A spatiotemporal dynamic evaluation method and system for habitat suitability of rice sheath blight

Abstract

This invention discloses a method and system for the spatiotemporal dynamic evaluation of habitat suitability for rice sheath blight. By integrating multi-source time-series information, including remote sensing, meteorological, and plant protection data, the method characterizes the spatiotemporal heterogeneity of key habitat factors sensitive to rice sheath blight. Based on this, a habitat suitability evaluation model for rice sheath blight is constructed using months as the time unit, enabling a refined spatiotemporal dynamic evaluation of rice sheath blight.

Description

Technical Field

[0001] The present invention belongs to the technical field of plant disease habitat suitability evaluation, and relates to a spatiotemporal dynamic evaluation method for disease habitat suitability based on multi-source time series information such as remote sensing, meteorology, and plant protection. Specifically, it relates to a spatiotemporal dynamic evaluation method and system for habitat suitability of rice sheath blight. Background Art

[0002] Currently, pest and disease control efforts, still dominated by pest and disease trend forecasting models based on meteorological plant protection, face challenges such as overexpansion of control areas and excessive pesticide application. This has led to a continuous increase in chemical pesticide inputs, increasing control costs while also contributing to agricultural non-point source pollution and threatening food safety. Regional habitat suitability assessments for pests and diseases are crucial for achieving accurate forecasts of crop pests and diseases across a wide range of areas, providing crucial baseline information for their prediction and control.

[0003] As a soil-borne disease, rice sheath blight relies primarily on irrigation water to carry soil sclerotia for transmission. Consequently, the occurrence and spread of sheath blight is closely linked to landscape patterns, such as farmland ditch networks. Furthermore, the mixed cropping pattern of single- and double-season rice further increases the spatiotemporal complexity of sheath blight's spread. However, current regional habitat suitability assessments for crop pests and diseases still primarily rely on meteorological and plant protection information, making it difficult to effectively capture the impact of complex spatial and temporal heterogeneity in habitats, greatly limiting the effectiveness of these assessments. In recent years, the deep integration of multidisciplinary technologies, such as remote sensing and computer science, has provided valuable opportunities for the innovative development of habitat suitability assessment technologies for crop pests and diseases.

[0004] The present invention fully considers the epidemic and spread characteristics of rice sheath blight, integrates multi-source time series information such as remote sensing, meteorology, and plant protection, and extracts habitat factors for rice sheath blight with a monthly time unit. The GCCM method, a spatiotemporal causal inference theory, is introduced to characterize the spatiotemporal heterogeneity of key habitat factors sensitive to rice sheath blight. On this basis, the classical species distribution model MaxEnt is used to construct a spatiotemporal dynamic evaluation model for the habitat suitability of rice sheath blight with fine spatiotemporal precision, realizing the monthly scale evaluation of the habitat suitability of rice sheath blight. This overcomes the technical bottleneck of insufficient characterization of the spatiotemporal heterogeneity of habitats in traditional pest and disease habitat suitability evaluation methods, improves the level of regional crop disease habitat suitability evaluation, and provides reliable background information for regional disease early warning modeling. Summary of the Invention

[0005] The purpose of the present invention is to address the shortcomings of existing technologies in effectively reflecting the complex impact of spatiotemporal heterogeneity in regional pest and disease habitats, and to provide a method for spatiotemporal dynamic evaluation of regional disease habitat suitability that fully considers the characteristics of rice sheath blight epidemic spread, including the construction of a habitat suitability evaluation model that fully considers the impact of spatiotemporal heterogeneity in the habitat of rice sheath blight epidemic spread under the mixed cropping model of single and double rice seasons. This method achieves the characterization of the spatiotemporal heterogeneity of key habitat factors sensitive to rice sheath blight by integrating multi-source time series information such as remote sensing, meteorology, and plant protection. On this basis, a rice sheath blight habitat suitability evaluation model is constructed with months as the time unit, thereby achieving spatiotemporal and spatially refined spatiotemporal dynamic evaluation of rice sheath blight.

[0006] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0007] In a first aspect, the present invention provides a spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight, comprising the following steps:

[0008] Extract monthly-scale habitat factors for rice sheath blight by integrating multi-source time series information.

[0009] Obtain the remote sensing, meteorological and plant protection time series data of the area to be evaluated over the years, use them to calculate the disease habitat factors at different temporal and spatial resolution scales, and resample each habitat factor to form a time series disease habitat factor with a unified spatial resolution.

[0010] The temporal disease habitat factors were divided into monthly units, and the mean values ​​of the same disease habitat factors in multiple phases within the same month were calculated. The monthly rice planting spatial distribution data were combined with masking to obtain the disease habitat factors of each month under the constraints of rice planting spatial distribution.

