EN中文

Journal of Geographical Sciences >

2024 , Vol. 34 >Issue 5: 921 - 941

DOI: https://doi.org/10.1007/s11442-024-2233-z

Research Articles

A 1000-year history of cropland cover change along the middle and lower reaches of the Yellow River in China

  • YANG Fan , 1 ,
  • ZHANG Hang 1 ,
  • HE Fanneng , 2, * ,
  • WANG Yafei 2 ,
  • ZHOU Shengnan 2 ,
  • DONG Guanpeng 1
Expand
  • 1. Key Research Institute of Yellow River Civilization and Sustainable Development, Henan University, Kaifeng 475001, Henan, China
  • 2. Key Laboratory of Land Surface Pattern and Simulation, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
*He Fanneng, Professor, specialized in historical geography and environmental change. E-mail:

Yang Fan (1991-), PhD and Associate Professor, specialized in long-term land use/cover change and their environmental effects. E-mail:

Received date: 2023-08-20

  Accepted date: 2024-03-06

  Online published: 2024-05-31

Supported by

National Natural Science Foundation of China(42201263)

National Key Research and Development Program of China on Global Change(2017YFA0603304)

Fold

Abstract

Landscape in the middle and lower reaches of the Yellow River in China has undergone significant changes for thousands of years due to agricultural expansion. Lack of reliable long-term and high-resolution historical cropland data has limited our ability in understanding and quantifying human impacts on regional climate change, carbon and water cycles. In this study, we used a data-driven modeling framework that combined multiple sources of data (historical provincial cropland area, historical coastlines, and satellite data-based maximum cropland extent) with a new gridding allocation model for croplands distribution to reconstruct a historical cropland dataset for the middle and lower reaches of the Yellow River at a 10-km resolution for 58 time points ranging from the period 1000 to 1999. The cropland area in the study area increased by 2.3 times from 21.87 million ha in 1000 to 50.64 million ha in 1999. Before 1393, the area of cropland increased slowly and was primarily concentrated in the Weihe and Fenhe plains. From 1393‒1820, the area of cropland increased rapidly, particularly on the North China Plain. Since 1820, cropland cover has tended to become saturated. Our newly reconstructed results agreed well with remotely sensed data as well as historical document-based facts regarding cropland distribution.

Key words： land use/cover change; gridding allocation; cropland cover; last millennium

Cite this article

YANG Fan , ZHANG Hang , HE Fanneng , WANG Yafei , ZHOU Shengnan , DONG Guanpeng . A 1000-year history of cropland cover change along the middle and lower reaches of the Yellow River in China[J]. Journal of Geographical Sciences, 2024 , 34(5) : 921 -941 . DOI: 10.1007/s11442-024-2233-z

1 Introduction

Over thousands of years, human land use and land cover change (LUCC) has significantly reshaped the landscape of the Earth’s terrestrial surface (Foley et al., 2005; Ellis et al., 2013; Stephens et al., 2019; Fluet-Chouinard et al., 2023) and profoundly influenced climate change through biogeophysical and biogeochemical processes (Scott et al., 2018; Mendelsohn and Sohngen, 2019). Therefore, long-term LUCC represents a key driver of past climate change at regional and global scales. Reconstructing historical LUCC provides primary data for modeling climate change (Vittorio et al., 2018; Nikiel and Eltahir, 2019), estimating carbon emissions (Houghton et al., 2012; Friedlingstein et al., 2022; Yang et al., 2023a), and assessing the ecological impacts of human activities (Mahbub et al., 2019; Ellis, 2021).
Croplands are an important land resource that serves as both the foundation of human survival and a primary manifestation of the LUCC (Fang et al., 2021; He et al., 2023). Changes in cropland cover represent a critical indicator for characterizing human activity intensity and provide insight into the relationship between environmental change and social rise and decline (Buntgen et al., 2011; Douglas et al., 2015; Wu et al., 2020b). In recent decades, substantial progress has been made in reconstructing historical cropland cover, including quantitative and spatial information, at both global (Ramankutty and Foley, 1999; Pongratz et al., 2008; Kaplan et al., 2011; Klein Goldewijk et al., 2017; Hurtt et al., 2020; Cao et al., 2021) and regional scales (Ye et al., 2009; Leite et al., 2012; Zumkehr and Campbell, 2013; Tian et al., 2014; Fuchs et al., 2015; Jin et al., 2015; Yang et al., 2015; Zhao et al., 2020; Guo et al., 2021; Wei et al., 2021; Yang et al., 2021; Li et al., 2023).
Gridding reconstruction, which includes spatial allocation and evolution models, pertains to the allocation of cropland areas to grids. Spatial allocation models, as opposed to spatial evolution models, are characterized by fewer parameters and ease of data acquisition and have been widely used in large-scale cropland cover reconstruction (He et al., 2023). Previous studies usually assumed that the contribution shares of various factors (such as altitude, slope, and climate) to the distribution of cropland cover are the same, and then adopted an allocation method with equal weight factors to build spatial allocation models (Li et al., 2016; He et al., 2023). However, considerable differences remain between factors that are not represented in existing spatial evolution models.
The middle and lower reaches of the Yellow River are important areas for the origin and development of dryland agriculture in Northern China. Throughout history, large-scale land reclamation activities have caused devastating damage to natural vegetation, including forests and grasslands, and have accelerated the deterioration of the ecological environment (Tan, 1962; Shi, 1985; Zheng et al., 2020). Over the past millennium, soil and water losses in the middle reaches of the Yellow River have intensified, and sediment runoff has increased rapidly (Zhao et al., 2013; Wu et al., 2020a). The riverbed in the lower reaches of the Yellow River rose rapidly, and the flooding intensity and frequency in this area have both increased significantly (Chen et al., 2012; Zhang and Fang, 2017). Frequent diversion of the Yellow River has resulted in the destruction and abandonment of numerous lakes on the North China Plain (Zhao, 1996). Conversely, records have shown that the continuous deterioration of the ecological environment seriously interfered with the development of the social economy in ancient China, leading to the separation of political and economic centers (Zheng et al., 2020). Consequently, the current ecological and environmental challenges in the middle and lower reaches of the Yellow River represent a continuation of historical problems, with land reclamation-dominated human activities playing a pivotal role in exacerbating these issues.
Reconstructing cropland cover for the middle and lower reaches of the Yellow River over the past millennium could provide valuable scientific data to: (1) reveal the ecological effects of past human activities, climate change, and carbon emissions; (2) deepen our understanding of the current ecological and environmental problems; (3) implement the major national strategy for ecological conservation and high-quality development in the Yellow River Basin; and (4) improve the Chinese historical LUCC dataset. Hence, this study used a data-driven modeling framework to reconstruct a historical cropland dataset for the middle and lower reaches of the Yellow River. Firstly, we utilized the Shapley value method to measure the contributions of various factors to the cropland cover distribution. Subsequently, a new evaluation model for land suitability for cultivation (LSC) was developed using unequal weight factors. Based on the LSC value, the historical atlas of China-obtained coastline data and the satellite data-based maximum extent of cropland cover were introduced to build a gridding allocation model for provincial croplands. The provincial cropland areas in the middle and lower reaches of the Yellow River over the past millennium were allocated to grids with a 10-km resolution.

