Information extraction and characteristic analysis of cultivated land abandonment in karst rocky desertification mountainous areas based on time-series vegetation index

Abstract

Due to the influence of rocky desertification, the phenomenon of abandoned farmland is serious in karst mountainous areas. The accelerated urbanization process, the transfer of rural labor force, and the obstruction of agricultural mechanization have aggravated the abandonment of cultivated land. Curbing the large-scale abandonment of cultivated land is of great significance to regional food security and sustainable agricultural development. Taking a typical karst rocky desertification mountainous area as an example, using Sentinel-2 A remote sensing images, the NDVI time series change detection coupling joint detection method was used to extract the abandoned land in Guanling County from 2019 to 2022, and its desertification degree and spatial heterogeneity were analyzed. The results show that: (1) The OA of the identification results of abandoned land and non-abandoned land from 2019 to 2022 is 0.86, and the Kappa coefficient is 70%, with good identification effect; (2) The abandoned area in Guanling County continued to increase from 2019 to 2022, but most of them were abandoned for two to three years, and a few were abandoned for four years. There were phenomena of recultivation and sudden abandonment of farmland; (3) Abandonment mostly occurred under conditions of no obvious desertification, mild desertification and moderate desertification, and the abandonment rate gradually decreased with the increase of desertification; (4) The abandoned land in Guanling County was mainly distributed in the direction of “northwest-southeast”, with high fragmentation, low spatial aggregation and strong heterogeneity. The study extracted information on abandoned farmland in karst desertification mountainous areas to provide a data basis for food security and protection of farmland quantity.

Food security is a national priority. To ensure national food security, we must strictly adhere to the red line of cultivated land and ensure the quality of cultivated land. Cultivated land is the foundation of agricultural development and the basis of farmers’ livelihoods¹. The transition of cultivated land from use to abandonment represents a land is a land use process, with abandoned cultivated land being a specific type of land cover resulting from “abandonment”^2,3. Abandoned cultivated land is defined as: cultivated land that was once used for planting or grazing is no longer in use, and the time is longer than the fallow period of local crop rotation, and herbaceous plants begin to grow, forming natural succession and forming abandoned cultivated land⁴.

In China, the overall quality of cultivated land is poor⁵. With the combined effects of rising agricultural labor costs^6,7, aging agricultural labor force⁸, and hindered agricultural mechanization, large-scale cultivated land abandonment has occurred in some areas, resulting in waste of cultivated land resources and affecting national food security. Therefore, obtaining the number of abandoned cultivated land and analyzing its spatial distribution are crucial for cultivated land protection. In addition, studies have shown that cultivated land abandonment has an impact on biodiversity⁹, soil stability¹⁰, carbon cycle¹¹, etc. Accurately understanding the situation of cultivated land abandonment in China can correctly guide the process of cultivated land abandonment, ensure food security, and assist in the formulation of land use policies¹².

At present, the main methods to obtain farmland abandonment information are sampling survey of regional farmers and remote sensing image monitoring¹³. Sampling survey of farmers has advantages in explaining the mechanism of farmland abandonment¹⁴, but the results are affected by the number, method, spatial distribution, topic sensitivity, subjectivity, etc. In addition, there are also cases of farmers under-reporting and over-reporting, which makes it difficult to obtain objective and comprehensive data¹⁵. Remote sensing monitoring methods have become a key means of monitoring and researching abandoned farmland with their advantages of low cost and high timeliness. This method can dynamically grasp the scope of abandoned farmland and obtain information on abandoned farmland at different scales, making up for the shortcomings of farmer sampling survey methods in obtaining spatiotemporal characteristics^16,17.