[0011] The disease habitat factors of each month obtained under the spatial distribution constraints of rice planting in multiple years are averaged according to the disease habitat factors of the same month in different years to complete the extraction of monthly rice sheath blight habitat factors based on comprehensive multi-source time series information.

[0012] Based on spatial causal inference, habitat factors are optimized to achieve the characterization of the spatiotemporal heterogeneity of key sensitive habitat factors of rice sheath blight.

[0013] Through spatial causal inference, the causal relationship between various habitat factors at the monthly scale and the diseased plant rate of rice sheath blight in that month was analyzed. The key sensitive habitat factors affecting the epidemic spread of rice sheath blight were initially screened based on whether the causal effect index was greater than the set threshold, and the initial screening habitat factors were obtained.

[0014] Spatial causal inference analysis was further used to preliminarily screen the possible causal relationships between habitat factors, draw a causal-oriented relationship diagram between the habitat factors of rice sheath blight, and combine the causal-oriented relationship diagram to achieve the optimization of key sensitive habitat factors of rice sheath blight in each month.

[0015] Construction and optimization of spatiotemporal dynamic evaluation model of habitat suitability for rice sheath blight based on species distribution model.

[0016] The selected key sensitive habitat factors of rice sheath blight in each month were used as the input of the species distribution model MaxEnt to construct a spatiotemporal dynamic evaluation model of habitat suitability for rice sheath blight (GCCM-MaxEnt).

[0017] The evaluation model parameters were optimized by combining the corrected AIC information criterion with the evaluation index AUC, forming a monthly-scale rice sheath blight habitat suitability evaluation model group optimized by month.

[0018] The monthly rice sheath blight habitat suitability evaluation model group was used to obtain the monthly rice sheath blight habitat suitability evaluation results during the critical period of rice in the target area.

[0019] Preferably, the monthly rice planting spatial distribution data is obtained based on the phenological period information of single-season and double-season rice: first, based on the transplanting period and harvesting period dates corresponding to the same spatial distribution of rice of each rice-growing type, the rice growth time range under the same spatial distribution of rice of each rice-growing type is obtained; on this basis, the rice growth time range is divided using the corresponding time range of each month as a threshold to obtain the rice planting spatial distribution data of each month.

[0020] Preferably, the rice sheath blight diseased plant rate data for each month is calculated from the plant protection time series data: first, representative fields where sheath blight occurs are selected as survey points to obtain the corresponding plant protection time series data, and the time series sheath blight diseased plant rate data for each survey point is obtained by calculating the ratio of the number of diseased plants to the total number of surveyed plants; secondly, the plant protection time series data of the sheath blight diseased plant rate of the survey point each year are divided into months as the time unit, and the maximum diseased plant rate value in each month is used as the sheath blight diseased plant rate of the survey point in that month; finally, the rice sheath blight diseased plant rate data for each month is obtained by calculating the average of the sheath blight diseased plant rates in the same month in all years.

[0021] Preferably, the causal effect index reflects the relationship between the actual observed value and the spatial causal inference predicted value by using the covariance of the normalized actual observed value and the spatial causal inference predicted value. A pre-set threshold value of the causal effect index is used to achieve initial screening of habitat factors.

[0022] Preferably, the direction of drawing the causal relationship diagram between the sheath blight habitat factors is as follows: first, spatial causal inference is used to quantify the causal relationship between the sheath blight habitat factors, and the causal effect index between the habitat factors is obtained; based on the calculated causal effect index, the causal relationship is defined in a directed acyclic graph using a directed acyclic graph tool, and an interactive, adjustable directed acyclic graph is obtained using an interactive network visualization tool; finally, a causal relationship diagram between the sheath blight habitat factors is drawn using a graphical grammar drawing tool.

[0023] As a preferred method, the evaluation model parameter optimization is achieved by combining the correction AIC information criterion with the evaluation index AUC. Specifically, first, the RM value range and step size are set, and the RM gradient experimental group is determined to quantitatively analyze the regulatory effect of parameter adjustment on model complexity; then, the RM with the smallest AICc value corresponding to model complexity under different RMs within the RM value range is selected; finally, the model AUC performance between the optimized RM and the default RM parameters of the species distribution model (RM=1) is compared, and the RM value with the largest AUC value is selected as the optimal RM of the model to complete the model optimization and establish a parameter optimization scheme suitable for the monthly scale evaluation model.

[0024] In a second aspect, the present invention provides a spatiotemporal dynamic evaluation system for habitat suitability of rice sheath blight, which includes the following modules:

[0025] Habitat factor extraction module: integrates multi-source time series information to extract monthly rice sheath blight habitat factors.

[0026] Habitat factor optimization module: Optimize key sensitive habitat factors based on spatial causal inference.