2 Data and methods

2.1 Study area

The middle and lower reaches of the Yellow River span the second and third tiers of China’s topography, connecting the Loess Plateau and North China Plain (Figure 1a). This region exhibits vast differences in geographical units, landforms, geology, climate, and resource endowments. The Loess Plateau, located in the middle reaches of the Yellow River, is the largest ecologically fragile region in China and is among the regions in the world with the most severe soil and water loss, which threatens the ecological security of the lower reaches of the Yellow River. Owing to the relatively low elevation of the Yellow River in the lower reaches, a significant accumulation of sediment was carried by the river, resulting in a suspended river that poses a major flood risk to towns along its banks.
Figure 1 Study area (a. the Yellow River Basin and the scope of historical influence in the lower reaches of the Yellow River; b. provincial units)
The scope of this study encompassed a wider area than the current middle and lower reaches of the Yellow River. It included not only the current middle and lower reaches, but also the regions that were affected by the swinging of the Yellow River over the past millennium (Figure 1a). Moreover, the estimates and statistics of historical cropland areas were primarily based on administrative districts. Therefore, we chose the following regions as basic units for gridding allocation of historical cropland cover: Gan-Ning (Gansu and Ningxia), Shaanxi, Shanxi, Henan, and Shandong (currently located in the middle and lower reaches of the Yellow River), Jing-Jin-Ji (Beijing, Tianjin, and Hebei), Anhui, and Hu-Ning (Shanghai and Jiangsu), which were part of the historical area of influence of the Yellow River (Figure 1b).

2.2 Data sources

The data sources primarily comprise historical archives-based data, encompassing historical provincial cropland area, historical coastlines, as well as remote sensing-based data, including the 1980-2010 LUCC products, Digital Elevation Model (DEM), and climate potential productivity (CPP), and data generated by these sources.

2.2.1 Historical provincial cropland area data

The data used in this study on provincial cropland areas since the Northern Song Dynasty were derived from literature published in recent years (Table 1). To reconstruct the provincial cropland areas of China, scholars have developed methods that enabled them to match data based on Chinese historical tax records for cropland areas and relevant historical documents (Ge et al., 2004; Li et al., 2016; He et al., 2017; Li et al., 2018a; 2018b; 2020). To solve the issue of changes in historical borders and administrative divisions, the provincial cropland areas of present-day China were created by unifying time slices and spatial ranges (He et al., 2023). A total of 58 time point-cropland data were collected over the past millennium. The temporal resolution of the data was 100 years from the Northern Song Dynasty (960-1127) to the Ming Dynasty (1368-1644), 1-39 years from the Qing Dynasty (1636-1911) to the Republic of China (1912-1949), and 1-9 years since the founding of the People’s Republic of China.
Table 1 Data sources for historical cropland cover in the middle and lower reaches of the Yellow River (DEM: digital elevation model; CPP: climate potential productivity)
Data variables Temporal coverage Spatial resolution Data type Data source/ Reference
Historical cropland Song, Liao, and Jin dynasties
(1000, 1066, 1078, 1162, 1215)		 Provincial level Reconstructed values Published literature He et al. (2017);
Li et al. (2018a)
Yuan Dynasty (1290) Li et al. (2018b)
Ming Dynasty (1393, 1583, 1620) Li et al. (2020)
Qing Dynasty to the present (1661-1999) (49 time points) Ge et al. (2004);
Li et al. (2016)
Historical coastlines Song Dynasty (1142) Reconstructed vector lines Historical Atlas of
China		 Tan (1982)
Yuan Dynasty (1280)
Ming Dynasty (1433)
Qing Dynasty (1820)
Remotely sensed land use data 1980, 1990, 2000, 2010 1 km Grid Resources and Environmental Sciences Data Center of the Chinese Academy of Sciences, http://www.resdc.cn/ (last access: 10 February 2022)
DEM 2000 90 m Grid Resources and Environmental Sciences Data Center of the Chinese Academy of Sciences, http://www.resdc.cn/ (last access: 10 August 2022)
CPP 1951-1980 1 km Grid Data Sharing Infrastructure of Earth System Science, http://www.geodata.cn/ (last access: 10 August 2022)

2.2.2 Historical coastline data

The coastlines of each dynasty were sourced from the Historical Atlas of China (Tan, 1982) (Figure 2 and Table 1). Considerable differences exist between the historical and present Chinese coastlines owing to climate change. Specifically, since the Song Dynasty (960-1279), the coastlines along the western coast of Bohai Bay (Figure 2a), northern coast of Jiangsu, and the Yangtze Estuary (Figure 2b) have been higher than their current counterparts. If these changes in coastlines are not considered and present-day coastlines are used as the baseline for gridding allocation, the provincial cropland area of each dynasty will be inaccurately allocated to the sea. Therefore, the present study introduced historical coastline data as a limiting factor for the gridding allocation of croplands.
Figure 2 Historical coastline over the past millennium (a. Coastlines in the western coast of Bohai Bay; b. northern coast of Jiangsu, and the Yangtze Estuary)

