Winter wheat drought remote sensing monitoring method

Abstract

The invention provides a winter wheat drought remote sensing monitoring method fused with downscaling sunlight-induced chlorophyll fluorescence (SIF) data. A winter wheat drought remote sensing monitoring method fused with downscaling SIF data comprises the steps that S1, input data of an SIF data reconstruction machine learning model are selected, and the input data comprise MODIS reflectivity, fPAR, LAI and ERA5 reanalysis data of temperature, saturation vapor pressure difference, photosynthetically active radiation and soil humidity data; s2, training and constructing a machine learning model; s3, construction and effect inspection of a long-time-sequence high-resolution downscaling SIF product data set are carried out; and S4, constructing a novel drought fluorescence monitoring index. The winter wheat drought remote sensing monitoring method provided by the invention has the advantages that the problem of how to fuse SIF information can be solved, the timeliness and accuracy of agricultural drought monitoring can be improved in mechanism, and the agricultural meteorological disaster prevention and reduction capability can be effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing monitoring, and in particular to a winter wheat drought remote sensing monitoring method that integrates downscaled sunlight-induced chlorophyll fluorescence data. Background Art

[0002] As one of the longest-lasting and most destructive extreme climate events, drought has long attracted the attention of scholars. The occurrence of large-scale drought events often weakens the carbon deposition of terrestrial ecosystems and exacerbates the contradiction between food demand and biofuel production. The occurrence of extreme drought events, even in areas with well-developed irrigation facilities, will have a great impact on food production and social economy. In recent years, the global yield loss of major crops (wheat, corn and rice) caused by drought has reached 1.82×10 5 10,000 tons, and the probability of drought will continue to increase under future climate scenarios, which will pose a major risk to crop growth. As the world's second most widely planted cereal crop, wheat is very susceptible to drought stress during the flowering and filling stages due to its poor drought resistance. As an important grain, cotton and oil production base in my country, the Huanghuaihai region accounts for more than 60% of the country's wheat production. However, with the increasing demand for food in the region and the further intensification of water resource conflicts, the intensity, frequency and scope of drought have increased, and the possibility of winter wheat suffering from agricultural drought risks during its growing season has also increased. Therefore, exploring methods for timely and accurate monitoring of agricultural drought during the growth period of winter wheat in the region and accurately estimating winter wheat yields under drought conditions are of great significance to the government's formulation of drought prevention and drought resistance measures and food production stabilization decisions, optimization of agricultural water resources management, and food security.

[0003] Traditional drought monitoring mainly uses temperature, precipitation, evaporation, runoff and other data observed by meteorological or hydrological stations and soil moisture data observed by agricultural meteorological stations to construct drought indices. Among them, the most widely used indices include the Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI), and Palmer Drought Index (PDSI). However, due to the limitations of the spatial density of stations, the drought index calculated solely by station data still cannot fully understand the occurrence and development process of drought. In addition, for terrestrial ecosystems, drought may also be affected by non-meteorological factors (such as groundwater storage, agricultural irrigation, etc.), and meteorological drought indices alone are still not enough to monitor agricultural drought. The rapid development of satellite remote sensing technology has opened up new ways to monitor and assess the impact of large-scale drought events on surface vegetation. Traditional vegetation indices (VIs) obtained from remote sensing data, such as the Normalized Difference Vegetation Index (NDVI), the Enhanced Vegetation Index (EVI), and the Vegetation Near-Infrared Reflectance Index (NIRv), have been widely used to monitor the physiological state of vegetation and assess the impact of stress on vegetation productivity. Common drought indices based on VIs include the Vegetation Condition Index (VCI), the Vegetation Health Index (VHI), and the Temperature Vegetation Drought Index (TVDI). However, traditional VIs can only reflect changes in vegetation pigment content, but not changes in instantaneous photosynthetic rate, and have no direct correlation with actual photosynthesis. Therefore, traditional VIs are slow to respond to vegetation water loss, and there is a significant lag in capturing drought information, which greatly limits its applicability in early drought monitoring.

