsetting the negative effects of a drought on croplands [30]. This may be caused by an increased trend of D20p (trend of D20p from 0.02 days/a in 1992–2010 to 0.03 days/a in 2010–2020; Figure A2c). An extreme precipitation event is an exception, and the area of significant correlation between annual extreme precipitation and the cropland area in FPEN is small (Figure 4c, p < 0.05), possibly due to few areas of a significant trend of annual extreme precipitation in the spatial extent (significant trend of annual extreme precipitation accounted for 6.3% of FPEN; Figure 4a,c, p < 0.05). Overall, these results directly demonstrate the impacts of extreme events, especially extreme temperature, on the cropland area trend compared to average climate change in FPEN.

4.3. Policy Recommendations

Although the existing cropland protection policies have a positive impact on the stability or increase in the cropland area, there will be great uncertainties in the future [3,70]. However, the increasing gradually negative contributions of extreme events, environmental conditions, economic development, and policies’ factors could not be neglected. According to the results, this study provides the following four aspect suggestions: (1) Strengthen intensive land utilization. The occupation of arable land for urban use is one of the main reasons for the decline in arable land. Therefore, improve the requirement of cropland protection policies, and strictly limit the conversion of high-quality cropland to other uses [70]. (2) Vigorously promote sustainable agricultural intensification, reduce pesticide and fertilizer use, and adopt intercropping/fallow forms of farming [71]. It could help increase food production while avoiding the ecological environment degradation of FPEN due to intensive intensification. (3) Increase financial investment related to agriculture, implement targeted agricultural payment policies, and attract large, medium, and small agricultural enterprises to participate in local agricultural development [10]. At the same time, it is suggested for land managers to appropriately raise prices of agricultural products to ensure farmers’ income, and promote for motivated farmers to engage in agricultural activities and appropriately maintain or increase cropland areas in FPEN in the future [19]. (4) Accelerate development of agricultural science and technology, such as promoting climate-smart agriculture, better coping with the impact of climate change and extreme events on food production, and reducing the risk of cropland loss caused by unfavorable climatic conditions [72].

Furthermore, considering the different climatic and environmental conditions in FPEN, the above recommendations should be precisely implemented according to location conditions [19]. For example, the eastern region has abundant black soil resources and flat terrain. Intensive agriculture should be vigorously developed, increasing financial investment, and appropriately increasing the area of cropland. On the contrary, in the central and western regions with a relatively vulnerable ecological environment, drought-resistant, heat-resistant, and cold-resistant crops should be promoted under the premise of ensuring the existing croplands, while avoiding large-scale expansion of cropland.

4.4. Limitations of This Study

In this study, the only consideration was the effect of the frequency of extreme events on the cropland area trend. However, the cropland area trend is also affected by other features of extreme events, such as the duration, intensity, and areal extent. Additionally, lag effects between driving factors and the cropland area trend were not considered in this study. More importantly, although there was an attempt to explore the impacts of drivers on the cropland area trend from as many perspectives as possible in this study, all relevant factors still cannot be taken into account. In addition, the interactions among drivers were not considered in this study. Therefore, addressing the above issues will be focused on in future work.

5. Conclusions

This study investigated the spatiotemporal dynamics of a cropland area and its driving factors during 1992–2020, and quantified the impacts of these drivers on the cropland area trend in three periods (1992–2020, 1992–2010, and 2010–2020). It was found that cropland changes showed spatial heterogeneity. Increased cropland areas were mainly located in the eastern region, while decreased cropland areas were focused in the central and western regions. Yet, the total cropland area showed a significant increasing trend with 333.5 km²/a (p < 0.05) in FPEN during 1992–2020. The results of ridge regression analyses presented that ecological construction and urban sprawl dominated cropland area change in FPEN. Ecological construction was always the largest contributor, even though relative contributions of urban sprawl increased by 33.7% during 2010–2020. Also, impacts of socioeconomic factors on the cropland area trend increased. Notably, extreme event effects on the cropland area trend evidently increased, and TH90p was the largest—~99.4%—largely due to an increased frequency of extreme weather events under climatic warming. Importantly, the dominant area of extreme events in 2010–2020 increased about six times compared to that in 1992–2010. TH90p and TL10p are the first two dominant extreme climatic factors. Therefore, more caution should be paid to increasing impacts of extreme events on croplands by policy and planning makers.