Method for identifying extreme agricultural drought events based on dynamic crop sensitive vegetation health index

Abstract

The application discloses a kind of extreme agricultural drought event identification method based on dynamic crop sensitive type vegetation health index, comprising S1, collection and arrangement meteorological, agricultural drought basic data;S2, calculate crop sensitive type weighted VHI data;S3, independent calculation dynamic double threshold value is pixel by pixel;S4, construct meteorological-agricultural coupling trigger mechanism;S5, wide range data parallel computing, identify extreme drought event;S6, post-processing and space-time analysis.The method improves the drought response sensitivity of key growth period, and avoids the problem that fixed threshold is poor in adaptability in different regions, different years, improves the regional pertinence of drought identification.Through establishing meteorological-agricultural coupling trigger mechanism, real agricultural drought and non-drought factor interference are effectively distinguished, false positive rate is reduced, and the accuracy of event identification is improved.The application improves the timeliness, adaptability and identification accuracy of drought monitoring, and is suitable for large-scale agricultural drought monitoring and evaluation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of remote sensing drought monitoring and identification, and particularly relates to an extreme agricultural drought event identification method based on a dynamic crop-sensitive vegetation health index. BACKGROUND

[0002] The agricultural drought identification method is an important basis for guiding agricultural disaster prevention and mitigation and fine management. Generally, remote sensing vegetation index, meteorological factor or soil moisture data are used as the basis, combined with the change trend in a certain period, to realize the monitoring and early warning of regional drought degree. Compared with the static evaluation method which completely relies on long-term historical trend modeling, the current drought identification method based on remote sensing and meteorological information can be updated in a shorter period, has certain real-time and practicality, and can be realized under the current remote sensing and meteorological data acquisition capability.

[0003] For the monitoring and identification of agricultural drought, the common method is to use meteorological drought index (such as SPI, PDSI), soil moisture and the like for single factor or weighted multi-factor drought evaluation based on traditional single index model or index fusion method. For example, Chinese patent application, application number: 202110257127.1, name: an agricultural drought monitoring method based on soil moisture and meteorological time lag; and paper, Feng Kai, et al. Spatial and temporal response relationship between agricultural drought and meteorological drought based on three-dimensional perspective [J]. Transactions of the Chinese Society of Agricultural Engineering, 2020, 36 (08): 103-113. However, the above method usually fails to fully consider the time sequence influence of drought on crop growth process, especially the response difference of water stress in the critical growth period of crops, and is easily disturbed by non-drought factors.

[0004] Therefore, it is necessary to introduce crop phenology information, to improve the drought identification model from static vegetation state analysis to dynamic growth period driven mechanism, to establish the mapping relationship between multi-source information such as remote sensing and meteorology and drought response in different growth stages of crops, and to use a multi-source remote sensing identification framework integrating phenology driving to improve the accuracy of drought monitoring and agricultural adaptability. SUMMARY

[0005] The application aims to provide an extreme agricultural drought event identification method based on a dynamic crop-sensitive vegetation health index, to solve the problems in the prior art, and to realize accurate identification and comprehensive evaluation of extreme agricultural drought events through crop coefficient weighting, double growth period determination and meteorological-agricultural coupling trigger mechanism.

[0006] To solve the above technical problems, the application adopts the following technical solutions:

[0007] The extreme agricultural drought event identification method based on the dynamic crop-sensitive vegetation health index comprises the following steps:

[0008] S1, collecting and sorting meteorological and agricultural drought basic data: the basic data includes per-pixel weekly vegetation health index VHI data and standardized precipitation index SPI3 data in the study area, and the spatial and temporal resolution of the standardized precipitation index SPI3 data is consistent with that of the weekly vegetation health index VHI data, ensuring the coupling of meteorological drought information and vegetation drought information.

[0009] S2, standardizing the vegetation health index VHI data to eliminate dimensional differences in different regions and years; dividing the whole year into two states of "growth period" and "non-growth period" per week, denoted as Zscore-VHI; using crop coefficient Kc to weight the vegetation health index VHI data per week to obtain crop coefficient weighted VHI, denoted as Kc-VHI; obtaining crop sensitive type weighted VHI data according to Zscore-VHI and Kc-VHI.

[0010] S3, independently calculating dynamic double thresholds per pixel: when the pixel is in the growth period, the extreme agricultural drought trigger threshold adopts the 5th percentile of the historical Kc-VHI of the pixel, and the recovery threshold adopts the 50th percentile of the historical Kc-VHI; if the pixel is in the non-growth period, the extreme agricultural drought trigger threshold adopts the 5th percentile of the historical Zscore-VHI of the pixel, and the recovery threshold adopts the 50th percentile of the historical Zscore-VHI.