[0004] Solar-Induced Chlorophyll Fluorescence (SIF) is the energy that is re-excited after vegetation photosynthesis absorbs sunlight. It can directly reflect the dynamic changes of actual photosynthesis of plants and is considered to be a direct probe of photosynthesis. It can make up for the shortcomings of traditional optical remote sensing observations such as vegetation indices based on "greenness", and provides a new method for large-scale vegetation photosynthesis monitoring. In recent years, global satellite SIF data have been successfully inverted from space-borne sensors such as SCIAMACHY, GOME-2, GOSAT, TanSat, OCO-2, and TROPOMI, making large-scale research on plant photosynthesis possible. It also provides a new data source for monitoring regional vegetation growth conditions and environmental stresses. Among them, et al. and Guanter et al. used the TROPOMI sensor on the Sentinel-5P to obtain global SIF data with a time resolution of 1 day and a spatial resolution of 3.5 km × 5.5 km, which is considered to be the most promising SIF satellite remote sensing product. Since there is no satellite sensor specifically used to detect SIF, the SIF data inverted by satellite still have problems such as low spatial resolution (SCIAMACHY and GOME-2), discontinuous spatial sampling (GOSAT and OCO-2), and short time series (TROPOMI). In view of the above problems, scholars have successively used machine learning algorithms to reconstruct the above SIF products to improve product accuracy, such as CSIF, SIFoco2_005, GOSIF, DSIF, RTSIF, etc. However, in practical applications such as crop growth monitoring and yield prediction at the regional scale (especially field scale), the spatial resolution (0.05°×0.05°) of existing SIF products is still relatively rough and cannot meet the needs of precision agricultural monitoring. Therefore, it is urgent to adopt new models or algorithms to rebuild and generate SIF product datasets with longer time series and higher resolution to make up for the shortcomings of existing products.

[0005] Studies have shown that when plants are subjected to environmental stress (such as water stress, heat stress, etc.), SIF can achieve rapid response and early warning compared with traditional VIs. At present, SIF data based on satellite inversion have been widely used for monitoring and assessment of drought events at the regional scale. Previous studies mainly selected typical drought events at the regional scale, used SIF to explore the occurrence and development mechanism of drought, and compared them with traditional VIs, thereby revealing the great potential of SIF in accurately and timely monitoring drought events. However, how to bring out the above "potential", that is, how to systematically apply SIF to monitor and assess drought events, and further improve the timeliness and accuracy of agricultural drought monitoring at the regional scale, still needs further study. In addition, the occurrence of agricultural drought is often affected by the three factors of soil-vegetation-atmosphere, which is not only closely related to soil moisture, vegetation growth conditions, precipitation, etc., but also inseparable from land surface temperature (LST) and evapotranspiration (ET). Due to the very complex coupling relationship between the various drought-causing factors, multi-source information fusion technology has become a new method and new trend in drought monitoring research. Therefore, considering the superiority of SIF in drought monitoring, it is feasible and necessary to innovatively construct a new drought fluorescence monitoring index by integrating high-resolution satellite SIF information (vegetation physiological parameters) and non-vegetation parameter information such as soil moisture, precipitation, and LST.

[0006] Therefore, it is necessary to provide a remote sensing monitoring method for winter wheat drought that integrates downscaled sunlight-induced chlorophyll fluorescence data to solve the above technical problems. Summary of the invention

[0007] The technical problem solved by the present invention is to provide a winter wheat drought remote sensing monitoring method that integrates downscaled sunlight-induced chlorophyll fluorescence data and can solve the problem of how to integrate SIF information to improve the timeliness and accuracy of agricultural drought monitoring from a mechanistic perspective, and can enhance the agricultural meteorological disaster prevention and mitigation capabilities.

