Spatial Distribution Characteristics and Influencing Factors of Neofusicoccum laricinum in China

Abstract

The long-term spatial–temporal variation in shoot blight of larch in China has not yet been clearly defined, and the mechanisms behind its long-distance spread remain unknown. This study, based on the historical occurrence dataset of shoot blight of larch in China, used spatial statistical analysis to describe the spatial changes in the disease across five stages since 1973. Subsequently, the study utilized Geo Detector and Random Forest models to investigate the relationship between the spread and occurrence of shoot blight of larch and seven influencing factors. The results revealed the following: (1) The spread of shoot blight of larch in China exhibits significant directionality, with the affected regions distributed along a northeast–southwest axis, and the epicenter of the spread is shifting southwestward; (2) Shandong and Jilin provinces served as the initial introduction points for shoot blight of larch, with most infected counties in other provinces experiencing outbreaks between 1989 and 1996, accompanied by a noticeable spread to neighboring provinces; (3) the occurrence of shoot blight of larch demonstrates a significant positive spatial clustering effect, forming a monocentric “core–periphery” structure centered in Liaoning Province, where kernel density values decrease gradually outward from the core. Geo Detector identified “seedling planting area” as a potential spatial driving factor for the disease. These findings underscore the critical influence of the combined effects of human activities and natural factors in shaping the spatiotemporal distribution patterns of shoot blight of larch.

1. Introduction

Larch (Larix) species are deciduous conifers of the Pinaceae family, widely distributed in the permafrost and seasonal frost zones of the Northern Hemisphere [1,2,3,4]. In China, larch is one of the important economic forest species, particularly in the northeastern regions (such as Heilongjiang, Jilin, and Inner Mongolia), where it is planted on a large scale. Its timber is extensively used in construction, papermaking, and furniture manufacturing. Additionally, larch is highly adaptable, grows rapidly, and is commonly used in large-scale shelterbelt construction and barren land afforestation, especially in areas affected by desertification and soil erosion in northern and northeastern China. Despite its high environmental adaptability, cold resistance, and fast growth rate, larch is susceptible to various biotic and abiotic stresses, with shoot blight of larch being a significant stressor impacting this species.

Shoot blight of larch is a fungal disease caused by the infection of Neofusicoccum laricinum (Sawada) Y. Hattori & C. Nakash in larch trees [5]. This disease has a long history of invasion in China, with a wide spread and significant harm. It is one of the two major tree diseases under strict management in China and has been listed in the country’s control list of invasive alien species and harmful forest pests [6,7]. In China, it mainly affects artificial larch forests. The disease was first reported in Japan in 1939 and spread to China in the early 1970s. Since then, it has affected 12 provinces, including Northeast and North China, with an epidemic area covering more than 500,000 ha, posing a huge threat to ecological security in northern China [8,9,10,11]. Additionally, the disease has been reported in Russia, North Korea, South Korea, the UK, Canada, and other countries [12]. This disease is highly contagious and can cause the death of new shoots in larch trees, leading to crown dieback when it recurs year after year. It poses a severe threat to the establishment of larch plantations, especially for young and middle-aged forests aged 6–15 years. Once it invades larch forests, it can deal a devastating blow to tree growth [8,10].

In recent years, scholars have conducted extensive research on this disease, most of which originates from China. For example, Bruda et al. used LC-MS/MS and weighted gene co-expression network analysis to investigate farrerol’s effects on Neofusicoccum laricinum; the research suggests that farrerol enhances disease resistance in larch [13]. Zhang et al. used the optimized Maximum Entropy (MaxEnt) and the Biomod2 ensemble (EM) model to predict the potential geographic distribution areas of shoot blight of larch in China. They found that about 20% of the total land area in China is a potential suitable distribution zone for the disease [14]. Zhou et al. also improved the Stacking model, which can predict infected areas of shoot blight of larch in Northeast China [9]. However, these studies lack an overall explanation of the spatiotemporal patterns of shoot blight of larch and research on its spread and diffusion mechanisms. Therefore, clarifying the spatiotemporal patterns of the disease’s occurrence and development in China and identifying the key factors that influence these patterns are crucial for understanding the disease’s occurrence patterns and for conducting disaster prediction and guiding production control measures.

A suitable external environment is a key factor influencing whether fungi can reproduce and spread widely [15]. Temperature and humidity can significantly affect fungal reproduction, and forests with high canopy density also favor the occurrence and development of tree diseases [8]. The conidia of the pathogen causing shoot blight of larch spread primarily through rainwater splash, which is significantly influenced by precipitation and wind speed [10]. On the other hand, the long-distance spread and diffusion of plant pests and diseases are believed to be closely related to human activities. Increasingly frequent trade of agricultural and forestry products has made the movement of plants and plant products between regions more active [16]. The transplantation of larch seedlings and the transportation of felled larch branches and other tree products can introduce pathogens into new areas. Once the pathogen establishes itself in the new area, it continues to spread through natural transmission and human planting of infected plants [12]. This is suspected to be the primary method of long-distance spread of the disease. However, the long-distance spread of shoot blight of larch remains a hypothesis, lacking sufficient evidence for confirmation.