[0011] S4, constructing a meteorological-agricultural coupling trigger mechanism: dynamically selecting corresponding thresholds according to the crop growth cycle, and combining vegetation health index and lagging meteorological drought signals to identify and track extreme drought events, and performing data quality inspection on the preliminary identified events.

[0012] S5, large-scale data parallel computing: dividing the study area into spatial blocks, using multi-thread parallel processing technology to execute extreme drought event identification per pixel, obtaining an initial list of extreme drought events containing event start and end time, drought duration, recovery duration, minimum Kc-VHI value point time, average VHI value during the event, and pixel center longitude and latitude, and outputting the structure to a standardized database.

[0013] S6, post-processing and spatio-temporal analysis: performing event post-processing and filtering on the initial list of extreme drought events, removing completely non-growth period events, outputting the final extreme drought event feature table in CSV format, and performing time series and spatial distribution analysis on the identification results.

[0014] Further optimization, the specific steps of standardizing the vegetation health index in step S2 to divide the growth period and non-growth period state are:

[0015] S2.1, convert the VHI data obtained in step S1 from TIFF format to NetCDF format, build a multi-dimensional spatio-temporal data structure, reorganize the discrete weekly TIFF files into a standardized spatio-temporal data set, set the time dimension as year y and week w w , w ∈[1,52]; the spatial resolution is r , unit: km; then the converted NetCDF data is stored as a matrix:

[0016] (1);

[0017] wherein, m, n are the number of grids in the longitude and latitude directions, respectively; represents the VHI original value at grid ( y w ) in the i, j year and the y week; y ∈[1,Y], Y is the total number of years.

[0018] S2.2, standardize the NetCDF format VHI data according to the week sequence to obtain Zscore-VHI:

[0019] (2);

[0020] wherein, .

[0021] S2.3, based on the crop type and growth stage, linear interpolation of the crop coefficient curve and the regional crop calendar is performed to calculate the dynamic Kc value in the growth period:

[0022] (3);

[0023] wherein, is the sowing week in the yth year, obtained according to the regional crop calendar; is the germination week in the yth year, is the initial growth stage in the yth year; is the mid-week in the yth year, is the rapid development stage in the yth year; is the mature week in the yth year, is the mid-growth stage in the yth year; is the harvest week in the yth year, is the mature stage in the yth year; represents the dynamic crop coefficient in the yth year and the wth week, represents the initial crop coefficient, ​​represents the medium crop coefficient, represents the pre-harvest crop coefficient.

[0024] S2.4, the crop sensitivity weighted VHI data is calculated by using the differentiated weighting strategy, wherein the Zscore-VHI is coupled with Kc to obtain Kc-VHI for the growth period, and the Zscore-VHI is maintained for the non-growth period, and the expression is as follows:

[0025] (4);

[0026] The growth period is coupled with the intensified crop sensitive period drought signal, and the sensitivity of drought identification in the critical growth period is improved; the non-growth period maintains the original value of the standardized VHI, and the integrity of the natural drought signal is ensured.

[0027] S2.5, a growth period identification matrix N is constructed to generate a global space identification system, which supports efficient time and space screening:

[0028] (5);

[0029] (6)。

[0030] S2.6, all original values, Kc interpolation results or coupling results are uniformly set to NaN due to the mask clipping of the pixel equal to -9999.0, spatial index and coordinate consistency verification is carried out, and x 、 y , the time dimension is completely consistent with the original TIFF, and the coordinate drift caused by resampling or mask is avoided; finally, the output file can be directly used for parallel threshold calculation.

[0031] Further optimization, the specific steps of calculating the starting and recovery thresholds of extreme agricultural drought in the growth period and the non-growth period in the step S3 are:

[0032] S3.1, when the pixel is in the growth period, and the spatial pixel ( i , j ) is calculated alone, then the growth period drought triggering threshold and drought recovery threshold calculation formula are as follows:

[0033] (7);

[0034] (8);

[0035] wherein, 、 are the growth period drought triggering threshold and recovery threshold respectively; Q p is p quantile function; v i,j,y,w is the positioni , j ) the Kc-VHI value of the yth year in the xth week; y w the Kc-VHI value of the yth year in the xth week; is the yth sowing week; is the yth harvesting week; Y is the total number of years of historical data; is the cross-year data aggregation.

[0036] S3.3, when the pixel is in the non-growth period, the time range covers the whole year non-crop growth week, the statistical characteristics of the original Zscore data are retained, and the non-growth period drought trigger and drought recovery threshold is calculated as follows:

[0037] (9) ;

[0038] (10) ;

[0039] , wherein, , are the non-growth period drought trigger threshold and the recovery threshold, respectively.