[0008] In order to solve the above technical problems, the present invention provides a winter wheat drought remote sensing monitoring method integrating downscaled sunlight-induced chlorophyll fluorescence data, comprising the following steps:

[0009] S1: Selection of input data for SIF data reconstruction machine learning model:

[0010] Based on the light energy utilization model, SIF can be expressed as:

[0011] SIF=PAR×fPAR×SIF yield

[0012] Where PAR is the photosynthetically active radiation reaching the vegetation canopy; fPAR is the photosynthetically active radiation ratio absorbed by the vegetation; SIF yield is the fluorescence quantum yield;

[0013] S2: Training and building of machine learning models:

[0014] Based on the model input data selected in S1, a machine learning algorithm such as a convolutional neural network, a support vector machine, a random forest, and an extreme learning tree is used to input the aggregated or resampled remote sensing product data into the machine learning model to train the model, thereby obtaining a trained machine learning model;

[0015] S3: Construction and effect verification of long-term high-resolution downscaled SIF product dataset:

[0016] Based on the machine learning model trained in S2, MODIS GPP, existing satellite SIF product data, GPP data of ecological station flux observations, and SIF data of farmland ecological station observations were used to test the comprehensive effect of the constructed SIF product data at the regional and station scales.

[0017] S4: Establish a drought fluorescence monitoring index construction plan:

[0018] Including: improving the drought fluorescence monitoring index based on the principle of traditional vegetation drought monitoring index or constructing the drought fluorescence monitoring index based on the principle of soil-vegetation-atmosphere system interaction;

[0019] S5: Unification of data temporal and spatial resolution;

[0020] S6: Validity test of drought fluorescence monitoring index;

[0021] S7: Spatial and temporal evolution characteristics of drought during the growing period of winter wheat.

[0022] Preferably, in S1, based on the SIF expression, the MCD43C4 reflectance data in the MODIS remote sensing product, the fPAR and LAI data in MCD15A2H and MCD15A3H, the temperature, saturated vapor pressure difference, photosynthetically active radiation, and soil moisture data in the ERA5 reanalysis data are selected as the explanatory variables of the model input, and the TROPOMI SIF data are used as the explained variables.

[0023] Preferably, in S2, the model input data selected in S1 are aggregated or interpolated and resampled to obtain 4-day and 8-day, 0.05° resolution MCD43C4 reflectance data from 2018 to 2021, fPAR and LAI data in MCD15A3H and MCD15A2H, and temperature, saturated vapor pressure difference, photosynthetically active radiation, and soil moisture data in ERA5 reanalysis data as the explanatory variable X for model training input; 4-day and 8-day, 0.05° resolution TROPOMISIF data obtained by interpolation and resampling are selected as the explained variable Y for model input, and the machine learning model is trained to obtain a trained machine learning model.

[0024] Preferably, in S3, the 4-day and 8-day, 500m resolution MCD43A4 reflectivity data from 2002 to 2024 obtained by arithmetic averaging in the region, the fPAR and LAI data in MCD15A3H and MCD15A2H, and the temperature, saturated vapor pressure difference, photosynthetically active radiation, and soil moisture data in the 4-day and 8-day, 500m resolution ERA5 reanalysis data obtained by interpolation and resampling are selected as explanatory variables X and input into the machine learning model trained in S2 to obtain the 4-day and 8-day, 500m resolution downscaled SIF product data set in the region from 2002 to 2024.

[0025] Preferably, in S4, the drought fluorescence monitoring index is improved and generated based on the principle of the traditional vegetation drought monitoring index, including: based on the construction principle of the traditional vegetation drought monitoring index, SIF is used to replace the traditional vegetation index NDVI, the drought fluorescence monitoring index that integrates SIF information is improved and generated, and its applicability in regional-scale agricultural drought monitoring is explored.

[0026] Preferably, in S4, the drought fluorescence monitoring index constructed based on the principle of interaction between the soil-vegetation-atmosphere system includes: based on the principle of interaction between the soil-vegetation-atmosphere system, introducing the soil moisture condition index SMCI calculated based on soil moisture data, the fluorescence condition index FCI generated based on the improvement of VCI, the temperature condition index TCI calculated based on the surface temperature LST, and the precipitation condition index PCI calculated based on precipitation data, and using the entropy weight method to determine the weights of the four indices at the pixel scale or using a machine learning algorithm to construct and generate a new drought fluorescence monitoring index DFMI.

