Crop meteorological disaster remote sensing monitoring method based on multi-source data and mechanism model

Abstract

This invention discloses a remote sensing and forecasting method for crop meteorological disasters based on multi-source data and a mechanistic model. The method integrates multi-source data, including Sentinel-2 multispectral data, surface temperature data, rainfall and temperature data from meteorological stations, and ground-measured crop growth and disaster location data. Based on crop growth mechanisms and phenological characteristics, the method accurately extracts crop planting areas by analyzing the vegetation index growth curves of maize and soybeans. Combining the mechanisms of flood and drought disasters, water sensitivity indices and drought sensitivity indices are calculated respectively, and surface temperature data is introduced to assist in judging the severity of disasters. Using a random forest model, with disaster sensitivity indices and surface temperature as feature variables, and combined with ground-measured disaster locations, the model is trained to achieve disaster range extraction, hierarchical assessment, and yield loss prediction. This invention can quickly help agricultural management departments formulate disaster prevention and mitigation measures.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing monitoring technology, specifically to a remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models. Background Technology

[0002] Global climate change and continuous changes in environmental conditions have led to a significant increase in the frequency and intensity of crop meteorological disasters such as floods and droughts, posing a serious threat to the stability of agricultural production. As a core element in ensuring food security and sustainable agricultural development, timely monitoring and accurate forecasting of crop meteorological disasters have become a critical issue that needs to be addressed in current agricultural production management. Major food crops such as corn and soybeans are extremely sensitive to water conditions during their growth cycle. Floods causing root hypoxia and rot, and drought causing leaf wilting and reduced photosynthetic efficiency, can both result in significant yield reductions or even crop failure, directly impacting regional food supply and farmers' economic benefits.

[0003] Remote sensing technology, with its unique advantages of large-scale coverage, real-time acquisition, and non-destructive detection, has been gradually applied to the field of crop meteorological disaster monitoring. However, existing remote sensing forecasting methods face the following problems:

[0004] (1) The data is mostly based on a single satellite data source. For example, vegetation indices are extracted from optical remote sensing images to judge disasters, or meteorological satellite data is used for macro-trend analysis. The complementary information from multiple data sources is not effectively integrated. Although optical remote sensing can reflect the state of crop canopy, it is easily affected by cloud interference; meteorological data can provide climate background, but lacks precise details at the field scale; ground measurement data can verify the results, but it is not systematically integrated into the forecasting model, resulting in fragmented data information and an inability to fully depict the entire process of disaster occurrence.

[0005] (2) Existing remote sensing and forecasting methods are mostly based on statistical models, lacking a deep integration with crop growth mechanisms and disaster occurrence mechanisms. For example, some models only establish correlations between historical disaster data and remote sensing indices, without considering the differences in crop response to water stress at different growth stages, such as the jointing stage of maize and the flowering stage of soybeans, or analyzing the influence mechanism of micro-processes such as physiological dehydration caused by soil hypoxia and drought due to flooding on spectral characteristics. Such statistical models, which are detached from the mechanism, show a significant decrease in adaptability and stability when facing complex climatic conditions or different crop types in different regions, and are prone to misjudgment and omission. For example, normal physiological senescence of crops may be misjudged as drought stress, or the extraction of flood range in low-lying areas may be biased due to the neglect of topographic factors.

[0006] (3) The existing disaster assessment system is imperfect. It focuses only on the statistics of disaster area and lacks a unified standard for the classification of disaster severity. It also fails to link disaster with crop yield loss, and cannot provide a quantitative basis for the priority formulation of subsequent disaster reduction measures. The results verification link is weak. It often lacks systematic ground measurement data support and determines the accuracy of the forecast by verifying only a small number of points, which makes it difficult to guarantee the reliability when applied on a large scale.

[0007] In summary, existing remote sensing methods for monitoring meteorological disasters face problems such as insufficient integration of multi-source data, weak integration of mechanisms, and poor adaptability. A remote sensing forecasting method capable of systematically integrating multi-source data and deeply fusing mechanistic models is needed. Summary of the Invention

[0008] To address the aforementioned problems in existing technologies, this invention integrates remote sensing data, meteorological data, ground-measured data, and basic geographic data, and fuses mechanistic models to propose a remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanistic models, comprising the following steps:

[0009] S1: Collect multi-source data and perform standardization processing on the multi-source data; the multi-source data includes remote sensing data, meteorological data, ground-measured data, and basic geographic data; the standardization processing includes remote sensing data preprocessing and meteorological data preprocessing;

[0010] S2: Based on crop growth mechanisms and phenological characteristics of remote sensing data, construct vegetation index growth curves, and combine extraction rules to screen and correct crop planting areas;

[0011] S3: Calculation of key meteorological disaster indicators. Based on the differences in the mechanisms of floods and droughts, the flood disaster sensitivity index and drought disaster sensitivity index are calculated separately, and surface temperature data is introduced to help determine the degree of disaster.

[0012] S4: Based on the flood and drought disaster sensitivity indices, a classification model is constructed to classify crop planting areas to extract the disaster range. The disaster level is divided by combining the disaster duration, disaster sensitivity index threshold and crop damage symptoms. A regression model is used to predict yield loss.

[0013] S5: Select ground verification points to compare the forecast results with the actual measurement results to verify the accuracy. At the same time, compare the forecast data with the statistical data of the agricultural department to carry out disaster spatial distribution, temporal variation analysis and disaster-causing factor analysis.

[0014] S6: Construct a disaster remote sensing and forecasting database to store various types of data and compile meteorological disaster remote sensing monitoring reports.

[0015] Preferably, step S1 includes:

[0016] S11: Multi-source data acquisition; the remote sensing data includes multispectral data and surface temperature data; the multispectral data includes green band, red band, near-infrared band, and shortwave infrared band, with a temporal resolution of 5 days; the surface temperature data has a spatial resolution of 1 km and a temporal resolution of 8 days; the meteorological data includes daily rainfall, daily average temperature, and daily maximum temperature data; the ground-measured data includes crop growth, disaster occurrence, disaster location coordinates, and crop leaf sample data; the basic geographic data includes administrative division vector maps and cultivated land distribution vector maps.

[0017] S12: Multi-source data preprocessing; the remote sensing data preprocessing includes radiometric correction, atmospheric correction, geometric correction, cloud removal, and data synthesis; the radiometric correction converts digital quantization values ​​into surface radiance values, the atmospheric correction eliminates the effects of aerosol scattering and water vapor absorption, the geometric correction uses topographic maps as a reference to control errors within a threshold, and the cloud removal data synthesis generates continuous cloud-free images; the meteorological data preprocessing includes removing outliers, interpolating station data into raster data, and converting the projection coordinate system.

[0018] Preferably, step S2 includes:

[0019] S21: Calculate vegetation index using near-infrared and red light reflectance; construct vegetation index curves based on vegetation indices of different crops at different growth stages;

[0020] S22: Establish planting area extraction rules and extract planting areas; the extraction rules include: (1) screening areas where the peak value of vegetation index is greater than the threshold and the growth cycle exceeds the threshold as the preliminary crop planting areas; (2) distinguishing the preliminary range of different crops based on the phenological differences of different crops; (3) superimposing the cultivated land distribution vector map to eliminate non-cultivated land areas, and correcting the crop type by combining the actual measured planting points on the ground.

