A near-real-time remote sensing monitoring method for crop phenology

Abstract

This invention discloses a near-real-time remote sensing monitoring method for crop phenology, belonging to the field of remote sensing monitoring technology. The method includes: acquiring the finite VI time-series curve of the current year and the complete VI time-series curve of historical years; generating a shape model for the current year; acquiring the complete VI time-series curves of N pixels of the same crop; selecting the top 10 complete VI time-series curves with the highest correlation coefficients between the finite VI time-series curve and the complete VI time-series curve of the target pixel; obtaining 10 first phenological estimates using the SMF-S method, and then combining them with the optimal translation step size to obtain 10 second phenological estimates for the current year of the target pixel; selecting the median of the second phenological estimates as the detection result. The near-real-time remote sensing monitoring method for crop phenology proposed in this invention, which integrates spatiotemporal information and shape model matching, can be used to detect the phenological stages of various crops during the growing season based on finite vegetation index time series, with significantly higher accuracy than other mid-season crop phenological detection methods.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing monitoring technology, and in particular relates to a near-real-time remote sensing monitoring method for crop phenology that integrates spatiotemporal information and shape model matching. Background Technology

[0002] Crop phenology defines the key stages in crop growth and development and provides crucial information for various agricultural activities, such as irrigation, fertilization, harvesting, and pest and disease control. Therefore, timely and accurate acquisition of crop phenological information at regional or global scales using satellite remote sensing data is essential. In recent years, increasing research has focused on real-time monitoring of crop phenology, also known as intra-season or growing season phenological monitoring. The key issue boils down to detecting crop phenology from incomplete VI (vegetation index) time-series curves. Assuming the VI time-series curve only covers up to a certain cutoff date, one solution is to predict the VI time-series curve after the cutoff date using historical years' VI time-series curves, and then use existing post-season detection methods to obtain estimated intra-season phenological periods. However, the accuracy of VI time-series curve reconstruction significantly impacts the performance of such methods. Another solution is to simulate the changing trends of local segments based on real-time acquired VI time-series curves to obtain phenological estimates. However, this method is only applicable to a few phenological periods (e.g., the greening stage) and has a detection time lag of several weeks.

[0003] Liu et al. (2022) proposed the SMF-S method based on the Shape Model Fitting (SMF) method (Reference 1: Liu, L., et al., "Detecting Crop Phenology from VegetationIndex Time-series Data by Improved Shape Model Fitting in Each PhenologicalStage." Remote Sensing of Environment 277(2022):113060. Web.). This method uses only local segments of the VI time-series curve for matching, which can better match the characteristics on the curves before and after the phenological period, obtaining high-precision post-season phenological estimates. However, this article did not consider the scenario of intra-season crop phenological detection. Figure 1 This paper demonstrates the differences in phenological detection using the SMF-S method in post-season and intra-season scenarios. The analysis shows that directly applying the SMF-S method to intra-season scenarios does not yield satisfactory results. In post-season detection scenarios, for a phenological period p0 on the shape model, the local segment of the VI time-series curve centered on p0 (i.e., Figure 1Curve matching is performed using the thick black line AB in A. The corresponding local segments in the VI time-series curve of the target pixel have very similar shapes. Therefore, the phenological estimate j′0 can be obtained by curve translation and stretching. In contrast, intra-season phenological detection can only use the local segments of the VI time-series curve of the target pixel before j0 (i.e., Figure 1 The thick black line A′B′ in B matches the shape model. However, the position of the phenological period of each target pixel in the VI time-series curve is different, which may lead to two challenges in intra-season phenological detection. First, for some target pixels, p0 is not in A′B′. In this case, a good match of curve A′B′ does not necessarily mean that p′0 is accurately estimated, because there are asynchronous vegetation growth during the growing season. Second, for some target pixels, curve A′B′ may lack features. In this case, curve matching is easily affected by noise, which leads to poor estimation of p′0. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a near-real-time remote sensing monitoring method for crop phenology, which solves the problem of limited application scenarios of existing mid-season detection methods and the shortcomings of directly using the SMF-S method in mid-season detection scenarios.

[0005] The technical solution adopted in this invention is as follows:

[0006] A near-real-time remote sensing monitoring method for crop phenology, used to detect phenological stages of various crops during the growing season based on finite VI (vegetation index) time-series curves, is characterized by comprising the following steps:

[0007] S1. Obtain the finite VI time series curve of the current year of the crop under study before a certain cutoff date and the complete VI time series curve of historical years; at the same time, obtain the spatial distribution of the crop under study in the current year and the true value of ground observation phenology, and generate a shape model.