[0027] Evaluation model group acquisition module: Construction and optimization of the spatiotemporal dynamic evaluation model of rice sheath blight habitat suitability based on the species distribution model.

[0028] Evaluation module: The key habitat factors sensitive to rice sheath blight in each month of the target area are input into the spatiotemporal dynamic evaluation model of rice sheath blight habitat suitability in each month to obtain the evaluation results of rice sheath blight habitat suitability in each month of the target area, that is, the spatial distribution range map of the potential distribution probability of rice sheath blight in each month.

[0029] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the implementation of the first aspect as described above.

[0030] In a fourth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the implementation of the first aspect above.

[0031] The present invention has the following beneficial effects:

[0032] 1. This method, coupled with GCCM causal inference theory, characterizes the spatiotemporal heterogeneity of habitat factors in disease-sensitive areas. By integrating multi-source data such as meteorology, remote sensing, and plant protection, it explores the mechanism by which spatiotemporal heterogeneity of habitats influences the spread of pests and diseases. Compared with traditional methods, it can analyze the impact of habitat differences on crop diseases at a finer spatiotemporal scale, while effectively improving the interpretability and reliability of selected habitat factors.

[0033] 2. This paper establishes a modeling strategy for the spatiotemporal dynamic evaluation of regional habitat suitability for crop diseases based on the causal inference theory GCCM coupled with the species distribution model MaxEnt. Through monthly-scale modeling, it can provide a refined response to the spatiotemporal process characteristics of the occurrence and epidemic of diseases within an interannual period. Compared with existing evaluation methods, it has a stronger mechanism and can effectively improve the effectiveness and reliability of habitat suitability evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a technical roadmap of an embodiment of the present invention.

[0035] Figure 2 This is a causal relationship diagram among monthly rice sheath blight habitat factors according to an embodiment of the present invention.

[0036] Figure 3 This is a comparison chart of the accuracy of the monthly rice sheath blight habitat suitability evaluation model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further explained below with reference to the accompanying drawings:

[0038] A spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight disease comprises the following steps:

[0039] Step 1: Integrate multi-source time series information to extract monthly rice sheath blight habitat factors.

[0040] 1-1. Obtain the remote sensing, meteorological and plant protection time series data of the area to be evaluated over the years, use them to calculate the disease habitat factors at different temporal and spatial resolutions, and resample each habitat factor to form a unified spatial resolution time series disease habitat factor.

[0041] 1-2. Divide the temporal habitat factors into monthly units, calculate the mean of the same habitat factor in multiple phases within the same month, and perform masking on the monthly rice planting spatial distribution data to obtain the disease habitat factors for each month of each year under the constraints of rice planting spatial distribution.

[0042] In one embodiment, the monthly rice planting spatial distribution data is obtained based on the phenological period information of single-season and double-season rice: first, based on the transplanting period and harvesting period dates corresponding to the same spatial distribution of rice of each rice-growing type, the rice growth time range under the same spatial distribution of rice of each rice-growing type is obtained; on this basis, the rice growth time range is divided using the corresponding time range of each month as a threshold to obtain the rice planting spatial distribution data of each month.

[0043] 1-3. The disease habitat factors of each month under the spatial distribution constraints of rice planting in multiple years are calculated according to the mean of the disease habitat factors of the same month in different years to complete the extraction of monthly rice sheath blight habitat factors based on comprehensive multi-source time series information.

[0044] Step 2: Based on the spatial causal inference GCCM, the optimal habitat factors were selected to characterize the spatiotemporal heterogeneity of key sensitive habitat factors of rice sheath blight.

[0045] 2-1. GCCM was used to analyze the causal relationship between monthly habitat factors and the rice sheath blight disease rate in that month. The causal effect index was determined to be greater than the set threshold to complete the initial screening of key sensitive habitat factors affecting the spread of rice sheath blight and obtain the initial screening habitat factors.

[0046] In one embodiment, monthly rice sheath blight diseased plant rate data is calculated from plant protection time-series data: first, representative fields where sheath blight occurs are selected as survey points, and in accordance with the national standard "GBT 15791-2011 Technical Specification for Rice Sheath Blight Monitoring and Forecasting", the prevalence of rice sheath blight for consecutive years at each survey point is obtained, and the total number of surveyed plants and the number of sheath blight diseased plants are recorded. The same survey point is surveyed once at set intervals during the critical rice season each year to obtain plant protection time-series data, and the time-series sheath blight diseased plant rate data for each survey point is calculated by calculating the ratio of the number of diseased plants to the total number of surveyed plants; second, the plant protection time-series data of the sheath blight diseased plant rate of each survey point is divided into months, and the maximum diseased plant rate value in each month is used as the sheath blight diseased plant rate of the survey point in that month; finally, the rice sheath blight diseased plant rate data for each month is obtained by calculating the average of the sheath blight diseased plant rates in the same month in all years.