[0021] Preferably, step S3 includes:

[0022] S31: Calculation of flood disaster indicators, wherein the flood disaster sensitivity index includes the normalized water body index, the improved normalized water body index, the automatic water body extraction index, and the WI2015 index;

[0023] S32: Calculation of drought disaster indicators, wherein the drought disaster sensitivity index includes shortwave infrared vertical water loss index, normalized differential water index, global vegetation water index, shortwave infrared water stress index, and shortwave angular slope index.

[0024] S33: Surface temperature auxiliary calculation, resample the surface temperature data to the required resolution, calculate the daily average surface temperature of the crop planting area and eliminate fluctuations by using the window moving average method; if at least two drought sensitivity indices reach the drought threshold and the daily average temperature is more than 3°C higher than normal, drought is preliminarily confirmed; if at least two flood sensitivity indices reach the flood threshold and the daily average temperature is more than 1°C lower than normal, flood is preliminarily confirmed.

[0025] Preferably, step S4 includes:

[0026] S41: Disaster range extraction; The classification model adopts a random forest model, with ground-measured disaster locations as labels, and the labels are divided into "affected" and "not affected"; The characteristic variables of flood disaster are flood disaster sensitivity index and surface temperature, and the characteristic variables of drought disaster are drought disaster sensitivity index and surface temperature; The dataset is divided into training set and test set to verify and optimize the parameters, and the splitting criterion is the Gini coefficient; Output a binary map of "not affected - affected";

[0027] S42: Disaster severity assessment; flood disaster classification based on normalized water index, improved normalized water index, duration of waterlogging, degree of crop leaf curling, and chlorophyll content; drought disaster classification based on shortwave infrared vertical water loss index, global vegetation moisture index, daily average surface temperature, degree of crop leaf curling, and chlorophyll content.

[0028] S43: Disaster impact assessment; Affected area statistics: The number of pixels for each disaster level is counted using the area calculation tool of the geographic information system, and the actual area is calculated by combining the spatial resolution of remote sensing data; Production loss prediction: Based on the correlation between production data and disaster level, a regression model is used to establish the relationship between production loss rate and disaster level; Level production loss equals the affected area of ​​that level × the average reduction rate of that level × the yield per unit area in a normal year; Total production loss equals the sum of production losses of each level.

[0029] Preferably, step S5 includes:

[0030] S51: Accuracy verification and data comparison verification; The accuracy verification selects ground verification points, covering various types of disasters including unaffected, lightly affected, moderately affected, and severely affected. A confusion matrix is ​​constructed by comparing the forecast results with the measured results, and the accuracy, recall, and F1 score are calculated. If the accuracy does not meet the standard, the exponential threshold is adjusted or the model parameters are optimized; The data comparison verification compares the disaster-affected area reported by remote sensing with the statistical data of the Agriculture and Rural Affairs Bureau, and the error is controlled within the threshold.

[0031] S52: The spatial distribution analysis uses a geographic information system to draw a spatial distribution map of disaster levels and overlays digital elevation model data to analyze the impact of terrain on disaster distribution; the temporal variation analysis compares disaster data from the same period in consecutive years to analyze the trend of disaster occurrence; the disaster-causing factor analysis combines meteorological data and uses principal component analysis to identify key driving factors of disaster occurrence.

[0032] Preferably, S6 includes:

[0033] S61: Construct a database to store data; the database includes a remote sensing data layer, a crop data layer, a disaster data layer, an indicator data layer, and a verification data layer; the remote sensing data layer stores preprocessed remote sensing images and resampled surface temperature images; the crop data layer stores vector data of crop planting areas and attributes such as planting area and plot number; the disaster data layer stores vector data of flood and drought disaster distribution and attributes such as disaster level, affected area, and occurrence time; the indicator data layer stores raster data of flood and drought sensitivity indices; the verification data layer stores vector data of ground verification points and attributes such as coordinates, measured disaster level, crop growth parameters, and sampling time.

[0034] S62 generates a remote sensing monitoring report on meteorological disasters; the remote sensing monitoring report on meteorological disasters includes an overview of the study area, data and methods, monitoring results, verification accuracy, and prevention and control recommendations.

[0035] Preferably, the flood disaster classification includes: (1) mild flood: chlorophyll content decreases by 5%-10%, and root activity decreases by 10%-15%; (2) moderate flood: chlorophyll content decreases by 10%-20%, and root activity decreases by 15%-25%; (3) severe flood: chlorophyll content decreases by >20%, and root activity decreases by >25%.

[0036] The drought disaster classification includes: (1) mild drought: leaf curling degree 1, photosynthetic efficiency decreases by 10%-15%; (2) moderate drought: leaf curling degree 2, photosynthetic efficiency decreases by 15%-25%; (3) severe drought: leaf curling degree 3, photosynthetic efficiency decreases by >25%, plant growth stagnates.

[0037] Preferably, the formulas for calculating the flood disaster sensitivity index are as follows:

[0038] Normalized Difference Water Index = (Green band reflectance - Near-infrared band reflectance) / (Green band reflectance + Near-infrared band reflectance);

[0039] Improved Normalized Difference Water Index = (Green band reflectance - Shortwave infrared band reflectance) / (Green band reflectance + Shortwave infrared band reflectance);

[0040] Automatic water extraction index = 4 × (green band reflectance - shortwave infrared band reflectance) - (0.25 × near-infrared band reflectance + 2.75 × shortwave infrared band reflectance);

[0041] WI2015 index = Green band reflectance - 0.1 × Near-infrared band reflectance - 0.9 × Short-wave infrared band reflectance;

[0042] The formulas for calculating the drought sensitivity index are as follows:

[0043] Shortwave infrared vertical water loss index = (shortwave infrared band reflectance - near-infrared band reflectance) / (shortwave infrared band reflectance + near-infrared band reflectance);

[0044] Normalized differential moisture index = (near-infrared reflectance - mid-infrared reflectance) / (near-infrared reflectance + mid-infrared reflectance), where the mid-infrared band is Sentinel-2 band 11;

[0045] Global vegetation moisture index = [(near-infrared reflectance + 0.1) - (short-wave infrared reflectance + 0.02)] / [(near-infrared reflectance + 0.1) + (short-wave infrared reflectance + 0.02)];

[0046] Shortwave infrared moisture stress index = shortwave infrared reflectance / near-infrared reflectance;

[0047] Shortwave angle slope index = (shortwave infrared band reflectance - red light band reflectance) / (shortwave infrared band reflectance + red light band reflectance).

[0048] Preferably, the vegetation index curves include a maize vegetation index curve and a soybean vegetation index curve; the peak value of the flowering period index in the maize vegetation index curve is 0.8-0.9, and the maturity period index is ≤0.4; the peak value of the flowering and pod-setting period index in the soybean vegetation index curve is 0.7-0.8, and the maturity period index is ≤0.3; the growth cycle threshold is ≥120 days, and the growth cycle is the duration from the first rise of the vegetation index to the first drop below 0.4.

[0049] The present invention has the following beneficial effects:

[0050] (1) This invention integrates remote sensing data, meteorological data, ground-measured data, and basic geographic data. Sentinel-2 multispectral data provides high-resolution crop canopy spectral information, surface temperature data supplements the temperature environment background for crop growth, meteorological station data supports the analysis of disaster-causing factors, ground-measured data verifies the model accuracy, and basic geographic data defines the crop planting range. The complementary fusion of multi-source data effectively avoids the shortcomings of optical remote sensing, such as susceptibility to cloud interference and lack of field details in meteorological data. It fully captures the entire process of disaster from occurrence to development, avoids monitoring bias caused by information fragmentation, and provides comprehensive data support for disaster forecasting.