[0027] Preferably, in S5, before constructing the drought fluorescence monitoring index, the spatiotemporal resolution of the different model input data sets selected in S1 and S2 need to be unified, that is, to be consistent with the SIF data, and the low-resolution data sets need to be interpolated and resampled through ArcGIS or programming software for further collaborative analysis for machine learning model construction.

[0028] Preferably, in S6, based on the traditional meteorological drought monitoring index SPEI and the soil moisture and soil humidity data observed at agricultural meteorological observation stations, the correlation coefficient method is used to systematically evaluate the comprehensive performance of the new drought fluorescence monitoring index constructed above, and determine the optimal drought fluorescence monitoring index constructed by integrating SIF information.

[0029] Preferably, in S7, the spatiotemporal evolution characteristics of drought during the winter wheat growing period in the region are analyzed based on the optimal drought fluorescence monitoring index, and the duration and intensity of typical drought events are determined using the run theory to achieve timely and accurate monitoring of agricultural drought at the regional scale.

[0030] Compared with the related art, the winter wheat drought remote sensing monitoring method provided by the present invention by integrating downscaled sunlight-induced chlorophyll fluorescence data has the following beneficial effects:

[0031] The present invention provides a remote sensing monitoring method for winter wheat drought by integrating downscaled sunlight-induced chlorophyll fluorescence data. The method uses a machine learning algorithm to reconstruct and generate a long-time series high-resolution downscaled SIF product data set. After that, the high-resolution SIF data is innovatively integrated to construct a new drought fluorescence monitoring index to timely and accurately monitor and evaluate drought events during the winter wheat growing period. The method can solve the scientific problem of how to integrate SIF information to improve the timeliness and accuracy of agricultural drought monitoring from a mechanistic perspective. The method can enhance the agricultural meteorological disaster prevention and mitigation capabilities, provide scientific and technological support for firmly guarding the bottom line of ensuring national food security, comprehensively promoting rural revitalization, accelerating agricultural and rural modernization, and accelerating the construction of an agricultural power. The method has both scientific significance and major national and social needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1A roadmap for the first technique of the winter wheat drought remote sensing monitoring method integrating downscaled sunlight-induced chlorophyll fluorescence data provided by the present invention;

[0033] Figure 2 A roadmap for the second technique of the winter wheat drought remote sensing monitoring method integrating downscaled sunlight-induced chlorophyll fluorescence data provided by the present invention;

[0034] Figure 3 The invention provides a drought monitoring index table based on NDVI and SIF for the winter wheat drought remote sensing monitoring method integrating downscaled sunlight-induced chlorophyll fluorescence data. DETAILED DESCRIPTION

[0035] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0036] Please refer to Figure 1 , Figure 2 and Figure 3 2. The winter wheat drought remote sensing monitoring method provided in this embodiment, which integrates downscaled sunlight-induced chlorophyll fluorescence data, adopts the technical solution of technology 1:

[0037] (1) Construction of long-term high-resolution downscaled satellite SIF product dataset:

[0038] Using TROPOMI SIF product data, MODIS MCD43C4 and MCD43A4 reflectivity data, MCD15A2H and MCD15A3H photosynthetically active radiation ratio (fPAR), leaf area index (LAI), and ERA5 reanalysis meteorological data, temperature, saturated vapor pressure deficit, photosynthetically active radiation, and soil moisture data, the optimal machine learning algorithm (such as convolutional neural network, support vector machine, random forest, extreme learning tree, etc.) was selected to reconstruct a long-term high-resolution downscaled SIF product dataset in China (2002-2024, with a temporal resolution of 4 days and 8 days, and a spatial resolution of 500m). MODIS GPP data and existing satellite SIF product data (such as CSIF and GOSIF) were used for regional scale verification, and ChinaFLUX flux site GPP data and ChinaSpec site SIF data were used for site scale verification.

[0039] (2) Technical route: Figure 1 shown.

[0040] (3) Specific technology - solution details

[0041] ①Selection of input data for machine learning models

[0042] Based on the light energy utilization model, SIF can be expressed as:

[0043] SIF=PAR×fPAR×SIF yield

[0044] Where PAR is the photosynthetically active radiation reaching the vegetation canopy; fPAR is the photosynthetically active radiation ratio absorbed by the vegetation; SIF yield is the fluorescence quantum yield.