[0040] S3.3, a spatial grid data set containing four types of thresholds is constructed, a threshold-phenology linkage rule is established, and according to the week sequence w , the corresponding period threshold is automatically matched, and uninterrupted drought monitoring throughout the year is realized.

[0041] Further optimization, the specific steps of constructing a meteorological-agricultural coupling trigger mechanism to identify extreme agricultural drought events in step S4 are:

[0042] S4.1, according to the Kc-VHI data, Zscore-VHI data, dynamic threshold and SPI3 data, the state transition is identified and the drought event is tracked; when the crop sensitive type weighted VHI value is lower than the corresponding starting threshold, and at least one week SPI3 value in the past 8 weeks is less than -2.0, it means that the drought state is entered:

[0043] (11) ;

[0044] , wherein, is the SPI3 value of the pixel ( i , j ) at the time k t ; L is 8 weeks; when the crop sensitive type weighted VHI value rises to the corresponding recovery threshold and lasts for more than the minimum recovery week, the drought state ends.

[0045] S4.2, in the event identification process, the threshold dynamic selection rule is: when the week sequence belongs to the growth period, the growth period threshold is selected; when the week sequence belongs to the non-growth period, the non-growth period threshold is selected, that is:​

[0046] (12);

[0047] wherein, is a drought initiation threshold, is a drought recovery threshold.

[0048] S4.3, performing data quality check on the event identified in step S4.1, specifically:

[0049] counting the proportion of missing values in the pixel-level data due to mask clipping and invalid observations in the duration of the drought event, if the proportion of missing values is greater than or equal to 50%, it means that the data integrity is insufficient, and the event is excluded; counting the proportion of valid observation points of vegetation health index VHI in the total observation points in the same time series, if the proportion of valid observation points is less than or equal to 80%, it means that the VHI data reliability is insufficient, and the event is excluded. Through these two checks, it can be ensured that the retained drought event data has sufficient integrity and reliability, avoiding misjudgment due to data missing or noise.

[0050] Further optimization, the step S5 specifically includes the following steps:

[0051] S5.1, dividing the study area into spatial blocks according to 64x64 pixels, using ThreadPoolExecutor multi-thread parallel processing each spatial block, executing the drought event identification of S1-S4 pixel by pixel, to obtain a preliminary list of extreme drought events; the preliminary list of extreme drought events includes event start time, end time, duration, recovery weeks, minimum Kc-VHI value point time, average VHI in the event, and pixel center position longitude and latitude.

[0052] S5.2, structuring the preliminary list of extreme drought events to generate a standardized drought event database, specifically:

[0053] ;

[0054] wherein, is the geographic location coordinates of the first event, is the geographic location coordinates of the g event, is the start time of the first event, is the start time of the g event; is the end time of the first event, is the end time of the g event; is the minimum Kc-VHI value point time of the first event, is the minimum Kc-VHI value point time of the gThe lowest Kc-VHI value point time for each event; The number of weeks that the first event lasted. For the first g The number of weeks the event lasted; The number of weeks for recovery after the first event. For the first g Recovery weeks for each event; The average VHI value for the first event. For the first g The average VHI value for each event.

[0055] Further optimization is achieved by defining the specific steps of step S6, post-processing and filtering for extreme agricultural drought events, as follows:

[0056] S6.1 Post-process and filter the standardized drought event database generated in step S5.2 to remove completely non-growing season events that have no impact on agricultural drought.

[0057] S6.2 Output the final event feature table processed in step S6.1 in CSV format. The feature table contains detailed attributes of each identified extreme drought event.

[0058] S6.3. Based on the event characteristic table, analyze the temporal evolution trend of extreme drought events, including statistical analysis of the average intensity of extreme drought events and their changes over time, as well as identifying periodic or trend patterns in the occurrence of drought events.

[0059] S6.4 Combine the geographical location information in the event list with drought characteristics to visualize and analyze the spatial distribution characteristics of extreme drought events, generate a drought event density map to intuitively display drought-prone areas, and verify the accuracy of drought event identification through spatiotemporal analysis methods.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1) Based on the standard crop coefficient curve and combined with the regional crop calendar for weekly interpolation, crop sensitivity weighting of drought identification indicators is realized, making the response to drought during key growth periods more sensitive and solving the problem of insufficient adaptability of traditional methods to differences in growth periods. A pixel-level dynamic dual-threshold strategy based on historical quantile calculation is introduced, which is used to determine the onset and recovery of drought during the "growing season" and "non-growing season" respectively. This avoids the problem of poor adaptability of fixed thresholds in different regions and years, and improves the regional specificity of drought identification.