Methods for extracting abandoned farmland using remote sensing technology include supervised classification, multi-temporal superposition, and remote sensing time series analysis¹⁸. In the application of supervised classification, Alcantara et al. used the support vector machine method to classify MODIS data, combined phenological indicators and NDVI to extract abandoned farmland in Eastern Europe and draw abandoned agricultural maps, indicating that MODIS 250 m data can be used to draw large-scale abandoned agricultural maps, and pointed out that abandoned agriculture is common in Eastern Europe¹⁹. Niu Jiqiang et al. used the support vector machine method to classify Landsat data, and then dynamically monitored the classification results, extracting that seasonal and perennial abandoned farmland in Zilu Town, Luoshan County accounted for 14.33% and 7.45% of the total cultivated land area, respectively²⁰. However, this method may have the phenomenon of different objects with the same spectrum or different spectra with the same object during the interpretation process, resulting in the misclassification or omission of abandoned farmland²¹. In the study of multi-temporal range stacking method, Xiao Guofeng et al. used decision tree method to extract abandoned cultivated land and studied its spatial distribution, duration and reclamation area²². Shi Tiechou et al. obtained the distribution of abandoned farmland by superimposed the cultivated land scope in 2002 and 2011 in the typical abandoned farmland area of Chongqing City, and then removed the farmland returned to forest and forest engineering map spots²³. However, the accuracy of abandoned farmland extraction by this method mainly depends on the accuracy of land use data, so the reliability of the extraction results is unstable.In the application of remote sensing time series analysis methods, Song Xianqiang et al. analyzed the NDVI differences in spring, summer and autumn in the cultivated land area of Puge County, Sichuan Province based on decision trees, and divided the thresholds to distinguish the abandoned cultivated land from the non-abandoned cultivated land²⁴. Jiang Tao sequentially used vegetation index change detection, post-classification change monitoring method and decision tree classification method to extract information of abandoned cropland in Qinghemen District, Fuxin City, and then verified the extraction accuracy by Kappa coefficients, and the results showed that the vegetation index change detection method was more reliable than the other two types of methods²⁵. This type of time series analysis method based on crop phenological cycle can accurately distinguish between fallow, abandoned and non-abandoned farmland, and the extraction results are reliable. However, existing research rarely focuses on rocky desertification and abandoned farmland in karst areas, and most of the research focuses on plains or non-karst hilly areas. Rocky desertification in the south-western region is one of the three major ecological disasters in China^26,27. The phenomenon of rocky desertification, which causes soil erosion and leads to the loss of surface soil and the exposure of rocks, thus depriving the surface of its value for agricultural use and the ecological environment, is particularly pronounced in the south-western part of China. Cultivated land is an important means of production for farmers, and the aggravation of rocky desertification will seriously compress the resources of cultivated land, the survival and development space of farmers, and threaten the sustainable development of the economy²⁸. Therefore, it is of great significance to grasp and analyse the quantity and distribution of abandoned farmland in rocky desertification areas.

Under the special background of karst geology, problems such as land fragmentation, decline in the quantity and quality of cultivated land are prominent, and the regional food security situation is severe²⁹. It is extremely important to strengthen the management of cultivated land quantity and the construction of cultivated land quality³⁰. In order to grasp the amount of abandoned farmland in the region, consolidate regional food security, and promote sustainable agricultural development, this study is based on Sentinel-2 A data. According to the spectral characteristics of remote sensing images and the phenological characteristics of crops, the maximum value composite (MVC) method is used to construct the quarterly time series NDVI data from 2019 to 2022. The NDVI time series change detection coupled joint detection method is used to extract the number and scale of abandoned farmland in the study area from 2019 to 2022. The rocky desertification data is superimposed, and the landscape pattern index of abandoned farmland is calculated. The rocky desertification degree and spatial landscape pattern of abandoned farmland in Guanling County are analyzed, which provides a decision-making basis for political tasks such as rectifying abandoned farmland and ensuring food security.

Study area

Guanling County in Guizhou Province is a typical mountainous agricultural county and was once designated as a national poverty-stricken county³¹ The fragile ecological environment and rapid urbanization process have led to the deterioration of planting conditions and the reduction of agricultural income, which in turn has led to the abandonment of cultivated land. This not only reduces the quantity of arable land resources, but also reduces their quality, thus posing a serious threat to the security of arable land resources and local food production³². Guanling County, Guizhou Province, is located in the centre of the East Asian Karst Distribution Area, situated in the slope zone tilted to the south of the ridged slopes of the eastern part of the Yunnan-Guizhou Plateau towards the hills of Guangxi, with a large degree of topographic undulation, a high terrain in the north-west, and a low terrain in the south-east, which is a typical karstic landform (Fig. 1). The total area of the county is 1464 km², with 4 streets, 9 towns and 1 township under the jurisdiction. In 2020, the rocky desertification area of Guanling County was 254.9 km², accounting for 17.4% of the county areaWith a predominantly humid climate in the mid-subtropical monsoon, the average annual temperature is 16.2 °C, the annual precipitation is 1205.1–1656.8 mm, the frost-free period reaches more than 270d, sufficient light resources, and rain and heat in the same season.The type of cultivated land is dominated by sloping cultivated land, with a small overall quantity and a large proportion of medium- and low-quality cultivated land^33,34, the main crops are corn, sorghum, rice, sweet potatoes, and vegetables.

Fig. 1

Overview of the study area. (A) Guizhou Province; (B) Cultivated land in Guanling County; (C) DEM of Guanling County; (a, b) UAV images of verification data.

Data source and methods

Data source

Remote sensing images

The Sentinel-2 A image has a spatial resolution of 10 m in visible and near-infrared bands and a revisit period of 10 days³⁵. The high spatial and temporal resolution is suitable for the dynamic observation of land use in county karst complex landforms. In this paper, Google Earth Engine (https://code.earthengine.google.com/), a Google geocomputing cloud platform, was programmed to acquire a total of 36 periods of Sentinel-2 A de-clouded NDVI imagery covering the complete climatic period of crops (1 March to 30 November) from 2019 to 2022. The maximum NDVI values for each season were calculated by the maximum value synthesis method, and finally a total of 12 periods of NDVI data were obtained for 2019–2022 (Fig. 2).

Fig. 2

NDVI data for spring, summer and autumn 2019–2022.