[0047] In one embodiment, the causal effect indicator is ρ, which reflects the relationship between the actual observation value and the GCCM prediction value by the standardized covariance between the actual observation value and the GCCM prediction value. A pre-set ρ value threshold is used to achieve preliminary screening of habitat factors.

[0048] 2-2. Further use GCCM to analyze the possible causal relationships between the pre-screened habitat factors, draw a causal relationship diagram between the habitat factors of rice sheath blight, and combine the causal relationship diagram to achieve the optimization of the key sensitive habitat factors of rice sheath blight in each month.

[0049] In one embodiment, the direction of drawing the causal relationship diagram between the sheath blight habitat factors is as follows: first, GCCM is used to quantify the causal relationship between each pair of sheath blight habitat factors to obtain the causal effect index ρ value between each pair of habitat factors; based on the calculated ρ value, the causal relationship is defined into a directed acyclic graph DAG using the directed acyclic graph tool dagitty, and an interactive and adjustable DAG is obtained using the interactive network visualization tool visNetwork; finally, the causal relationship diagram between the sheath blight habitat factors is drawn using the graphical grammar drawing tools ggdag and ggplot2.

[0050] Step 3: Construct and optimize the spatiotemporal dynamic evaluation model of rice sheath blight habitat suitability based on the species distribution model.

[0051] 3-1. The key sensitive habitat factors for rice sheath blight selected by GCCM in each month were used as inputs to the species distribution model MaxEnt to construct a spatiotemporal dynamic evaluation model for habitat suitability for rice sheath blight (GCCM-MaxEnt).

[0052] 3-2. The evaluation model parameters were optimized by combining the calibrated AIC information criterion with the evaluation index AUC, forming a monthly-scale rice sheath blight habitat suitability evaluation model group that was optimized by month.

[0053] In one embodiment, the regularization coefficient RM optimization strategy of the GCCM-MaxEnt model that combines the correction AIC information criterion with the area under the evaluation index curve AUC is specifically as follows: first, the RM value range and step size are set, and the RM gradient experimental group is determined to quantitatively analyze the regulatory effect of parameter adjustment on model complexity; then, the RM with the smallest AICc value corresponding to model complexity under different RMs within the RM value range is selected; finally, the model AUC performance between the optimized RM and the default RM parameters (RM=1) of the MaxEnt model is compared, and the RM value with the largest corresponding AUC value is selected as the optimal RM of the model to complete the model optimization and establish a parameter optimization scheme suitable for the monthly scale evaluation model.

[0054] Step 4: Input the key habitat factors sensitive to rice sheath blight in each month of the target area into the spatiotemporal dynamic evaluation model of rice sheath blight habitat suitability for each month to obtain the evaluation results of rice sheath blight habitat suitability for each month in the target area, that is, the spatial distribution range map of the potential distribution probability of rice sheath blight in each month.

[0055] like Figure 1 As shown in Figure 1, a spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight that integrates multi-source temporal information is presented. The specific steps are as follows:

[0056] Step 1: Integrate multi-source information to obtain the spatiotemporally continuous habitat factors of rice sheath blight, and realize the extraction of habitat factors of rice sheath blight on a monthly scale.

[0057] 1-1 Acquire multi-source time series data, including remote sensing, meteorological, and plant protection data, covering the period from 2008 to 2015. Due to the lack of plant protection survey data for 2012, all data for 2012 were removed in subsequent studies. The target area is located from the middle and lower reaches of the Yangtze River to the southwestern rice-growing areas.

[0058] 1-2 Habitat factors characterizing rice growth conditions include normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), leaf area index (LAI), fractional vegetation cover (FVC), fraction of photosynthetic radiation absorbed by vegetation (FPAR), and net primary productivity (NPP), all of which are standardized MODIS vegetation products; habitat factors characterizing environmental meteorological parameters include maximum temperature (TMX), minimum temperature (TMN), mean temperature (TMP), sunshine hours (NSUN), relative humidity (RH), precipitation (PRE), land surface temperature (LST), and elevation information (DEM). Meteorological data are from the National Tibetan Plateau Data Center, the National Earth System Data Center, and the Central The national daily surface climate data set (V3.0) was used. The land surface temperature (LST) was derived from the standardized MODIS product, and the elevation information (DEM) was obtained from the Resources and Environmental Science Data Center of the Chinese Academy of Sciences. Habitat factors characterizing the spatial landscape pattern include patch density (PD), landscape shape index (LSI), edge density (ED), patch richness density (PRD), Shannon diversity index (SHDI), Simpson diversity index (SIDI), maximum patch index (LPI), and mean fractal dimension (FRAC_MN). All of these factors are based on the 30-m resolution land use product CLCD and are calculated using the landscape spatial pattern quantitative analysis software FRAGSTATS.