Random Forest is a classic machine learning algorithm known for handling large datasets. It introduces randomness into the algorithm, which reduces the likelihood of overfitting compared to typical machine learning algorithms [17]. In addition to making predictions, it can rank the importance of various influencing factors and is widely used in fields like ecological protection and environmental studies [18,19]. Despite its strength, Random Forest results often lack sufficient interpretability, and it does not directly reveal interactions between variables. Geo Detector is a statistical method used to analyze spatial heterogeneity and spatial differentiation mechanisms, primarily to explore the relationship between the spatial distribution of a geographic phenomenon and its potential driving factors. It is particularly suited for analyzing spatiotemporal patterns, especially in fields such as geography, epidemiology, and ecology [20,21,22]. Geo Detector reveals differences between regions by detecting spatial heterogeneity between variables based on geographic information. In recent years, Geo Detector has achieved significant success, especially in analyzing the spatial distribution characteristics of diseases, the impact of environmental and socio-economic factors, multi-factor interaction analysis, and predicting disease spread and helping to formulate strategies [23,24]. Combining Geo Detector with Random Forest can provide a deeper and more accurate understanding of spatial analysis of influencing factors.

Therefore, this study uses methods like Geo Detector and Random Forest, focusing on aspects such as the occurrence of shoot blight of larch and the infection process of the pathogen, which have not been considered in previous studies. The research will take as an example all the districts and counties in China where shoot blight of larch is present. By applying new computer modeling techniques, the study aims to comprehensively analyze 18 potential environmental conditions that may influence the spread of this disease. The goal is to explain the true causes of shoot blight of larch on a broader scale, predict the possibility of future disease outbreaks, and propose effective control measures. The main research contents of this paper include the following: (1) a detailed description of the spatiotemporal pattern of the spread and diffusion of shoot blight of larch in China over the past 50 years; (2) clarification of the mechanisms of long-distance spread of shoot blight of larch; and (3) exploration of the combined effects of multiple factors on the spread and diffusion of shoot blight of larch.

2. Materials and Methods

2.1. Focal Species

The pathogen causes shoot blight symptoms by infecting and damaging the tree, and in severe cases, this can lead to stunted growth or even death of the tree [25]. Its distribution is mainly concentrated in temperate regions, particularly in northeastern China, the Russian Far East, and parts of North America, often occurring in areas with dense larch populations [12]. The pathogen spreads through conidia and asexual spores, which are typically formed under moist conditions and dispersed by wind. Higher humidity and precipitation facilitate the release and spread of the spores. The pathogen can also spread to surrounding larch trees via rainwater and wind, leading to rapid regional dispersal.

During the autumn and winter months, the pathogen usually forms large numbers of conidia at the infected lesions, which are released again in spring under warm and humid conditions, starting a new infection cycle. The adaptability of the shoot blight of larch pathogen is reflected in the pathogen’s high dependence on moist environments and its ability to survive on larch trees for extended periods [26]. With climate change, the spread of the pathogen may expand, particularly in warmer and more humid regions [14]. The ecological function of the shoot blight of larch pathogen is mainly as a pathogen affecting the growth and reproduction of larch trees. Its widespread dissemination may lead to the death of larch trees in forests, thereby impacting the overall structure and function of the ecosystem.

2.2. Materials

This study focuses on all counties with larch distribution within the provinces in China where shoot blight of larch occurs. The study area is located in northern China (31°42′–53°33′ N, 91°20′–135°2′ E) and includes 488 counties across 12 provinces: Hebei, Shanxi, Inner Mongolia, Liaoning, Jilin, Heilongjiang, Shandong, Henan, Shaanxi, Gansu, Qinghai, and Ningxia Hui Autonomous Region. The region spans a vast area with significant latitudinal and longitudinal variations, encompassing diverse topographies such as plateaus, plains, hills, and mountains. The climate types are equally varied, including temperate monsoon, subtropical monsoon, and temperate continental climates.

The baseline county-level data on shoot blight of larch were provided by the National Forestry and Grassland Administration’s Biological Disaster Prevention and Control Center. These data include the annual number, names, and administrative codes of newly affected counties from 1973 to 2021. Larch distribution data were extracted and organized from the “2020 Forest Resource Management Map” by selecting the “dominant tree species” field for “larch species”. This includes species such as Larix gmelinii var. principis-rupprechtii, Larix olgensis, Larix gmelinii, and Larix kaempferi. These data cover counties with larch distribution across the 12 provinces, including their names and administrative codes [27].