Sample data

The study went to Guanling County in 2021 and 2022 to collect drone image data and field ground survey data, and vectorized abandoned and non-abandoned farmland in spring, summer and autumn, and determined the criteria for distinguishing abandoned and non-abandoned farmland (Fig. 3): In the UAV image data of the plot, if there are signs of cultivation in any season, such as signs of crop growth and traces of the use of farming tools, the plot is identified as cultivated land, and seasonal abandonment is not excluded. On the contrary, if the land is bare soil or weeds in the UVA image data and there are no signs of cultivation, it is abandoned farmland. 92 abandoned farmland plots in Guanling County in 2021, 118 abandoned farmland plots in 2022, 77 continuously abandoned farmland plots from 2021 to 2022, and 100 continuously cultivated plots were successfully extracted. The 2021 sample data are used to construct abandoned farmland extraction rules, and the 2021–2022 sample data are used for accuracy verification.

Fig. 3

Explanatory diagram of the rules for the discrimination of abandoned land. Abandoned 1, Abandoned 2, Abandoned 3: Abandoned all year round; Fallow 1: Abandoned in spring, cultivated in summer and autumn; Fallow 2: Abandoned in summer, cultivated in spring and autumn; Fallow 3: Abandoned in autumn, cultivated in spring and summer; Fallow 4: Abandoned in spring and summer, cultivated in autumn.

Data on cultivated land plots and desertification levels

Cultivated land plot data and rocky desertification grade data come from the State Engineering Technology Institute for Karst Desertification Control. The number of cultivated land parcels totalled 265,154, with a total area of 258.51 km²; the spatial resolution of the rocky desertification grade data was 2.5 m, and they were divided into five grades: no obvious rocky desertification, mild rocky desertification, moderate rocky desertification, strong rocky desertification and very strong rocky desertification.

Methods

Phenological differences are key to distinguishing between abandoned and non-abandoned farmland. Due to the influence of farmers’ planting and harvesting behaviors, crops will show obvious periodic phenological changes, which can be summarized as the “sowing period”, “maturity period” and “harvest period” of crops. The normalized difference vegetation index (NDVI) is one of the important parameters reflecting crop growth and nutritional information. Due to human activities, the changes in NDVI of crops between mature and immature periods are more obvious than those of natural vegetation, which means that the difference in NDVI between mature and immature periods is greater³⁶. This difference can be used to identify abandoned and non-abandoned farmland based on the difference in NDVI values in different seasons, The technical process is shown in Fig. 4.

At present, the academic community is still unclear about the time threshold for abandoned farmland. Some scholars regard farmland that has been idle for one year as abandoned farmland, while others believe that farmland that has been idle for two years or more is abandoned farmland³⁷. Based on previous experience, this paper considers farmland that has been idle for two years or more as abandoned farmland. The data collection period for this study is 2019–2022. In order to analyze the data change trend more carefully, the data is divided into three consecutive time periods for analysis: 2019–2020, 2020–2021, and 2021–2022. In addition, this paper analyzes the data of two cumulative time periods, namely 2019–2021 and 2020–2022, as well as the total data of the entire research period (2019–2022), to more comprehensively understand the dynamic changes of abandoned farmland in Guanling County.

The NDVI time series data is masked using the cultivated land vector boundary to remove interference from other land objects and obtain the NDVI time series data at the plot scale. Difference operations were performed on the spring-summer and autumn-summer NDVI time-series data of the same year, respectively, and combined with the sample data of the field ground survey, which were used to obtain the segmentation thresholds of abandoned and non-abandoned farmland. The plots with NDVI difference greater than the threshold in spring-summer and autumn-summer of the same year were defined as plots with one year of fallow, and the fallow plots of the adjacent two years were spatially superimposed and intersected to obtain the abandoned farmland plots, and to calculate the abandonment rate in the study area, i.e., the proportion of abandoned farmland area to the background farmland area²². The specific expression is as follows:

$$P_{a} = \frac{{A_{i} }}{{A_{0} }}*100\%$$

(1)

where $\:{\text{P}}_{\text{a}}$ denotes the abandonment rate, $\:{\text{A}}_{\text{i}}$ denotes the abandoned farmland area in $\:\text{i}$th year, $\:{\text{A}}_{0}$ denotes the baseline cultivated land area.

Fig. 4

Technical flow chart of abandoned land extraction.

Some studies have pointed out that cultivated land is more likely to be abandoned in a karst geological context³⁸. This paper uses the above method to extract abandoned farmland plots and superimpose rocky desertification grade data, calculates the distribution of abandoned farmland with different grades of rocky desertification, calculates the landscape pattern index, draws the standard deviation ellipse, and explains the current status of abandoned farmland in Guanling County from the perspective of abandoned farmland rocky desertification classification and spatial landscape pattern.

