Impact of Huanglongbing on Citrus Orchards: A Spatiotemporal Study in Xunwu County, Jiangxi Province

Abstract

Due to human activities and changes in land use, the spatiotemporal pattern of citrus has undergone significant changes after the outbreak of Huanglongbing (HLB). We selected time-series Sentinel-2 images to delineate citrus orchard areas following the onset of HLB. This was conducted to extract citrus orchards in Xunwu County between 2017 and 2022. The spatial and temporal patterns and their influencing factors were investigated by spatial analysis. Results show (1) a notable decline in total citrus orchard area by 216.70 km², primarily witnessed in orchards without insect-proof screens (IPS), shifting towards cropland, bush, and IPS areas. Contrastingly, citrus orchards with IPS exhibited a modest increase from 7.82 km² to 111.39 km², predominantly converting from areas lacking IPS, cropland, and bare land. (2) Spatial distribution patterns revealing a “cold in the south and hot in the north” trend. Orchards without IPS are concentrated in central and northern regions, while those with IPS are clustered predominantly in the north, with a recent shift towards the northeast. (3) Landscape analysis indicating a trend of fragmentation of citrus orchards, while a gradual dispersion of orchards without IPS and those with IPS showcased enhanced concentration and aggregation. (4) Orchards with IPS predominantly occupy regions characterized by an elevation ranging between 300 m and 400 m, primarily in the southeast, southwest, and southern directions. These areas exhibit slopes averaging between 10° and 15°, with surface temperatures ranging from 18 °C to 26 °C. Additionally, these orchards tend to be situated in proximity to impervious surfaces and roads.

1. Introduction

Citrus fruits, a high-value crop, stand among the most widely cultivated and cherished fruit trees globally. Acknowledged for their nutritional value as sources of vitamins, fiber, and minerals, they hold a commendable place for their benefits to human health. In 2021, North America produced approximately 6.3 million tons of citrus, while Asia’s citrus fruit production surpassed 83.5 million tons (accessed on 20 January 2023, https://www.statista.com). As one of the significant origins of citrus, China boasts diverse citrus resources [1], leading the world with a production volume of 44.6 million tons. Over recent decades, China’s citrus cultivation area has expanded continuously, reaching 5648.55 square kilometers by 2021 [2]. Southern China’s subtropical regions, encompassing provinces such as Jiangxi, Hubei, Sichuan, and Guangdong, serve as primary habitats for citrus growth. Among these, Ganzhou City in Jiangxi Province stands as a prominent citrus production hub, situated in one of the ecologically sensitive and fragile hilly regions of southern China. However, in recent years, the outbreak of Huanglongbing (HLB) disease, caused by Candidatus Liberibacter asiaticus (CLas) primarily transmitted by Diaphorina citri Kuwayama, has posed a significant threat. It stands as one of the most devastating diseases in the global citrus industry. Symptoms in affected citrus trees include yellowing of new shoots, spotting on fully matured old leaves, or zinc-deficiency-like symptoms in fresh mature leaves. Additionally, the fruits appear smaller, discolored, and imbalanced, and acquire a bitter taste [3]. Post-infection, citrus trees witness a severe decline in yield and quality [4], with the disease being challenging to cure, rendering the produce economically nonviable. HLB has spread globally across 40 countries in Asia [5], Africa [6], North America, South America, and Oceania [3]. In Florida, USA, HLB causes annual losses exceeding USD 1 billion, consistent with a 74% drop in citrus production [7].

Xunwu County, as a prominent citrus cultivation area in southern Jiangxi, China, stands as one of the severely affected regions by HLB. Since the outbreak in 2012, citrus production in Xunwu County has plummeted by 71,300 tons, accompanied by a reduction in orchard area by 107,100 mu (accessed on 20 January 2023, http://www.xunwu.gov.cn/). To mitigate losses caused by HLB, local authorities have resorted to various measures such as tree felling, burning, uprooting, and pesticide application on infected trees. Additionally, they have implemented insect-proof screens (IPS) on newly planted citrus trees to curb the spread of HLB, all of which have significantly impacted citrus orchard landscapes. Primarily, a considerable number of trees have been felled, leading to reclamation and renovation of orchards. This practice has exacerbated soil erosion in the southern hilly regions [8] and escalated greenhouse gas emissions [9]. Subsequently, post-clearance, some orchards remain barren, while others have been abandoned, transforming into overgrown bush. New citrus trees have been cultivated in certain areas, shielded by IPS. These disparate landscape changes following HLB’s spread have resulted in unprecedented complexity and spatial heterogeneity in citrus orchards, further fragmenting the landscape pattern. Moreover, the widespread use of IPS, primarily composed of polyethylene, to combat HLB, poses challenges in recycling and degradation. The substantial increase in IPS usage potentially leads to issues such as white pollution and soil contamination. Therefore, precise comprehension of temporal and spatial changes in citrus orchards post-HLB outbreak holds paramount importance in understanding the current state of the citrus industry, evaluating its impact on the ecological environment, and assessing associated effects.

The current estimation of citrus orchard acreage primarily relies on manual surveys or on-site measurements, incurring substantial costs annually while struggling to ensure data accuracy. Remote sensing technology, owing to its timeliness and objectivity, offers the ability to extract and monitor crop planting areas on a large and continuous scale [10]. Research on monitoring citrus orchards through remote sensing has mainly focused on the selection of remote sensing data sources, image characteristics, and classification methods. Regarding data sources, extraction utilizes Landsat-8 [11,12], Sentinel-2 [13], CBERS-2B [14], GF-2 [15], among other datasets. Feature selection involves employing texture features [16,17], vegetation indices, water indices [11,12,13], and terrain characteristics [11,12]. In terms of classification methods, machine learning algorithms such as decision trees [18], support vector machines [19], and random forests [20] have been primarily used for extraction, all yielding satisfactory results. However, current studies on citrus orchard extraction mainly focus on feature extraction of orchard land without IPS coverage, neglecting to include IPS in the extraction scope. This oversight may lead to distorted actual citrus orchard acreage statistics. To accurately extract the distribution and area of citrus orchards under IPS coverage, researchers have leveraged the physical similarity between IPS and plastic film mulching. Based on Sentinel-2 imagery [21], scholars have introduced the plastic-mulched landcover index (PMLI) [22], a specific spectral index, achieving favorable results in remote sensing extraction of citrus orchards covered by IPS. This method effectively reflects the spatial distribution of citrus orchards under IPS coverage.