2.2.3 Maximum cropland allocation extent

The maximum extent of cropland allocation represents an alternative significant limiting factor for gridding allocation of cropland. Historically, the cropland area in China has shown an increasing trend, and in recent decades, there has been a rapid increase in construction land, mainly at the expense of cropland (He et al., 2023). Therefore, most historical cropland cover distributions in China did not exceed the extent of modern croplands and construction land. This is particularly true in the middle and lower reaches of the Yellow River, which are the traditional agricultural areas in China. Based on this finding, we merged the remotely sensed cropland and construction land data for the years 1980, 1990, 2000, and 2010 to determine the maximum extent of cropland allocation in the middle and lower reaches of the Yellow River (Figure 3). Remotely-sensed land use data were obtained from the Resources and Environmental Sciences Data Center of the Chinese Academy of Sciences (http://www.resdc.cn) (Table 1).
Figure 3 Remote sensing land use data-derived provincial maximum cropland allocation extent in the middle and lower reaches of the Yellow River

2.2.4 Other basic data

This study employed several parameters, namely altitude, slope, degree of relief (DR), and CPP, to assess the LSC. Altitude data were sourced from a DEM with a spatial resolution of 90 m × 90 m, which was obtained from the Resources and Environmental Sciences Data Center of the Chinese Academy of Sciences (http://www.resdc.cn) (Table 1). Slope and DR were calculated using the DEM data. Data on 1-km CPP were acquired from the Data Sharing Infrastructure of Earth System Science (http://www.geodata.cn) (Table 1).

2.3 Methods

Existing historical cropland gridding allocation methods typically adopt a simplified assumption that all factors contribute equally to the cropland distribution, and thus use spatial allocation models with equal weight factors. To address this issue, we proposed a data-driven cropland gridding allocation model with unequal weight factors (Figure 4). The model consists of the following six steps. (1) The Shapley value method was used to quantify the contributions of four factors (altitude, slope, CPP, and DR) to the modern cropland distribution. (2) Based on the principle of uniformity, we assumed that the statistical relationships between different factors and the distribution of cropland cover obtained in step (1) were also applicable to the historical period. Subsequently, an evaluation model for the LSC with unequal weight factors was devised. (3) Historical coastline data and the maximum extent of cropland allocation were introduced as limiting factors for cropland gridding allocation. (4) A gridding allocation model for provincial croplands was established by combining steps (2) and (3). (5) The provincial cropland areas in the middle and lower reaches of the Yellow River were allocated to grids with a 10-km resolution using the developed model. (6) The allocation method was evaluated and the reconstruction results were analyzed.
Figure 4 Scheme for gridding reconstruction of historical cropland cover in the middle and lower reaches of the Yellow River

2.3.1 Quantifying the contribution of different factors to the cropland distribution

The Shapley value method used in the present study was introduced to quantify the influences of altitude, slope, CPP, and DR on the distribution of cropland cover (Zhang et al., 2023; Nandlall and Millard, 2020). The Shapley value possesses three desirable properties for a fair and simple factor-contribution metric. Firstly, it is nondiscriminatory, meaning that the labels or orders of the factors do not affect the Shapley values. Factors that contributed equally to the distribution of cropland cover had identical Shapley values. Secondly, the Shapley value is marginal, assigning higher values to factors that contribute more to the objective. Finally, the Shapley value is efficient, with the sum of all Shapley values equal to the score obtained when all factors participate.
Four factors were selected: altitude, slope, CPP, and DR, and remotely sensed cropland cover data were obtained for four time slices, especially 1980, 1990, 2000, and 2010. The basic geographical data were then resampled to a target spatial resolution of 10 km. The contribution of each factor to the distribution of cropland cover was quantified using the Shapley value method (Table 2).
Table 2 Flow of Shapley value calculation.
Item Definition Formula
Step 1: Input profiles (T) T, the universe of input profiles, represents the set of all possible input profiles. The universe with Nv factors will have 2Nv input profiles. For example, for a three-factor model, the universe is:
$\text{T}=\left\{ \left\{ 0,0,0 \right\};\left\{ 0,0,1 \right\};\left\{ 0,1,0 \right\};\left\{ 0,1,1 \right\} \right.;$
$\left. \left\{ 1,0,0 \right\};\left\{ 1,0,1 \right\};\left\{ 1,1,0 \right\};\left\{ 1,1,1 \right\} \right\}$
Step 2: Shapley set (Q) Q is defined the set of all input profiles in the universe of the model for which a certain factors is marked as a nonparticipant. For example, in a three-factor model, the Shapley set for the third factor would be:
$\begin{aligned}&\mathrm{m}\Big(\mathrm{v}\big(\vec{\mathrm{x}}\big),\mathrm{v}_{\mathrm{p}}\Big)=\mathrm{R}^{2}\Big(\mathrm{v}\big(\vec{\mathrm{x}}\big)+\mathrm{v}_{\mathrm{p}}\Big)-\mathrm{R}^{2}\Big(\mathrm{v}\big(\vec{\mathrm{x}}\big)\Big);\\&\mathrm{p\in1,2,...,N_{v}}\end{aligned}$
Step 3: Marginal
contribution (m)		 When vp is added as a participant, the added value of R-squared is defined as the marginal contribution. $\mathrm{m}\left(\mathrm{v}(\overrightarrow{\mathrm{x}}), \mathrm{v}_{\mathrm{p}}\right)=\mathrm{R}^{2}\left(\mathrm{v}(\overrightarrow{\mathrm{x}})+\mathrm{v}_{\mathrm{p}}\right)-\mathrm{R}^{2}(\mathrm{v}(\overrightarrow{\mathrm{x}})) ;$
$\text{p}\in 1,2,...,{{\text{N}}_{\text{v}}}$
$\mathrm{v}(\overrightarrow{\mathrm{x}})$denotes the variable set of the input profile.
Step 4: Shapley value (S) $\mathrm{S\big(v_p\big)=\sum_{\vec{x}\in Q(T,p)}\frac{\big(\big|\vec{x}\big|\big)!\big(N_v-\big|\vec{x}\big|-1\big)!}{N_v !}m\big(v\big(\vec{x}\big),v_p\big)}$
Step 5: Relative important of factors (S%) $S_{\%}\left(v_{p}\right)=\frac{S\left(v_{p}\right)}{\sum S(v)}$
The contribution of each factor to the distribution of cropland cover is presented in Table 3. These results indicated differences in the influence of each factor on the distribution of cropland cover at the provincial scale. This finding suggested that the LSC evaluation model, which employs equal weight factors as devised in a previous study, may deviate from actual observations.
Table 3 Contribution shares of factors on the distribution of cropland cover
Jing-Jin-Ji Shaanxi Henan Shanxi Shandong Anhui Hu-Ning Gan-Ning
Altitude 0.21 0.45 0.39 0.27 0.27 0.33 0.13 0.25
CPP 0.11 0.17 0.08 0.05 0.06 0.08 0.15 0.59
DR 0.41 0.24 0.33 0.42 0.39 0.30 0.51 0.11
Slope 0.27 0.14 0.20 0.26 0.28 0.30 0.20 0.05