Previous studies have shown that the occurrence of shoot blight of larch is significantly correlated with factors such as temperature and humidity, rainfall in June–August, wind speed in May–June, as well as stand age, topographic slope, stand density, tree species, and soil type. In this study, considering the systematic nature of the influencing factors and the availability of data, a total of 18 variables were selected as the influencing factors for the research [28,29]. Meteorological and surface data were obtained from the National Earth System Science Data Center (http://www.geodata.cn (accessed on 28 February 2025)). Monthly precipitation, temperature, and wind speed data from 1901 to 2022 at a 1 km resolution were downloaded from the database. Canopy closure data were derived from the 2020 GLASS (Global Land Surface Satellite) product provided by the same data center [30,31]. Canopy closure refers to the extent to which the tree canopy in a forest covers the ground, serving as an indicator of stand density. The calculation method is typically the ratio of the vertical projection area of the canopy to the forest land area, with no units. It is usually expressed as a decimal, where complete ground coverage is represented by a value of 1.0.

2.3. Spatial Statistical Analysis

2.3.1. Standard Deviation Ellipse

The standard deviation ellipse is a classic spatial statistical method used to reveal the spatial distribution characteristics of geographic features. It measures the direction and distribution of a dataset [32]. The basic parameters of the standard deviation ellipse, such as the centroid, rotation angle, major axis, minor axis, and area, can be calculated using the “Geographic Distribution Measurement” tool in ArcGIS 10.8 (Esri, Redlands, CA, USA). This method helps to depict the centrality, extent, orientation, and spatial morphological characteristics of the infected counties of shoot blight of larch.

2.3.2. Global Moran’s Index

Moran’s index is a spatial autocorrelation measurement method used to describe the degree of dispersion or aggregation among all spatial objects in a study area, as well as the average spatial association, spatial distribution patterns, and significance levels [33]. To assess the spatial autocorrelation of shoot blight of larch across the entire study area, the research presented in this paper utilizes the global Moran’s index. The calculation formula is as follows:

$$I={\displaystyle \frac{N}{W}}{\displaystyle \frac{\sum _{i=1}^N\sum _{j=1}^N{w}_{ij}(x_i-\overline{x})(x_j-\overline{x})}{\sum _{i=1}^N{(x_i-\overline{x})}^2},}$$

(1)

where $N$ is the number of counties in the study area, $i$ and $j$ represent the $i$-th and $j$-th study areas, $x$ is the value of the research target in the study area, $\overline{x}$ is the mean of $x$, $w_{ij}$ is the spatial weight, and $W$ is the sum of the spatial weights. The value of $I$ ranges from [−1, 1]: $I$ > 0 indicates that similar observations are spatially clustered, $I$ < 0 indicates that dissimilar observations are spatially clustered, and $I$ = 0 indicates no spatial autocorrelation, representing a random spatial distribution in the study area.

2.3.3. Kernel Density Estimation

Kernel density estimation is a non-parametric method used to estimate the probability density function of a random variable [34]. It is used to calculate the density in the surrounding areas of the infected counties of shoot blight of larch, identifying the core distribution characteristics of the infected regions. The calculation formula is as follows:

$$F\left(x\right)={\displaystyle \frac{1}{Nd}}{\sum }_{i=1}^{n}K\left({\displaystyle \frac{x-x_i}{d}}\right),$$

(2)

where $F\left(x\right)$ represents the kernel density estimate, $K$ is the Gaussian kernel function, $x$ is the estimation point, $x_i$ is the $i$-th infected county of shoot blight of larch, $d=100$ km is the distance, and $N$ is the number of infected counties within the bandwidth range. The implementation of spatial statistical methods such as the standard deviation ellipse, global Moran’s I index, and kernel density estimation, as well as the map creation, were all completed using ArcGIS 10.8.

2.4. Geo Detector

To examine the correlation between the spatial distribution of shoot blight of larch and various influencing factors, the Geo Detector method was employed to analyze spatial heterogeneity. Geo Detector is a spatial statistical method capable of detecting spatial heterogeneity and uncovering its underlying driving forces. Additionally, it can explore interactions among different factors [21,35]. The calculation formula is as follows:

$$q=1-{\displaystyle \frac{\sum _{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}},$$

(3)

where $L=12$ represents the number of provinces in the study area. $N_h$ and $N$ refer to the number of counties in province $h$ and the entire study area, respectively. ${\sigma }_h^2$ and ${\sigma }^2$ represent the variance in the occurrence of shoot blight of larch within province $h$ and across the entire region. The value of q ranges from [0, 1], indicating the explanatory power of each influencing factor on the spatial heterogeneity of shoot blight of larch. A larger q value indicates a stronger explanatory power of the factor, and its statistical significance can be assessed using the Geo Detector method. In this study, version GeoDetector_1.0-5 was employed (https://cran.r-project.org/web/packages/geodetector/index.html (accessed on 28 February 2025)) and executed within the R version 4.1.2 (R Core Team, Vienna, Austria) software environment.