[0008] S2. Determine N pixels with the same crop type as the target pixel within the local spatial window of the historical year, and obtain the complete VI time series curves of these N pixels; then perform translation matching only on the finite VI time series curves of the target pixel and the N complete VI time series curves in the local time window, and record the highest correlation coefficient as the optimal translation step size, and then select the top 10 optimal translation step sizes and complete VI time series curves with the highest correlation coefficients.

[0009] S3. Using the SMF-S method, post-season phenological detection was performed on the 10 complete VI time series curves obtained in S2 to obtain 10 relatively accurate first phenological estimates.

[0010] S4. Add the 10 optimal translation step sizes obtained in S2 to the corresponding first phenological estimates obtained in S3 to obtain 10 second phenological estimates for the target pixel in the current year. For stability considerations, the median of the 10 second phenological estimates is selected as the near-real-time phenological detection result for the target pixel in the current year.

[0011] Furthermore, in step S2, the phenological period of target pixel i in the current year j is determined. In other words, take the finite VI time series curve a segment x i Simultaneously, the complete VI time-series curve of the same crop pixel k within a local spatial window of a historical year is obtained. a segment x k (where yearj∈[j-1, j-2, j-3…jm]); then stable real-time curve matching is achieved using only linear translation:

[0012]

[0013] yearj∈[j-1, j-2, j-3…jm]

[0014] Among them, tshift k Let R(.) be the optimal translation step size, and let R(.) represent the Pearson correlation coefficient. Let the one with the highest correlation coefficient be the optimal translation step size, and then select the top 10 optimal translation step sizes based on the correlation coefficient ranking.

[0015] Assuming phenological periods For example, if the cutoff date of the finite VI time series curve for the current year is t0, then the local matching segment between the target VI time series curve and the VI time series curves for historical years is represented as follows:

[0016] x i =[t0-2t] w ,t0]

[0017] x k ∈[t0-3t w ,t0+t w ],with|x k |=2t w

[0018] Among them, t w It is a predefined time window.

[0019] Then, the top 10 complete VI time series curves ranked by Pearson correlation coefficient were selected.

[0020] Furthermore, in step S3, the SMF-S method is used to perform post-phenological seasonal detection on the 10 complete VI time series curves obtained in S2; assuming the target phenological period in the shape model is phen0, the SMF-S method is used to obtain the historical year time series curves. The first estimate of the phenological period

[0021]

[0022] Among them, tshift SMF-S The optimal linear translation parameters are obtained through curve matching using the SMF-S method; finally, 10 relatively accurate estimates of the first phenological phenomena are obtained.

[0023] Furthermore, in step S4, the results of steps S2 and S3 are combined, through... right The formula for phenological estimation is as follows:

[0024]

[0025] Obtain 10 second phenological estimates for the target pixel in the current year. The median of the 10 second phenological estimates is taken as the final phenological estimate of target pixel i in the current year j, as shown in the following formula:

[0026]

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] This invention presents a near-real-time remote sensing monitoring method for crop phenology that integrates spatiotemporal information and shape model matching. It addresses the limitations of existing mid-season detection methods in terms of application scenarios and overcomes the shortcomings of the SMF-S method when directly applied to mid-season detection scenarios. This method enables real-time and accurate mid-season phenological detection for multiple crops and multiple phenological points. First, this invention only performs translation operations during curve matching, ensuring the accuracy of curve matching in mid-season scenarios. Second, in the output results, the median of the post-season phenological estimates from the top 10 curves in the correlation coefficient ranking is retained, improving the algorithm's stability. Simultaneously, this invention incorporates the idea of ​​shape model matching, enabling better detection of crop phenological periods as defined by agronomy, rather than simply detecting changes in the VI time-series curve. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of curve matching between the SMF-S method in the post-season detection scenario (A) and the mid-season detection scenario (B).

[0030] Figure 2 This is a flowchart of near-real-time remote sensing monitoring of crop phenology based on the fusion of spatiotemporal information and shape model matching, according to an embodiment of the present invention.

[0031] Figure 3 The spatial distribution map of winter wheat in the North China Plain (top) and the NDVI time series and main phenological stages of winter wheat (bottom) are shown in the embodiments of the present invention.

[0032] Figure 4 This is a spatial distribution of US corn and soybean sites on the PhenoCam website, as well as a display of the location and planting area of ​​Dekalb County, according to an embodiment of the present invention.