NDVI time series construction

The growth process of crops involves human activities such as sowing, fertilizing, weeding, and harvesting, which makes the vegetation coverage higher than that of natural vegetation. Therefore, the NDVI change range and speed of crops are also higher than that of natural vegetation. Natural vegetation is not disturbed by human activities, and the NDVI change range and speed are relatively gentle³⁹. The main crop grown in Guanling County are corn, sorghum, rice, sweet potatoes, etc. Among them, the sowing period of corn is usually from March to April, the maturity period is from May to August, and the harvest period is September; the sowing period of sorghum is April, the maturity period is from May to August, and the harvest period is September; the sowing period of rice is May, the maturity period is from June to August, and the harvest period is September; the sowing period of sweet potatoes is from April to May, the maturity period is from June to August, and the harvest period is September. Based on the above growing season characteristics of crops, the stages of the year are mainly divided into three: March to May is the sowing period, June to August is the maturity period, and September to November is the harvest period, corresponding to spring, summer, and autumn respectively (Fig. 5). In order to carry out NDVI change detection, the GEE platform was used to obtain NDVI data from March to November, and the maximum value synthesis method was used to process the NDVI data of each month, which was used as the basic data for quarterly change detection.

Fig. 5

Phenological cycle diagram of major crops.

Change detection

Change detection (Fig. 6) is the key to extract land cover change, in which threshold is used to determine whether the monitoring object attributes change, and the segmentation accuracy of threshold also directly affects the overall classification accuracy⁴⁰. There are various threshold segmentation algorithms, and this study chooses the threshold segmentation algorithm based on sample statistics to determine the segmentation threshold.

Spring is the crop sowing period, the cultivated land is semi-bare, and the NDVI value is low; summer is the crop maturity period, and the NDVI value reaches the annual peak; autumn is in the crop harvesting period, and the cultivated land is in the naked state, with low NDVI values. Based on this principle, the NDVI difference between spring and summer and autumn and summer was calculated with the cultivated land as the limit range and the plot as the unit. If the difference is positive, it indicates that there is seasonal abandonment behavior in the plot. If the difference is negative, it indicates that the plot has planting behavior.The sample data is superimposed on the NDVI image of the corresponding period in 2021 to obtain the NDVI difference of different phases and different crops, and the difference is analyzed to determine the best segmentation point. The NDVI difference image is reclassified according to the segmentation point, so that all non-abandoned cultivated land falls in the non-abandoned cultivated land area in the image, and all abandoned cultivated land falls in the abandoned cultivated land area in the image, and the threshold is determined.

Joint change detection includes intra-year and inter-annual detection[25]. Intra-annual detection extracted the cropland plots that did not have planting behaviours in a single year through the threshold value, and inter-annual detection was to analyse the difference of summer NDVI in the same plots in adjacent years, and the weeds in long-term abandoned plots were denser than that in a single year, so the NDVI value of abandoned farmland plots would increase over time, and the farmland plots whose summer NDVI was significantly smaller than that of the previous year were extracted on the basis of intra-annual detection and defined as abandoned farmland plots. In order to avoid the influence of special circumstances on the inter-annual NDVI difference (e.g. drought caused by sudden high temperature in the following year, resulting in weed withering), this paper chooses to repeat the intra-annual detection, and superimpose the plots obtained from the intra-annual detection in the two adjacent years in which no cultivation occurred, and the spatial superposition of the overlapping plots is defined as abandoned cultivated land.

Fig. 6

Schematic diagram of joint change detection. (red: abandoned farmland, green: farmland under cultivation, blue: single season fallow, yellow: single year fallow).

Accuracy evaluation

The accuracy of remote sensing image classification results is reflected by the accuracy evaluation after classification. The most common evaluation methods are qualitative evaluation and quantitative evaluation. In current research, quantitative evaluation is the most commonly used method. This method evaluates the classification results through calculated data indicators. Among them, the confusion matrix is the most widely used quantitative evaluation method. The confusion matrix is also called the error matrix. The confusion matrix can be used to calculate accuracy evaluation factors such as overall accuracy OA (Overall Accuracy), producer accuracy PA (Producer’s Accuracy), user accuracy UA (User’s Accuracy) and Kappa coefficient KC (Kappa Coefficient). Since OA and Kappa coefficient reflect the overall classification effect, this paper selects OA and Kappa coefficient as the main factors for accuracy evaluation⁴¹. The calculation formula is as follows:

$${\text{OA}} = \mathop \sum \limits_{{i = 1}}^{k} \frac{{x_{{ii}} }}{N}$$

(2)

$$P_{e} = \frac{{a_{1} \times b_{1} + a_{2} \times b_{2} + \cdots + a_{k} \times b_{k} }}{{N^{2} }}$$

(3)

$${\text{K}} = \frac{{P_{0} - P_{e} }}{{1 - P_{e} }}$$

(4)

In formulas (2)–(4), Where $\:{\text{P}}_{0}$ denotes the overall classification accuracy, $\:\text{K}$ denotes the total number of categories of objects to be classified, $\:\text{N}$ denotes the total number of samples, $\:{\text{x}}_{\text{i}\text{i}}$ denotes the number of correctly classified objects on the diagonal of the matrix, $\:{\text{P}}_{\text{e}}$ denotes the accidental consistency error,$\:{\text{a}}_{1},{\text{a}}_{2},\ldots,{\text{a}}_{\text{k}}$, denote the number of measured samples of each category,$\:{\text{b}}_{1},{\text{b}}_{2},\ldots,{\text{b}}_{\text{k}\:}$, denote the predicted number of samples of each category.