[0062] 2) Remote sensing vegetation index and meteorological drought index SPI3 are coupled in time sequence, through setting lagging meteorological drought trigger condition, real agricultural drought and non-drought factor interference are effectively distinguished, false positive rate is reduced, and the accuracy of event identification is improved. In terms of drought recovery identification, the recovery trend of Kc-VHI and Zscore-VHI and the minimum recovery week requirement are comprehensively considered, so that the event termination judgment is more logical and timely.

[0063] 3) The "64*64 pixel space block + ThreadPoolExecutor multithreading parallel processing" strategy is adopted, the distributed calculation is carried out on the pixel-by-pixel drought event identification process, the processing efficiency of large-scale data is greatly improved, meanwhile, the standardized database containing event start and end time, spatial coordinates and other attributes is output, efficient support is provided for subsequent spatio-temporal analysis, and the efficiency bottleneck of traditional methods in large-scale application is solved. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 is a flowchart of the method of the present application;

[0065] Figure 2 is a comparison chart of one-pixel 1982 vegetation health index VHI data standardization and Kc weighting processing of the present application;

[0066] Figure 3 is a dynamic threshold calculation logic diagram of the present application;

[0067] Figure 4 is a relationship chart of extreme drought event frequency and corn yield interannual variation of the embodiment of the present application. DETAILED DESCRIPTION

[0068] The technical solutions of the present application will be further specifically described below by combining with the drawings.

[0069] As shown in Figure 1 , the extreme agricultural drought event identification method based on dynamic crop sensitive vegetation health index comprises the following steps:

[0070] S1, collect and arrange meteorological and agricultural drought basic data: including the pixel-by-pixel week-scale vegetation health index VHI data and the standardized precipitation index SPI3 data in the research area; wherein, the time and space resolution of the standardized precipitation index SPI3 data is consistent with that of the VHI data.

[0071] S2, standardize the vegetation health index VHI data, and divide the whole year into "growth period" and "non-growth period" two states per week, denoted as Zscore-VHI, weight the VHI data per week using the crop coefficient Kc to obtain the crop coefficient weighted VHI, denoted as Kc-VHI, Zscore-VHI and Kc-VHI are collectively denoted as crop sensitive weighted VHI data, which specifically includes the following steps:

[0072] S2.1, convert the vegetation health index VHI data obtained in step S1 from TIFF format to NetCDF format, construct a multi-dimensional spatio-temporal data structure, reorganize the discrete weekly TIFF files into a standardized spatio-temporal data set, set the time dimension as year y and week sequence w , w ∈ [1, 52]; the spatial resolution is r , unit: km, then the converted NetCDF data is stored as a matrix, as formula (1).

[0073] S2.2, standardize the NetCDF format VHI data in step S2.1 to obtain Zscore-VHI, calculate the standardized value according to the week sequence, as formula (2).

[0074] S2.3, based on crop type and growth stage, linear interpolation of crop coefficient curve and regional crop calendar, distinguish the growth period and calculate the dynamic Kc value in the growth period, as formula (3).

[0075] S2.4, coupling of growth period Kc and non-growth period processing of input data after standardization in step S2.2 and crop coefficient Kc calculated in step S2.3, using differentiated weighting strategy, optimizing the whole time series drought signal, as formula (4). The growth period coupling strengthens the crop sensitive period drought signal, and improves the sensitivity of drought identification in the critical growth period; the non-growth period maintains the original value of the standardized VHI, ensuring the integrity of the natural drought signal.

[0076] S2.5, using the growth period coupling and non-growth period processing VHI data in step S2.4, constructing a growth period identification matrix N, as formula (5), (6), generating a global spatial identification system, supporting efficient spatio-temporal filtering.

[0077] S2.6, all original values, Kc interpolation results or coupling results are uniformly set to NaN if the pixel value is equal to -9999.0 due to mask clipping, spatial index and coordinate consistency check is performed to ensure that the x, y, time dimensions are completely consistent with the original TIFF, avoiding coordinate drift caused by resampling or mask; the final output file can be directly used for parallel threshold calculation.

[0078] S3. Calculate dynamic dual thresholds independently for each pixel: When a pixel is in its growth phase, the extreme agricultural drought trigger threshold is the 5th percentile of the pixel's historical Kc-VHI, and the recovery threshold is the 50th percentile of the historical Kc-VHI; if a pixel is not in its growth phase, the extreme agricultural drought trigger threshold is the 5th percentile of the pixel's historical Zscore-VHI, and the recovery threshold is the 50th percentile of the historical Zscore-VHI. Specifically, this includes the following steps:

[0079] S3.1 Load the crop-sensitive weighted VHI dataset and read the Kc-VHI and Zscore-VHI spatiotemporal datasets from the NetCDF file output in step S2 as input data.