In recent years, the study of the spatiotemporal dynamics and influencing factors of citrus orchards has emerged as a focal point. Presently, spatiotemporal dynamic research predominantly employs methods such as the Location Quotient, Moran’s Index [23], Industrial Centroids [24], hot-spot/cold-spot analysis [25], concentration index, and landscape pattern indices. The coverage of IPS may exacerbate the fragmentation of agricultural landscapes. Analyzing the spatiotemporal pattern changes in citrus orchard land with and without IPS coverage contributes to evaluating the stability of the study area’s ecosystem. It further unveils complex human–environment relationships, providing data support for optimizing land-use planning and formulating sustainable land-use strategies and management plans [26]. Studies on the influencing factors of spatiotemporal changes in citrus orchards currently utilize methods such as spatial autocorrelation [27], spatial econometric models, geographic weighted regression models [28], and geographical detectors [29]. Factors considered primarily involve natural elements such as topography, temperature, and proximity to water sources. The construction of IPS is closely related to human activities, with transportation accessibility being a key human activity influencer. Hence, this study further explores the relationship between IPS and human factors such as distance to roads and impervious surfaces. These factors’ relationship with citrus orchards often displays nonlinearity and complexity, challenging traditional empirical statistical methods to explain their correlations [30]. Random forest regression, as an advanced machine learning algorithm, can rank the relative importance of various variables [31]. It exhibits stronger analytical and explanatory capabilities regarding influencing factors on citrus orchards. Therefore, it is employed to explore the underlying mechanisms and distribution patterns of citrus orchards under IPS coverage.

To accurately delineate the spatial distribution of citrus orchards and IPS in Xunwu County over recent years, this study aimed to utilize Sentinel-2 satellite remote sensing data. Texture features, spectral indices, and terrain characteristics were employed to establish feature sets. Classification was conducted using random forest to map land-use alterations resulting from HLB impact between 2017 and 2022. In investigating the repercussions of HLB on citrus cultivation, as well as summarizing recent trends in citrus orchard development, spatial data analysis, hot-spot analysis, landscape pattern analysis, and random forest regression analysis are proposed. The study aimed to furnish a reliable data foundation and technical support for sustainable development in the local citrus industry. Furthermore, it sought to offer rational suggestions for optimizing citrus orchard planting structures, pest and disease control, and ecological environment management in the future.

2. Materials and Methods

2.1. Study Area

Xunwu County is situated in the southern part of Jiangxi Province, at the southeastern end of Ganzhou City. It shares boundaries with Fujian, Guangdong, and Hunan on its east, south, and west sides respectively, and is located between 24°30′40″ to 25°12′10″ north latitude and 115°21′22″ to 115°54′25″ east longitude. Comprising 14 townships, the county covers a total area of 2351 km² (Figure 1).

Xunwu County falls within the subtropical humid climate zone, characterized by moderate temperatures, abundant heat, ample rainfall, a lengthy frost-free period, an average annual temperature of 18.9 °C, and an average annual sunshine duration of 1823.8 h. The region experiences significant diurnal temperature variations and a rainy–hot concurrent climate, receiving an average annual rainfall of 1650.3 mm. Dominated by red soil, the county features predominantly hilly terrain with limited fields. The topography mainly exhibits intermediate lowlands with higher elevations in the northwest and southeast, offering favorable geographic, soil, and climatic conditions for citrus cultivation.

Ganzhou City in Jiangxi Province serves as a vital citrus production base in China, with Xunwu County being one of the key origins of citrus in southern Jiangxi. Since the 1970s, policies such as “Fruit Development to Enrich People” and the cultivation of a “Clustered Industry surpassing 10 billion Yuan” proposed by the municipal government have significantly propelled the rapid development of the citrus industry in Xunwu County. Citrus has become a pivotal sector driving rural economic rejuvenation. However, around 2010, due to declining profitability in the citrus industry, the number of mismanaged orchards increased, leading to a drastic rise in populations of the Asian citrus psyllid. Consequently, from 2012 to 2016, a severe outbreak of HLB significantly impacted the citrus industry in Xunwu County [32]. According to the statistical bulletin on the county’s economic and social development issued by the Xunwu County People’s Government (Figure 2), during the period from 2012 to 2016, both orchard acreage and citrus production witnessed a significant decline. Orchard acreage decreased from 449,500 mu to 342,400 mu, while citrus production dropped from 295,100 tons to 223,800 tons. Prevention and control measures were then implemented, resulting in an increase in citrus planting areas in 2017. Although HLB has been systematically removed, IPS use has continued for a new round of citrus breeding and protection of disease-free fruit trees. The orchard area covered by IPS has gradually increased since 2017. Consequently, two distinct types of orchard landscapes have emerged: healthy and open-air citrus orchards (Figure 3a), and citrus orchards covered with IPS (Figure 3c).

2.2. Data Source

2.2.1. Land Cover Reference Data

In Xunwu County, citrus orchards exhibit two distinct forms: one is uncovered, exposing the citrus trees, while the other is covered with IPS. In our land cover classification, these were respectively labeled as “No IPS” and “IPS”. Combining the current local land use, the study area was divided into eight primary land cover types: impervious surfaces, forest, water, cropland, No IPS, bush, bare land, and IPS.