[0045] Based on the above SIF expression, considering that too many input variables (inconsistency of data temporal and spatial resolution) may increase the uncertainty of machine learning model simulation, this study intends to select the MCD43C4 and MCD43A4 reflectance data in MODIS remote sensing products, the LAI and fPAR data in MCD15A3H and MCD15A2H, and the temperature, saturated water vapor pressure difference, photosynthetically active radiation, and soil moisture in the ERA5 reanalysis data as the explanatory variables of the model input, and the TROPOMI SIF data as the explained variable. The main reasons for selecting the above three explanatory variables are: first, reflectance and meteorological data have been proven in previous studies to be able to be used in the construction of SIF data; second, fPAR is an important part of the SIF calculation process; third, LAI and SIF yield The specific machine learning model input data information is shown in Table 1:

[0046] Table 1 Machine learning model input data information

[0047]

[0048] ② Training and construction of machine learning models

[0049] Based on the model input data selected in ①, we first select the daily 0.05° resolution MCD43C4 band 1-7 reflectivity data in China from 2018 to 2021, the 4-day 500m resolution MCD15A3H fPAR and LAI data or the 8-day 500m resolution MCD15A2H fPAR and LAI data, and the hourly 0.1° resolution ERA5 reanalysis data of temperature, saturated vapor pressure deficit, photosynthetically active radiation, and soil moisture data as the explanatory variable X of the model input, and the daily 3.5km×5.5km resolution TROPOMI SIF data from 2018 to 2021 are used as the explained variable Y of the model input. Before inputting the model data, considering the low resolution of the TROPOMISIF data, it is first necessary to interpolate and resample it to 4-day and 8-day, 0.05° resolution grids; secondly, the above explanatory variables (MCD43C4 reflectivity, MCD15A3H or MCD15A2H fPAR and LAI data, hourly ERA5 reanalysis data) need to be aggregated to 4-day or 8-day, 0.05° resolution grids to match the resampled TROPOMI SIF data. Afterwards, the aggregated or resampled remote sensing product data are input into the machine learning model using machine learning algorithms such as convolutional neural networks, support vector machines, random forests, and extreme learning trees to train the model and obtain a trained machine learning model.

[0050] ③ Construction and effect verification of the long-term high-resolution downscaling TROPOMI SIF product dataset

[0051] Based on the machine learning model trained in ②, we first select the 4-day and 8-day 500m resolution MCD43A4 reflectivity, 4-day 500m resolution MCD15A3H fPAR and LAI data, 8-day 500m resolution MCD15A2H fPAR and LAI data calculated from daily data in China from 2002 to 2024, and the temperature, saturated water vapor pressure deficit, photosynthetically active radiation, and soil moisture data from the 4-day and 8-day 500m resolution ERA5 reanalysis data resampled and recalculated as explanatory variables X and input them into the trained machine learning model to obtain the 4-day and 8-day 500m resolution downscaled TROPOMI SIF product dataset from 2002 to 2024; then, we use MODIS GPP (MOD17A2H), existing satellite SIF product data (such as CSIF and GOSIF, etc.), 30 ecological station flux observation GPP data shared by the China Flux Observation Research Network (ChinaFLUX), and 5 farmland ecological station observation SIF data shared by the China Ecosystem Spectral Observation Network (ChinaSpec) were used to test the comprehensive effect of the constructed high-resolution downscaled SIF product data from the regional and station scales. Among them, the ChinaFLUX and ChinaSpec station information is shown in Table 2:

[0052] Table 2 ChinaFLUX and ChinaSpec ecosystem observation site information

[0053]

[0054]

[0055] The winter wheat drought remote sensing monitoring method provided in this embodiment that integrates downscaled sunlight-induced chlorophyll fluorescence data adopts the technical solution of technology 2:

[0056] (1) Construction of a new drought fluorescence monitoring index integrating downscaled SIF data