[0080] S3.2. Based on the growth stage identifier matrix output in step S2.5, separate the data and use only the growth stage data. And each spatial cell ( i , j ) Calculated separately, the growth period trigger threshold and drought recovery threshold are as shown in equations (7) and (8).

[0081] S3.3 When a pixel is in a non-growth phase, The time range covers the entire year of non-crop growing seasons, and retains the statistical characteristics of the original Zscore data. The thresholds for drought triggering and drought recovery during the non-growing season are calculated as shown in equations (9) and (10).

[0082] S3.4 Construct a spatial grid dataset containing four types of thresholds, establish threshold-phenological linkage rules, and base them on the weekly sequence. w Automatically match the threshold for the corresponding period to achieve uninterrupted drought monitoring throughout the year.

[0083] S4. Construct a meteorological-agricultural coupling triggering mechanism, dynamically select thresholds based on crop growth cycles, and combine vegetation health index and lagged meteorological drought signals to identify and track extreme drought events. Perform data quality checks on the initially identified events, specifically including the following steps:

[0084] S4.1. Based on Kc-VHI data, Zscore-VHI data, dynamic thresholds, and SPI3 data, state transition identification and drought event tracking are performed. When the crop-sensitive weighted VHI value is lower than the corresponding start threshold and the SPI3 value is less than -2.0 for at least one week in the past 8 weeks, it indicates that a drought state has been entered, as shown in Equation (11). When the crop-sensitive weighted VHI value rises back to the corresponding recovery threshold and continues to exceed the minimum recovery week, it indicates that the drought state has ended.

[0085] S4.2 During the event recognition process, the dynamic threshold selection rule is as follows: when the cycle belongs to the growth period, the growth period threshold is selected; when the cycle belongs to the non-growth period, the non-growth period threshold is selected.

[0086] S4.3, performing data quality check on the event identified in step S4.1, specifically:

[0087] Statistically counting the proportion of missing values in the pixel-level data caused by mask clipping and invalid observations in the duration of the drought event, if the proportion of missing values is greater than or equal to 50%, it indicates that the data integrity is insufficient, and the event is excluded; statistically counting the proportion of the number of valid observation points of vegetation health index VHI in the total number of observation points in the same time series, if the proportion of valid observation points is less than or equal to 80%, it indicates that the VHI data reliability is insufficient, and the event is excluded.

[0088] S5, dividing the study area into spatial blocks, using multi-thread parallel processing technology to execute extreme drought event identification pixel by pixel, obtaining an initial list of extreme drought events containing event start and end time, drought duration, recovery duration, minimum point time, average VHI value during the event, and pixel center longitude and latitude, etc. and outputting it to a standardized database, including the following steps:

[0089] S5.1, dividing the study area into spatial blocks with 64x64 pixels, using ThreadPoolExecutor multi-thread parallel processing of each spatial block, executing drought event identification of S1-S4 pixel by pixel, obtaining an initial list of extreme drought events; the initial list of extreme drought events contains event start and end time, drought duration, recovery duration, minimum Kc-VHI value point time, average VHI during the event, and pixel center longitude and latitude.

[0090] S5.2, structuring the initial list of extreme drought events to generate a standardized drought event database, specifically:

[0091] ;

[0092] wherein, is the geographic location coordinates of the first event, is the geographic location coordinates of the g th event, is the start time of the first event, is the start time of the g th event; is the end time of the first event, is the end time of the g th event; is the minimum Kc-VHI value point time of the first event, is the minimum Kc-VHI value point time of the g th event; is the duration of the first event in weeks, is the duration of theg the number of weeks of the event; the number of weeks of recovery for the 1st event, the number of weeks of recovery for the 1st event; g the number of weeks of recovery for the 1st event; the average VHI value for the 1st event, the average VHI value for the 1st event; g the average VHI value for the 1st event.

[0093] S6, post-processing and filtering of the preliminary list of extreme drought events, outputting a final extreme drought event feature table in CSV format, and performing time series and spatial distribution analysis on the identified results, specifically including the following steps:

[0094] S6.1, post-processing and filtering of the standardized drought event database generated in step S5.2, removing completely non-growing season events that have no impact on agricultural drought.

[0095] S6.2, outputting the final event feature table after step S6.1 in CSV format, which contains detailed attributes of each identified extreme drought event.

[0096] S6.3, based on the event feature table, analyzing the time evolution trend of extreme drought events, including statistical analysis of the average intensity of extreme drought events and its change over time, and identifying periodic or trend patterns of drought event occurrence.