We used Google Earth Pro software to visually interpret high-resolution images and marked the 2020 samples on them. Since certain land cover types exhibit limited changes over short periods (such as forest and bush in mountainous regions, rivers, and urban areas), we used the 2020 samples as a reference, adjusting them with imagery from other years available on Google Earth to acquire samples for each year. In cases where Google Earth’s imagery was unavailable for certain years, Sentinel-2 original images were used as substitutes for visual interpretation. To ensure the credibility of the classification results, the number of samples for each year was determined based on the approximate area proportions of different land cover types within the research area. The sample points within the study area between 2017 and 2022 were collected (

Table 1. Number of sample points in Xunwu County from 2017 to 2022.

	2017	2018	2019	2020	2021	2022
Impervions	145	163	163	182	175	181
Forest	799	799	788	787	787	781
Water	127	123	121	131	131	131
Cropland	384	409	409	463	452	452
No IPS	512	412	369	376	357	355
Bush	310	322	310	282	283	283
Bare land	176	188	209	115	137	137
IPS	51	90	174	203	250	252
Total	2504	2506	2543	2539	2572	2572

). We allocated 70% of these points for training and the remaining 30% for validation.

2.2.2. Sentinel-2

Compared to the Landsat and SPOT series satellite imagery, Sentinel-2 satellite data possess 13 multispectral bands with a maximum spatial resolution of 10 m and a revisit period of 5 days, making the data more advantageous for vegetation cover classification. Additionally, the data provide continuous observations of vegetation cover types. The image data used in this study, Sentinel-2A (dataset ID: “COPERNICUS/S2_SR”), were acquired from the GEE platform (accessed on 25 January 2023, https://earthengine.google.com). Due to the high cloud cover and rainfall in the region, a cloud threshold of less than 1% was chosen to screen the images. The entire year’s images were initially processed to remove cloud cover, and then a median compositing method was applied to create a single image for each year. This procedure resulted in a total of 6 composite Sentinel-2 images from 2017 to 2022, each representing the median value for the respective year. The number of images utilized for each year is outlined in

Table 2. The number and date of Sentinel-2 images used for the lowest cloud composite.

Year	Number of Images	Path/Row	Image Acquisition Date
2017	10	T50RLP	11 February 2017, 2 April 2017 3 November 2017, 18 December 2017
		T50RLN	11 February 2017, 2 April 2017, 20 August 2017 24 October 2017, 3 November 2017, 18 December 2017
2018	8	T50RLP	23 March 2018, 4 October 2018, 29 October 2018
		T50RLN	12 January 2018, 23 March 2018, 4 October 2018 29 October 2018, 18 December 2018
2019	19	T50RLP	22 January 2019, 27 January 2019 20 August 2019, 19 September 2019 24 September 2019, 19 October 2019 8 November 2019, 23 November 2019 3 December 2019, 8 December 2019, 13 December 2019
		T50RLN	22 January 2019, 19 September 2019 24 September 2019, 8 November 2019 23 November 2019, 3 December 2019 8 December 2019, 13 December 2019
2020	9	T50RLP	21 February 2020, 16 April 2020, 23 October 2020 12 November 2022, 22 December 2022
		T50RLN	21 February 2020, 16 April 2020 12 November 2022, 22 December 2022
2021	17	T50RLP	1 January 2021, 15 February 2021, 20 February 2021 27 November 2012, 2 December 2021, 7 December 2021
		T50RLN	1 January 2021, 21 January 2021 31 January 2021, 5 February 2021 15 February 2021, 20 February 2021 1 April 2021, 12 November 2021 27 November 2021, 2 December 2021, 7 December 2021
2022	11	T50RLP	13 September 2022, 13 October 2022 23 October 2022, 22 December 2022
		T50RLN	3 September 2022, 13 September 2022 18 September 2022, 13 October 2022 23 October 2022, 22 December 2022, 27 December 2022

.

2.2.3. Other Data

The other data utilized in this study primarily include terrain data, land surface temperature data, road data, and administrative boundary data. The terrain data were obtained from GEE in the form of a 30 m resolution SRTM DEM (dataset ID: “USGS/SRTMGL1_003”). Land surface temperature data were derived from Landsat 8 TIRS data provided by the GEE platform, calculated using a single window algorithm. Road data were acquired from OpenStreetMap (accessed on 25 January 2023, https://extract.bbbike.org). Administrative boundary data for the townships in Xunwu County were downloaded from the Geographical Information Monitoring Cloud Platform (accessed on 25 January 2023, http://www.dsac.cn/).

2.3. Methods

2.3.1. Classification Feature Extraction

Xu Hanzeyu and his team conducted research on citrus remote sensing extraction methods based on spectral and terrain characteristics. Their efforts resulted in an overall accuracy of 93.15% and a kappa value of 0.9. Their findings demonstrated significant differences in reflectance among land cover types such as forest, water, cropland, bare land, and impervious surfaces, particularly in bands B4, B8, and B11. Extracting the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and modified normalized difference water index (MNDWI) proved effective in classifying these land cover types. Citrus orchards without IPS coverage exhibited soil background color due to plant spacing, emphasizing their spectral characteristics using the soil-adjusted vegetation index (SAVI). Yu et al. observed that in IPS-covered areas, orchards with IPS showed higher reflectance in visible light bands compared to those without IPS, while near-infrared and shortwave infrared bands, especially B11, displayed similar values. Consequently, the researchers introduced the plastic mulch index (PMLI), derived from normalizing red band (B4) and shortwave infrared (B11) data, achieving favorable results in distinguishing citrus orchards with and without IPS. Additionally, in orchards without IPS, trees followed a regular pattern forming distinct stripe-like features. However, after IPS installation, orchards appeared more uniform, masking this pattern and resulting in differences in texture features in the images. Therefore, texture features were incorporated to better differentiate citrus orchards with and without IPS. Citrus, being a sun-loving and warm-temperature plant, requires specific elevations to ensure accumulated temperature, a certain slope for drainage, and adequate sunlight on the sunny slope. Introducing terrain features helped better distinguish citrus orchards from other land cover types. Furthermore, elevation and slope characteristics could differentiate between IPS and vegetable greenhouses in valley plains. Overall, we obtained four sets of 24 features, which are detailed in

Table 3. All image features for classification.