[0051] (2) This invention deeply integrates crop physiological response patterns and disaster action mechanisms into the forecasting process. For example, considering the different sensitivities to water stress at different growth stages such as the jointing stage of maize and the flowering stage of soybean, the selection of disaster sensitivity indices and threshold settings are optimized by combining physiological processes such as root hypoxia caused by flooding and leaf dehydration caused by drought. At the same time, when constructing the forecasting model based on the random forest model, feature variables are selected based on crop growth mechanisms. This mechanism-based design enables the method to maintain stable performance in different crop types and different climatic regions, effectively reducing problems such as misjudging normal crop aging as drought and ignoring topography leading to deviations in flood range, and significantly improving the accuracy and regional adaptability of forecasting.

[0052] (3) Relying on the advantages of remote sensing technology in large-scale coverage and rapid acquisition, combined with the efficient data processing capabilities of cloud computing platforms, this method can complete disaster monitoring of tens of thousands of hectares of farmland in a short time, generate a spatial distribution map of disasters across the entire region within 1-3 days after a disaster occurs, and promptly capture the dynamic spread of floods and droughts, providing timely support for the rapid formulation of disaster prevention and mitigation measures.

[0053] (4) This invention constructs a multi-level disaster assessment system encompassing scope, severity, and impact. This system not only accurately counts the affected area but also classifies disasters into multiple levels based on sensitivity index thresholds and crop damage symptoms. Furthermore, it incorporates a correlation model between crop yield and disaster level to predict yield losses. This multi-dimensional assessment overcomes the limitations of existing methods that only count the affected area, providing agricultural management departments with precise quantitative data to prioritize disaster prevention and mitigation, thus helping to reduce economic losses caused by disasters.

[0054] This invention can provide continuous data and technical support for the agricultural sector to formulate long-term disaster prevention and mitigation plans and study the impact of climate change on crop disasters. Attached Figure Description

[0055] Figure 1 This is a step diagram of an embodiment of the present invention;

[0056] Figure 2 This is a technical roadmap for an embodiment of the present invention;

[0057] Figure 3 This is a remote sensing monitoring map of flooding in key areas of Inner Mongolia Autonomous Region, as described in an embodiment of the present invention.

[0058] Figure 4 This is a remote sensing monitoring map of drought in key areas of Inner Mongolia Autonomous Region, as described in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] This study area covers Jungar Banner in Ordos City, Arukorqin Banner in Chifeng City, and Siziwang Banner in Ulanqab City, Inner Mongolia Autonomous Region. This region is a typical dryland farming area in northern China, primarily cultivating corn and soybeans. The average annual rainfall is 200-400 mm, concentrated in July and August, making seasonal flooding likely. May and June often experience prolonged periods of low rainfall and high temperatures, leading to frequent droughts. The study area possesses complete agricultural statistical data and ground observation stations, meeting the needs for disaster forecasting and verification. Figure 1 and Figure 2 As shown, this embodiment combines multi-source data with a mechanistic model to achieve accurate monitoring and assessment of flood and drought disasters affecting corn and soybeans in this region. The method includes the following steps:

[0061] S1: Multi-source data acquisition and preprocessing.

[0062] S11: Multi-source data acquisition. To meet the needs of the entire disaster forecasting and reporting process, core data covering multiple dimensions including remote sensing, meteorology, surface data, and geography will be collected to ensure spatiotemporal matching and information complementarity, including:

[0063] 1. Remote Sensing Data Acquisition. The core data source is multispectral data from the Sentinel-2 satellite at the European Space Agency's Copernicus Data Center, with spatial resolutions of 10m for the green, red, and near-infrared bands, and 20m for the shortwave infrared band. Several key bands are included: the green band (band 3, center wavelength 542.4nm, bandwidth 27nm) for identifying sensitive signals related to water bodies and vegetation chlorophyll; the red band (band 4, center wavelength 665nm, bandwidth 31nm) for reflecting differences in vegetation photosynthesis; the near-infrared band (band 8, center wavelength 842nm, bandwidth 145nm) for capturing vegetation canopy structure and biomass information; and the shortwave infrared band (band 11, center wavelength 1610nm, bandwidth 143nm) for monitoring soil and vegetation moisture changes. Data covers the entire crop growth period from April to October, with one scene acquired every 5 days to ensure the capture of critical points in the event of a disaster.

[0064] To acquire land surface temperature (LST) data from NASA's MODIS (Moderate-resolution Imaging Spectroradiometer) Earth Observation System (MOD11A2), with a spatial resolution of 1 km and a temporal resolution of 8 days, noise from single-day data was eliminated through 8-day data synthesis to obtain the temperature environment background for crop growth, with a focus on extracting temperature data during the high-temperature and drought period from May to August and the flood period from July to August.

[0065] 2. Meteorological Data Acquisition. Daily-scale data for three consecutive years from 20 national meteorological stations in the study area and surrounding regions were collected, including daily rainfall, daily average temperature, and daily maximum temperature. Daily rainfall was used to determine the flood threshold (cumulative rainfall ≥ 50 mm over three consecutive days prior to flooding) and the drought threshold (no effective rainfall for 15 consecutive days during a drought period, with daily rainfall < 0.1 mm). Daily average temperature and daily maximum temperature were used to assist in drought level classification; daily average temperature ≥ 25℃ was required during a drought period. Data was sourced from the China Meteorological Administration's meteorological data sharing service platform. Data integrity was ensured, with a missing data rate < 5%. Missing data was supplemented using linear interpolation from adjacent stations.

[0066] 3. Ground measurement data acquisition. During the key growth stages of corn and soybean, namely corn jointing stage (June-July), flowering stage (July-August), and grain filling stage (August-September), and soybean branching stage (June-July), flowering and pod setting stage (July-August), and grain filling stage (August-September), 120 ground verification points were set up using a grid-based method, with 40 points in each banner / county, covering plots of high, medium, and low yield levels, and each verification point had a control range of 100m×100m. The recorded content includes: (1) Crop growth, plant height is measured every 10 days and the average value of 5 plants is taken; leaf moisture content is determined by drying and weighing method, and the top 3-5 functional leaves are taken; (2) Disaster occurrence, the number of days of flooding is recorded by combining on-site inspection and farmer interviews; the degree of leaf curling in drought is divided into 0-3 levels, level 0 no curling, level 1 slight curling, level 2 moderate curling, and level 3 severe curling; (3) Disaster location coordinates are located by GNSS with an accuracy of 1m; (4) Crop leaf samples are collected simultaneously, 10 functional leaves are collected at each point, chlorophyll content is measured by SPAD-502 chlorophyll meter, 3 points are measured for each leaf and the average is taken, nitrogen content is determined by Kjeldahl method, and used to verify the effectiveness of remote sensing index.

[0067] 4. Acquisition of Basic Geographic Data. A 1:50,000 administrative division vector map of the study area was acquired, including the boundaries of banners, counties, and townships, to define the administrative scope for disaster monitoring and reporting. A 1:10,000 vector map of cultivated land distribution from the Ministry of Natural Resources' Land Use Status Database was also acquired, distinguishing cultivated land from non-cultivated land such as forest land, grassland, and building land, ensuring that disaster monitoring and reporting focus solely on crop-growing areas. Both types of data were projected using the 2000 National Geodetic Coordinate System and the Albers equal-area conic projection, consistent with the remote sensing data projection, to avoid spatial bias.