Landscape pattern index calculation

Combined with the results of abandoned farmland extraction, this study selected some landscape pattern indices according to patch type and landscape level to analyze the landscape structure characteristics of abandoned farmland in Guanling County. The landscape pattern index includes Mean Patch Size (MPS), Morphological Spatial Index (MSI), Mean Patch Fractal Dimension (MPFD), Aggregation Index (AI), and Patch Density (PD), which analyze the morphological characteristics and aggregation degree of abandoned farmland in space. The calculation methods and meanings of different indexes are as follows:

$$MPS = \frac{{\sum A_{i} }}{N}$$

(5)

where $\:\text{M}\text{P}\text{S}$ is used to reflect the degree of heterogeneity of the landscape, $\:{\text{A}}_{\text{i}}$ denotes the area of the $\:\text{i}$th patch, $\:\text{N}$ denotes the total number of patches.

$$MSI = \frac{{\sum P_{i} }}{{\sum A_{i} }}$$

(6)

where $\:\:\text{M}\text{S}\text{I}$ is used to measure the complexity of patch morphology and the diversity of shapes, $\:{\text{P}}_{\text{i}}$ denotes the boundary length of the $\:\text{i}$th patch, $\:{\text{A}}_{\text{i}}$ denotes the area of the $\:\text{i}$th patch.

$$MPFD = \frac{{\sum D_{i} }}{N}$$

(7)

where $\:\text{M}\text{P}\text{F}\text{D}\:$is used to describe the morphological complexity of the patch and reflect its fractal characteristics,$\:{\:\text{D}}_{\text{i}}$ denotes the fractal dimension of the $\:\text{i}$th patch,$\:\:\text{N}$denotes the total number of patches.

$$AI = \frac{{\sum A_{i} ^{2} }}{{A_{{total}} }}$$

(8)

where $\:\text{A}\text{I}\:$is used to measure the degree of aggregation of patches in the landscape. The higher the value, the higher the degree of aggregation,$\:{\text{A}}_{\text{i}}$ denotes the area of the $\:\text{i}$th patch,$\:{\:\text{A}}_{\text{t}\text{o}\text{t}\text{a}\text{l}}$ denotes the sum of the areas of all patches.

$$PD = \frac{N}{{A_{{total}} }}$$

(9)

where $\:\text{P}\text{D}\:$represents the number of patches per unit area, and reflects the degree of landscape fragmentation,$\:\text{N}\:$denotes the total number of patches,$\:{\text{A}}_{\text{t}\text{o}\text{t}\text{a}\text{l}}$ denotes the total area of the study area.

Standard deviation ellipse

Standard Deviation Ellipse (SDE), as a spatial data analysis tool, is mainly used to describe the distribution characteristics of spatial data and its degree of diffusion. This tool can effectively analyze the spatial relationship of data sets and is widely used to study the aggregation pattern of geographic events and changes in their spatial distribution. The ellipsoid generated by the standard deviation ellipse is achieved by calculating the mean and variance of the data, aiming to show the distribution characteristics of the data in a specific direction. Among them, the center of the ellipse represents the average position or center of gravity of the spatial data, reflecting the relative position of the distribution of geographic elements in two-dimensional space. The long and short axes of the ellipse represent the main direction and secondary direction of the distribution of geographic elements, respectively. The azimuth of the ellipse reflects the main trend of the spatial distribution of geographic elements, and the flattening of the ellipse is the ratio of the long and short axes, reflecting the morphology of the spatial distribution of geographic elements. That is, SDE mainly analyzes the spatiotemporal change pattern and trend of the center of gravity, shape, direction, range, etc. of the research object within different time ranges through the center, long and short axes, azimuth, and flattening of the ellipsoid^42,43. In this study, the Standard Deviation Ellipse method was used to portray the spatial differentiation of abandoned cropland in Guanling County from 2019 to 2022, respectively.

Results

Time series change analysis and threshold determination

The NDVI time series curves of major crops and abandoned farmland in Guanling County are shown in Fig. 7. The results of the maximum NDVI difference of different crops in different time phases are shown in Table 1. The difference between spring and summer of abandoned farmland is -0.13, and the difference between autumn and summer is -0.08. The threshold value was determined to be -0.1 by the threshold segmentation algorithm, that is, the plots with NDVI difference greater than − 0.1 were determined to be cultivated plots with no planting behavior temporarily.

Fig. 7

NDVI curve chart of main crops during their growth period.

Table 1 Calculation results of NDVI difference (maximum value) of land cover at different time periods.

Accuracy verification

In order to verify the accuracy of remote sensing monitoring, 77 abandoned farmland plots and 100 non-abandoned farmland plots obtained from field measurements in 2021–2022 were used to verify the accuracy of remote sensing monitoring results, The confusion matrix is shown in Table 2. The overall classification accuracy OA was calculated to be 0.86 and the Kappa coefficient was 0.70 through the confusion matrix. The verification results show that it is feasible to extract abandoned farmland based on the multi-temporal NDVI difference method.