Feature Type	Abbreviation	Feature Name	Formula	Reference
Spectral bands	B2	Blue
	B3	Green
	B4	Red
	B5	Red-edge 1
	B6	Red-edge 2
	B7	Red-edge 3
	B11	SWIR1
	B12	SWIR2
Conventional spectral index	NDVI	Normalized difference vegetation index	(NIR − R)/(NIR + R)	[33]
	EVI	Enhanced vegetation index	[2.5 × (NIR − R) − NIR]/(NIR + 6 × R − 7.5 × B + 1)	[34]
	MNDWI	Modified normalized difference water index	(G − MIR)/(G + MIR)	[35]
	SAVI	Soil-adjusted vegetation index	[1.5 × (NIR − R)]/(NIR + R + 0.5)	[36]
	PMLI	Plastic-mulched landcover index	(SWIR1 − Red)/(SWIR1 + Red)	[22]
Textural features	SAVG	Sum average	${\Sigma }_{\mathrm{i},j=0}^{N-1}ip\left(\mathrm{i},j\right)=u$
	VAR	Variance	${\Sigma }_{\mathrm{i},j=0}^{N-1}\mathrm{p}\left(\mathrm{i},j\right) {\left(\mathrm{i}-u\right)}^2$
	IDM	Inverse difference moment	${\Sigma }_{\mathrm{i},j=0}^{N-1}\frac{P\left(\mathrm{i},j\right)}{1+{\left(\mathrm{i}-j\right)}^2}$
	CON	Contrast	${\Sigma }_{\mathrm{i},j=0}^{N-1}\mathrm{p}\left(\mathrm{i},j\right) {\left(i-j\right)}^2$
	DISS	Dissimilarity	${\Sigma }_{\mathrm{i},j=0}^{N-1}\mathrm{p}\left(\mathrm{i},j\right) \left|i-j\right|$
	ENT	Entropy	${\Sigma }_{\mathrm{i},j=0}^{N-1}-p\left(\mathrm{i},j\right)log\left(p\left(\mathrm{i},j\right)\right)$
	ASM	Angular second moment	${\Sigma }_{\mathrm{i},j=0}^{N-1}\mathrm{p}{\left(\mathrm{i},j\right)}^2$
	CORR	Correlation	${\Sigma }_{\mathrm{i},j=0}^{N-1}\frac{\mathrm{ij}p \left(\mathrm{i},j\right)-{\mu }_x{\mu }_y}{{\sigma }_x{\sigma }_y}$
Topographic-temperature features	DEM	DEM
	SLOPE	SLOPE
	Aspect	Aspect

.

2.3.2. Random Forest

Random forest is a machine learning algorithm that aggregates predictions from multiple individual decision trees [37]. These trees are trained on different subsets of data using bootstrap aggregating, commonly known as bagging, aiding in mitigating overfitting and enhancing the overall model accuracy. The algorithm introduces randomness in the selection of data points for training each tree and in the feature consideration during tree construction. This stochastic nature fosters diversity among trees, reinforcing the model’s ability to generalize to new data. Consequently, it is acknowledged for demonstrating robust performance in handling multi-dimensional classification features, finding extensive applications in domains such as land-use classification mapping [38,39,40]. In this study, leveraging the GEE platform, a classification approach employing 100 decision trees was applied to input feature sets for each target year, aiming to attain reliable classification results.

2.3.3. Accuracy Assessment

Accuracy assessment refers to the statistical analysis of the results obtained after image classification, aiming to assess the quality of the classification outcomes and the reliability of the classification method. In this study, the classification accuracy was evaluated via computing the overall accuracy (OA), Kappa coefficient, and confusion matrix.

2.3.4. Landscape Pattern Analysis

Landscape pattern involves the quantification and analysis of spatial distribution, shapes, sizes, and interrelationships among different elements (such as forest, water, impervious surfaces, and cropland) on the Earth’s surface. It provides structural information about land cover and reveals the changing patterns of surface landscape elements. Following the outbreak of HLB, various landscape types within citrus orchards in Xunwu County became intermingled spatially, exacerbating the fragmentation of the citrus orchard landscape pattern. Therefore, investigating the structural characteristics and evolutionary trends of citrus orchard landscapes over an extended period is crucial. This includes analyzing the impact of citrus orchard changes post-HLB outbreak on the landscape pattern of Xunwu County. In this study, Fragstats 4 software was employed, utilizing patch number, mean patch size, largest patch index, and area-weighted mean shape index as indicators to assess changes in the landscape pattern of citrus orchards in Xunwu County.

2.3.5. Random Forest Regression Analysis

Random forest regression, initially proposed by Breiman, is a novel machine learning algorithm based on classification and regression decision trees. It consists of multiple decision trees, each serving as a classification or regression model. By recursively splitting input data features, it generates a predictive outcome. The final prediction is obtained by averaging the predictions of multiple decision trees. Compared to other regression models, random forest offers advantages such as high accuracy, reduced susceptibility to overfitting, and robust handling of high-dimensional data. It can also be used to assess and rank the importance of feature variables. This study adopts the random forest regression analysis method to quantitatively analyze and rank the factors influencing the distribution of citrus orchards under IPS coverage.