[0068] S12: Multi-source data preprocessing. Professional software, such as the remote sensing image processing platform ENVI, the geographic information system ArcGIS, and Google Earth Engine, is used to perform data standardization processing, eliminate interference factors, and ensure data consistency. Remote sensing data preprocessing includes:

[0069] (1) Radiometric correction: The Sentinel-2 data is radiometrically calibrated to convert the digital quantization values ​​received by the sensor into surface radiance values. The absolute radiometric calibration coefficients provided by the European Space Agency are used. The calculation formula is: Radiance value = Digital quantization value × Calibration coefficient gain + Calibration coefficient offset, which eliminates the radiation deviation caused by sensor response differences.

[0070] (2) Atmospheric correction: Using the FLAASH module of ENVI, based on the MODTRAN4 atmospheric radiative transfer model, the atmospheric profile data obtained from the NASA AERONET site on the same day for the study area were input, including aerosol optical thickness and water vapor content. The aerosol type was selected as continental type, and the water vapor inversion method adopted was the near-infrared band ratio method to eliminate the influence of aerosol scattering and water vapor absorption on spectral reflectance, so as to obtain the true surface reflectance data. The reflectance error after correction was controlled within 5%.

[0071] (3) Geometric correction: Based on the 1:50,000 topographic map of the study area, 15-20 ground control points were selected, including road intersections, river bends, and reservoir dams. The control points were required to be evenly distributed within the image area and to avoid being located in cloud or shadow areas. A quadratic polynomial transform model with a fitting accuracy of R²≥0.95 and bilinear interpolation were used for resampling to control the geometric accuracy error of the Sentinel-2 data to within 1 pixel 10m, ensuring spatial matching of data from different time phases.

[0072] (4) Cloud Removal and Data Synthesis: Using the Sentinel-2 quality control QA60 band, cloud areas with a QA60 value ≥ 1024 and cloud shadow areas with a QA60 value ≥ 2048 were identified by pixel values. Median synthesis was performed on multiple Sentinel-2 images from the same growth stage, such as the corn jointing stage from June 20 to July 20, to retain the reflectivity of cloudless pixels and generate continuous cloudless image data. This avoids monitoring loopholes caused by cloud coverage in single-image data. The cloud cover in the synthesized image is < 5%.

[0073] Meteorological data preprocessing includes:

[0074] (1) Conduct quality control and eliminate outliers. Daily rainfall > 200 mm and daily average temperature > 40℃ or < 0℃ are considered outliers. Outliers are identified using the 3-times standard deviation method.

[0075] (2) The meteorological data of the stations were interpolated into raster data with a spatial resolution of 1 km using the Kriging interpolation method. A spherical semivariogram model was selected for interpolation, with a fitting accuracy of R²≥0.9 and a search radius of 50 km to ensure that the interpolation results conform to the regional meteorological spatial distribution pattern.

[0076] (3) Convert the interpolated meteorological raster data into a projection coordinate system consistent with the Sentinel-2 data to facilitate subsequent multi-source data joint analysis.

[0077] S2: Precise Extraction of Crop Planting Areas. Based on crop growth mechanisms and phenological characteristics from remote sensing data, corn and soybean planting areas were extracted using the GEE (Google Earth Engine) platform, eliminating interference from non-planting areas. The specific process is as follows:

[0078] S21: NDVI (Normalized Difference Vegetation Index) Curve Analysis during Phenological Periods. Sentinel-2 NDVI data for the study area were obtained for two consecutive years. The normalized vegetation index is calculated using the formula: NDVI = (Near-infrared reflectance - Red reflectance) / (Near-infrared reflectance + Red reflectance). NDVI values ​​range from -1 to 1, with negative values ​​representing water bodies and positive values ​​representing vegetation; higher values ​​indicate higher vegetation cover. The mean NDVI was calculated at 16-day intervals, and NDVI curves for the growth cycles of maize and soybean were constructed.

[0079] Maize NDVI curve: During the sowing period from mid-April to early May, NDVI < 0.3, indicating the crop is in the seedling stage with vegetation cover < 10%; during the jointing period from early June to early July, NDVI rapidly rises to 0.6-0.8, indicating the crop enters a rapid growth stage with stem elongation and increased leaf area index; during the flowering period from mid-July to early August, NDVI reaches its peak at 0.8-0.9, with vegetation cover > 80% and strongest photosynthesis; during the grain-filling period from mid-August to early September, NDVI slowly decreases to 0.6-0.8, indicating nutrient transfer to the grains; during the maturity period from mid-September to early October, NDVI drops below 0.4, with leaf yellowing and weakened photosynthesis.

[0080] Soybean NDVI curve: Sowing period (early to late May): NDVI < 0.2, vegetation cover < 5% during emergence stage; Branching period (mid-June to mid-July): NDVI rises to 0.5-0.7, branching growth and leaf area expansion; Flowering and pod setting period (late July to mid-August): NDVI reaches peak value of 0.7-0.8, vegetation cover > 70%; Grain filling period (late August to mid-September): NDVI drops to 0.5-0.7, grain filling; Maturity period (late September to early October): NDVI drops below 0.3, plant withering.

[0081] S22: Construction of Planting Area Extraction Rules. Combining NDVI curve characteristics and basic geographic data, a three-level extraction rule system is established to ensure extraction accuracy, including:

[0082] 1. Preliminary screening of crop planting areas: Screen areas with an NDVI peak value ≥ 0.6 and a growth cycle (the duration from the first rise of NDVI to 0.3 to the first drop below 0.4) ≥ 120 days. An NDVI peak value ≥ 0.6 ensures that the area has high-density vegetation with a coverage > 50%, and a growth cycle ≥ 120 days excludes short-term weeds with a growth cycle < 90 days, thus preliminarily identifying crop planting areas.

[0083] 2. Differentiate between corn and soybean planting areas: Select NDVI data from June 20 to June 30. At this time, corn is in the late jointing stage with NDVI ≥ 0.6; soybean is in the early branching stage with NDVI ≤ 0.5. Use this difference in NDVI thresholds to preliminarily divide the two types of crops.

[0084] 3. Optimize Extraction Results: Overlay a 1:10,000 farmland distribution vector map to remove misclassified areas outside the farmland boundaries. For example, forest land may have a high NDVI but not be farmland, while building land may have a low NDVI but be non-farmland. Combine this with ground-measured crop planting points to correct the extraction results. If an area is extracted as corn but actually measured as soybean, the crop type for that area is adjusted. Finally, vector data for corn and soybean planting areas are obtained. The extraction accuracy is calculated using ground verification points, with an overall accuracy ≥90% and a Kappa coefficient ≥0.85.

[0085] S3: Calculation of Key Meteorological Disaster Indicators. Based on the principles of electromagnetic wave reflection and absorption, and combined with the differences in the mechanisms of floods and droughts, sensitivity indices for the two types of disasters are calculated to quantify the characteristics of disaster occurrence. Surface temperature data is also introduced to assist in judging the severity of disasters.

[0086] S31: Calculation of Flood Disaster Indicators. The core of flood disasters is water body expansion and crop flooding stress. Multiple water sensitivity indices can be selected to highlight the spectral differences between water bodies, vegetation, and soil, including:

[0087] 1. Normalized Difference Water Index (NDWI): This index enhances water body identification by utilizing the difference in reflectance between the green band and the near-infrared band. The calculation formula is: NDWI = (Green band reflectance - Near-infrared band reflectance) / (Green band reflectance + Near-infrared band reflectance). In flood-prone areas, due to high water reflectance, the NDWI is ≥0.1; in healthy vegetation areas, due to high near-infrared reflectance, the NDWI is ≤0; and in soil areas, the NDWI is between 0 and 0.1.