CPP: Climatic potential productivity; DR: Degree of relief

2.3.2 Assessing the land suitability for cultivation

In addition to altitude, slope, and CPP, DR was introduced in the present study to devise an LSC evaluation model and proved to be a critical factor affecting the distribution of cropland cover (Table 2). Specifically, contribution shares of DR in Jing-Jin-Ji, Shanxi, Shandong, and Hu-Ning were higher than those of the remaining factors. However, previous studies have paid little attention to the use of DR.
Based on the selected factors and their respective contribution shares, an LSC evaluation model with unequal weight factors was developed and calculated as follows:
$\begin{align} & \text{ln}{{L}_{suit}}\left( i,j \right)=\alpha \times \text{ln}{{V}_{norm\_alti}}\left( i,j \right)+\beta \times \text{ln}{{V}_{norm\_slop}}\left( i,j \right)+ \\ & \text{ }\gamma \times \text{ln}{{V}_{norm\_DR}}\left( i,j \right)+\delta \times \text{ln}{{V}_{norm\_CPP}}\left( i,j \right) \end{align}$
where Lsuit (i, j) denotes the LSC value of grid j of province i; the weights of altitude, slope, DR, and CPP in province i, are denoted by α, β, γ, and δ respectively, and were obtained from Table 2; and Vnorm (i, j) represents the normalized factor value of grid j in province i (note: the specific methods used for normalization are described in Table S1).
Table S1 Normalization formulas for altitude, CPP, DR, and slope
Factor Relationship with cropland distribution Formula
Altitude Negative ${{V}_{norm\_alti}}\left( i,j \right)=\frac{{{V}_{alti}}\left( i,max \right)-{{V}_{alti}}\left( i,j \right)}{{{V}_{alti}}\left( i,max \right)}$
CPP Positive ${{V}_{norm\_CPP}}\left( i,j \right)=\frac{{{V}_{CPP}}\left( i,j \right)}{{{V}_{CPP}}\left( i,max \right)}$
DR Negative ${{V}_{norm\_DR}}\left( i,j \right)=\frac{{{V}_{DR}}\left( i,max \right)-{{V}_{DR}}\left( i,j \right)}{{{V}_{DR}}\left( i,max \right)}$
Slope Negative ${{V}_{norm\_slop}}\left( i,j \right)=\frac{{{V}_{slop}}\left( i,max \right)-{{V}_{slop}}\left( i,j \right)}{{{V}_{slop}}\left( i,max \right)}$

Note: positive characterizes that the larger the factor value, the more cropland distribution; negative characterizes that the larger the factor value, the less cropland distribution. Vnorm(i,j) represents the factor value of grid j in province i after the normalization; V(i,max) denotes the maximum factor value in province i; V(i,j) refers to the factor value of grid j in province i.

2.3.3 Devising gridding allocation model for provincial cropland

Historical coastlines and the maximum extent of cropland allocation were regarded as limiting factors. Subsequently, based on the LSC values calculated as described in Section 2.3.2, a gridding allocation model for croplands was developed using Eq. (2). Using this model, we generated a 10-km cropland cover dataset for the middle and lower reaches of the Yellow River over the past millennium.
$Cro{{p}_{grid}}\left( i,j,t \right)=\rho \times \varphi \times \frac{{{L}_{suit}}\left( i,j \right)}{\mathop{\sum }^{}{{L}_{suit}}\left( i,j \right)}\times Cro{{p}_{prov}}\left( i,t \right)$
where Cropgrid (i, j, t) represents the cropland area in grid j of province i in year t, and Cropprov (i, t) denotes the provincial cropland area of province i in year t. ρ and φ refer to the indexes of the maximum cropland allocation extent and historical coastlines, respectively. If grid j in province i exceeds the boundary of the maximum cropland allocation extent or historical coastlines, then the values of ρ and φ are assigned to zero, respectively; otherwise, their values are one.

2.3.4 Reliability assessment of the results and cropland datasets comparison

Evaluating the reliability of our reconstruction results is challenging because of the lack of historical data related to the actual grid-scale cropland distribution (Li et al., 2023). Thus, we conducted an evaluation of the reconstruction results and compared them with existing global datasets to increase the confidence of the newly developed cropland dataset.
An indirect evaluation method was employed in the present study. Specifically, the 1980 provincial cropland areas based on a remotely sensed land use dataset were allocated to a grid scale using our proposed method. Subsequently, the allocation results were compared to the corresponding period-cropland grid data based on remote sensing images to verify the feasibility of our allocation method (Eq. [3]).
$Diffe{{r}_{grid}}\left( i,j,1980 \right)=Cro{{p}_{grid\_reconstruct}}\left( i,j,1980 \right)-Cro{{p}_{grid\_RS}}\left( i,j,\text{1980} \right)$
where Cropgrid_reconstruct(i, j, 1980) represents the 1980 allocation cropland area in grid j of province i; Cropgrid_RS (i, j, 1980) denotes the remotely sensed cropland area in grid j of province i in 1980; and Differgrid (i, j, 1980) refers to the absolute difference in grid j of province i in 1980.
Additionally, cropland cover data covering the study area and time period are available in the following global land use datasets: the history of the global environment database-HYDE (Klein Goldewijk et al., 2017) and the global cropland dataset for 10,000 to 2100-GCD (Cao et al., 2021). Cropland data from these two global datasets were compared to our reconstruction results at provincial and grid scales.