[0033] Figure 5 Phenological stages of U.S. corn as a visual representation of the PhenoCam website in this embodiment of the invention;

[0034] Figure 6 Phenological periods for visual interpretation of US soybean sites on the PhenoCam website, as described in this embodiment of the invention;

[0035] Figure 7 The figure shows the phenological estimation performance when different spatiotemporal window sizes are set in this embodiment of the invention. The MAE value in the figure is the average value of the estimation error for all phenological periods.

[0036] Figure 8 This is a scatter plot showing the relationship between ground phenological values ​​and estimated phenological values ​​for winter wheat in the North China Plain, according to an embodiment of the present invention. The MAE value in each plot is the average of all samples.

[0037] Figure 9 This is a scatter plot showing the relationship between ground phenological values ​​and estimated phenological values ​​for maize in the United States, according to an embodiment of the present invention. The MAE value in each plot is the average of all samples.

[0038] Figure 10 This is a scatter plot showing the relationship between ground phenological values ​​and estimated phenological values ​​for soybeans in the United States, according to an embodiment of the present invention. The MAE value in each plot is the average of all samples.

[0039] Figure 11 The regional mapping results of winter wheat in the North China Plain during the emergence period in 2016, as an embodiment of the present invention, were compared with the SMF-S post-season detection results and the SMF-S mid-season method. The scatter density map shows the comparison between post-season and intra-season phenological estimates.

[0040] Figure 12 The regional mapping results of winter wheat in the North China Plain at maturity in 2016, as an embodiment of the present invention, were compared with the SMF-S post-season detection results and the SMF-S mid-season method. The scatter density map shows the comparison between post-season and intra-season phenological estimates.

[0041] Figure 13 The regional mapping results of the 2022 V7 phenological period of US maize in this embodiment of the invention were compared with the SMF-S post-season detection results and the SMF-S mid-season method. The scatter density map shows the comparison between the post-season and intra-season phenological estimates.

[0042] Figure 14 The regional mapping results of the 2022 R6 phenological period of US maize in this embodiment of the invention were compared with the SMF-S post-season detection results and the SMF-S mid-season method. The scatter density map shows the comparison between the post-season and intra-season phenological estimates.

[0043] Figure 15 The regional mapping results of the 2022 V2 phenological period of US soybean in this embodiment of the invention were compared with the SMF-S post-season detection results and the SMF-S mid-season method. The scatter density map shows the comparison between the post-season and intra-season phenological estimates.

[0044] Figure 16 The regional mapping results of the 2022 R7 phenological period of US soybean in this embodiment of the invention were compared with the SMF-S post-season detection results and the SMF-S mid-season method. The scatter density map shows the comparison between the post-season and intra-season phenological estimates. Detailed Implementation

[0045] To make the objectives and technical solutions of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0046] Taking winter wheat in the North China Plain and US corn and soybeans as examples, in this embodiment, the process of the near-real-time remote sensing monitoring method for crop phenology, which integrates spatiotemporal information and shape model matching, is as follows: Figure 2 As shown, it includes the following steps:

[0047] S1. Obtain the finite VI time series curve of the current year of the crop under study before a certain cutoff date and the complete VI time series curve of historical years; at the same time, obtain the spatial distribution of the crop under study in the current year and the true value of ground observation phenology, and generate a shape model.

[0048] (1) Ground phenological observations of winter wheat in the North China Plain. The North China Plain (29°-42°N, 105°-122°E) is an important winter wheat growing area in China, accounting for 60% of the country's wheat production. Winter wheat has a relatively long growing season, mainly sown in October of the previous year and harvested in June of the same year. Ground phenological observation records from 2008 to 2016 can be obtained from the National Meteorological Information Center of the China Meteorological Administration. Figure 3The spatial distribution of ground phenological observation stations in the North China Plain is shown. The data records nine phenological stages of winter wheat, including sowing, emergence, tillering, overwintering, greening, jointing, heading, milk stage, and maturity (see Table 1 for relevant agronomic descriptions). Figure 3 The table also shows the positions of the nine phenological stages on the NDVI timeline of winter wheat. Due to the fact that some observations of certain phenological stages at some ground stations were not conducted consecutively, the number of observational data for the nine phenological stages varies, ranging from 159 to 236 (Table 1).