[0059] 1-3 The habitat factors of different temporal and spatial resolutions obtained in each year were selected in time, and the sensitive period of rice sheath blight from April to October was selected. The average of each habitat factor was calculated by month for time-month standardization, and the spatial resolution was uniformly resampled to 1 km for standardization to obtain the temporal and spatial standardized multidimensional habitat factors for each year and month. On this basis, the habitat factors of the same month from 2008 to 2015 (except 2012) were further averaged to obtain the monthly scale rice sheath blight habitat factors in the target area.

[0060] 1-4 Based on the phenological period time information of single- and double-season rice in the ChinaCropArea1km dataset, the spatial distribution data of rice planting in the target area was obtained for each month from April to October, with the time of each month as the phenological period threshold range. The obtained monthly-scale rice sheath blight habitat factors were masked, and finally the monthly-scale rice sheath blight habitat factors were extracted under the constraints of the spatial distribution of rice planting.

[0061] The spatial distribution data of rice planting are obtained based on the phenological period information of single-season and double-season rice: first, the transplanting period and harvesting period corresponding to the same spatial distribution of rice of each rice-growing type are used as the basis to obtain the rice growing time range under the same spatial distribution of rice of each rice-growing type; on this basis, the rice growing time range is divided using the corresponding time range of each month as the threshold to obtain the spatial distribution data of rice planting for each month.

[0062] Step 2: Introduce spatial causal inference GCCM to optimize habitat factors and realize the spatiotemporal heterogeneity characterization of key sensitive habitat factors of rice sheath blight.

[0063] 2-1 The rice sheath blight plant protection survey data is the diseased plant rate data every 5-7 days from 2008 to 2015 (excluding 2012). The diseased plant rate DI is calculated as follows:

[0064]

[0065] Among them, M is the total number of plants in the plant protection data, M disease The number of diseased plants.

[0066] For the rice sheath blight disease rate data for each month (April to October) each year, the plant protection survey data on sheath blight disease rate at the same plant protection station every 5-7 days were divided into monthly time units, and the maximum diseased plant rate value in each month was obtained as the sheath blight disease rate of the station in that month; the seven-year average diseased plant rate for the same month from 2008 to 2015 (except 2012) was calculated and used as the final rice sheath blight disease rate for each month for causal inference analysis.

[0067] 2-2 Considering the district / county characteristics of the plant protection survey data, the monthly-scale rice sheath blight habitat factors obtained under the constraints of the spatial distribution of rice planting were taken as units, and the districts and counties where the rice sheath blight survey and plant protection stations were located were used as units. The mean values ​​of the habitat factors within each district and county were calculated to match the rice sheath blight plant protection survey while ensuring the representativeness of the habitat factors. Ultimately, a dataset of key sheath blight habitat factors that is suitable for GCCM analysis and is compatible with the sheath blight plant protection survey data was formed.

[0068] 2-3 Use GCCM to quantify the causal relationship between the rice sheath blight disease rate and various habitat factors at the monthly scale. Based on the generalized embedding theorem, GCCM proposes to reconstruct the state space based on spatial lag, and realize causal inference based on the state space cross-mapping prediction. It can identify the causal direction and estimate the causal effect in weak coupling relationships, and identify the dominant causal direction and estimate the causal effect in strong coupling relationships. The generalized embedding theorem is a theory about the mapping between differentiable manifolds M and Euclidean space R, that is, M can be mapped to R through observation functions, and points in R can be used to construct M, so that a state space containing spatial cross-sectional data can be constructed. Specifically, GCCM defines the prediction based on the nearest mutual neighbors as a cross-mapping prediction, and the calculation formula is as follows:

[0069]

[0070] Among them, S is the spatial unit of the variable Y that needs to be predicted (rate of sheath blight diseased plants / habitat factor); is the shadow manifold based on the habitat factor x The predicted value of is the spatial unit used for prediction; For space unit The observation value of; L is the embedding dimension; For space unit The corresponding weights are determined by an exponential decay function:

[0071]

[0072] in is the weight function between two states in the shadow manifold, which is defined as follows:

[0073]

[0074] Where exp is the exponential function, dis(*,*) is the distance function between two states in the defined shadow manifold, and its definition formula is as follows

[0075]

[0076] Where |*| represents the absolute value, abs[*,*] represents the distance function between two vectors, The first element h in si (x) corresponds to the spatial focal unit, and The other elements in correspond to vectors with multiple spatial units. For raster data, since the number and position of neighbors in a certain order are fixed, abs[*,*] is defined as follows:

[0077]

[0078] in is the spatial unit of the k-th spatial lag of si in direction d, and D is the number of spatial units at order k.