3. Results

3.1. Accuracy Assessment of Citrus Orchard Remote Sensing Extraction

We conducted accuracy assessments on image classification results from 2017 to 2022 using 30% of sample points chosen for validation each year. Figure 4 illustrates an average overall accuracy (OA) of 88.62% and an average Kappa coefficient of 0.85, affirming the reliability of the classification results. The fluctuation in precision across these years mainly occurs due to variations in image and sample quality. This is coupled with variability introduced during the feature and sample selection in the classification process driven by the random forest algorithm. These fluctuations exhibit a trend of variability, with standard deviations of 1.24% for OA and 0.02% for the Kappa coefficient. The accuracy of the classification results across each year exhibits a certain level of stability.

3.2. Analysis of Citrus Orchard Area Dynamic Changes from 2017 to 2022

3.2.1. Land-Use Analysis in Xunwu County

By comparing the classification results from 2017 to 2022 in the study area (Figure 5), it is observed that the largest land cover in the entire county is forest, mainly distributed in the southern-central region of Xunwu County. Following this, the citrus orchards without IPS are predominant, primarily situated in the hilly areas of the central and northern parts of Xunwu County. The spatial distribution of citrus orchards with and without IPS shows a significant overlap, appearing intermixed. However, orchards with IPS are concentrated more prominently in the northern hilly regions. In recent years, their distribution has increasingly concentrated, notably evident by distinctive red patches representing orchards with IPS in the northern region of Xunwu County. This observation indicates a clear expansion trend in orchards with IPS in recent years.

A further analysis of the proportional changes in land-use types within the study area (Figure 6) reveals that in 2017, the primary land-use types in Xunwu County were forest and citrus orchards without IPS. Due to a significant outbreak of HLB, the area of citrus orchards without IPS rapidly declined from 621.55 km² in 2017 to 301.29 km² in 2022. Conversely, the most substantial growth occurred in citrus orchards covered by IPS, rising from 7.82 km² in 2017 to 111.39 km² in 2022. The most notable increase, by 85.87%, was observed between 2019 and 2020 in the IPS-covered orchards, predominantly originating from areas previously occupied by citrus orchards without IPS, cropland, and bare land. Except for a slight increment in cropland, no other land-use types exhibited significant fluctuations in area.

3.2.2. Analysis of Temporal Trends in Citrus Orchards

The area of citrus orchards in the study region exhibited a declining trend from 2017 to 2022 (Figure 7), witnessing an overall reduction of 216.70 km². This decline primarily occurred in citrus orchards without IPS, which decreased by 320.27 km². The areas previously occupied by these diminished orchards predominantly transitioned into cropland, bush, and citrus orchards covered by IPS (Figure 8). Notably, the most significant transitions occurred between 2017 and 2018, with a shift of 71.88 km² to cropland, and between 2018 and 2019, where transitions to bush and orchards with IPS were prominent, accounting for 39.07 km² and 16.12 km², respectively. Orchards under IPS showed a considerable expansion, increasing by 103.57 km². These expansions mostly derived from areas previously occupied by orchards without IPS, cropland, and bare lands (Figure 9). Particularly, the transitions from orchards without IPS and cropland to orchards under IPS were most evident between 2018 and 2019, amounting to 16.12 km² and 10.38 km², respectively. Furthermore, the transition from barren land to orchards under IPS was notably prominent in 2019–2020, accounting for 1.06 km². Despite an overall reduction in citrus orchard area in Xunwu County, the rate of decline slowed between 2018 and 2021, showing instances of resurgence. This suggests that the use of IPS has to some extent limited the spread of HLB, thereby mitigating losses in citrus orchards.

3.2.3. Spatial Changes Trend Analysis of Citrus Orchards

By comparing the classification results of Xunwu County in 2017 and 2022, a quantitative analysis of the transformation between various land classes within the study area was conducted (Figure 10). During the period from 2017 to 2022, the area undergoing land type transitions accounted for 26.68% of Xunwu County’s land area. Among these transitions, the conversion of uncovered citrus orchards (19.62%) to forest was mainly distributed in the peripheral areas of the orchards. The transformation of uncovered citrus orchards (18.81%) to cropland was predominantly located in the plains or valleys near water bodies in the central part of Xunwu County. The shift from uncovered citrus orchards (12.83%) to covered ones was mainly observed in the hilly areas of the central and northern parts of Xunwu County, with a more concentrated occurrence in the northern hilly regions. It is speculated that citrus orchards in the northern hilly areas have been more severely affected by HLB.

3.3. Citrus Orchard Spatial Layout Changes Analysis

Analyzing how citrus orchard layouts change gives important insights to improve their production spaces. In this study, we utilized hot-spot analysis to explore the spatiotemporal distribution features of citrus orchards in Xunwu County and investigated the migration of the spatial distribution center under IPS coverage through the IPS concentration index.

3.3.1. Spatiotemporal Distribution Characteristics of Citrus Orchards

To understand the spatial clustering of citrus orchards, a six-year dataset (2017–2022) of citrus orchards was overlaid using ArcGIS for hot-spot analysis. The results indicate a significant geographical clustering of citrus orchards, showcasing an overall spatial structure of “colder in the south and hotter in the north”. Uncovered citrus orchards are widely distributed (Figure 11a), mainly concentrated in the hilly areas of the central and northern regions of Xunwu County, specifically in Wenfeng Township, Chengjiang Town, and Jitan Town. These areas, with moderate elevation, adequate sunlight and accumulated temperature, proximity to water bodies, a certain slope aiding drainage and irrigation, and near residential areas, facilitate management and operations. Additionally, a small number of orchards are present in the southwestern and eastern mountainous valley regions. Covered citrus orchards exhibit a more concentrated distribution (Figure 11b), primarily in the northern part of Xunwu County, with the majority clustered in Chengjiang Town. The local warm and humid climate in this area favors the reproduction of Diaphorina citri and pest activity, thereby increasing the risk of citrus orchard infections. Moreover, the relatively developed citrus industry in the area with extensive fruit tree plantations provides more opportunities for disease spread. Hence, local efforts should strengthen monitoring and pest control to curb the spread of HLB.