[0049] (2) PhenoCam phenological data for US corn and soybean. The PhenoCam website provides digital photographs taken every ten minutes from hundreds of vegetation sites, which can record seasonal changes in vegetation. PhenoCam data can be used to generate quantitative observations of vegetation phenology. This embodiment collected PhenoCam data from US corn and soybean growing areas from 2019 to 2022. Unlike natural vegetation, the ground reference phenological period of corn and soybean cannot be estimated from the greenness index time series. Therefore, relevant technicians generated ground reference phenological data through visual interpretation of digital photographs.

[0050] The agronomical interpretation characteristics of phenological stages for maize and soybean are summarized in Table 1. Phenological data from a total of 28 maize stations and 22 soybean stations were collected. Figure 4 The locations of these sites are shown. For maize, four key phenological stages can be identified through visual interpretation: V2.5 (germination), V7 (jointing), VT (tasseling), and R6 (maturity). For soybeans, three key phenological stages can be identified through visual interpretation: VC (cotyledon stage), V2 (second three-leaf stage), and R7 (maturity). Figure 5 , Figure 6 This paper demonstrates the characteristics of four phenological stages of maize and three phenological stages of soybean on NDVI time series curves, and provides examples of digital photographs for different phenological stages. Furthermore, due to the lack of digital photographs, the true values ​​for certain phenological stages at some ground stations could not be obtained.

[0051]

[0052] Table 1

[0053] (3) NDVI Time Series Curve and Preprocessing. MODIS NDVI time series curves were used for winter wheat in the North China Plain because the fields there are relatively large. First, MODIS surface reflectance data (MOD09Q1) from 2008 to 2016 were collected, with a spatial resolution of 250 meters and a temporal resolution of 8 days. Second, NDVI was calculated using red and near-infrared surface reflectance to generate the original NDVI time series curve. Third, linear interpolation was performed using temporally adjacent cloud-free values ​​to fill in missing values ​​caused by cloud contamination in the original NDVI time series curve. Fourth, Savitzky-Golay filtering was used to correct negatively offset NDVI noise, smoothing the NDVI time series curve.

[0054] Since the specific locations of winter wheat agricultural meteorological stations are not provided, the target curve generation method of Liu et al. (2022) is followed to reduce the uncertainty of spatial consistency between ground observations and satellite observations. The steps are as follows: First, winter wheat for 2008-2016 is identified using the method of Qiu et al. (2017) (Reference 3: Qiu, B., et al., "Winter Wheat Mapping Combining Variations before and after Estimated Heading Dates." ISPRS Journal of Photogrammetry and Remote Sensing 123(2017):35-46. Web.), and only pixels identified as winter wheat in all nine years are retained as the spatial distribution of winter wheat. Second, the average NDVI time series data of winter wheat pixels within a 20km × 20km radius around each station is calculated as the NDVI time series data of the station. Third, only ground stations with a percentage of winter wheat pixels greater than 20% within a 20km × 20km radius are retained. The effectiveness of this process has been experimentally verified in previous studies.

[0055] The corn and soybean fields at the PhenoCam site are relatively small, so in the US region, 10-meter NDVI time-series data were generated by fusing 10-meter Sentinel-2 NDVI data and 250-meter MODIS NDVI data. The fusion method was GapFilling and Savitzky-Golay filtering (GF-SG) (Reference 4: Chen, Y., et al., "A Practical Approach to Reconstruct High-quality Landsat NDVI Time-series Data by GapFilling and the Savitzky–Golay Filter." ISPRS Journal of Photogrammetry and Remote Sensing 180(2021):174-90. Web.).

[0056] S2. Determine N pixels within the local spatial window of the historical year that have the same crop type as the target pixel, and obtain the complete VI time series curves of these N pixels; then perform translation matching only on the finite VI time series curves of the target pixel and the N complete VI time series curves within the local time window, and record the highest correlation coefficient as the optimal translation step size, and then select the top 10 complete VI time series curves in the correlation coefficient ranking.

[0057] Specifically, for the phenological period of target pixel i in the current year j. In other words, take the finite VI time series curve a segment x i Simultaneously, the complete VI time-series curve of the same crop pixel k within a local spatial window of a historical year is obtained. a segment x k (where yearj∈[j-1,j-2,j-3…jm]); then stable real-time curve matching is achieved using only linear translation:

[0058]

[0059] yearj∈[j-1,j-2,j-3…jm]

[0060] Among them, tshift k Let R(.) be the optimal translation step size, and let R(.) represent the Pearson correlation coefficient. Let the one with the highest correlation coefficient be the optimal translation step size, and then select the top 10 optimal translation step sizes based on the correlation coefficient ranking.