[0079] Finally, the ability of GCCM to cross-map predictions is measured by the Pearson correlation coefficient between the true observations and the corresponding predicted values, i.e., the causal effect indicator ρ:

[0080]

[0081] in, represents the covariance between the observed value and the GCCM predicted value, and denote the variances of the observed values ​​and the GCCM predicted values, respectively.

[0082] In the embodiment, the sensitive habitat factors of sheath blight were preliminarily screened with ρ>0.2 as the threshold value, and the sensitive habitat factors of each month from April to October were obtained as shown in Table 1 below:

[0083] Table 1: Sensitive habitat factors for each month from April to October

[0084] month Sheath blight-sensitive habitat factors April LSI, PRD, TMX, PRE, RH, NDVI, LST May LSI, PRD, TMX, PRE, RH, LST, NSUN June LSI, TMP, PRE, RH, NSUN, LST, LAI, NDVI, FPAR, EVI, NPP July LSI, TMP, PRE, RH, NSUN, LST, LAI, NDVI, FPAR, EVI, NPP August LSI, TMP, PRE, RH, NSUN, LST, LAI, NDVI, FPAR, EVI, NPP September LSI, TMX, PRE, RH, NSUN, LST, NDVI, EVI, FPAR October LSI, TMX, PRE, RH, NSUN, LST, NDVI, EVI, FPAR

[0085] 2-4 To reduce the redundant information between various factors, the causal relationship between each pair of sensitive habitat factors of sheath blight obtained by the initial screening was further analyzed by GCCM, and a causal relationship diagram between the key habitat factors of sheath blight was drawn ( Figure 2 ): Based on the calculated causal effect index ρ value between each pair of rice sheath blight habitat factors using GCCM, dagitty was used to define the causal relationship into a directed acyclic graph (DAG), and visNetwork was used to obtain an interactive and adjustable DAG. Finally, ggdag and ggplot2 were used to draw a causal-oriented relationship diagram between the sheath blight habitat factors. On this basis, combined with the causal-oriented relationship diagram, for each causal relationship chain in the relationship diagram, the source habitat factor of all associated habitat factors was selected to achieve the optimization of the key habitat factors for rice sheath blight in each month. The key sensitive habitat factors of rice sheath blight with spatiotemporal heterogeneity finally obtained are shown in Table 2 below:

[0086] Table 2: Key sensitive habitat factors of rice sheath blight with temporal and spatial heterogeneity

[0087] month Key habitat factors sensitive to sheath blight April LSI, PRD, TMX, PRE, NDVI May LSI, PRD, PRE, TMX, NSUN June to August LSI, TMP, PRE, NSUN, LAI September to October LSI, TMX, PRE, NSUN, EVI

[0088] Step 3. Using the key habitat factors for rice sheath blight selected by GCCM for each month as input to the species distribution model MaxEnt, a spatiotemporal dynamic evaluation model for rice sheath blight habitat suitability, GCCM-MaxEnt, was constructed based on the species distribution model MaxEnt. The model parameters were optimized by adjusting the AIC information criterion to form a set of monthly rice sheath blight habitat suitability evaluation models corresponding to April–October.

[0089] 3-1 The MaxEnt model is a probability model based on the maximum entropy principle. It establishes a non-random relationship between the spatial distribution data of diseases and the habitat factor layer through ecological principles, and outputs the potential distribution probability of diseases, thereby constructing a disease habitat suitability evaluation model with strong mechanistic rationality. MaxEnt maximizes the conditional probability distribution under the entropy constraint:

[0090]

[0091]

[0092] in, characteristic function, is the weight of the feature function.

[0093] 3-2 The ENMeval toolkit was used in combination with the corrected AIC information criterion to optimize the regularization coefficient RM of the MaxEnt model: an RM gradient experimental group with a range of 0.5-4.0 and a step size of 0.5 was set to quantitatively analyze the regulatory effect of parameter adjustment on model complexity; the RM with the smallest AICc value corresponding to model complexity under different RMs in the range of 0.5-4.0 was selected; the model AUC performance between the optimized RM and the default RM parameters of the MaxEnt model (RM=1) was compared to reveal the marginal effect of model complexity on the accuracy of species distribution prediction, and a parameter optimization scheme suitable for the monthly scale evaluation model was established.

[0094] AUC is the area under the curve, which represents the model's ability to distinguish between the presence and pseudo-absence of pests and diseases, and its value ranges from 0 to 1. The evaluation criteria are: AUC < 0.7 indicates insufficient model accuracy, 0.7 ≤ AUC < 0.9 indicates an effective prediction model, and AUC ≥ 0.9 indicates a high-precision model. The AUC calculation formula is as follows:

[0095]

[0096] Where TPR: true positive rate; FPR: false positive rate.