3 Results

3.1 Total and provincial cropland area

Changes in cropland area for the middle and lower reaches of the Yellow River between 1000 and 1999 are illustrated in Figure 5. Overall, the cropland area for this region increased by 2.3 times from 21.87 million ha in 1000 to 50.64 million ha in 1999, with a peak of 55.98 million ha in 1975. Cropland areas during the past millennium have continuously increased over time and can be divided into the following three phases: 1000-1393 (Phase I), 1393-1724 (Phase II), and 1724-1999 (Phase III). Between 1000 and 1393, the cropland area fluctuated without considerable net change, accounting for 19.22-30.13 million ha. From 1393 to 1724, the cropland area rapidly increased, represented by the total cropland area increasing from 25.73 million ha in 1393 to 47.63 million in 1724, with an increase of 21.90 million ha over 331 years. From 1724 to 1999, the cropland area increased slowly, with a recorded increase of 3.01 million ha over 275 years.
Figure 5 Total cropland area in the middle and lower reaches of the Yellow River over the past millennium (Phase I: 1000-1393; Phase II: 1393-1724; Phase III: 1724-1999)
Figure 6 illustrates the changes in cropland areas over the past millennium in eight provinces located in the middle and lower reaches of the Yellow River. Although there were variations in cropland areas among the provinces, the trend of cropland in each province closely followed the overall trend in Figure 5. This trend was characterized by a slight increase, followed by a rapid increase, and then a slow increase.
Figure 6 Provincial cropland area in the middle and lower reaches of the Yellow River over the past millennium

3.2 Spatial distribution of cropland cover

Figure 7 shows the spatial distribution of cropland cover in the middle and lower reaches of the Yellow River over the past millennium. Prior to 1393, the areas of croplands increased slowly, with more than 70% of all grid cells featuring cropland coverage below 30% (Figures 7a-7g). Cropland cover was primarily concentrated on the Weihe and Fenhe Plains, with 74% and 72% of all grid cells having coverage <30% in 1215 and 1393, respectively. Subsequently, cropland cover expanded rapidly, especially on the North China Plain (Figures 7h-7k). The proportion of grid cells with cropland coverage exceeding 50% increased from 2.7% in 1393 to 16.8% in 1583 and 31.50% in 1724. Since then, owing to the rapid utilization of arable land, cropland cover entered a slow-growth stage, reaching saturation and stabilizing its spatial distribution (Figures 7l-7o). The proportion of grid cells with cropland coverage exceeding 50% was 32.4% in 1820 and 34.1% in 1999, peaking at 36.4% in 1975.
Figure 7 Cropland cover maps for the middle and lower reaches of the Yellow River over the past millennium (Panels a-o denote 1000, 1066, 1078, 1162, 1215, 1290, 1393, 1583, 1620, 1661, 1724, 1820, 1910, 1949, and 1999, respectively.)

3.3 Results validation and method evaluation

Provincial cropland data were obtained from several studies (Table 1). These cropland data were reconstructed based on historical tax records for cropland areas from different dynasties along with population data. As a traditionally agricultural country, croplands are a significant source of tax revenue in China. Historical tax records for cropland areas served as the basis for the official taxation of croplands in each dynasty. Historical tax record-based reconstruction provides a more accurate estimate of historical croplands that closely reflects the actual cropland areas. Therefore, the reconstructed provincial cropland area data were regarded as reliable data sources among the currently available historical data. Second, the trends and characteristics of provincial croplands in various dynasties are closely correlated with relevant records in the historical literature, such as population data, agricultural policies, social events, and agricultural development (He et al., 2023). Based on this, we believe that the input data for grid reconstruction of croplands, i.e., the historical provincial cropland area data, are reliable.
Moreover, we employed an indirect evaluation method to validate the results of the gridding reconstruction. Figure 8 illustrates the spatial patterns of the allocation results, remotely sensed cropland cover in 1980, and the absolute differences between them were calculated using Eq. (3). The spatial pattern of cropland cover obtained by our proposed model generally agreed with that derived from the remote sensing data (Figures 8a and 8b). Positive differences (i.e., the allocated cropland area was significantly higher than that obtained from the remote sensing data) were scattered in the eastern and southeastern Gan-Ning region (Figure 8c), whereas negative differences were mainly concentrated in the mountainous areas in the northwestern part of Jing-Jin-Ji (Figure 8c). Only 1.55% of all grid cells had absolute differences exceeding 50% (Table 4). If an absolute difference value within 20% is considered as indicative of effective allocation, our allocation accuracy can reach 77% (Table 4). Therefore, a gridding allocation model with unequal weight factors is feasible. Our assessment indicated that our reconstruction results objectively revealed the changes in cropland cover for the middle and lower reaches of the Yellow River between 1000 and 1999.
Figure 8 Spatial patterns of the allocation result (a), remote sensing-derived cropland cover in 1980 (b), and the differences between them (c)
Table 4 Statistical value of differences in cropland cover in 1980 between allocation results and remote sensing-derived cropland cover
Difference (%) Number of grids (%) Difference (%) Number of grids (%)
<-80 0.02 0-10 15.21
-80 to -70 0.07 10-20 4.19
-70 to -60 0.46 20-30 1.70
-60 to -50 0.90 30-40 0.60
-50 to -40 1.88 40-50 0.28
-40 to -30 4.88 50-60 0.07
-30 to -20 11.72 60-70 0.03
-20 to -10 18.71 70-80 0.00
-10 to 0 39.30 >80 0.00