3.3.2. Analysis of IPS Concentration Index

The IPS concentration index refers to the proportion of citrus orchard areas covered by IPS in the total net-covered area within a specific region during a certain period. This index is pivotal in portraying the spatial layout variations of net coverage from both temporal and spatial perspectives. A higher concentration index indicates a significant concentration of net-covered citrus orchards in a specific area during the period, signifying the spatial gravity center of net coverage within the entire study area. The calculation formula for the IPS concentration index is expressed as:

$${\mathrm{P}}_{\mathrm{i}}={\mathrm{S}}_{\mathrm{i}}/\mathrm{S}$$

Here, P_i represents the concentration index of town i, S_i denotes the area of net-covered citrus orchards in town i, and S represents the total area of net-covered citrus orchards in the study area. The computed IPS concentration indices for each town within the research area from 2017 to 2022 are presented in

Table 4. The IPS concentration index in each township.

Township	2017	2018	2019	2020	2021	2022
Sanbiao	0.0486 (7)	0.0324 (9)	0.0258 (10)	0.0231 (10)	0.0302 (11)	0.0281 (8)
Jitan	0.0608 (6)	0.0978 (3)	0.1507 (2)	0.1554 (3)	0.1503 (2)	0.1709 (2)
Chenguang	0.0713 (4)	0.0413 (6)	0.0401 (7)	0.0412 (8)	0.0479 (6)	0.0329 (6)
Guizhumao	0.0449 (9)	0.0284 (10)	0.0270 (9)	0.0271 (9)	0.0319 (9)	0.0226 (11)
Nanqiao	0.0912 (3)	0.0591 (4)	0.0488 (5)	0.0566 (4)	0.0510 (5)	0.0301 (7)
Chengjiang	0.2033 (2)	0.3755 (1)	0.3623 (1)	0.3194 (1)	0.3051 (1)	0.3555 (1)
Longting	0.0094 (13)	0.0131 (14)	0.0150 (13)	0.0123 (14)	0.0163 (14)	0.0079 (14)
Xiangshan	0.0087 (14)	0.0168 (12)	0.0223 (12)	0.0184 (12)	0.0215 (12)	0.0232 (10)
Luoshan	0.0212 (12)	0.0268 (11)	0.0463 (6)	0.0439 (7)	0.0464 (8)	0.0468 (5)
Wenfeng	0.2600 (1)	0.1734 (2)	0.1316 (3)	0.1612 (2)	0.1431 (3)	0.1538 (3)
Changpu	0.0236 (11)	0.0158 (13)	0.0134 (14)	0.0167 (13)	0.0195 (13)	0.0116 (13)
Danxi	0.0418 (10)	0.0369 (7)	0.0240 (11)	0.0228 (11)	0.0304 (10)	0.0137 (12)
Liuchu	0.0694 (5)	0.0467 (5)	0.0394 (8)	0.0451 (6)	0.0466 (7)	0.0249 (9)
Shuiyuan	0.0460 (8)	0.0361(8)	0.0532 (4)	0.0566 (5)	0.0599 (4)	0.0778 (4)

Numbers in parentheses represent the ranking of IPS concentration index for each township.

.

From Table 4, the changes in the pest control net concentration index from 2017 to 2022 can be categorized into four types:

Stable Type: Regions where the pest control net concentration index remained relatively consistent across different periods. This includes five townships: Chengjiang, Wenfeng, Longting, Guizhumaomao, and Changpu. Among them, Longting, Guizhumaomao, and Changpu maintained their index ranking within the top two positions from 2017 to 2022, while Chengjiang alternated between the first and second positions, indicating consistent concentration in these areas. Regions in a stable phase require the government’s focused attention, emphasizing the necessity of maintaining regular pest monitoring and preventing potential plastic and soil pollution.

Increasing Type: Areas where the pest control net concentration index showed an increase in the later years compared to the initial years. This encompasses four townships: Xiangshan, Luoshan, Shuiyuan, and Jitan. These townships saw their index rankings rise from positions 14, 12, 8, and 6 in 2017 to positions 10, 5, 4, and 2 in 2022, respectively. In regions experiencing increasing trends, HLB might not be entirely eradicated, posing a risk of further spread, necessitating strengthened government efforts in detecting and managing HLB.

Decreasing Type: Regions where the pest control net concentration index consistently decreased in the later years compared to earlier years. This category includes three townships: Chenguang, Liuche, and Nanqiao. These townships observed a decline in their index rankings from positions 4, 5, and 3 in 2017 to positions 6, 9, and 7 in 2022, respectively. For areas in a decreasing phase, authorities should monitor potential trends of local farmers abandoning orchards, providing industry support and guiding proactive measures against HLB.

Fluctuating Type: Areas where the pest control net concentration index showed significant fluctuations between 2017 and 2022. This mainly pertains to Sanbiao and Danxi Townships. Sanbiao Township was ranked 7th in 2017, dropped to 11th in 2021, and rose again to 8th in 2022. Danxi Township shifted from 10th in 2017, rose to 7th in 2018, and then dropped to 12th in 2022. In fluctuating regions, dynamic management strategies are essential for adjusting citrus orchard management based on specific circumstances.

Observations of IPS distribution across various townships in Xunwu County (Figure 12) indicate a gradual shift in the spatial centroid of citrus orchards with IPS from the central to the northeastern parts. Initially concentrated in Chengjiang Township and Wenfeng Township, the coverage expanded to surrounding areas. This shift reflects an adjustment in the county’s planting structure, signifying the spatial evolution of HLB prevention strategies. With these changes, orchard management needs greater flexibility to adapt. Effective management and monitoring of the expanding coverage area are crucial to ensure comprehensive implementation of disease prevention and control strategies. The widespread adoption of IPS in adjacent townships demonstrates the feasibility and effectiveness of this control measure. It also underscores farmers’ consensus and commitment to combating HLB, reflecting, to some extent, a pattern of balanced and collective development across Xunwu County. The local government needs to enhance communication and collaboration between townships, improve resource allocation, and collectively develop and implement a more comprehensive disease management plan. This aims to more effectively and comprehensively manage the HLB relief strategy county-wide, better addressing the dynamic changes in HLB within the county.