[0057] Two methods are used to construct drought fluorescence monitoring index. One is based on the principle of traditional vegetation drought monitoring index. The traditional vegetation index (VCI, VHI and TVDI) is improved by integrating downscaled TROPOMI SIF data to generate new drought fluorescence monitoring index (FCI, FHI and TFDI). The other is based on the principle of interaction between soil-vegetation-atmosphere system (SPAC). The soil moisture condition index (SMCI), the fluorescence condition index (FCI) generated based on the improvement of VCI, the temperature condition index (TCI) and the precipitation condition index (PCI) are introduced. The entropy weight method or machine learning / deep learning algorithm is used to construct a new drought fluorescence monitoring index (DFMI). After that, the performance of the new drought fluorescence monitoring index constructed above is effectively evaluated using meteorological drought index SPEI, soil moisture and soil moisture data to determine the optimal index. Finally, based on the optimal index, the spatiotemporal evolution characteristics of drought during the growth period of winter wheat in the Huanghuaihai region are analyzed, the duration and intensity of typical drought events are determined, and the effect is tested using meteorological disaster yearbook data.

[0058] (2) Technical route, such as Figure 2 As shown,

[0059] (3) Details of the second technical solution

[0060] ① Construction scheme of new drought fluorescence monitoring index

[0061] Scheme 1: Improve and generate a new drought fluorescence monitoring index based on the principle of traditional vegetation drought monitoring index

[0062] At present, agricultural drought monitoring indices based on NDVI mainly include VCI, VHI, TVDI, etc. In view of the superiority of SIF in drought monitoring, this project intends to use SIF to replace the traditional vegetation index NDVI based on the construction principle of the above-mentioned traditional vegetation drought monitoring index, improve the generation of drought fluorescence monitoring index (FCI, FHI and TFDI) integrating SIF information, and explore its applicability in regional agricultural drought monitoring. Among them, the weight β in the VHI and FHI indexes is determined by the entropy weight method; the dry edge (LST) involved in the TVDI (TFDI) index is max ) and the wet edge equation (LST min ) Linear and nonlinear equations are used for fitting, and the equation with better fitting effect is selected as the dry side and wet side fitting scheme. The specific calculation formula is as follows Figure 3 As shown in Table 3:

[0063] Scheme 2: Drought fluorescence monitoring index based on the interaction principle of soil-vegetation-atmosphere system

[0064] Based on the principle of interaction between soil-vegetation-atmosphere system, attempts are made to introduce drought monitoring indexes such as soil moisture condition index (SMCI) calculated based on soil moisture data, fluorescence condition index (FCI) generated based on VCI improvement, temperature condition index (TCI) calculated based on surface temperature LST, and precipitation condition index (PCI) calculated based on precipitation data. The entropy weight method is used to determine the weights of the four indexes at the pixel scale or machine learning / deep learning algorithms are used to construct and generate new drought fluorescence monitoring index DFMI. The entropy weight method calculation process is as follows:

[0065] The entropy weight method was first introduced into information systems by Shannon et al. from thermodynamics. At present, the entropy weight method is a popular method for determining parameter weights, and has been widely used in engineering technology, social economy, remote sensing monitoring and other fields. In the communication process, the uncertainty of the signal is called "information entropy". The higher the weight, the lower the information entropy. The present invention mainly uses the entropy weight method to determine the weights of the four indexes at the pixel scale during the model construction process.

[0066] Assume there are m objects to be evaluated and n evaluation indicators, X = (x ij ) m×n is the initial decision matrix for the evaluation problem, where i = 1, 2, 3, …, m; J = 1, 2, 3, …, n.

[0067] The original data is then standardized to eliminate the influence of different dimensions.

[0068]

[0069] In the formula is the average value of the jth index, s j is the standard deviation of the jth indicator.