4 Discussion

4.1 Cropland datasets comparison

Cropland data for the years 1000, 1400, 1720, and 2000 were extracted from two global datasets: HYDE and GCD. The cropland areas of three datasets (HYDE, GCD, and our reconstruction) were then compared at total, provincial, and grid scales. As depicted in Figure 9, discrepancies are evident in the total cropland area for the middle and lower reaches of the Yellow River, except for the HYDE data in 1400 and 2000 and the GCD data in 1720, which closely align with our reconstruction. At the provincial scale, although the cropland areas for the three datasets in Shandong were similar, significant differences in area were observed among the seven provinces between the two global datasets and our reconstruction, particularly in the years 1000, 1400, and 1720. Among these provinces, Jing-Jin-Ji exhibited the largest differences in cropland area between the three time points of our reconstruction and those of HYDE and GCD, with the amount of cropland in our reconstruction being 3-5 times higher. In particular, the cropland area in our reconstruction for this region in 1000 was 5.4 times greater than that in HYDE. Similarly, the cropland areas of Hu-Ning in 1000, 1400, and 1720 were 1.5-2.5 times greater in our reconstruction compared to those in HYDE and GCD, with the cropland area in our reconstruction for this region in 1720 being 2.5 times higher than that in HYDE. Moreover, the cropland areas in the GCD for Shandong, Gan-Ning, Henan, and Anhui were greater than those in the HYDE and our reconstruction. For instance, in 1400, the cropland areas in the GCD for Gan-Ning and Henan were three and two times greater than those in our reconstruction, respectively.
Figure 9 Comparison of total and provincial cropland area from the HYDE, GCD, and our reconstruction (HYDE: global environment database; GCD: global cropland dataset. Total denotes the entire middle and lower reaches of the Yellow River.)
A significant deviation in the spatial pattern was observed among HYDE, GCD, and our reconstruction, as shown in Figure 10. In HYDE (Figures 10a-10c) and GCD (Figures 10e-10g), croplands were primarily distributed near major modern rivers (Figure 10m), indicating that these two global datasets placed greater weight on factors such as river system in the cropland gridding allocation model. However, this allocation strategy may result in high uncertainty when reconstructing historical cropland cover because cropland distribution in Chinese history was not near major rivers and was sometimes located far away to avoid floods (Wang, 1980). Historical cropland distribution was closely linked to river systems, which comprised numerous small tributaries and artificial irrigation canals. River systems frequently undergo changes due to natural and human activities. In the past, the course of the Yellow River frequently changed in its lower reaches. For instance, during the Song Dynasty, the Yellow River flowed through the Jing-Jin-Ji region into the sea, and during the Ming and Qing dynasties, it flowed southward into the Huaihe River; whereas, it currently enters the sea through Shandong (Figure 10n). Moreover, no evidence supporting the distribution of croplands in the aforementioned global datasets was found in the historical literature or previous studies. Therefore, the strategies used in the HYDE and GCD datasets to allocate croplands to major rivers, particularly around the modern Yellow River, were inappropriate. These results do not accurately reflect the historical spatial distribution of cropland cover in the middle and lower reaches of the Yellow River. Our proposed method for allocating cropland cover data effectively avoided this issue (Figures 10i-10k).
Figure 10 Comparison of cropland maps among three datasets (a-d. HYDE; e-h. GCD; and i-l. our reconstruction. Panels (m) and (n) represent modern rivers and Yellow Rivers in different periods. HYDE: global environment database; GCD: global cropland dataset.)

4.2 Advantages of natural factor-based cropland gridding allocation method

Previous studies have utilized altitude, slope, and CPP to devise a gridding allocation method for cropland cover distribution, as many natural and human environmental factors that impact this distribution have undergone frequent changes throughout history and were arduous to obtain. These three factors are well established as significant contributors to cropland cover distribution and possess the added benefit of remaining unchanged or changing slightly over time, which can be replaced with modern data. Therefore, altitude, slope, and CPP were incorporated in our study. Subsequently, a previously overlooked factor, the DR, was also introduced. Our results indicated that DR holds greater relative importance to cropland distribution than the other three factors (Table 3). Moreover, the devised model is based on natural factors, such as altitude, slope, CPP, and DR, and can therefore be referred to as a natural factor-based cropland gridding allocation model.
Previous studies adopted gridding allocation models with equal weight factors. In this study, we utilized the Shapley value method to quantify the contribution shares of these natural factors affecting the distribution of cropland cover and developed a new gridding allocation model with unequal weight factors. Furthermore, we observed that the influence of the selected factors on cropland distribution varied across provinces, and this regional heterogeneity was effectively captured by our model. Consequently, this optimized model was more reliable for determining historical cropland distribution patterns. Given these advantages, this gridding allocation model with unequal weight factors has great potential for application at local, regional, and even global scales.
Additionally, owing to multi-source data, such as high-precision remote sensing images and field survey data, many high-resolution modern LUCC datasets have been developed. However, the scarcity of historical data limits the information that is available for reconstructing historical cropland cover datasets. Therefore, we cautiously chose a target resolution of 10 km for this study. A resolution of 10 km or 5 min has been widely adopted by numerous historical LUCC reconstructions at regional and global scales. Technically, our reconstruction method could obtain historical cropland cover data at a resolution of 1 km or higher. However, this process may lead to high uncertainty in the dataset. This is because existing studies, including the present study, generally use natural-factor-based grid allocation models. For example, in a 1-km grid, several adjacent 1-km grids in a certain region may not have significant differences in natural environmental conditions, and the allocation of cropland areas may be more influenced by human factors such as settlement points. Unfortunately, data on these human factors are generally lacking in current historical LUCC studies. Forcibly increasing the resolution of historical cropland cover datasets in the absence of sufficient historical basic data may deviate from the actual distribution of croplands.

4.3 Uncertainties

Due to changes in historical borders and administrative divisions, the original reconstructed cropland area data in Table 1 were not spatially consistent across different dynasties. To achieve consistent cropland area data for spatial units during grid-based reconstruction of cropland cover, the original reconstructed cropland area data required adjustment (He et al., 2023). This adjustment process primarily involved merging; however, in some cases, a small portion of the original cropland area data was split, leading to some uncertainty.
It should be pointed out that although the curve formed by connecting the collected 58 time points cropland in this study could objectively reflect the trends and characteristics of the changes in the cropland area at provincial scales over the past millennium, the “peaks” and “valleys” of the curve may not necessarily correspond to actual turning points in history, and there may be phase and quantity differences. Understanding this requires a comprehensive analysis in future studies after examining more historical materials to enrich and improve the reconstructed results.
Moreover, although we utilized historical coastlines and the maximum extent of cropland cover as limiting factors, natural factors may have allocated a small amount of cropland area to grids where croplands could not have existed historically. In addition, there are noticeable differences in the distribution of cropland cover on either side of provincial boundaries at some time points. In reality, the cropland distribution should exhibit a gradual transition. In the past, human factors such as settlement points and artificial irrigation channels have played an equally (if not more) important role in the distribution of cropland cover. Compared to the global historical cropland datasets and the gridding allocation models with equal weighting factors used in previous studies, our reconstruction method yielded significant improvements. However, it is essential to acknowledge that the current approach still has some limitations. In the future, digital acquisition and full adoption of all these factors will undoubtedly improve the reliability of our reconstruction results.
The several shifts in the course of the Yellow River downstream have substantially impacted the distribution of cropland cover. Notably, significant alterations occurred in the years 1128 and 1855 within the past millennium. However, at present, there is a lack of effective methodologies for quantifying the spatiotemporal ramifications of these major course changes on the cropland cover in the surrounding regions, so it is crucial to exercise caution when utilizing our dataset. Moreover, we conducted a reassessment of the cropland cover data at 58 points over the past millennium. Among these, the reconstruction closest to the year 1128 is that of the year 1162, with a 34-year interval between the two points. We posit that the impact of this Yellow River course change on the distribution of cropland cover should be minimal 30 years later. The nearest reconstruction point to the year 1855 is that of 1873, with an 18-year gap between the two points. Therefore, the cropland cover reconstructed for 1873 would be significantly influenced by the Yellow River course change, necessitating further refinement of the data for this point.