3.4. Analysis of Citrus Orchards’ Landscape Pattern Changes

To investigate the landscape changes between two categories of citrus orchards—those with and without IPS coverage in Xunwu County—we utilized ArcGIS to extract the land cover patches, computed various landscape indices, and quantitatively analyzed the long-term landscape changes in citrus orchards. The changes in patch numbers and average patch sizes for these two types of landscapes are presented in

Table 5. Changes in the number and average patch area of citrus orchard.

	Patch Numbers		Average Patch Sizes (m2)
	No IPS	IPS	No IPS	IPS
2017	79,134	22,759	7100	300
2018	106,780	23,394	4300	400
2019	112,216	97,644	3500	500
2020	129,410	129,974	2600	700
2021	150,739	139,544	2300	800
2022	143,228	110,652	1900	900

.

The patch number of citrus orchards without IPS coverage showed an overall increasing trend from 2017 to 2022. The most significant increase occurred between 2017 and 2018, followed by a steady rise until reaching a peak in 2021 and subsequently declining. Simultaneously, the average patch size consistently decreased, hitting its lowest value in 2022. These findings illustrate a continuous decline in the area of citrus orchards without IPS coverage, coupled with a considerable fragmentation of patches.

In contrast, citrus orchards under IPS coverage witnessed a rapid growth in patch numbers from 2017 to 2022, with the fastest growth observed between 2018 and 2019. The number of patches reached its peak in 2021 and displayed a decreasing trend between 2021 and 2022. Additionally, the average patch size consistently increased, reaching its maximum value in 2022. As the coverage of IPS expanded, there was an improvement in patch fragmentation levels.

The maximum patch index and area-weighted average shape index reflect the aggregation and dominance of landscape patch shapes. According to Figure 13 and Figure 14, citrus orchards without IPS coverage show a marked decline in both the maximum patch index and area-weighted mean shape index, while those with IPS coverage exhibit a slow overall increase in these indices. Throughout 2017 to 2022, a significant number of citrus orchards were covered by IPS to prevent the spread and diffusion of HLB. This led to a gradual concentration and contiguous expansion of the IPS coverage, significantly enhancing its aggregation and dominance within the land use of Xunwu County. In contrast, citrus orchards without IPS coverage transformed from a concentrated and contiguous distribution to a more fragmented layout. This resulted in an increase in the number of patches, a decrease in the average patch size, and a gradual dispersal of patch shapes, visibly diminishing their dominance.

Overall, the citrus orchard landscape in Xunwu County continues to exhibit a trend towards fragmentation, potentially heightening the local ecosystem’s vulnerability. This fragmented landscape, compared to the previously uniform and cohesive citrus orchards, struggles to provide adequate ecological functions and services. Moreover, it is prone to soil erosion and land degradation, exerting adverse effects on soil quality and water resources. Furthermore, this fragmentation extends beyond individual orchards and may result in reduced biodiversity and disrupted ecological balance, impacting local plant and animal communities. Agriculturally, fragmentation can restrict orchard crop growth, cause unstable yields, and potentially lead to issues such as pest infestation. Managing and operating fragmented citrus orchards also pose increased challenges, raising farmers’ management costs and resource commitments. Local governments could formulate policies encouraging land consolidation, promoting the amalgamation of citrus orchards covered by IPS to create larger, contiguous orchard areas. Simultaneously, providing technical support and subsidies could incentivize farmers to participate in land consolidation and orchard optimization initiatives.

3.5. Analysis of Factors Influencing the Distribution of Citrus Orchards under IPS Coverage

To explore the influencing factors behind the distribution of citrus orchards under IPS coverage, we selected five natural factors—aspect, slope, elevation, temperature, and distance to water source—as well as two anthropogenic factors—distance to roads and impervious surfaces. We conducted an importance analysis using random forest regression to assess the significance of these seven categories of influencing factors. The results, as shown in

Table 6. Ranking of the importance of each impact factor.

Ranking	Factors	Importance
1	aspect	23.4%
2	temperature	20.3%
3	distance from road	19.4%
4	slope	12.5%
5	elevation	12.3%
6	distance from water	9.5%
7	distance from impervious surfaces	2.6%

, highlight aspect and temperature as the primary factors.

A quantitative analysis was conducted on the aforementioned influencing factors, as depicted in Figure 15. Citrus orchards covered by IPS are predominantly situated in the low-altitude zones ranging from 300 m to 400 m. The growth of citrus fruits necessitates specific heat accumulations; therefore, the low temperatures at higher altitudes are unfavorable for citrus growth, and the topography at higher elevations is unsuitable for constructing IPS. Given citrus plants’ preference for sunlight, orchards covered by IPS are mainly distributed in the southeastern, southwestern, and southern regions. Additionally, these orchards are predominantly found on slopes ranging from 10° to 15°, favoring gentle slopes for adequate drainage, a crucial aspect for citrus cultivation. Optimal surface temperatures for citrus orchards covered by IPS range from 18 °C to 26 °C. Interestingly, their distribution does not exhibit a significant correlation with proximity to water sources.

Regarding distance, the areas covered by IPS in citrus orchards are highest within 0–50 m from impervious surfaces, decreasing as the distance increases. Similarly, concerning roads, the covered areas are highest within 0–500 m, gradually diminishing with distances exceeding 3000 m, eventually displaying minimal coverage. This variation in the distribution of citrus orchards covered by IPS concerning distance could be attributed to differing levels of farmer engagement or land-use planning, signifying the notable influence of human activities on citrus cultivation and the installation of IPS.