[0097] After optimizing the RM parameters, the optimal RM parameters in the MaxEnt model for the monthly rice sheath blight habitat suitability evaluation model from April to October were 1.5, 0.5, 3.0, 3.5, 0.5, 1.0, and 1.5, respectively. After optimizing the RM parameters, the AUC values ​​of the rice sheath blight habitat suitability evaluation model from April to October were 87.1%, 85.8%, 83.1%, 80.1%, 81.1%, 82.6%, and 86.2%, respectively. Except for the AUC value in September, which remained unchanged, the AUC values ​​of the evaluation models after the RM parameter optimization increased by 1.8%, 1.6%, 2.1%, 2.7%, 4.3%, and 1.9%, respectively, compared with the default parameters. Figure 3 ).

[0098] Step 4: Input the key habitat factors sensitive to rice sheath blight in each month of the target area into the monthly habitat suitability evaluation model GCCM-MaxEnt to obtain the habitat suitability evaluation results of rice sheath blight in each month of April to October in the target area, that is, the spatial distribution range map of the potential distribution probability of rice sheath blight in each month.

[0099] By comparing the habitat suitability results of rice sheath blight in each month, it was found that the overall habitat suitability area of ​​rice sheath blight showed the characteristics of expanding from south to north and then gradually retreating to the south over time.

[0100] The marginal contribution of each key habitat factor to the evaluation model was quantified using substitution importance. It was found that the leading driving habitat factors of the rice sheath blight habitat suitability evaluation model were TMX and PRE in April, TMX in May, TMP in June-August, and TMX in September-October.

[0101] Ecological response curves were further constructed using the MaxEnt model software's built-in plotting module. Using the standardized habitat factor as the abscissa and the risk probability of sheath blight as the ordinate, the "Cloglog" plotting module within MaxEnt software was used to plot the probability of habitat suitability for sheath blight under different values ​​of the dominant driving habitat factor for each month. The probability of habitat suitability for sheath blight ranged from 0 to 1. Based on these ecological response curves, the nonlinear interaction between the dominant driving habitat factor and habitat suitability for sheath blight was analyzed. The results revealed that the dominant driving habitat factor PRE peaked at 221.6 mm in April, the dominant driving habitat factor TMX peaked at 28.6°C in May, the dominant driving habitat factor TMP peaked at 29.5°C, 29.5°C, and 28.4°C from June to August, and the dominant driving habitat factor TMX peaked at 32.0°C and 30.6°C from September to October, respectively.

[0102] Compared with traditional disease habitat suitability evaluation methods that rely on meteorological plant protection information, expert experience, and statistical methods to select habitat factors, the spatiotemporal dynamic evaluation method for rice sheath blight habitat suitability proposed in this method fully considers the impact of mixed single- and double-season rice cropping on the prevalence and spread of rice sheath blight, and can analyze the impact of habitat differences on crop diseases at a finer spatiotemporal scale. The introduction of spatial causal inference theory can effectively improve the interpretability and reliability of habitat factors. At the same time, modeling disease habitat suitability evaluation with months as the time unit can provide a refined response to the spatiotemporal process characteristics of disease occurrence and epidemic within an interannual period, has stronger mechanistic rationality, and improves the effectiveness and reliability of habitat suitability evaluation. It provides new ideas and methods for the habitat suitability evaluation of rice sheath blight, and provides important background information for large-scale crop disease and insect pest forecasting and prevention, and has application potential.

Claims

1. A spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight, characterized in that: The following steps are involved: Extract monthly rice sheath blight habitat factors by integrating multi-source temporal information; Based on spatial causal inference, the optimal habitat factors were selected to characterize the spatiotemporal heterogeneity of key sensitive habitat factors of rice sheath blight. Construction and optimization of a spatiotemporal dynamic evaluation model for habitat suitability of rice sheath blight based on a species distribution model; Using the monthly rice sheath blight habitat suitability assessment model set, we obtained monthly rice sheath blight habitat suitability assessment results during the critical rice season in the target area. The monthly habitat factors of rice sheath blight were extracted by integrating multi-source time series information, as follows: Obtain the remote sensing, meteorological and plant protection time series data of the area to be evaluated over the years, use them to calculate the disease habitat factors at different temporal and spatial resolutions, and resample each habitat factor to form a time series disease habitat factor with a unified spatial resolution; The disease habitat factors of each month were divided into time series. The mean of the same disease habitat factor in multiple time phases within the same month was calculated. The monthly rice planting spatial distribution data were combined with masking to obtain the disease habitat factors of each month under the constraints of rice planting spatial distribution. The disease habitat factors of each month under the spatial distribution constraints of rice planting in multiple years were averaged according to the disease habitat factors of the same month in different years, and the monthly rice sheath blight habitat factors were extracted by integrating multi-source time series information. The optimization of habitat factors is based on spatial causal inference, as follows: The causal relationship between the monthly habitat factors and the rice sheath blight disease rate was analyzed through spatial causal inference. The key sensitive habitat factors affecting the spread of rice sheath blight were screened based on whether the causal effect index was greater than the set threshold. The primary screening habitat factors were obtained. Spatial causal inference analysis was used to preliminarily screen the possible causal relationships between habitat factors, and a causal-oriented relationship diagram between the habitat factors of rice sheath blight was drawn. The causal-oriented relationship diagram was then combined to achieve the optimization of the key sensitive habitat factors of rice sheath blight in each month.