[0070] The information entropy H of each indicator is defined as follows:

[0071]

[0072] Finally, the entropy weight ω obtained from the information entropy j It is expressed as:

[0073]

[0074] Where 0≤ω j ≤1,

[0075] ② Unification of data temporal and spatial resolution

[0076] Before constructing a new drought fluorescence monitoring index, the temporal and spatial resolutions of different model input data sets need to be unified, that is, consistent with the SIF data, mainly through ArcGIS or programming software to interpolate and resample the low-resolution data sets for further collaborative analysis. Specific data information is shown in Table 4:

[0077] Table 4 Construction of drought fluorescence monitoring index dataset

[0078]

[0079] ③Validation of the new drought fluorescence monitoring index

[0080] Taking the traditional meteorological drought monitoring index SPEI, soil moisture and soil humidity data of agricultural meteorological observation stations as benchmarks, the correlation coefficient method was used to systematically evaluate the comprehensive performance of the above-mentioned new drought fluorescence monitoring indices (FCI, FHI, TFDI and DFMI) to determine the optimal drought fluorescence monitoring index constructed by integrating SIF information.

[0081] ④ Temporal and spatial evolution characteristics of drought during the growing period of winter wheat in the Huang-Huai-Hai region

[0082] Based on the optimal drought fluorescence monitoring index, the spatiotemporal evolution characteristics of drought during the winter wheat growing period in the Huanghuaihai region from 2002 to 2024 were analyzed, and the run theory was used to determine the duration and intensity of typical drought events, so as to achieve timely and accurate monitoring of agricultural drought at the regional scale.

[0083] Compared with the related art, the winter wheat drought remote sensing monitoring method provided by the present invention by integrating downscaled sunlight-induced chlorophyll fluorescence data has the following beneficial effects:

[0084] The present invention provides a remote sensing monitoring method for winter wheat drought that integrates downscaled sunlight-induced chlorophyll fluorescence data. A machine learning algorithm is used to reconstruct and generate a long-time series high-resolution downscaled SIF product data set (2002-2024, with a time resolution of 4 days and 8 days, and a spatial resolution of 500m); then, the high-resolution downscaled SIF data is innovatively integrated to construct a new drought fluorescence monitoring index to timely and accurately monitor and evaluate drought events during the winter wheat growth period. It can solve the scientific problem of how to integrate SIF information to improve the timeliness and accuracy of agricultural drought monitoring from a mechanistic perspective. It can enhance the ability of agricultural meteorological disaster prevention and mitigation, and provide scientific and technological support for firmly guarding the bottom line of ensuring national food security, comprehensively promoting rural revitalization, accelerating agricultural and rural modernization, and accelerating the construction of an agricultural power. It has both scientific significance and major national and social needs.

[0085] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A remote sensing monitoring method for winter wheat drought by integrating downscaled sunlight-induced chlorophyll fluorescence data, characterized in that: The following steps are involved: S1: Selection of input data for SIF data reconstruction machine learning model: Based on the light energy utilization model, SIF can be expressed as: SIF=PAR×fPAR×SIF yield Where PAR is the photosynthetically active radiation reaching the vegetation canopy; fPAR is the photosynthetically active radiation ratio absorbed by the vegetation; SIF yield is the fluorescence quantum yield; S2: Training and building of machine learning models: Based on the model input data selected in S1, a machine learning algorithm such as a convolutional neural network, a support vector machine, a random forest, and an extreme learning tree is used to input the aggregated or resampled remote sensing product data into the machine learning model to train the model, thereby obtaining a trained machine learning model; S3: Construction and effect verification of long-term high-resolution downscaled SIF product dataset: Based on the machine learning model trained in S2, MODIS GPP, existing satellite SIF product data, GPP data of ecological station flux observations, and SIF data of farmland ecological station observations were used to test the comprehensive effect of the constructed SIF product data at the regional and station scales. S4: Establishment of drought fluorescence monitoring index construction scheme: Including: improving the drought fluorescence monitoring index based on the principle of traditional vegetation drought monitoring index or constructing the drought fluorescence monitoring index based on the principle of soil-vegetation-atmosphere system interaction; S5: Unification of data temporal and spatial resolution; S6: Validity test of drought fluorescence monitoring index; S7: Spatial and temporal evolution characteristics of drought during the growing period of winter wheat.

2. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1 is characterized in that: In S1, based on the expression of SIF, the daily and 0.05° resolution MCD43C4 reflectivity data of MODIS remote sensing product data, the 4-day and 500m resolution MCD15A3H and the 8-day and 500m resolution MCD15A2H fPAR and LAI data, and the hourly and 0.1° resolution ERA5 reanalysis data temperature, saturated water vapor pressure difference, photosynthetically active radiation, soil moisture data are selected as the explanatory variables of the model training input, and the daily and 3.5km×5.5km resolution TROPOMI SIF data are selected as the explained variables of the model training input.

3. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1 is characterized in that: In S2, the model input data selected in S1 are aggregated or interpolated and resampled to obtain the MCD43C4 reflectance data with a time resolution of 4 days and 8 days and a spatial resolution of 0.05° from 2018 to 2021, the fPAR and LAI data in MCD15A3H and MCD15A2H, and the temperature, saturated water vapor pressure difference, photosynthetically active radiation, and soil moisture data in the ERA5 reanalysis data as the explanatory variable X of the model training input; the 4-day and 8-day TROPOMI SIF data with a resolution of 0.05° obtained by interpolation and resampling are selected as the explained variable Y of the model input, and the machine learning model is trained to obtain a trained machine learning model.

4. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1 is characterized in that: In S3, the 4-day and 8-day, 500-m resolution MCD43A4 reflectivity data from 2002 to 2024 obtained by arithmetic averaging, the fPAR and LAI data in MCD15A3H and MCD15A2H, and the temperature, saturated vapor pressure difference, photosynthetically active radiation, and soil moisture data in the 4-day and 8-day, 500-m resolution ERA5 reanalysis data obtained by interpolation and resampling are selected as explanatory variables X and input into the machine learning model trained in S2 to obtain the 4-day and 8-day, 500-m resolution downscaled SIF product dataset in the region from 2002 to 2024.

5. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1, characterized in that: In the S4, the drought fluorescence monitoring index is improved and generated based on the principle of the traditional vegetation drought monitoring index, including: based on the construction principle of the traditional vegetation drought monitoring index, SIF is used to replace the traditional vegetation index NDVI, and the drought fluorescence monitoring index that integrates the downscaled high-resolution SIF information is improved and generated, and its applicability in regional-scale agricultural drought monitoring is explored.

6. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1, characterized in that: In the S4, the drought fluorescence monitoring index constructed based on the principle of interaction between the soil-vegetation-atmosphere system includes: based on the principle of interaction between the soil-vegetation-atmosphere system, the soil moisture condition index SMCI calculated based on soil moisture data, the fluorescence condition index FCI generated based on the improved VCI, the temperature condition index TCI calculated based on the surface temperature LST, and the precipitation condition index PCI calculated based on precipitation data are introduced. First, the entropy weight method is used to determine the weights of the four indexes at the pixel scale. Second, a machine learning algorithm is used to construct and generate a new drought fluorescence monitoring index DFMI based on the measured soil moisture data.

7. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1, characterized in that: In S5, before constructing the drought fluorescence monitoring index, the spatiotemporal resolution of the different model input data sets selected in S1 and S2 needs to be unified, that is, consistent with the SIF data, and the low-resolution data sets need to be interpolated and resampled through ArcGIS or programming software for further collaborative analysis for machine learning model construction.

8. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1, characterized in that: In S6, based on the traditional meteorological drought monitoring index SPEI and the soil moisture and soil humidity data observed at agricultural meteorological observation stations, the correlation coefficient method is used to systematically evaluate the comprehensive performance of the new drought fluorescence monitoring index DFMI constructed above, and determine the optimal drought fluorescence monitoring index constructed by integrating the downscaled high-resolution SIF information.

9. The winter wheat drought remote sensing monitoring method of fusing downscaled sunlight-induced chlorophyll fluorescence data according to claim 1, characterized in that: In S7, the spatiotemporal evolution characteristics of drought during the growing period of winter wheat in the Huanghuaihai region are analyzed based on the optimal drought fluorescence monitoring index, and the duration and intensity of typical drought events are determined by the run theory. At the same time, the effectiveness is verified by using data such as the meteorological disaster yearbook, in order to achieve timely and accurate monitoring of agricultural drought at the regional scale.

Images