[0088] 2. Modified Normalized Difference Water Index (MNDWI): This index replaces the near-infrared band with the short-wave infrared band to reduce vegetation interference (vegetation has low reflectivity in the short-wave infrared band). The calculation formula is: Modified Normalized Difference Water Index = (Green band reflectivity - Short-wave infrared band reflectivity) / (Green band reflectivity + Short-wave infrared band reflectivity). It is suitable for areas with high vegetation cover (such as during the corn flowering period). In flood-prone areas, the modified normalized difference water index should be ≥0.2, and in non-flood-prone areas ≤0.1.

[0089] 3. Automated Water Extraction Index (AWEI): This index suppresses background noise from soil and vegetation through multi-band combination. The calculation formula is: AWEI = 4 × (Green band reflectance - Shortwave infrared band reflectance) - (0.25 × Near-infrared band reflectance + 2.75 × Shortwave infrared band reflectance). In flood-prone areas, the AWEI is ≥-100; in non-water areas, it is ≤-100. This index has high accuracy in identifying shallow water areas (water depth <10cm).

[0090] 4. Water Index 2015 (WI2015): This index optimizes the distinction between water and soil. The calculation formula is: WI2015 = Green band reflectance - 0.1 × Near-infrared band reflectance - 0.9 × Short-wave infrared band reflectance. In flood-prone areas, the WI2015 index is ≥0.05, while in soil areas it is <0.05. This effectively eliminates interference from arid soils, as arid soils have high reflectance in the short-wave infrared band, resulting in a lower WI2015 index.

[0091] S32: Calculation of Drought Disaster Indicators. The core of drought disaster is soil moisture deficit and crop water stress. Five drought sensitivity indices are selected to cover soil and vegetation moisture status:

[0092] 1. Shortwave Infrared Perpendicular Water Stress Index (SPSI): Based on the water absorption characteristics of the shortwave infrared band, water has strong absorption and low reflectivity in the shortwave infrared band, used to monitor soil moisture loss. The calculation formula is: Shortwave Infrared Perpendicular Water Stress Index = (Shortwave Infrared Band Reflectivity - Near-Infrared Band Reflectivity) / (Shortwave Infrared Band Reflectivity + Near-Infrared Band Reflectivity). In arid areas, due to low soil moisture, shortwave infrared reflectivity is high, and the SPSI is <-0.3; in normal areas, soil moisture is sufficient, shortwave infrared reflectivity is low, and the SPSI is >-0.1.

[0093] 2. Normalized Difference Moisture Index (NDMI): Reflects the leaf moisture content of vegetation. The calculation formula is: NDMI = (Near-infrared reflectance - Mid-infrared reflectance) / (Near-infrared reflectance + Mid-infrared reflectance), where the mid-infrared band uses Sentinel-2 band 11 (1610nm). Under drought stress, crop leaf moisture content decreases, mid-infrared reflectance increases, and the NDMI is ≤-0.1; healthy crops have high leaf moisture content and low mid-infrared reflectance, and the NDMI is ≥0.

[0094] 3. Global Vegetation Moisture Index (GVMI): Enhances vegetation moisture signal and reduces background interference. The calculation formula is: Global Vegetation Moisture Index = (Near-infrared reflectance + 0.1) - (Short-wave infrared reflectance + 0.02) / [(Near-infrared reflectance + 0.1) + (Short-wave infrared reflectance + 0.02)]. In arid regions, vegetation moisture is insufficient, with a GVMI < 0.4; in healthy regions, it > 0.6. This index is sensitive to drought during the soybean grain-filling stage.

[0095] 4. Shortwave Infrared Water Stress Index (SIWSI): This index reflects vegetation water stress by comparing the ratio of shortwave infrared to near-infrared reflectance. The formula is: Shortwave Infrared Water Stress Index = Shortwave Infrared Reflectance / Near-Infrared Reflectance. In arid regions, vegetation experiences severe water stress, resulting in a relatively higher shortwave infrared reflectance and an SIWSI > 0.5; in normal regions, the SIWSI < 0.3.

[0096] 5. Shortwave Angle Slope Index (SASI): This index assesses vegetation health based on changes in the slope of the shortwave infrared band. The formula is: Shortwave Angle Slope Index = (Shortwave Infrared Band Reflectance - Red Band Reflectance) / (Shortwave Infrared Band Reflectance + Red Band Reflectance). In arid regions, damaged vegetation leaves result in a large rate of change in shortwave infrared reflectance, with a Shortwave Angle Slope Index < -0.2; in normal regions, it > -0.1.

[0097] S33: Surface Temperature Auxiliary Calculation. Based on MODIS surface temperature data, the spatial resolution was first resampled to 20m using bilinear interpolation to match the Sentinel-2 data. Then, the daily average surface temperature of the crop-growing area was calculated, and a 3-day moving average method was used to eliminate daily temperature fluctuations. During droughts, the daily average surface temperature is 3-5℃ higher than the average of the same period in the past 5 years. For example, the normal daily average temperature during the corn jointing stage is 22℃, but it can reach 25-27℃ during droughts. During floods, water evaporates and cools the water, resulting in a daily average surface temperature 1-2℃ lower than the normal year. Combining the daily average surface temperature with drought and flood sensitivity indices, if at least two drought sensitivity indices reach the drought threshold and the daily average temperature is more than 3℃ higher than the normal year, drought can be preliminarily confirmed; if at least two flood sensitivity indices reach the flood threshold and the daily average temperature is more than 1℃ lower than the normal year, flooding can be preliminarily confirmed, reducing misjudgments based on a single index.

[0098] S4: Disaster Assessment and Spatial Extraction. Based on the above indicators, and combining a random forest model with spatial analysis techniques, the area, severity, and impact of floods and droughts are assessed, generating spatial distribution maps of the disasters. The steps include:

[0099] S41: Disaster range extraction.

[0100] 1. Random Forest Model Construction.

[0101] Sample preparation was conducted. Disaster locations measured on the ground were used as labels. Flood disaster labels were divided into "affected" (≥1 day of water accumulation) and "unaffected" (no water accumulation). Drought disaster labels were divided into "affected" (leaf curling degree ≥1) and "unaffected" (leaf curling degree 0). 1000 sample points were selected for each disaster type: 700 affected points and 300 unaffected points. Sample points were randomly selected from within the control range of 120 ground verification points to ensure sample representativeness.

[0102] Feature variables were selected. Feature variables for flood disasters include the Normalized Water Index (NDI), the Modified Normalized Water Index (MDI), the Automatic Water Extraction Index (ADI), the WI2015 Index, and surface temperature. Feature variables for drought disasters include the Shortwave Infrared Vertical Water Loss Index (SWISE), the Normalized Differential Moisture Index (NDM), the Global Vegetation Moisture Index (GVM), the Shortwave Infrared Moisture Stress Index (SWISE), the Shortwave Angle Slope Index, and surface temperature.

[0103] Model parameter optimization was performed. The dataset was divided into a 7:3 ratio (70% training set and 30% test set), and 5-fold cross-validation was used to optimize the parameters. The number of decision trees was tested using grid search, ranging from 50 to 200. The model exhibited the lowest out-of-bag error (OOB) when the number of decision trees was 100, with flood OOB errors at 8.2% and drought OOB errors at 9.5%. The number of features was set to the square root of the total number of features. For floods, 2 out of 5 features were selected, and for droughts, 2 out of 6 features were selected. Random feature selection reduced the correlation between decision trees. The Gini coefficient was used as the splitting criterion for the decision trees, as this coefficient measures node purity; a smaller Gini coefficient indicates higher node purity.