[0061] Assuming phenological periods For example, if the cutoff date of the finite VI time series curve for the current year is t0, then the local matching segment between the target time series curve and the historical time series curves is represented as follows:

[0062] x i =[t0-2t] w ,t0]

[0063] x k ∈[t0-3t w ,t0+t w ],with|x k |=2t w

[0064] Among them, t w It is a predefined time window, a fixed value of 30. This value is acceptable because the difference in phenological periods within a spatial local window or between pixels from different years is typically less than 30 days.

[0065] Then, the top 10 complete VI time series curves ranked by Pearson correlation coefficient were selected.

[0066] S3. Using the SMF-S method, post-season phenological detection was performed on the 10 complete VI time series curves obtained in S2 to obtain 10 relatively accurate first phenological estimates.

[0067] Specifically, the SMF-S method was used to perform post-season phenological detection on the 10 complete VI time-series curves obtained in S2. The SMF-S method used an iterative procedure to match the shape model curve with the studied VI curve for each target phenological period. Assuming the target phenological period in the shape model is phen0, the SMF-S method was used to obtain the historical year time-series curves. The first estimate of the phenological period

[0068]

[0069] Among them, tshift SMF-S The optimal linear translation parameters are obtained through curve matching using the SMF-S method; finally, 10 relatively accurate estimates of the first phenological phenomena are obtained.

[0070] S4. Add the 10 optimal translation step sizes obtained in S2 to the corresponding first phenological estimates obtained in S3 to obtain 10 second phenological estimates for the target pixel in the current year. For stability considerations, the median of the 10 second phenological estimates is selected as the near-real-time phenological detection result for the target pixel in the current year.

[0071] Specifically, combining the results of steps S2 and S3, through right The formula for phenological estimation is as follows:

[0072]

[0073] Obtain 10 second phenological estimates for the target pixel in the current year. The median of the 10 second phenological estimates is taken as the final phenological estimate of target pixel i in the current year j, as shown in the following formula:

[0074]

[0075] This invention searches for similar curves within a certain spatiotemporal range. Figure 7 The accuracy of results from different combinations of time and spatial windows is demonstrated. For the winter wheat ground station in the North China Plain, the time window varied from 1 year to 5 years, while the spatial window ranged from 250 meters to 40 kilometers. For the PhenoCam stations for maize and soybeans in the United States, due to the limited accumulation of Sentinel-2 data, the time window varied from 1 year to 3 years. When calculating on imagery with a spatial resolution of 10 meters, the computational load is very high. Considering practical applications, a maximum of 500 pixels of the same crop type are randomly selected for curve matching within each window. Furthermore, it is not necessary to use all pixels because some pixels may be in the same field at a spatial scale of 10 meters. Based on the experimental results, a stable phenological estimate can be generated with a time window of 3 years and a spatial window of 5 km. Considering the balance between estimation accuracy and computational efficiency, this set of parameter values ​​was adopted in subsequent validation.

[0076] In order to clarify the advantages of the present invention, it is compared with four other phenological detection methods: (1) The SMF-S method performs post-season detection on the complete NDVI, which can be regarded as a reference value for comparison with other intra-season phenological estimates; (2) The SMF-S method is directly applied to the intra-season NDVI time series to detect phenology, and curve matching is performed on a local segment of the NDVI time series before the cutoff date; (3) Similar to (2), the SMF method is applied to the incomplete intra-season NDVI time series to detect phenology; (4) Similar to (2), but the stretching coefficient is removed in the curve matching. The comparison with (4) highlights the importance of the present invention in integrating spatiotemporal information.

[0077] The following three experiments are set up to evaluate the effectiveness of the near real-time remote sensing monitoring method for crop phenology that integrates spatiotemporal information and shape model matching of the present invention.

[0078] The design and implementation results of each experiment are as follows:

[0079] Experiment 1: Quantitative Evaluation of Phenological Estimates from Ground Stations. For a given phenological period, assuming that the NDVI time series curve for the current year only yields the dates of phenological points corresponding to the shape model, phenological estimates from different methods are obtained (i.e., SMF-S mid-season, SMF-S mid-season without stretching, SMF mid-season, and spatiotemporal fusion shape model matching method (the method of this invention)). A quantitative evaluation is performed by calculating the mean absolute error (MAE) between the ground-observed phenological values ​​and the phenological estimates from different methods.