2.5. Random Forest

Random Forest is an ensemble learning classification algorithm composed of multiple decision trees. The algorithm primarily involves three main steps: random sampling, random feature selection, and majority voting [36]. By introducing a small random subset of variables at each node during the construction of decision trees, each node exhibits randomness in variable selection, while the decision trees themselves are independent of one another. Once parameter optimization is complete, Random Forest determines the importance of each feature by calculating its average influence across various decision trees. Within each tree, the improvement in accuracy (e.g., the reduction in Gini impurity) brought by each feature at the splitting nodes is recorded. Subsequently, the average accuracy improvement for this feature across all trees is calculated. Finally, the importance of features is standardized to facilitate comparison and analysis [37,38]. This level of computation provides insights into the significant impact of each feature on prediction results, enabling feature selection and model optimization. Notably, compared to many existing classification algorithms, Random Forest excels in avoiding overfitting, making it a robust and reliable choice for classification tasks.

The Random Forest regression model, based on Python 3.10 (Python Software Foundation, Wilmington, DE, USA), was constructed using the RandomForestRegressor from the Sklearn machine learning library. The study employed cross-validation and grid search within a defined range to determine the optimal values of the model parameters that yielded the best performance. This research addresses a binary classification problem, and given the application scenario’s high requirements for the model, its performance is assessed using accuracy, recall, precision, and F1 score. Specifically, accuracy refers to the proportion of correctly predicted samples among the total samples; precision represents the proportion of actual positive samples among those predicted as positive by the model; recall indicates the proportion of predicted positive samples among the actual positive samples; and the F1 score is the harmonic mean of precision and recall, serving as a comprehensive indicator that considers both precision and recall. When these four evaluation metrics exceed 80%, the model is considered reliable.

2.6. Random Forest Elimination with Cross-Validation

RFECV, which stands for Recursive Feature Elimination with Cross-Validation, is a commonly used feature selection method that combines RFE (Recursive Feature Elimination) and CV (Cross-Validation) [39]. The feature selection process using RFECV consists of two stages. In the RFE stage, the initial feature set includes all available features. The model is built using the current feature set, and the importance of each feature is calculated. Features with low weights are removed, and the feature set is updated. This process is repeated through multiple iterations until all features are ranked based on their importance. In the CV stage, based on the feature importance determined in the RFE phase, different numbers of features are selected, and cross-validation is performed on the selected feature set. The number of features that results in the highest average score is chosen, completing the feature selection process.

2.7. Statistical Analysis

In this study, the relationship between the spread of shoot blight of larch and larch seedling afforestation was assessed using correlation coefficient statistical methods, including Logistic regression models, Spearman’s rank correlation coefficient, Kendall’s Tau correlation coefficient, and the Mann–Whitney U test, which are classical non-parametric statistical methods.

The Logistic regression model is primarily used to analyze the causal relationship between binary dependent variables and independent variables. Coefficients, odds ratios (ORs), and p-values are used to assess the effect of independent variables on the dependent variable. The size and sign of the regression coefficients reflect the direction and strength of the effect of each independent variable on the dependent variable. OR is obtained by exponentiating the regression coefficients and is always positive, with a range of [0, ∞) [40].

Spearman’s rank correlation coefficient and Kendall’s Tau are used to analyze the monotonic (non-linear) relationship between variables and examine the correlation between independent variables. The results are evaluated mainly by the correlation coefficient value and p-value. The correlation coefficient ranges from −1 to +1, with +1 indicating perfect positive correlation, −1 indicating perfect negative correlation, and 0 indicating no correlation [41].

The Mann–Whitney U test is used to compare the distribution differences between two independent samples and is suitable for non-normal distributions. The results are typically reported using the U statistic and p-value. The U value itself does not have a direct “good or bad” standard but reflects the difference in rank order between the two groups. Smaller U values generally indicate greater differences between the groups [42].

The p-values from all four methods are used to verify statistical significance. If the p-value is less than 0.05, it indicates that the correlation is statistically significant. The construction of the Random Forest and RFECV models, as well as the correlation analysis, was completed using PyCharm 2022.2.1 (JetBrains, Prague, Czech Republic).

3. Results

3.1. Spatial Differentiation in Affected Counties

3.1.1. Spatiotemporal Characteristics of Shoot Blight of Larch Development

From 1970 to 2021, the increase in the number of counties infected with shoot blight of larch in China exhibits a certain periodicity, with significant surges in the number of infected counties occurring at regular intervals (Figure 1). Based on this pattern, the spread and occurrence of shoot blight of larch in China is divided into five stages, using the years of significant increases in the number of infected counties as breakpoints. The historical occurrence stages of shoot blight of larch are shown in Figure 2.

Stage 1 (1973): Shoot blight of larch was first discovered in China, with only five coastal counties in Jilin and Shandong provinces being affected.

Stage 2 (1973–1989): This was the primary phase of the epidemic. The disease gradually spread inland, with a slow increase in the number of affected counties, mostly concentrated in the three northeastern provinces.