5 Conclusions

In this study, altitude, slope, CPP, and DR were selected as factors affecting cropland cover, and their contribution shares were quantified using the Shapley value method. Based on this analysis, a new cropland gridding allocation model with unequal weight factors was developed. By combining this model with historical provincial cropland area, historical coastlines, and satellite data-based maximum cropland extent, cropland cover maps for 58 time points in the middle and lower reaches of the Yellow River were reconstructed for the past millennium at a 10-km resolution. This dataset is available at https://doi.org/10.5281/zenodo.7807784 (Yang et al., 2023b). Significant quantitative differences were found between the global dataset (HYDE and GCD)-derived croplands and historical document-based reconstruction results. In the two global datasets, historical croplands were primarily distributed around major modern rivers, a strategy that does not accord with historical facts regarding cropland distribution in China.

References

Publishing order | Descend order by publishing year | Descend order by cited within
[1]
Buntgen U, Tegel W, Nicolussi K et al., 2011. 2500 years of European climate variability and human susceptibility. Science, 331(6017): 578-582.

DOI PMID

[2]
Cao B W, Yu L, Li X C et al., 2021. A 1 km global cropland dataset from 10 000 BCE to 2100 CE. Earth System Science Data, 13(11): 5403-5421.

[3]
Chen Y Z, Syvitski J P, Gao S et al., 2012. Socio-economic impacts on flooding: A 4000-year history of the Yellow River, China. Ambio, 41(7): 682-698.

DOI PMID

[4]
Douglas P M, Pagani M, Canuto M A et al., 2015. Drought, agricultural adaptation, and sociopolitical collapse in the Maya Lowlands. Proceedings of the National Academy of Sciences of the United States of America, 112(18): 5607-5612.

DOI PMID

[5]
Ellis E C, 2021. Land use and ecological change: A 12,000-year history. Annual Review of Environment and Resources, 46(1): 1-33.

[6]
Ellis E C, Kaplan J O, Fuller D Q et al., 2013. Used planet: A global history. Proceedings of the National Academy of Sciences of the United States of America, 110(20): 7978-7985.

DOI PMID

[7]
Fang X Q, He F N, Wu Z L et al., 2021. General characteristics of the agricultural area and fractional cropland cover changes in China for the past 2000 years. Acta Geographica Sinica, 76(7): 1732-1746. (in Chinese)

DOI

[8]
Fluet-Chouinard E, Stocker B D, Zhang Z et al., 2023. Extensive global wetland loss over the past three centuries. Nature, 614(7947): 281-286.

[9]
Foley J A, Defries R, Asner G P et al., 2005. Global consequences of land use. Science, 309(5734): 570-574.

DOI PMID

[10]
Friedlingstein P, Jones M W, O'Sullivan M et al., 2022. Global Carbon Budget 2021. Earth System Science Data, 14(4): 1917-2005.

[11]
Fuchs R, Herold M, Verburg P H et al., 2015. Gross changes in reconstructions of historic land cover/use for Europe between 1900 and 2010. Global Change Biology, 21(1): 299-313.

DOI PMID

[12]
Ge Q S, Dai J H, He F N et al., 2004. Spatiotemporal dynamics of reclamation and cultivation and its driving factors in parts of China during the last three centuries. Progress in Natural Science, 14(7): 605-613.

[13]
Guo R, Liu F G, Chen Q et al., 2021. Reconstruction of cultivated land pattern in the upper reaches of the Yellow River in the late Northern Song Dynasty: Take Hehuang Valley as an example. Journal of Natural Resources, 36(1): 27-37. (in Chinese)

[14]
He F N, Li M J, Li S C et al., 2017. Reconstruction of Lu-level cropland areas in the Northern Song Dynasty (AD976-1078). Journal of Geographical Sciences, 27(5): 606-618.

[15]
He F N, Yang F, Zhao C S et al., 2023. Spatially explicit reconstruction of cropland cover for China over the past millennium. Science China Earth Sciences, 66(1): 111-128.

[16]
Houghton R A, House J I, Pongratz J et al., 2012. Carbon emissions from land use and land-cover change. Biogeosciences, 9(12): 5125-5142.

[17]
Hurtt G C, Chini L, Sahajpal R et al., 2020. Harmonization of global land use change and management for the period 850-2100 (LUH2) for CMIP6. Geoscientific Model Development, 13(11): 5425-5464.

[18]
Jin X B, Cao X, Du X D et al., 2015. Farmland dataset reconstruction and farmland change analysis in China during 1661-1985. Journal of Geographical Sciences, 25(9): 1058-1074.

[19]
Kaplan J O, Ruddiman W F, Crucifix M C et al., 2011. Holocene carbon emissions as a result of anthropogenic land cover change. The Holocene, 21(5): 775-791.

[20]
Klein Goldewijk K, Beusen A, Doelman J et al., 2017. Anthropogenic land use estimates for the Holocene-HYDE 3.2. Earth System Science Data, 9(2): 927-953.

[21]
Leite C C, Costa M H, Soares-Filho B S et al., 2012. Historical land use change and associated carbon emissions in Brazil from 1940 to 1995. Global Biogeochemical Cycles, 26(2): 1-13.

[22]
Li M J, He F N, Li S C et al., 2018a. Reconstruction of the cropland cover changes in eastern China between the 10th century and 13th century using historical documents. Scientific Reports, 8: 13552. doi: 10.1038/s41598-018-31807-6.

[23]
Li M J, He F N, Yang F et al., 2018b. Reconstructing provincial cropland area in eastern China during the early Yuan Dynasty (AD1271-1294). Journal of Geographical Sciences, 28(12): 1994-2006.

[24]
Li M J, He F N, Yang F et al., 2020. Reconstruction of provincial cropland area and its spatial-temporal characteristics in the Ming Dynasty. Geographical Research, 39(2): 447-460. (in Chinese)

DOI

[25]
Li S C, He F N, Zhang X Z, 2016. A spatially explicit reconstruction of cropland cover in China from 1661 to 1996. Regional Environmental Change, 16(2): 417-428.