Table 2 Confusion matrix and accuracy.

Distribution of abandoned farmland

The number and area of abandoned farmland were counted, and the abandonment rate of farmland was calculated according to the definition of abandonment rate (Table 3). It was found that the abandonment rate in the study area was the highest in 2021–2022, with an abandonment rate of 20.92%, and the lowest in 2019–2020, with an abandonment rate of 11.76%.Based on the abandoned farmland in 2019, the number of plots abandoned for three consecutive years in 2019–2021 and 2020–2022 gradually decreased; the number of plots abandoned for four consecutive years dropped to 11,597 plots, but the overall abandoned area showed an upward trend, and the phenomenon of farmland abandonment did not seem to be under control, indicating that some plots had the behavior of rehabilitation during 2019–2022. Meanwhile, compared with reclaiming land, more farmers have abandoned farmland, and the number of abandoned farmland plots has decreased, but the area of abandoned farmland has continued to increase. As can be seen from Fig. 8, the aforementioned plots of continuously abandoned farmland and the plots of sudden abandoned farmland are mainly distributed in the northern and central regions of the study area with higher elevations, while the abandonment rates and duration of abandonment are relatively low in the southern and southwestern regions with lower elevations.

Fig. 8

Spatiotemporal distribution of abandoned land in the study area from 2019 to 2022.

Table 3 Statistical table of abandonment in the study area from 2019 to 2022.

Status of rocky desertification of abandoned farmland

Guanling County has non-karst and karst landforms. The grades of abandoned farmland with rocky desertification in the karst landform area are divided into abandoned farmland with no obvious rocky desertification, abandoned farmland with slight rocky desertification, abandoned farmland with moderate rocky desertification, abandoned farmland with severe rocky desertification, and abandoned farmland with extremely severe rocky desertification⁴⁴. As shown in Table 4, the number of abandoned farmland with rocky desertification in the karst area accounted for an average of 83.5% of the total number of abandoned farmland plots from 2019 to 2022, while the number of abandoned farmland plots with non-karst rocky desertification accounted for 16.5%. It is worth noting that the proportion of abandoned rocky desertification plots that have been abandoned for four consecutive years is 92.6%, which is significantly higher than the average proportion. This shows that abandonment mainly occurs in rocky desertification plots, especially in plots with no obvious rocky desertification, mild rocky desertification, and moderate rocky desertification where the soil layer is thinner, the rocks are more exposed, and the land productivity is lower. The rocky desertification problem is common in Guanling County, but the degree of rocky desertification in most areas has not yet reached severe rocky desertification or extremely severe rocky desertification. Therefore, the number of abandoned farmland with severe rocky desertification and extremely severe rocky desertification is relatively small⁴⁵. Non-karst plots have better terrain conditions and planting conditions, and the number of abandoned farmland is relatively small.

Table 4 Statistics of abandoned land blocks in the research area due to rock desertification.

Standard ellipse deviation distribution of abandoned farmland

The distribution ellipse of abandoned farmland was calculated by standard deviation ellipse method. The spatial distribution ellipse of abandoned farmland in Guanling County is shown in Fig. 9, and the standard deviation related parameters are shown in Table 5. The overall distribution of abandoned farmland in 2019–2022 showed a “northwest-southeast” spatial distribution pattern; the main spatial distribution was located in Longtan Street, Baili Street, and Yongning Town in Guanling County; the longitude change range of the abandoned farmland center was similar to the latitude change range, indicating that the abandonment force in the north-south direction was equal to that in the east-west direction. Among them, the standard deviation ellipse of abandoned farmland in 2021–2022 changed most significantly. Compared with 2019–2020 and 2020–2021, the change in the short axis of the ellipse was greater than that in the long axis, and the spatial distribution obviously expanded in the east-west direction, which was similar to the direction of the abandoned farmland ellipse that had been abandoned for four consecutive years, and the flattening rate decreased, indicating that the abandonment direction was more obvious and the spatial distribution was more discrete.

Fig. 9

Standard deviation ellipse and center of gravity of abandoned land from 2019 to 2022.

Analysis of the landscape pattern of deserted arable land

The calculation results of the landscape pattern index of abandoned farmland in Guanling County from 2019 to 2022 are shown in Tables 6 and 7 respectively. From a spatial perspective, the MPS value of abandoned cultivated land is significantly smaller than that of non-abandoned cultivated land, and the PD value of abandoned cultivated land is significantly greater than that of non-abandoned cultivated land. That is, the average plot area of abandoned cultivated land is much smaller than that of non-abandoned cultivated land, and the plot fragmentation is greater and heterogeneous. The quality is stronger; the MSI and MPFD values of abandoned cultivated land are greater than that of non-abandoned cultivated land, that is, compared with non-abandoned cultivated land, abandoned cultivated land in the study area is more fragmented, the shape of the plot is more complex, and the degree of deviation from square and circle is higher; abandoned cultivated land The AI value is lower than that of non-abandoned cultivated land, which means that the spatial distribution of abandoned cultivated land in the study area is more dispersed. From a time point of view, the MPS value of two-year abandonment increased, and the PD and AI values decreased compared with the three- and four-year abandonment period; the MPS and AI values of non-abandoned cultivated land increased, and the PD value decreased, indicating that with the continuous abandonment time With the increase of, the landscape fragmentation of abandoned cultivated land increases, the area of plots becomes smaller and scattered in different areas, and the overall landscape structure becomes more scattered and complex. On the contrary, as the planting time increases, the fragmentation of non-abandoned cultivated land weakens, and plots The complexity of the shapes is also reduced and the landscape structure is more concentrated.