Stage 3 (1989–1996): This was the explosive outbreak period. During this time, the disease spread rapidly, and the number of affected counties increased dramatically before slowing down. The outbreak spread nationwide, with Heilongjiang Province particularly affected due to its extensive larch forests and delayed control measures. Between 1973 and 1996, the disease spread extensively in Heilongjiang, affecting as many as 59 counties.

Stage 4 (1996–2007): This was the stabilization period. After peaking in 1996, the disease spread slowly. During this stage, outbreaks were mostly confined to the northeastern region, with sporadic occurrences in other provinces.

Stage 5 (2007–present): This is the decline period. As an invasive species, shoot blight of larch now occurs sporadically in China, with only 24 counties reporting disease information.

The reasons for this distribution pattern are the driving effect of human factors, which facilitated the spread of shoot blight of larch within China, and the fundamental constraint imposed by differences in natural factors on the distribution of affected counties.

3.1.2. The Spatial Distribution of Affected Counties Is Uneven, with a “Northeast–Southwest” Directional Trend

According to the results of the standard deviation ellipse (Figure 3), the concentrated areas of shoot blight of larch are predominantly distributed in the northeast–southwest direction. Between 1973 and 1989, the epicenter of the blight remained in Jilin Province, shifting 140.5 km northwest. The ellipse’s coverage was largely confined to the three northeastern provinces, indicating that the spread of the disease was concentrated in this region. From 1989 to 1996, the epicenter moved southwest to Liaoning Province, with a migration distance of 460.6 km. The ellipse’s rotation angle was approximately 50.44°, and its coverage expanded, marking the spread of the disease to Hebei, Shanxi, Shaanxi, Gansu, Qinghai, and Inner Mongolia.

Between 1996 and 2007, 80.56% of shoot blight of larch cases occurred within the three northeastern provinces and Inner Mongolia. The epicenter shifted 172.0 km northeast, within Inner Mongolia. The ellipse’s rotation angle was about 51.03°, and its coverage area further increased. During this period, the majority of newly affected counties were located in the three northeastern provinces and Inner Mongolia, accounting for more than 84% of new cases.

From 2007 to the present, over 50% of newly affected counties have been outside the three northeastern provinces and Inner Mongolia. The epicenter moved to Baoding City in Hebei Province, with the ellipse’s semi-minor axis measuring 41.67 km, its semi-major axis 124.83 km, and a rotation angle of 43.77°. Spatially, the distribution has contracted in the “northwest–southeast” direction while expanding in the “northeast–northwest” direction. The spatial pattern resembles the division along the “Hu Line”, with 42.9% of the land area on the southeast side containing 90.29% of affected counties, while 57.1% of the land area on the northwest side contains only 9.71% of affected counties [43].

3.2. Spatial Clustering of Affected Counties

3.2.1. The Spatial Clustering Effect of Affected Counties Is Significant

The calculation results of the global Moran’s index are shown in

Table 1. Summary of global Moran index.

Results	1973–1989	1989–1996	1996–2007	2007–Present
Moran’s index	0.104945	0.144843	0.151849	0.113876
Expected index	−0.020833	−0.006803	−0.005556	−0.004878
Variance	0.003180	0.000988	0.000655	0.000434
Z score	2.230628	4.824397	6.148643	5.697596
p value	0.025706	0.000001	0.000000	0.000000
Distribution types	Clustered	Clustered	Clustered	Clustered

. From 1973 to the present, during the five development periods of shoot blight of larch, the global Moran’s index for affected counties was positive and passed the significance test (p < 0.01). This indicates a significant positive spatial clustering effect, where the disease exhibits a clustered distribution, lacking independence and randomness. The occurrence of the disease is associated with similar conditions in affected regions.

During the three periods from 1973 to 2007, the spread of shoot blight of larch was primarily within provinces, leading to higher Moran’s index values and an increasing degree of clustering among affected counties. In the fifth period, however, the disease spread to the northwest regions of China, causing the distribution of affected counties to become more dispersed, which explains the lower Moran’s index compared to the third and fourth stages.

3.2.2. The Density of Affected Counties Gradually Decreases from East to West, with Significant Regional Differentiation Characteristics

Using the kernel density analysis tool to identify the core areas of clustering for counties affected by shoot blight of larch (Figure 4), it was found that the disease is primarily concentrated in four regions in China: Shandong Province, the three northeastern provinces, southwestern Gansu Province, and northern Shanxi Province. Among these, the density of affected counties in the northeastern region is significantly higher than in the other three areas. This region, with a history of shoot blight of larch outbreaks spanning over 40 years, has numerous infected counties that are densely distributed, making it the core region for the disease’s prevalence and development in China.