4. Discussion

4.1. Discussion of Different Citrus Orchard Analysis Methods

The outbreak of HLB highlighted a previously overlooked critical issue: the impact of IPS on the spectral characteristics of citrus orchards. Upon covering the orchards with IPS, there was a shift in color from light green to gray-white, altering their spectral features. Previous studies primarily focused on extracting features from orchards without IPS, neglecting their impact on spectral characteristics. This oversight might have distorted the actual statistical results concerning the orchard areas. Recognizing this issue, we introduced the plastic mulch index (PMLI) and texture features in citrus orchard extraction, successfully identifying orchards covered by IPS. This approach largely restored the actual distribution of citrus orchards in Xunwu County. Prior studies often utilized low-resolution satellite images (such as Landsat or CBERS-2B) for long-term mapping of citrus orchards or higher-resolution images (such as Sentinel-2) for single-year mapping. However, these methods struggled to simultaneously achieve high-precision results and comprehensive spatiotemporal analysis. Our research employed Sentinel-2 satellite data with a 10 m resolution, generating multi-year high-precision classification results. This method not only accurately reflected the current land use in Xunwu County but also facilitated a better analysis of spatiotemporal trends. Given the spatiotemporal distribution characteristics of citrus orchards after the HLB outbreak, we extensively investigated the migration of orchard distribution under IPS. We provided corresponding management measures and considerations for orchards in different types of townships. Additionally, we conducted landscape pattern analysis of citrus orchards, discussing potential ecological threats posed by fragmented orchards and suggesting relevant solutions for government consideration. This holistic research approach not only offers a fresh perspective for the precise assessment of citrus orchards but also provides practical recommendations for future land-use planning and ecological conservation in Xunwu County.

4.2. Significance of Citrus Orchard Study in the Context of HLB

We utilized satellite imagery with a high spatial resolution of up to 10 m to construct a long-term spatial distribution map of citrus orchards in Xunwu County. This addressed the gap in analyzing spatiotemporal changes in orchards covered by IPS, offering a more comprehensive and accurate depiction of their spatiotemporal characteristics. This comprehensive dataset support aids local governments in refining land-use planning and evaluating the citrus industry’s development more accurately. Simultaneously, it provides a reliable data foundation for quantitatively analyzing the ecological challenges posed by citrus orchards. We precisely delineated the distribution range and area changes in IPS, facilitating local governments in tracing and managing HLB outbreaks. Additionally, considering the limited lifespan of IPS (3–5 years), our analysis of land-use changes in citrus orchards covered by these screens will assist authorities in controlling potential future issues of plastic and soil pollution, thereby supporting environmental conservation efforts.

4.3. Limitations and Future Work

Our research also encounters certain limitations. Firstly, the distribution of IPS does not comprehensively represent areas affected by HLB as its deployment in citrus orchards is closely linked to individual preferences of local farmers or orchard managers. The extensive coverage of IPS demands substantial financial investments, and the use of IPS enclosures in citrus orchards increases labor costs. Additionally, the IPS may have adverse effects on citrus pollination and fertilization processes. Consequently, even in the presence of HLB in citrus orchards, growers might resort to chemical treatments or opt to abandon diseased trees. Future studies could benefit from utilizing drones to identify diseased trees left uncleared, serving as a crucial supplementary method for monitoring HLB.

An analysis of the changes in citrus orchard area indicates a decreasing trend in the area of orchards covered by IPS in Xunwu County in 2022. One potential reason for this might be the lapse in replacing screens upon their expiration, leading to uncovered areas around citrus orchards. Another possibility could be effective local control of HLB, prompting farmers to remove screens to reduce management costs, consequently reducing the area of orchards covered by IPS. These aspects necessitate continuous monitoring and research into Xunwu County’s citrus orchards. With the reduction in the number of screens, it becomes essential to consider the location of these screens and their potential impact on the local ecological environment. Integrating hydrological, soil erosion, and sedimentation data to analyze the impact of orchards covered by IPS on hydrology and ecology is a future research direction.

5. Conclusions

This paper focuses on Xunwu County, Ganzhou City, Jiangxi Province, utilizing the Google Earth Engine platform and a random forest-based citrus image extraction algorithm. It completes remote sensing mapping of citrus orchards from 2017 to 2022, analyzing the spatiotemporal pattern changes and influencing factors. The conclusions are as follows:

(1) Using Sentinel-2 imagery to analyze the occurrence of HLB for citrus orchard extraction resulted in a classification accuracy and Kappa coefficient exceeding 87% and 0.83, respectively, demonstrating commendable classification performance.

(2) The total area of citrus orchards in the study area decreased by 216.70 km². This decline primarily occurred in the area without IPS, which decreased by 320.27 km² and transitioned mainly into cropland, bush, and orchards under IPS. Orchards under IPS exhibited a minor expansion trend, increasing by 103.57 km², sourced primarily from areas without IPS, cropland, and bare land.

(3) The spatial distribution of citrus orchards in the study area exhibits a “hot in the north, cold in the south” pattern. Orchards without IPS are predominantly distributed in the central and northern hilly areas, while orchards with IPS are concentrated in the northern region. Recently, the spatial focus of IPS-covered orchards has shifted northeastward.

(4) From a landscape distribution perspective, orchards without IPS are gradually dispersing, while IPS-covered orchards are tending towards a more concentrated and contiguous distribution. However, the overall situation still indicates a trend towards fragmentation in Xunwu County’s citrus orchards.

(5) Regarding the influencing factors of citrus orchard distribution under IPS, the order of importance from high to low is as follows: aspect > temperature > distance from road > slope > elevation > distance from water > distance from impervious surfaces. IPS-covered orchards are primarily located at elevations of 300 m–400 m, with southeast, southwest, and southern aspects, slopes ranging from 10° to 15°, surface temperatures between 18 °C and 26 °C, and proximity to impervious surfaces and roads, while showing no apparent relationship with distance to water sources.