[0104] Model training was performed. Model training was implemented using Python's Scikit-learn library. Each decision tree was constructed based on the bootstrap samples of the training set. Each node selected the optimal split point from chosen features, with the split point selection criterion being the maximum reduction in the Gini coefficient. After training, key features were selected based on feature importance evaluation, specifically the sum of Gini coefficient reductions. In flood disasters, the improved normalized water index had the highest importance, accounting for 35%, while in drought disasters, the shortwave infrared vertical water loss index had the highest importance, accounting for 32%.

[0105] 2. Disaster Range Extraction. The optimized random forest model was applied to the preprocessed remote sensing data to classify the maize and soybean planting areas in the study area, outputting a binary map of "unaffected - affected," where 1 represents affected and 0 represents unaffected. Post-processing was performed using spatial analysis tools of the Geographic Information System (GIS). A removal tool was used to eliminate scattered affected patches smaller than 0.1 hectares (100 pixels at 10m resolution), as these patches were often misclassified, such as small puddles in the field or drought-affected individual crops. A fusion tool was used to merge adjacent affected patches to form continuous affected areas. Finally, spatial distribution vector data of floods and droughts was obtained, including the location, area, and crop type information of the affected areas, such as... Figure 3 , Figure 4 As shown.

[0106] S42: Disaster Severity Assessment. Based on the duration of the disaster, sensitivity index thresholds, and crop damage symptoms, floods and droughts are classified into three levels: mild, moderate, and severe. The grading standards are as follows:

[0107] 1. Classification of flood disasters.

[0108] Mild: Normalized water index 0.1-0.2, improved normalized water index 0.2-0.3, waterlogging lasts for 1-3 days; crops show slight yellowing of leaves, chlorophyll content decreases by 5%-10%, no obvious wilting, mild root hypoxia, root activity decreases by 10%-15%, with little impact on yield.

[0109] Moderate: Normalized water index 0.2-0.3, improved normalized water index 0.3-0.4, waterlogging lasts for 3-7 days; crops show large-area yellowing of leaves, chlorophyll content decreases by 10%-20%, some lower leaves wither, root system is moderately hypoxic, root activity decreases by 15%-25%, and the impact on yield is moderate.

[0110] Severe: Normalized water index > 0.3, improved normalized water index > 0.4, waterlogging lasts for more than 7 days; crops show overall wilting, chlorophyll content decreases by more than 20%, root rot, root vitality decreases by more than 25%, and has a serious impact on yield.

[0111] 2. Drought disaster classification.

[0112] Mild: Shortwave infrared vertical water loss index -0.3 to -0.1, global vegetation moisture index 0.4-0.6, daily average surface temperature 3°C higher than normal; crops show mild leaf curling, curling degree 1, chlorophyll content decreased by 10%-15%, and photosynthetic efficiency decreased by 10%-15%.

[0113] Moderate: Shortwave infrared vertical water loss index -0.5 to -0.3, global vegetation moisture index 0.2-0.4, daily average surface temperature 3-5℃ higher than normal; crops show severe leaf curling, curling degree 2, chlorophyll content decreased by 15%-25%, and photosynthetic efficiency decreased by 15%-25%.

[0114] Severe: Shortwave infrared vertical water loss index < -0.5, global vegetation moisture index < 0.2, daily average surface temperature more than 5°C higher than normal; crops show leaf drying, curling degree 3, plant growth stagnation, chlorophyll content decrease > 25%, photosynthetic efficiency decrease > 25%.

[0115] S43: Disaster impact assessment.

[0116] 1. Disaster Area Statistics. In the Geographic Information System (GIS), the number of pixels for each level of disaster is calculated using area calculation tools. Combined with the spatial resolution of Sentinel-2 data (e.g., 10m×10m resolution equals 0.01 hectares / pixel, 20m×20m resolution equals 0.04 hectares / pixel), this is converted to the actual affected area: 1 mu = 0.0667 hectares. For example, in the flood disaster in the corn-growing area of ​​Arukorqin Banner: 82,000 mu were lightly affected, 51,000 mu were moderately affected, and 23,000 mu were severely affected, for a total affected area of ​​156,000 mu.

[0117] 2. Yield Loss Prediction. Based on the correlation analysis of crop yield data and disaster levels in the study area over the past five years obtained from the local agricultural and rural affairs bureau, a linear regression model was used to establish the relationship between yield loss rate and disaster level.

[0118] Corn: Mild flooding reduces yield by 5%-8%, with a regression coefficient R²=0.82; moderate flooding reduces yield by 8%-15%, with R²=0.85; severe flooding reduces yield by 15%-30%, with R²=0.88. Mild drought reduces yield by 5%-10%, with R²=0.80; moderate drought reduces yield by 10%-20%, with R²=0.83; severe drought reduces yield by 20%-40%, with R²=0.86.

[0119] Soybeans: Mild flooding reduces yield by 8%-12%, R²=0.812; moderate flooding by 12%-20%, R²=0.84; severe flooding by 20%-35%, R²=0.87. Mild drought reduces yield by 10%-15%, R²=0.79; moderate drought by 15%-25%, R²=0.82; severe drought by 25%-45%, R²=0.85.

[0120] Based on the affected area and corresponding yield reduction rate at each level, the overall yield loss in the study area is calculated as follows: Overall yield loss = Σ (Affected area at a certain level × Average yield reduction rate at that level × Normal year yield per unit area). For example, in the Jungar Banner soybean planting area, the normal year yield per unit area is 150 kg / mu. The total drought-affected area is 125,000 mu, of which 30,000 mu are lightly affected with a yield reduction of 12.5%, 60,000 mu are moderately affected with a yield reduction of 20%, and 35,000 mu are severely affected with a yield reduction of 35%. Therefore, the overall yield loss = 3 × 150 × 12.5% ​​+ 6 × 150 × 20% + 3.5 × 150 × 35% = 56.25 + 180 + 183.75 = 4,200,000 kg = 4,200 tons.

[0121] S5: Result Verification and Comprehensive Analysis. Through ground verification and multi-dimensional analysis, the reliability of the monitoring and forecasting results is ensured, the patterns of disaster occurrence are revealed, and a basis for prevention and control is provided.

[0122] S51: Result verification.

[0123] 1. Accuracy Verification. Sixty ground verification points were selected, 20 from each banner / county, covering four types of disasters: unaffected, lightly affected, moderately affected, and severely affected, with 15 points for each type. The disaster levels reported by remote sensing were compared with the ground-based measurement results to construct a confusion matrix. Precision, recall, and F1 score were calculated: Precision = (Number of correctly reported samples) / (Total number of samples), reflecting overall reporting accuracy; Recall = (Number of correctly reported affected samples) / (Actual affected samples), reflecting the false negative rate in the affected area; F1 score = 2 × (Precision × Recall) / (Precision + Recall), comprehensively measuring precision and recall.

[0124] Validation criteria: Flood monitoring accuracy ≥ 85%, recall ≥ 82%, F1 score ≥ 83%; drought monitoring accuracy ≥ 80%, recall ≥ 78%, F1 score ≥ 79%. If the accuracy does not meet the criteria, return to S3 to adjust the index thresholds, such as adjusting the drought index shortwave infrared vertical water loss index threshold from -0.3 to -0.25, or return to S4 to optimize the random forest model parameters, such as increasing the number of decision trees to 120, and then re-perform the monitoring and reporting.