The density in Shandong Province is second only to the northeastern region. Although Shandong is one of the initial introduction points of the disease, the fragmented landscape connectivity of larch forests in this area has resulted in the spread being primarily driven by natural transmission. This has led to the disease in Shandong exhibiting a broad temporal span and scattered outbreak patterns. In contrast, southwestern Gansu and northern Shanxi, where shoot blight of larch was introduced more recently, have formed lower-density clustering areas. The clustering of shoot blight of larch decreases progressively outward from the core areas, forming a monocentric “core–periphery” structure centered on Liaoning Province.

3.3. Analysis of Influencing Mechanisms

The spatiotemporal distribution pattern of shoot blight of larch is the result of long-term interactions between human activities and natural environmental factors. The results of RFECV (

Table 2. Seven selected research variables.

Variables	Description	Unit	Whether to Use for Modeling
X1	Planting area of seedlings	Mu *	Yes
X2	Canopy density	—	Yes
X3	Maximum wind speed in May	m/s	No
X4	Maximum wind speed in June	m/s	Yes
X5	Average temperature in June	°C	No
X6	Average temperature in July	°C	No
X7	Average temperature in August	°C	Yes
X8	Annual average temperature	°C	No
X9	The highest temperature in June	°C	No
X10	The highest temperature in July	°C	No
X11	The highest temperature in August	°C	No
X12	Annual maximum temperature	°C	Yes
X13	Minimum temperature in June	°C	No
X14	Minimum temperature in July	°C	No
X15	Minimum temperature in August	°C	No
X16	Annual minimum temperature	°C	No
X17	Average precipitation in June	mm	Yes
X18	Annual average precipitation	mm	Yes

* Note: 1 mu = 666.67 m².

) indicate that the model achieves the highest score when the following 7 variables are selected as the influencing factors of the study, including planting area of seedlings (X1), canopy density (X2), maximum wind speed in June (X4), average temperature in August (X7), annual maximum temperature (X12), average precipitation in June (X17), and annual average precipitation (X18).

Using the presence of shoot blight of larch in prefecture-level units as the dependent variable and the seven primary influencing factors as independent variables, the natural breaks classification method was applied to discretize the independent variables into nine levels. Geo Detector was subsequently used to calculate the explanatory power of each independent variable for the spatial heterogeneity of shoot blight of larch. The results (Figure 5) showed that all variables passed the significance test (p < 0.05) and significantly influenced the spatial pattern of the disease, although their explanatory power varied. The q-values in descending order were q(X1) > q(X7) > q(X12) > q(X17) > q(X2) > q(X4) > q(X18), with q(X1) = 0.446, indicating that seedling planting area had a markedly higher explanatory power than the other variables and was the controlling influencing factor in the formation of the disease’s spatial pattern.

Interaction detector results demonstrated that the explanatory power of dual-factor interactions was significantly stronger than that of single factors, all showing nonlinear enhancement. The interaction explanatory power of q(X1∩X2), q(X1∩X7), q(X1∩X12), q(X1∩X18), q(X1∩X17), q(X1∩X4), q(X7∩X18), q(X7∩X17), q(X7∩X4) all exceeded 0.5. Among these, seedling planting area (X11) exerted the strongest influence on factor interactions, followed by average temperature in August (X13). These two variables were the dominant factors driving the enhanced interaction explanatory power for the spatial pattern of shoot blight of larch. Furthermore, the explanatory power of other factors significantly increased after interacting with any of the selected factors, indicating that the spatial differentiation in forest villages is the result of long-term, multifactorial interactions.

The results of the Random Forest and Geo Detector analyses are presented in Figure 6. In the Random Forest model, the optimal performance was achieved when the number of decision trees (n_estimators) was set to 7, the maximum number of features for the best split (max_depth) was 51, the random state for the generated forest control mode (random_state) was 3, the minimum sample size required for decision tree splitting (min_samples_split) was 0.001, and the minimum sample size for leaf nodes (min_samples_leaf) was 0.0045. At this configuration, the model’s accuracy, recall, precision, and F1 score were 90.76%, 85.11%, 90.91%, and 87.91%, respectively. These evaluation metrics indicate that the model performs well, demonstrating its reliability.

The ranking of importance among the seven influencing factors for disease occurrence was as follows: average temperature in August > average precipitation in June > annual average precipitation > annual minimum temperature > seedling planting area > canopy closure > maximum daily wind speed in June. Among these, the average temperature in August had the greatest impact on disease occurrence, with a feature importance score of 0.4608. The other factors also made significant contributions to the occurrence of shoot blight of larch, with importance scores all exceeding 0.05. This indicates that the occurrence of shoot blight of larch is predominantly driven by natural factors.

The results of the statistical analysis are shown in

Table 3. Results of correlation coefficient statistical analysis.

Statistical Method	Correlation Coefficient	p Value
Logistic	0.000027	0.001
Spearman	0.261	0.001
Kendall’s	0.214	0.001
Mann–Whitney	1436.000	0.001

. The accuracy of the Logistic regression model is 70%, indicating a certain level of correlation. Additionally, the OR is 1.000027, which suggests that for every 1 mu increase in larch seedling afforestation area, the probability of shoot blight of larch occurrence increases by 0.0027%. Although the correlation coefficients were low, both Spearman’s rank correlation coefficient and Kendall’s tau correlation coefficient indicated a significant positive correlation between continuous and binary variables. Furthermore, the Mann–Whitney U test results supported the statistical significance of median differences between groups.