2. The spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight according to claim 1, characterized in that: The construction and optimization of a spatiotemporal dynamic evaluation model for habitat suitability of rice sheath blight based on a species distribution model are as follows: The selected key sensitive habitat factors of rice sheath blight in each month were used as the input of species distribution model to construct a spatiotemporal dynamic evaluation model of habitat suitability for rice sheath blight. The evaluation model parameters were optimized by combining the corrected AIC information criterion with the evaluation index AUC, forming a monthly-scale rice sheath blight habitat suitability evaluation model group optimized by month.

3. The spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight according to claim 1, characterized in that: The monthly rice planting spatial distribution data is obtained based on the phenological period information of single- and double-season rice: first, based on the transplanting period and harvesting period dates corresponding to the same spatial distribution of rice of each rice-growing type, the rice growing time range under the same spatial distribution of rice of each rice-growing type is obtained; on this basis, the rice growing time range is divided using the corresponding time range of each month as a threshold to obtain the rice planting spatial distribution data of each month.

4. The spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight according to claim 1, characterized in that: The monthly rice sheath blight diseased plant rate data were calculated from the plant protection time-series data: first, representative fields where sheath blight occurred were selected as survey points to obtain the corresponding plant protection time-series data, and the time-series sheath blight diseased plant rate data of each survey point were obtained by calculating the ratio of the number of diseased plants to the total number of surveyed plants; secondly, the annual sheath blight diseased plant rate plant protection time-series data of the survey points were divided into months, and the maximum diseased plant rate value in each month was used as the sheath blight diseased plant rate of the survey point in that month; finally, the monthly rice sheath blight diseased plant rate data were obtained by calculating the average of the sheath blight diseased plant rates in the same month in all years.

5. The spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight according to claim 1, characterized in that: The causal effect index reflects the relationship between the real observation value and the spatial causal inference prediction value through the covariance of the normalized real observation value and the spatial causal inference prediction value; the initial screening of the habitat factor is achieved through the pre-set causal effect index threshold.

6. The spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight according to claim 1, characterized in that: The direction of drawing the causal relationship diagram between the habitat factors of the stripe blight disease is as follows: first, spatial causal inference is used to quantify the causal relationship between the habitat factors of the stripe blight disease, and the causal effect index between the habitat factors is obtained; based on the calculated causal effect index, the causal relationship is defined in a directed acyclic graph using a directed acyclic graph tool, and an interactive and adjustable directed acyclic graph is obtained using an interactive network visualization tool; finally, a causal relationship diagram between the habitat factors of the stripe blight disease is drawn using a graphical grammar drawing tool.

7. The spatiotemporal dynamic evaluation method for habitat suitability of rice sheath blight according to claim 2, characterized in that: The evaluation model parameters are optimized by combining the correction AIC information criterion with the evaluation index AUC. Specifically, the RM value range and step size are set first, and the RM gradient experimental group is determined to quantitatively analyze the regulatory effect of parameter adjustment on model complexity; then the RM with the smallest AICc value corresponding to model complexity under different RMs within the RM value range is selected; finally, the model AUC performance between the optimized RM and the default RM parameters of the species distribution model is compared, and the RM value with the largest corresponding AUC value is selected as the optimal RM to complete the model optimization and establish a parameter optimization scheme suitable for the monthly scale evaluation model.

8. A spatiotemporal dynamic evaluation system for habitat suitability of rice sheath blight, characterized by: It includes the following modules: Habitat factor extraction module: extracts monthly rice sheath blight habitat factors by integrating multi-source time series information; Habitat factor optimization module: optimize key sensitive habitat factors based on spatial causal inference; Evaluation model group acquisition module: Construction and optimization of spatiotemporal dynamic evaluation model of rice sheath blight habitat suitability based on species distribution model; Evaluation module: The key habitat factors sensitive to rice sheath blight in each month of the target area are input into the spatiotemporal dynamic evaluation model of rice sheath blight habitat suitability in each month to obtain the evaluation results of rice sheath blight habitat suitability in each month of the target area, that is, the spatial distribution range map of the potential distribution probability of rice sheath blight in each month.

Images