[0097] S6.4, combining geographical location information with drought characteristics in the event list, visualizing and analyzing the spatial distribution characteristics of extreme drought events, generating a drought event density map to visually display high-risk areas of drought, and verifying the accuracy of drought event identification through spatio-temporal analysis methods.

[0098] The steps of the present application will be described in detail below with specific implementations.

[0099] The embodiment takes the main rain-fed maize production area in Africa as the research object. The maize planting area in Africa is mainly concentrated in the west, center and south of Africa. In these areas, the main maize sowing period is usually the 18th week, the harvest period is the 38th week, the secondary rainy season sowing period is the 42nd week, and the harvest period is the 7th week of the next year. The research period is selected from 1981 to 2024, considering the two seasons of crop planting, a total of 88 crop planting seasons. The remote sensing data uses the vegetation health index VHI data provided by the satellite application and research center of the National Oceanic and Atmospheric Administration (NOAA), with a spatial resolution of 4km and a time interval of 7 days. The SPI3 data calculated by the monthly precipitation data in the CRUTS data set made by the UK National Atmospheric Science Center (NERC Centres for Atmospheric Science (UK), NCAS) is used as the meteorological data, with a spatial resolution of 0.5°. The crop growth information refers to the regional maize crop coefficient Kc curve and maize distribution map published by the Food and Agriculture Organization (FAO).

[0100] As shown in the accompanying Figure 1 In the embodiment, first, the Kc weighted processing of VHI is performed according to the Kc standard curve published by the FAO, as shown in the accompanying Figure 2 As shown in the accompanying Figure 3 As shown in the accompanying

[0101] In order to realize the scientific determination of extreme drought events, the standardized precipitation index SPI3 is introduced as a meteorological drought trigger condition in the embodiment. After the SPI3 data is interpolated to the same spatial and temporal resolution as the VHI data, when the Kc weighted VHI in any pixel is lower than the dynamic trigger threshold, and the SPI3 corresponding to the 1-8 week lag window is less than-2.0, it is marked as the start of the extreme agricultural drought event. If the VHI is higher than the recovery threshold for two consecutive periods, it is marked as the end of the drought.

[0102] In this embodiment, in order to improve the calculation efficiency, the regional space block division and parallel processing strategy are adopted. The research area is divided into 64*64 pixel units, and the large-area remote sensing data is processed through the multi-thread parallel recognition technology, and the drought start time, duration, minimum point time and other event attributes are output.

[0103] During the period of 1981-2024, a total of 237295 extreme agricultural drought events occurred in the main maize production area of Africa. Due to space constraints, only some of the data are shown, as shown in Table 1, which shows the top 10 extreme agricultural drought event information. The extreme agricultural drought events in this region not only occur frequently, but also last for a long time, posing a serious challenge to agricultural production. According to the event statistical information, the average duration of these drought events is 21.0 weeks, the median is 15.0 weeks, the average recovery time is 16.0 weeks, and the median is 12.0 weeks. The crop coefficient weighted VHI index shows that these extreme agricultural drought events have a significant impact on crop health, with a minimum crop coefficient weighted VHI ranging from -6.1 to -0.4, and an average crop coefficient weighted VHI ranging from -3.4 to 0.0. These characteristics indicate that the extreme agricultural drought events in the main maize production area of Africa have long duration and strong impact, posing a serious threat to local agricultural production and food security.

[0104] Table 1 Partial extreme agricultural drought event data in the main maize production area of Africa

[0105]

[0106] As Figure 4 shown, the frequency of extreme agricultural drought events identified by the method fluctuates significantly and has climate response characteristics. Overall, maize yield shows a long-term upward trend, but is affected by drought events and fluctuates, reflecting the effects of comprehensive agricultural development such as technological progress and variety improvement; but in years with high frequency of extreme drought (such as 1985, 1992, 2013, 2014), maize yield shows a stage decline, which is highly corresponding to the drought identification result. The interannual fluctuation of the frequency of extreme drought events is roughly synchronous with the period of maize yield decline, indicating that the method has high accuracy in both spatial and temporal scales.

[0107] Therefore, through the above embodiments, it is further illustrated that the scheme described in the present application improves the key period response and regional adaptability of drought identification; at the same time, coupling the SPI3 meteorological trigger mechanism and the VHI recovery trend, the accuracy and stability of the identification of extreme agricultural drought events are enhanced.

[0108] The above detailed description merely illustrates the technical solutions of the present application and its implementation details, and is not construed as limiting the scope of the present application. Based on the above description, those skilled in the art can make appropriate adjustments, modifications or replacements to the technical features in the above embodiments without departing from the essential principles and core ideas of the present application. All the changes or improvements made on the basis of the spirit of the present application shall be considered to fall within the scope of protection of the present application.