4. Discussion

4.1. The Transmission and Diffusion Mechanism of Shoot Blight of Larch

The human activities of transplanting infected seedlings are the dominant factor in shaping the spatial distribution of shoot blight of larch, with seedling planting area showing a significant positive correlation with the occurrence of the disease. From 1989 to 1996 and 1996 to 2007, China experienced the outbreak and stable periods of shoot blight of larch, respectively. The development of the disease during these two periods largely shaped the current spatiotemporal pattern of shoot blight of larch, which is now distributed across 12 provinces in China. After 1978, China launched the “Three-North Shelterbelt Project”, during which large-scale transportation of seedlings for afforestation likely carried larch seedlings infected with the blight pathogen from the three northeastern provinces to North China and Northwest China, serving as the primary cause of the disease’s introduction to these regions. Guo [25] noted that between 1981 and 1983 alone, afforestation in the Loess Hill area covered 2800 mu, involving the planting of over 600,000 trees, most of which were North China larch.

This study analyzed the relationship between the area of larch seedlings planted and the occurrence of larch shoot blight (Figure 7). A linear relationship was observed between the area of larch seedlings planted and the proportion of counties affected by larch shoot blight. As the area of larch seedlings planted within counties increased, the proportion of counties affected by larch shoot blight also increased. When the area of larch seedling planting exceeded 50,000 mu, the proportion of affected counties reached 64.29%. This indicates that the area of larch seedling planting can influence the spread of larch shoot blight. Furthermore, the study evaluated the relationship between the spread of larch shoot blight and the area of larch seedling planting using Logistic regression models and correlation coefficient statistical analysis methods.

The natural environment is a fundamental factor influencing the spatial pattern of shoot blight of larch. The occurrence and development of shoot blight of larch are directly influenced by ecological factors, with temperature, precipitation, canopy closure, and wind speed being the main contributing factors. Suitable temperatures are critical for the germination of ascospores and conidia of the blight [44]. This study identified average temperature in August as the most important factor influencing the outbreak. Symptoms of shoot blight of larch are most pronounced from mid-August to early September, and from late August to early September, perithecia gradually form on diseased branches, laying the groundwork for outbreaks in the following year. Analysis revealed that 92.23% of affected counties have an average August temperature between 9.50 °C and 22.28 °C, which aligns closely with previous studies suggesting the optimal mean temperature range for the warmest quarter is 10.1 °C to 24.0 °C [14].

Precipitation is also a critical factor influencing the outbreak of shoot blight of larch. Regions with high annual precipitation and relative humidity experience more severe infections in new larch shoots [26]. Furthermore, this study found that average precipitation in June significantly affects the spread of the disease. This period coincides with the spore dispersal stage and the active development of the disease. Research by Pan Xueren et al. indicated that spore dispersal peaks significantly following continuous rainfall. This finding underscores that adequate precipitation and humidity favor spore germination, intensifying the spread and severity of the disease.

The primary means of shoot blight of larch infection and development involve spore dispersal by wind and penetration through wounds [45]. Yu and Zhao [29] suggested that a daily maximum wind speed exceeding 4 m/s in May and June is a necessary condition for the outbreak of shoot blight of larch. Compared to Yu’s study, this research found that fewer than 10% of the affected counties met this condition, and all these counties were within 500 km of the coastline. This is consistent with the geographic limitation of Yu’s study, which focused on the three northeastern provinces, all located within 500 km of the coast.

Regarding the relationship between shoot blight of larch and canopy closure, previous studies have shown that dense forest stands with high canopy closure, which limit ventilation and light penetration, favor disease occurrence and development [46]. Disease severity tends to increase with canopy closure, and stands with low density exhibit lighter disease incidence compared to dense stands [47]. This study found a positive correlation between the number of affected counties and canopy closure, with over 80% of affected counties having a canopy closure greater than 40%.

4.2. Comparison and Selection of Models

This study represents the first application of two distinct modeling approaches, Geo Detector and Random Forest, to identify the key factors influencing the spatiotemporal dynamics of shoot blight of larch. By employing these two complementary modeling techniques, we found that several factors yielded p-values below 0.05 in the Geo Detector analysis, indicating statistically significant results that are both interpretable and reliable. In the results of the Random Forest algorithm, the model’s accuracy, recall, precision, and F1 score reached 90.76%, 85.11%, 90.91%, and 87.91%, respectively. These favorable evaluation metrics indicate that the model performs well, suggesting that the results are highly reliable.