[0125] 2. Data Comparison and Verification. Compare the disaster-affected area reported by remote sensing with the disaster statistics from the local agricultural and rural affairs bureau, ensuring the error is within 10%. For example, in Siziwang Banner, the drought-affected soybean area was reported as 58,000 mu by remote sensing, while the agricultural and rural affairs bureau's statistics showed 62,000 mu. The error is |58,000 - 62,000| / 62,000 × 100% = 6.5%, which meets the accuracy requirements. If the error exceeds 10%, the representativeness of the ground-based measured samples or the cloud cover of the remote sensing data needs to be checked, and the extraction process needs to be optimized again.

[0126] S52: Comprehensive Analysis.

[0127] 1. Spatial Distribution Analysis. A spatial distribution map of disaster levels was drawn using a Geographic Information System (GIS). A color-coded system was employed, with light yellow, moderate orange, and severe red used for different levels. This data was then overlaid with 30m resolution DEM (Digital Elevation Model) data to analyze the impact of topography on disaster distribution. For example, in Arukorqin Banner, flooding is mainly concentrated in the low-lying areas of the central and southern parts, at an altitude of <500m and a slope of <5°. Due to the flat terrain and drainage slope of <2‰, drainage is poor after rainfall. In Jungar Banner, drought is mainly concentrated in the northern planting area, at an altitude of 1000-1200m, with an annual rainfall of 200-250mm. Due to the high altitude and high evaporation rate (2000-2200mm annually), soil moisture is easily lost.

[0128] 2. Time Variation Analysis. Disaster data from July 2021 to July 2023 were compared to analyze disaster trends. For example, rainfall in the study area in July 2023 was 40% higher than in the same period of 2022. The average rainfall in July 2022 was 45 mm, while the average rainfall in July 2023 was 63 mm. This resulted in an increase in the flood-affected area of ​​the corn-growing region in Arukorqin Banner from 98,000 mu in 2022 to 156,000 mu in 2023, an increase of 60%. In June 2022, the study area experienced sustained high temperatures, with an average daily temperature of 28℃, and the drought-affected area increased by 35% compared to the same period in 2021.

[0129] 3. Disaster-causing factor analysis. Principal component analysis was used to identify key driving factors for disasters, based on meteorological data. For example, in June 2024, Jungar Banner experienced 18 consecutive days without effective rainfall, with an average daily temperature of 28℃. The soil in this area is sandy loam with poor water retention capacity, and the field water holding capacity is only 18%-22%, resulting in severe drought damage affecting 35% of the soybean planting area. In Arukorqin Banner, from July 5th to 7th, 2023, the rainfall reached 82mm over three consecutive days, and the drainage facility density in this area was <0.1km / km², leading to concentrated flooding.

[0130] S6: Data storage and monitoring report generation.

[0131] S61: Data Storage. A remote sensing and forecasting database for crop meteorological disasters in the study area was constructed using ArcGIS FileGeodatabase format to ensure data compatibility and security. The database contains the following data layers:

[0132] 1. Remote Sensing Data Layer: Preprocessed Sentinel-2 images are stored in folders by month, named "YYYYMM_Sentinel2.tif". Resampled MODIS land surface temperature images are named "YYYYMM_MOD11A2.tif".

[0133] 2. Crop Data Layer: Vector data for maize planting areas, with attributes including planting area and plot number. Vector data for soybean planting areas, with attributes similar to maize.

[0134] 3. Disaster Data Layer: Flood disaster distribution vector data, with attributes including disaster level, affected area, and occurrence time; drought disaster distribution vector data, with attributes the same as flood.

[0135] 4. Indicator Data Layer: Flood-sensitive index raster data, including Normalized Difference Water Index (NDDI), Improved Normalized Difference Water Index (MDDI), etc. Drought-sensitive index raster data, including Shortwave Infrared Vertical Water Loss Index (SDI), Normalized Differential Moisture Index (MDM), etc.

[0136] 5. Validation data layer: Ground validation point vector data, with attributes including coordinates, measured disaster level, crop growth parameters, sampling time, etc.

[0137] Database access permissions are set, with multiple levels of permissions. Data services are published through ArcGIS Server, supporting remote access and querying. Data is also backed up quarterly to prevent data loss.

[0138] S62: Monitoring report generation. The report uses a combination of text and graphics to visually present the disaster situation and provide a scientific basis for agricultural management departments to formulate disaster prevention and mitigation measures and for farmers to carry out field management. The "Meteorological Disaster Remote Sensing Monitoring Report" is compiled, including: (1) Overview of the study area: geographical location, crop planting structure, and climate background; (2) Data and methods: multi-source data list, preprocessing process, index calculation formula, and model parameters; (3) Monitoring results: disaster area and spatial distribution charts by crop and level, with flood monitoring map, drought monitoring map, and yield loss estimate; (4) Verification accuracy: confusion matrix results and comparison with ground statistical data; (5) Prevention and control recommendations: for flood areas, it is recommended to build drainage ditches and select flood-resistant varieties; for drought areas, it is recommended to promote drip irrigation technology and adjust the sowing period to avoid drought periods.

[0139] This embodiment improves forecasting accuracy and regional adaptability by integrating multi-source data and combining crop physiological responses with disaster mechanisms. It can quickly help agricultural management departments formulate disaster prevention and mitigation measures, which has a good effect on ensuring food security and promoting sustainable agricultural development.

[0140] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanistic models, characterized in that, Includes the following steps: S1: Collect multi-source data and perform standardization processing on the multi-source data; the multi-source data includes remote sensing data, meteorological data, ground-measured data, and basic geographic data; the standardization processing includes remote sensing data preprocessing and meteorological data preprocessing; S2: Based on crop growth mechanisms and phenological characteristics of remote sensing data, construct vegetation index growth curves, and combine extraction rules to screen and correct crop planting areas; S3: Calculation of key meteorological disaster indicators. Based on the differences in the mechanisms of floods and droughts, the flood disaster sensitivity index and drought disaster sensitivity index are calculated separately, and surface temperature data is introduced to help determine the degree of disaster. S4: Based on the flood and drought disaster sensitivity indices, a classification model is constructed to classify crop planting areas to extract the disaster range. The disaster level is divided by combining the disaster duration, disaster sensitivity index threshold and crop damage symptoms. A regression model is used to predict yield loss. S5: Select ground verification points to compare the forecast results with the actual measurement results to verify the accuracy. At the same time, compare the forecast data with the statistical data of the agricultural department to carry out disaster spatial distribution analysis, temporal change analysis and disaster-causing factor analysis. S6: Construct a disaster remote sensing and forecasting database to store various types of data and compile meteorological disaster remote sensing monitoring reports; Step S2 includes: S21: Calculate vegetation index using near-infrared and red light reflectance; construct vegetation index curves based on vegetation indices of different crops at different growth stages; S22: Establish planting area extraction rules and extract planting areas; the extraction rules include: (1) screening areas where the peak value of vegetation index is greater than the threshold and the growth cycle exceeds the threshold as the initially identified crop planting areas; (2) distinguishing the preliminary range of different crops based on the phenological differences of different crops; (3) superimposing the cultivated land distribution vector map to eliminate non-cultivated land areas, and correcting the crop type by combining the actual measured planting points on the ground. The vegetation index curves include the maize vegetation index curve and the soybean vegetation index curve; the peak value of the flowering period index in the maize vegetation index curve is 0.8-0.9, and the maturity period index is ≤0.4; The peak value of the soybean vegetation index during the flowering and pod-setting stage is 0.7-0.8, and the index during the maturity stage is ≤0.

3. The growth cycle threshold is ≥120 days, where the growth cycle is the duration from the first rise of the vegetation index to the first fall below 0.

4.

2. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 1, characterized in that, Step S1 includes: S11: Multi-source data acquisition; the remote sensing data includes multispectral data and land surface temperature data; the multispectral data includes green band, red band, near-infrared band, and shortwave infrared band, with a temporal resolution of 5 days; the land surface temperature data has a spatial resolution of 1 km and a temporal resolution of 8 days; The meteorological data includes daily rainfall, daily average temperature, and daily maximum temperature data; The ground-based measured data includes crop growth, disaster occurrence, disaster location coordinates, and crop leaf sample data; the basic geographic data includes administrative division vector maps and cultivated land distribution vector maps. S12: Multi-source data preprocessing; the remote sensing data preprocessing includes radiometric correction, atmospheric correction, geometric correction, cloud removal, and data synthesis; the radiometric correction converts digital quantization values ​​into surface radiance values, the atmospheric correction eliminates the effects of aerosol scattering and water vapor absorption, the geometric correction uses topographic maps as a reference to control errors within a threshold, and the cloud removal data synthesis generates continuous cloud-free images; the meteorological data preprocessing includes removing outliers, interpolating station data into raster data, and converting the projection coordinate system.

3. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 1, characterized in that, Step S3 includes: S31: Calculation of flood disaster indicators, wherein the flood disaster sensitivity index includes the normalized water body index, the improved normalized water body index, the automatic water body extraction index, and the WI2015 index; S32: Calculation of drought disaster indicators, wherein the drought disaster sensitivity index includes shortwave infrared vertical water loss index, normalized differential water index, global vegetation water index, shortwave infrared water stress index, and shortwave angle slope index. S33: Surface temperature auxiliary calculation, resample the surface temperature data to the required resolution, calculate the daily average surface temperature of the crop planting area and eliminate fluctuations by using the window moving average method; if at least two drought sensitivity indices reach the drought threshold and the daily average temperature is more than 3°C higher than normal, drought is preliminarily confirmed; if at least two flood sensitivity indices reach the flood threshold and the daily average temperature is more than 1°C lower than normal, flood is preliminarily confirmed.

4. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 3, characterized in that, Step S4 includes: S41: Disaster range extraction; The classification model adopts the random forest model, with the actual measured disaster points on the ground as labels, and the labels are divided into "affected" and "not affected"; The characteristic variables of flood disaster are flood disaster sensitivity index and surface temperature, and the characteristic variables of drought disaster are drought disaster sensitivity index and surface temperature; The dataset is divided into training and testing sets to validate the optimized parameters. The Gini coefficient is used as the splitting criterion. A binary map of "unaffected - affected" is output. S42: Disaster severity assessment; flood disaster classification based on normalized water index, improved normalized water index, duration of waterlogging, degree of crop leaf curling, and chlorophyll content; drought disaster classification based on shortwave infrared vertical water loss index, global vegetation moisture index, daily average surface temperature, degree of crop leaf curling, and chlorophyll content. S43: Disaster impact assessment; Affected area statistics: The number of pixels for each disaster level is counted using the area calculation tool of the geographic information system, and the actual area is calculated by combining the spatial resolution of remote sensing data; Production loss prediction: Based on the correlation between production data and disaster level, a regression model is used to establish the relationship between production loss rate and disaster level; Level production loss equals the affected area of ​​that level × the average reduction rate of that level × the yield per unit area in a normal year; Total production loss equals the sum of production losses of each level.

5. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 1, characterized in that, Step S5 includes: S51: Accuracy verification and data comparison verification; The accuracy verification selects ground verification points, covering various types of disasters including unaffected, lightly affected, moderately affected, and severely affected. A confusion matrix is ​​constructed by comparing the forecast results with the measured results, and the accuracy, recall, and F1 score are calculated. If the accuracy does not meet the standard, the exponential threshold is adjusted or the model parameters are optimized; The data comparison verification compares the disaster-affected area reported by remote sensing with the statistical data of the Agriculture and Rural Affairs Bureau, and the error is controlled within the threshold. S52: The spatial distribution analysis uses a geographic information system to draw a spatial distribution map of disaster levels and overlays digital elevation model data to analyze the impact of terrain on disaster distribution; the temporal variation analysis compares disaster data from the same period in consecutive years to analyze the trend of disaster occurrence; the disaster-causing factor analysis combines meteorological data and uses principal component analysis to identify key driving factors of disaster occurrence.

6. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 1, characterized in that, S6 includes: S61: Construct a database to store data; the database includes a remote sensing data layer, a crop data layer, a disaster data layer, an indicator data layer, and a verification data layer; the remote sensing data layer stores preprocessed remote sensing images and resampled surface temperature images; the crop data layer stores vector data of crop planting areas and attributes such as planting area and plot number; the disaster data layer stores vector data of flood and drought disaster distribution and attributes such as disaster level, affected area, and occurrence time; the indicator data layer stores raster data of flood and drought sensitivity indices; the verification data layer stores vector data of ground verification points and attributes such as coordinates, measured disaster level, crop growth parameters, and sampling time. S62: Generate a meteorological disaster remote sensing monitoring report; the meteorological disaster remote sensing monitoring report includes an overview of the study area, data and methods, monitoring results, verification accuracy, and prevention and control recommendations.

7. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 4, characterized in that, The flood disaster classification includes: (1) mild flood: chlorophyll content decreases by 5%-10%, and root activity decreases by 10%-15%; (2) moderate flood: chlorophyll content decreases by 10%-20%, and root activity decreases by 15%-25%; (3) severe flood: chlorophyll content decreases by >20%, and root activity decreases by >25%; The drought disaster classification includes: (1) mild drought: leaf curling degree is level 1, and photosynthetic efficiency decreases by 10%-15%; (2) moderate drought: leaf curling degree is level 2, and photosynthetic efficiency decreases by 15%-25%; (3) severe drought: leaf curling degree is level 3, photosynthetic efficiency decreases by >25%, and plant growth stagnates.

8. The remote sensing and forecasting method for crop meteorological disasters based on multi-source data and mechanism models according to claim 3, characterized in that, The formulas for calculating the flood disaster sensitivity index are as follows: Normalized Difference Water Index = (Green band reflectance - Near-infrared band reflectance) / (Green band reflectance + Near-infrared band reflectance); Improved Normalized Difference Water Index = (Green band reflectance - Shortwave infrared band reflectance) / (Green band reflectance + Shortwave infrared band reflectance); Automatic water extraction index = 4 × (green band reflectance - shortwave infrared band reflectance) - (0.25 × near-infrared band reflectance + 2.75 × shortwave infrared band reflectance); WI2015 index = Green band reflectance - 0.1 × Near-infrared band reflectance - 0.9 × Short-wave infrared band reflectance; The formulas for calculating the drought sensitivity index are as follows: Shortwave infrared vertical water loss index = (shortwave infrared band reflectance - near-infrared band reflectance) / (shortwave infrared band reflectance + near-infrared band reflectance); Normalized differential moisture index = (near-infrared reflectance - mid-infrared reflectance) / (near-infrared reflectance + mid-infrared reflectance), where the mid-infrared band is Sentinel-2 band 11; Global vegetation moisture index = [(near-infrared reflectance + 0.1) - (short-wave infrared reflectance + 0.02)] / [(near-infrared reflectance + 0.1) + (short-wave infrared reflectance + 0.02)]; Shortwave infrared moisture stress index = shortwave infrared reflectance / near-infrared reflectance; Shortwave angle slope index = (shortwave infrared band reflectance - red light band reflectance) / (shortwave infrared band reflectance + red light band reflectance).

Images