Claims

1. A method for extreme agricultural drought event identification based on dynamic crop sensitive vegetation health index, characterized in that, The method comprises the following steps: S1, collecting and arranging meteorological and agricultural drought basic data: the basic data includes pixel-by-pixel weekly vegetation health index VHI data and standardized precipitation index SPI3 data in the research area, and the time and space resolutions of the standardized precipitation index SPI3 data and the weekly vegetation health index VHI data are consistent; S2, calculating crop-sensitive weighted VHI data: the weekly vegetation health index VHI data is standardized, the whole year is divided into two states of "growth period" and "non-growth period" by weeks, and is recorded as Zscore-VHI; the vegetation health index VHI data is weekly weighted by using the crop coefficient Kc to obtain the crop coefficient weighted VHI, which is recorded as Kc-VHI; the crop-sensitive weighted VHI data is obtained according to Zscore-VHI and Kc-VHI; S3, independently calculating dynamic double thresholds pixel by pixel: when the pixel is in the growth period, the extreme agricultural drought trigger threshold adopts the 5th percentile of the historical Kc-VHI of the pixel, and the recovery threshold adopts the 50th percentile of the historical Kc-VHI; if the pixel is in the non-growth period, the extreme agricultural drought trigger threshold adopts the 5th percentile of the historical Zscore-VHI of the pixel, and the recovery threshold adopts the 50th percentile of the historical Zscore-VHI; S4, constructing a meteorological-agricultural coupling trigger mechanism: dynamically selecting the corresponding threshold according to the crop growth cycle, and combining the vegetation health index and the lagging meteorological drought signal to identify and track the extreme drought event, and performing data quality inspection on the preliminarily identified event; S5, large-scale data parallel calculation: the research area is divided according to spatial blocks, and the extreme drought event identification is performed pixel by pixel by using multi-thread parallel processing technology to obtain an extreme drought event preliminary list containing the start time, end time, duration, recovery week number, minimum Kc-VHI value point time, average VHI value during the event, and pixel center longitude and latitude, and the extreme drought event preliminary list is structured and output as a standardized database; S6, post-processing and spatial-temporal analysis: performing event post-processing and filtering on the extreme drought event preliminary list, and removing the completely non-growth period events, outputting the final extreme drought event feature table in CSV format, and performing time series and spatial distribution analysis on the identification result.

2. The method for identifying extreme agricultural drought events based on dynamic crop sensitive vegetation health index according to claim 1, characterized in that, Step S2 specifically comprises the following steps: S2.1 Convert the weekly vegetation health index (VHI) data obtained in step S1 from TIFF format to NetCDF format, construct a multidimensional spatiotemporal data structure, and reorganize the discrete weekly TIFF files into a standardized spatiotemporal dataset, assuming the time dimension is year y and week order. w , w ∈[1,52]; spatial resolution is r Unit: km; then the converted NetCDF data is stored as a matrix: (1)； wherein, m, n are the number of grids in the longitude and latitude directions, respectively; denotes the VHI raw value at grid ( y ) in the w week of the year; i,j y is the year number, y ∈ [1, Y], Y is the total number of years.​ S2.2, standardizing the VHI data in NetCDF format according to the week sequence to obtain Zscore-VHI: (2)； wherein ; ; S2.3, based on the crop type and the growth stage, linear interpolation is performed on the crop coefficient curve and the regional crop calendar to calculate the dynamic Kc value in the growth period: (3)； wherein, is the week of planting in year y, obtained from the regional crop calendar; is the week of germination in year y, is the initial growth stage in year y; is the mid-week in year y, is the rapid development stage in year y; is the week of maturity in year y, is the mid-growth stage in year y; is the week of harvest in year y, is the maturity stage in year y; denotes the dynamic crop coefficient for week w in year y, denotes the initial crop coefficient, denotes the mid-crop coefficient, denotes the pre-harvest crop coefficient; S2.4, the crop-sensitive weighted VHI data is calculated by using a differential weighting strategy, wherein Zscore-VHI and Kc are coupled to obtain Kc-VHI in the growth period, and Zscore-VHI remains the original value in the non-growth period, and the expression is as follows: (4)； S2.5, constructing a growth period identification matrix N to generate a global spatial identification system: (5)； (6)； S2.6, Unify all the original values, Kc interpolation results or coupling results to NaN which are equal to-9999.0 due to mask clipping, and check the consistency of spatial index and coordinate.