Given that each model algorithm emphasizes different aspects, each has its inherent strengths and limitations. A singular Random Forest approach, by itself, is insufficient for further elucidating the contributions of individual influencing factors. To address this, numerous studies have opted to enhance the algorithm, such as improving classification and regression trees [37], reducing the correlation between regression trees [38]. Ultimately, the integration of multi-dimensional evaluation methods and model optimization contributes to enhancing the reliability and generalizability of Random Forest. Furthermore, the application of multiple algorithms in tandem facilitates a more comprehensive, multi-angle analysis of the influencing factors. Methods such as Principal Component Analysis (PCA) [48], Exploratory Factor Analysis [49], and Geo Detector can be effectively integrated with Random Forest to deepen the analysis [21]. Hence, the selection of appropriate, mutually reinforcing research methodologies based on specific research objectives is essential.

For a given system and problem, the insights and predictions derived from a model often rely more on the modeling team involved than on the scenario being analyzed [50]. For example, in the case of global and European land cover projection models [51], as well as the predictions of Antarctic krill growth in the Southern Ocean made by eight different models, which are contradictory in many regions [52,53]. Therefore, selecting an appropriate model based on the specific characteristics of the research objectives is crucial.

Species distribution models (SDMs) are widely used in ecology, integrating environmental variables (such as temperature, precipitation, and elevation) with known species distribution data to predict the potential distribution of species in other unstudied areas. These models can also be combined with models such as Random Forest for improvement and possess geographical transferability for species with similar distribution environments [54,55]. Unlike the complex statistical modeling and hypothesis testing of species distribution models, Geo Detector primarily focuses on spatial distribution differences between variables to assess the strength of factors influencing species distribution, without requiring extensive model parameter adjustments. Geo Detector is particularly effective in identifying interactions between multiple factors, especially in complex environmental contexts, revealing which factor combinations have a significant impact on spatial distribution. Through repeated testing and analysis of interactions at different levels, it can provide more detailed information. While species distribution models can also incorporate interaction terms, they often rely more on prior hypotheses, and the selection and interpretation of interaction terms require strong domain knowledge, making them more challenging to intuitively grasp under the influence of multiple factors.

The results indicate that both the Random Forest and Geo Detector models provide similar findings regarding the primary factors influencing the spatiotemporal distribution of shoot blight of larch. The key determinants identified include the planting area of seedlings, canopy density, maximum wind speed in June, average temperature in August, annual maximum temperature, average precipitation in June, and annual average precipitation. The Random Forest algorithm, focusing primarily on the occurrence of the disease, highlights the importance of temperature, a factor directly influencing fungal proliferation. In contrast, Geo Detector places greater emphasis on the spatial and temporal impact of these factors, as well as their interactions, suggesting that human activities—particularly the planting area of seedlings, which facilitates the long-distance spread of the pathogen—constitute the most significant factor. This further underscores the value of combining these two models to achieve a higher degree of reliability and accuracy in identifying the key factors driving shoot blight of larch in China. Such insights are critical for predicting the future progression of the disease and for formulating effective, timely mitigation strategies.

4.3. The Limitations and Prospects of the Study

First, the influencing factors considered in this study remain incomplete. In real-world environments, the spread and prevalence of shoot blight of larch are influenced by more complex factors, such as tree species, forest age, slope aspect, elevation, and interspecies interactions. Since the mechanisms by which these factors affect the disease are unclear or difficult to quantify, the results of this study can serve as a reference for understanding the relationship between shoot blight of larch and environmental variables but cannot fully encapsulate their interactions. Second, this study used seedling planting area as a proxy for human activities to explore its relationship with the spread of shoot blight of larch. Due to the lack of precise annual data, this factor can only roughly describe the influence of human activities on the disease’s spread and diffusion during certain periods. Despite these limitations, this study represents a meaningful preliminary exploration that contributes to understanding the spatiotemporal patterns of shoot blight of larch’s spread and diffusion in China. The several key factors identified in this study can be used for further predictions regarding the occurrence and development of shoot blight of larch. Based on the aforementioned findings, the following measures can be implemented for effective management of the pathogen: (1) improve nursery sanitation by removing plant debris, regularly cleaning containers with hot water, and promptly removing infected plants that produce spores; (2) develop resistant varieties by selecting and breeding larch varieties with strong resistance, thereby reducing the risk of disease outbreaks; and (3) conduct regular disease monitoring and early warning to ensure early detection and timely management, thus achieving integrated control and ensuring the healthy growth of larch seedlings.

5. Conclusions

This study provides a comprehensive and accurate description of the spatial–temporal patterns of shoot blight of larch spread and diffusion across China over the past five decades, from 1973 to the present. The results indicate that the human activity of transplanting infected seedlings is the dominant factor driving the formation of these patterns. Based on real occurrence data, this study fills the gap in research on the spatial–temporal patterns of shoot blight of larch, particularly regarding the invasion process of this pathogen in China. The key findings of this research will provide detailed, reliable data support and theoretical foundation for future guidance on the prevention and control of shoot blight of larch.