Table 5 Calculation results of landscape pattern index of abandoned land in the study area from 2019 to 2022.

Table 6 Calculation results of landscape pattern index of non-abandoned land in the study area from 2019 to 2022.

The study discussed the cases of abandoned farmland areas of less than 0.5 hm², between 0.5 and 1.5 hm², and greater than 1.5 hm² in Guanling County from 2019 to 2022⁴⁶. The results showed that in 2019–2020, the average proportion of abandoned farmland with a unit area of less than 0.5 hm² was 72.86%; the average proportion of abandoned farmland with a unit area of between 0.5 and 1.5 hm² was 24.58%; and the average proportion of abandoned farmland with a unit area of greater than 1.5 hm² was 2.66%. The proportion of abandoned farmland with a unit area of less than 0.5 hm² was the largest, indicating that the abandoned farmland in Guanling County is mainly small-scale abandoned farmland, which verifies the theory of fragmented abandoned farmland plots in Guanling County.

Table 7 Distribution of abandoned land areas of different grades in the research area.

As shown in Table 8, there are significant differences in the number of units with an area less than 0.5 hm² in each township in different periods from 2019 to 2022. Most of the land that has been abandoned for four consecutive years is located in Yongning Town, Duanqiao Town, and Huajiang Town. In comparison, the area of abandoned farmland in Baihe Street is smaller. From the perspective of transportation, the provincial highway crosses Huajiang Town and passes through the northwest of Duanqiao Town, and the national highway crosses Yongning Town and Duanqiao Town. Compared with other towns, Yongning Town, Duanqiao Town, and Huajiang Town have convenient transportation and lower migrant work costs. Low, coupled with rising agricultural labor costs, farmers are more inclined to go out to work, resulting in an increase in farmland abandonment. From the perspective of natural conditions, the altitude range of Yongning Town is 1036–1818 m, and the average altitude within the township is high; the lowest altitude of Duanqiao Town is 621 m, the highest altitude is 1472 m, and the terrain is highly undulating; the degree of rocky desertification in Huajiang Town is relatively serious, and most of them are Small patches. Therefore, complex topography, large differences in topography within small areas, and low productivity of land make it difficult to cultivate cultivated land, resulting in a large number of land plots that continue to be abandoned for cultivation in Yongning Town, Duanqiao Town, and Huajiang Town. The area of Lily Street is small, the total cultivated land area is small, the corresponding abandoned area is small, and the elevation difference within the street is small, so it is relatively difficult to plant cultivated land. In general, the unit area of abandoned farmland in various townships in Guanling County is mainly less than 0.5 hm², and is generally distributed in the northwest direction (Fig. 10).

Table 8 The distribution of abandoned land less than 0.5 hectares in each Township from 2019 to 2022.

Fig. 10

Spatio-temporal distribution of abandoned land of different grades from 2019 to 2022.

Discussion

Overview of abandoned farmland in Guanling County

In this paper, the NDVI data of Sentinel-2 A images were used to extract abandoned farmland in the rocky desertification landscape of Guanling County. The OA was 0.86 and the Kappa coefficient was 0.70, indicating that the extraction results were reliable. Some studies believe that the abandonment rate of farmland in the mountainous areas of southwest China is 10–25%^47,48. Guanling County is a typical karst mountainous area in southwest China. The average abandonment rate from 2019 to 2022 was 11.83%, indicating that the experimental results of this study have a certain degree of reliability. The rocky desertification situation is a key factor in determining the abandonment of farmland. This study superimposed the rocky desertification grade data and calculated that the rocky desertification abandoned farmland in Guanling County accounted for 83.5% of the abandoned farmland, which is much higher than the non-karst abandoned farmland. The reason for this phenomenon is that the soil layer of rocky desertification farmland is thin, which is more prone to soil erosion; the farmland is fragmented and the planting conditions are poor, which is more likely to be abandoned. From the results of landscape index and unit area division, it can be seen that abandoned farmland is more fragmented and less concentrated than non-abandoned farmland. Therefore, the government should take measures such as continuously promoting land consolidation projects, encouraging farmers to participate in the market-oriented transfer of farmland, and increasing grain subsidies to curb the abandonment of farmland with better terrain conditions^{49,50,51,52,53}.