[26]
Li X Y, Tian H Q, Lu C Q et al., 2023. Four-century history of land transformation by humans in the United States (1630-2020): Annual and 1 km grid data for the HIStory of LAND changes (HISLAND-US). Earth System Science Data, 15(2): 1005-1035.

[27]
Mahbub R B, Ahmed N, Rahman S et al., 2019. Human appropriation of net primary production in Bangladesh, 1700-2100. Land Use Policy, 87: 104067. doi: 10.1016/j.landusepol.2019.104067.

[28]
Mendelsohn R, Sohngen B, 2019. The net carbon emissions from historic land use and land use change. Journal of Forest Economics, 34(3-4): 263-283.

DOI

[29]
Nandlall S D, Millard K, 2020. Quantifying the relative importance of variables and groups of variables in remote sensing classifiers using Shapley values and Game Theory. IEEE Geoscience and Remote Sensing Letters, 17(1): 42-46.

[30]
Nikiel C A, Eltahir E A B, 2019. Summer climate change in the Midwest and Great Plains due to agricultural development during the twentieth century. Journal of Climate, 32(17): 5583-5599.

[31]
Pongratz J, Reick C, Raddatz T et al., 2008. A reconstruction of global agricultural areas and land cover for the last millennium. Global Biogeochemical Cycles, 22(3): GB3018.

[32]
Ramankutty N, Foley J A, 1999. Estimating historical changes in global land cover: Croplands from 1700 to 1992. Global Biogeochemical Cycles, 13(4): 997-1027.

[33]
Scott C E, Monks S A, Spracklen D V et al., 2018. Impact on short-lived climate forcers increases projected warming due to deforestation. Nature Communications, 9(1): 157.

[34]
Shi N H, 1985. Changes of Forest and Grassland in the Loess Plateau for China. Xi’an: Shaanxi People’s Press. (in Chinese)

[35]
Stephens L, Fuller D, Boivin N et al., 2019. Archaeological assessment reveals Earth’s early transformation through land use. Science, 365(6456): 897-902.

DOI PMID

[36]
Tan Q X, 1962. Why the Yellow River was in a long period safe and sound condition after the Eastern Han Dynasty. Academic Monthly, (2): 23-35. (in Chinese)

[37]
Tan Q X, 1982. The Historical Atlas of China. Beijing: SinoMaps Press. (in Chinese)

[38]
Tian H Q, Banger K, Bo T et al., 2014. History of land use in India during 1880-2010: Large-scale land transformations reconstructed from satellite data and historical archives. Global Planet Change, 121: 78-88. doi: 10.1016/j.gloplacha.2014.07.005.

[39]
Vittorio A V D, Mao J, Shi X et al., 2018. Quantifying the effects of historical land cover conversion uncertainty on global carbon and climate estimates. Geophysical Research Letters, 45(2): 974-982.

[40]
Wang Y H, 1980. Some experiences and lessons of historical land use in China. Scientia Agricultura Sinica, (1): 90-96. (in Chinese)

[41]
Wei X Q, Li Y F, Guo Y et al., 2021. Spatio-temporal analysis of cropland change in the Guanzhong area, China, from 1650 to 2016. Journal of Geographical Sciences, 31(9): 1381-1400.

DOI

[42]
Wu X, Wang H, Bi N et al., 2020a. Climate and human battle for dominance over the Yellow River’s sediment discharge: From the Mid-Holocene to the Anthropocene. Marine Geology, 425: 9. doi: 10.1016/j.margeo.2020. 106188.

[43]
Wu X T, Wei Y, Fu B et al., 2020b. Evolution and effects of the social-ecological system over a millennium in China’s Loess Plateau. Science Advances, 6(41): eabc0276.

[44]
Yang F, He F N, Li S C et al., 2023a. A new estimation of carbon emissions from land use and land cover change in China over the past 300 years. Science of the Total Environment, 863: 160963. doi: 10.1016/j.scitotenv.2022. 160963.

[45]
Yang F, Zhang H, He F N et al., 2023b. A cropland cover dataset for the middle and lower reaches of the Yellow River over the past millennium. Zenodo [dataset], https://doi.org/10.5281/zenodo.7807784.

[46]
Yang X H, Jin X B, Guo B B et al., 2015. Research on reconstructing spatial distribution of historical cropland over 300 years in traditional cultivated regions of China. Global and Planetary Change, 128: 90-102. doi: 10.1016/j.gloPlacha.2015.02.007.

[47]
Yang X H, Jin X B, Xue Q F et al., 2021. Reconstruction of the spatial distribution of historical farmland in the Taiwan province of China for 1659-1945. Land Use Policy, 114: 105951. doi: 10.1016/j.landusepol.2021. 105951.

[48]
Ye Y, Fang X Q, Ren Y et al., 2009. Cropland cover change in Northeast China during the past 300 years. Science in China Series D: Earth Sciences, 52(8): 1172-1182.

[49]
Zhang H, Dong G P, Wang J F et al., 2023. Understanding and extending the geographical detector model under a linear regression framework. International Journal of Geographical Information Science, 37(11): 2437-2453.

[50]
Zhang X, Fang X, 2017. Temporal and spatial variation of catastrophic river floodings in the Lower Yellow River from AD 960 to 1938. The Holocene, 27(9): 1359-1369.

[51]
Zhao G, 1996. Environmental Changes in China’s History. Beijing: China Environmental Science Press. (in Chinese)

[52]
Zhao G, Mu X, Wen Z et al., 2013. Soil erosion, conservation, and eco-environment changes in the Loess Plateau of China. Land Degradation & Development, 24(5): 499-510.

[53]
Zhao Z L, Fang X Q, Ye Y et al., 2020. Reconstruction of cropland area in the European part of Tsarist Russia from 1696 to 1914 based on historical documents. Journal of Geographical Sciences, 30(8): 1307-1324.

DOI

[54]
Zheng J Y, Wen Y J, Fang X Q, 2020. Changes of climate and land cover in the middle and lower reaches of the Yellow River over the past 2000 years. Resources Science, 42(1): 3-19. (in Chinese)

DOI

[55]
Zumkehr A, Campbell J E, 2013. Historical U.S. cropland areas and the potential for bioenergy production on abandoned croplands. Environmental Science & Technology, 47(8): 3840-3847.

Options
Outlines

/

〈 〉