3. The method for extreme agricultural drought event identification based on dynamic crop sensitive vegetation health index according to claim 2, characterized in that, The step S3 specifically comprises the following steps: S3.1 When a pixel is in the growth phase, and the spatial pixel ( i , j If calculated separately, the formulas for calculating the drought trigger threshold and drought recovery threshold during the growing season are as follows: (7)； (8)； in, , These are the drought trigger threshold and recovery threshold during the growing season, respectively; Q p for p Quantile function; v i,j,y,w For position ( i , j ) No. y Year w Weekly Kc-VHI value; This is the sowing week of year y; Let Y be the harvest week of year y; Y is the total number of years of historical data. For cross-year data aggregation; S3.3, When the pixel is in the non-growing period, the time range covers the whole year non-crop growing week, the statistical characteristics of the original Zscore data are retained, and then the non-growing period drought trigger and drought recovery threshold are calculated as follows: (9)； (10)； wherein , are the non-growth period drought trigger threshold and recovery threshold, respectively; S3.3, Constructing the spatial grid dataset containing four types of thresholds, establishing the threshold-phenology linkage rules, and determining the weekly sequence w Automatic matching of corresponding period thresholds, realizing uninterrupted drought monitoring throughout the year.

4. The method for extreme agricultural drought event identification based on dynamic crop sensitive vegetation health index according to claim 3, characterized in that, The step S4 specifically comprises the following steps: S4.1, According to the Kc-VHI data, Zscore-VHI data, dynamic threshold and SPI3 data, the state transition is identified and the drought event is tracked; when the crop sensitive weighted VHI value is lower than the corresponding starting threshold and at least one week SPI3 value in the past 8 weeks is less than-2.0, it indicates that the drought state is entered: (11)； wherein, is a pixel ( i , j ) at time k t SPI3 value; L is 8 weeks; When the crop sensitive weighted VHI value rises to the corresponding recovery threshold and lasts for more than the minimum recovery weeks, it indicates that the drought state ends; S4.2, In the event identification process, the threshold dynamic selection rule is: when the week sequence belongs to the growing period, the growing period threshold is selected; when the week sequence belongs to the non-growing period, the non-growing period threshold is selected; S4.3, The data quality of the event identified in step S4.1 is checked, specifically: The proportion of missing values in the pixel-level data due to mask clipping and invalid observation in the drought event duration is counted, and if the proportion of missing values is greater than or equal to 50%, it indicates that the data integrity is insufficient, and the event is removed; The proportion of effective observation points of the vegetation health index VHI in the total observation points in the same time sequence is counted, and if the proportion of effective observation points is less than or equal to 80%, it indicates that the VHI data reliability is insufficient, and the event is removed.

5. The method for identifying extreme agricultural drought events based on dynamic crop- sensitive vegetation health index according to claim 4, characterized in that, The step S5 specifically comprises the following steps: S5.1, The research area is divided into spatial blocks according to 64*64 pixels, ThreadPoolExecutor multi-thread parallel processing is used for each spatial block, and the drought event identification of S1-S4 is executed pixel by pixel to obtain a preliminary list of extreme drought events; the preliminary list of extreme drought events includes event start time, end time, duration, recovery weeks, lowest Kc-VHI value point time, average VHI in the event and pixel center position longitude and latitude; S5.2, The preliminary list of extreme drought events is structured to generate a standardized drought event database, specifically: ； in, The geographical coordinates of the first event. For the first g The geographical coordinates of the event This is the start time of the first event. For the first g The start time of each event; The end time of the first event. For the first g The end time of each event; The time point at which the lowest Kc-VHI value of the first event is reached. For the first g The lowest Kc-VHI value point time for each event; The number of weeks that the first event lasted. For the first g The number of weeks the event lasted; The number of weeks for recovery after the first event. For the first g Recovery weeks for each event; The average VHI value for the first event. For the first g The average VHI value for each event.

6. The method for identifying extreme agricultural drought events based on dynamic crop- sensitive vegetation health index according to claim 5, characterized in that, The step S6 specifically comprises the following steps: S6.1, The standardized drought event database generated in step S5.2 is post-processed and filtered to remove completely non-growing period events which have no impact on agricultural drought; S6.2, The final event feature table after step S6.1 is output in CSV format, and the feature table includes detailed properties of each identified extreme drought event; S6.3, Based on the event feature table, the time evolution trend of the extreme drought event is analyzed, including the average intensity of the extreme drought event and its change over time, and the periodic or trend pattern of the drought event is identified; S6.4 Combine the geographical location information in the event list with drought characteristics to visualize and analyze the spatial distribution characteristics of extreme drought events, generate a drought event density map to intuitively display drought-prone areas, and verify the accuracy of drought event identification through spatiotemporal analysis methods.

Images