Shortcomings and prospects

The multi-temporal NDVI monitoring method has expanded the time dimension information to some extent, but the following problems still exist:

• (1) Through field research, it can be seen that most of the cultivated land in Guanling County is planted by farmers themselves, and large-scale planting has not been formed. Further analysis combined with the crop planting situation shows that some types of crops lack completely consistent phenological laws. For example, corn is one of the main crops in Guanling County and is sown in March every year, but a few farmers sow in April, forming the distinction between “early season corn” and “late season corn”. Due to the delayed sowing time, the maturity period is also delayed to early September, which in turn affects the autumn NDVI value and causes the misjudgment of abandoned farmland;

• (2) Through field research and UVA images, it can be observed that some farmlands in the study area are covered with plastic film. Plastic film-covered crops in Guizhou Province include tobacco and vegetables⁵⁴. If crops need to be covered with plastic film, farmers need to cover the farmland with plastic film at different times to control the water and heat balance of the soil. Different crops have different sowing periods, so the time of plastic film covering is also different. Most crops in Guanling County mature in summer, and the NDVI value in summer is significantly higher than that in spring and autumn. However, it is inevitable that some crops and vegetables are sown in summer, or even covered with plastic film, which affects the NDVI value. The NDVI value in summer is significantly lower than that in spring and autumn. Therefore, the cultivated land plots covered with plastic film in summer may be misclassified;

• (3) Therefore, improvements will be made in the following aspects in the future: ① Further distinguish abandoned farmland by combining its spectral and texture features to improve recognition accuracy; ② Use multiple methods to extract abandoned farmland information, conduct horizontal comparison between methods, analyze the advantages and disadvantages of various methods, and propose a new abandoned farmland information extraction method combining the advantages of multiple methods; ③ Combine natural, social and other factors to increase relevant data to study the driving forces of abandonment.

Conclusion

This paper takes Guanling County in Southwest China as the study area and extracts information on abandoned farmland from 2019 to 2022. The NDVI time series data is combined with crop phenology to analyze the NDVI differences between abandoned farmland and non-abandoned farmland in different phenological periods. The monitoring accuracy is verified by combining field survey sample data to obtain the distribution map of abandoned farmland in Guanling County at different periods from 2019 to 2022. The rocky desertification data is used for overlay analysis to calculate the number of abandoned farmland plots with different levels of rocky desertification. The standard deviation ellipse is drawn, the landscape pattern index is calculated, and the abandonment, spatial distribution and landscape pattern of farmland with different rocky desertification levels at different periods are analyzed. The following conclusions are drawn:

• (1) In terms of the abandoned area, the number of abandoned farmland and the area of abandoned land in Guanling County showed an overall “increasing” trend from 2019 to 2022, with the abandoned area expanding from 30.53 km2 to 54.33 km² in 2021–2022, and the abandonment rate gradually increasing from 11.76 to 20.92%. As the duration of abandonment increases, the abandoned area gradually decreases. Among them, the abandoned area that has been abandoned for four years is 17.71 km², and the abandonment rate is 6.28%, which is distributed in the central and northern parts with higher altitudes. The reasons for abandonment are that the marginal plots such as fragmented plots, shallow soil layers, and steep slopes have low farming benefits and greater difficulty in cultivating land, resulting in changes in the utilization status of cultivated land in mountainous areas⁵⁵.

• (2) In terms of desertification, the number of abandoned farmland plots in karst areas with desertification was much higher than that in non-karst areas at any time during 2019–2022, and the number of abandoned farmland plots was mainly in the range of no obvious desertification, mild desertification, and moderate desertification. The results show that farmland in karst areas is more likely to be abandoned than non-karst farmland, and as the degree of desertification increases, the area of abandoned farmland gradually decreases⁴⁹.

• (3) From the perspective of spatial distribution, the spatial distribution center of abandoned farmland in any period from 2019 to 2022 is located in Longtan Subdistrict, and is mainly distributed in the “northwest-southeast” direction. Compared with the distribution ellipse of abandoned farmland that was abandoned for two years, the distribution ellipse of abandoned farmland that was abandoned for four consecutive years obviously moved northwestward, with a shorter major axis, longer minor axis, and a higher flattening rate, indicating that the distribution of abandoned farmland is more dispersed, fully demonstrating the spatial differentiation and spatial aggregation characteristics of abandoned farmland.

• (4) From the perspective of the landscape pattern index, the abandoned farmland is fragmented and irregular in shape. Combined with the distribution of abandoned farmland at different levels in each township, the distribution area of abandoned farmland with a unit area of less than 0.5hm² is much larger than that of other levels, which is consistent with the conclusion of the landscape pattern index. This is due to the complexity of the karst mountain area system and the cultivated land system. Natural conditions are one of the main factors leading to the abandonment of farmland. The scale of farmland in hilly and mountainous areas is generally small, and the terrain is complex and the shape is irregular. In addition, human activities have caused the continuous farmland to be divided into smaller plots, resulting in further aggravation of the degree of fragmentation, increasing the difficulty of planting, increasing the cost of planting, and gradually decreasing the income from growing grain, which has aggravated the abandonment of farmland^56,57,58,59.