[0080] Figure 8 , Figure 9 , Figure 10 Scatter plots are shown for all samples of ground-based phenological observations and phenological estimates from different methods for winter wheat, maize, and soybean. At the winter wheat site, the average accuracy (MAE) for all phenological periods using the four methods was 9.8 days, 12.4 days, 13.0 days, and 27.1 days, respectively. The method of this invention exhibited the smallest estimation error and was comparable to the accuracy of post-season estimates (MAE: 9.8 days vs. 9.7 days). Similarly, at the maize site, the MAEs for the four methods were 8.4 days, 14.9 days, 15.5 days, and 55.3 days, respectively, while at the soybean site, the MAEs were 7.9 days, 12.4 days, 13.3 days, and 64.6 days, respectively. The method of this invention outperformed other intra-season detection methods. Furthermore, the estimation of all samples by the method of this invention was relatively stable, with no outliers, i.e., no significant large errors, indicating that the method of this invention is robust.

[0081] Experiment 2: Comparison of Regional Phenological Estimates. Regional mapping of phenological estimates was performed using the SMF-S post-season method, the method of this invention, and the SMF-S mid-season method. The SMF-S mid-season method without stretching and the SMF mid-season method were excluded in this experiment because the SMF mid-season method performed significantly worse, while the SMF-S mid-season method without stretching performed similarly to the SMF-S mid-season method. Since there was no real ground data at the pixel level over large-scale regions, the SMF-S post-season estimates were used as a reference phenological map, and comparisons were made between the methods. A 250-meter spatial resolution winter wheat phenological map for 2016 was generated on the North China Plain. In the United States, DeKalb County, Illinois, was selected as the study area, and 10-meter spatial resolution maize and soybean phenological maps for 2022 were generated. DeKalb County was chosen because corn and soybeans are the main crops in the county, and the corn planting date in 2019 was delayed by 3-4 weeks due to heavy spring rainfall, resulting in unusual phenological differences between 2019 and other years. This allowed for a comprehensive evaluation of the spatiotemporal fusion shape model matching method (see Experiment 3 for details).

[0082] This section presents a comparison of regional phenological estimation using the SMF-S postseason method, the method of this invention, and the SMF-S midseason method. For each crop type, results for one early phenological stage and one late phenological stage are shown. Figure 11 The regional mapping results of winter wheat emergence period are presented. Figure 12 The regional mapping results for the winter wheat maturity period are presented. Figure 13 The regional mapping results of the phenological stages of maize V7 are presented. Figure 14 The regional mapping results of the phenological stages of maize R6 are presented. Figure 15 The regional mapping results of soybean V2 phenological stages are presented. Figure 16 Regional mapping results of soybean R7 phenological stages are presented. The results show that the mapping results of the method of this invention are more consistent with those of the SMF-S post-season detection method compared to the mid-season method. For example, the MAE value and correlation coefficient for winter wheat emergence are 4.2 days and 0.86 for the method of this invention, and 6.5 days and 0.75 for the mid-season method of SMF-S. The regional mapping experimental results further highlight the potential of the method of this invention in practical applications.

[0083] Experiment 3: Performance of the spatiotemporal fusion shape model matching method in years with anomalous phenological changes

[0084] This invention integrates spatiotemporal information from spatially adjacent pixels of historical years to investigate whether its performance significantly degrades when crop phenology varies considerably from year to year. In 2019, due to heavy spring rainfall in Illinois, maize planting was delayed by 3-4 weeks, making it an anomalous year for crop phenology. A comparison of the regional phenological estimation performance for maize and soybeans in DeKalb County between 2019 and 2022 was conducted, and the differences in regional phenological estimation between this invention's method, the SMF-S mid-season method, and the SMF-S post-season method were calculated.

[0085] The results show that the accuracy of the two mid-season phenological detection methods was significantly lower in 2019 than in 2022. The statistical results in Tables 2 and 3 indicate that extreme phenological anomalies significantly impacted mid-season phenological detection. The performance decline in 2019 can be explained by the target phenological period being significantly later than the corresponding period on the shape model. However, the method of this invention outperformed the SMF-S mid-season method in both 2019 and 2022. Although three years of data from 2019 to 2021 were used to detect the phenology in 2022, the extreme phenological anomalies in 2019 had a relatively small impact on the performance in 2022, demonstrating the robustness of the method of this invention.

[0086]

[0087] Table 2

[0088]

[0089] Table 3

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for crop phenology near real-time remote sensing monitoring, for detecting phenology stage based on limited VI time series curve for multiple crops in growing season, characterized in that, Includes the following steps: S1. Obtain the finite VI time-series curve of the current year of the crop under study before a certain cutoff date and the complete VI time-series curve of historical years; at the same time, obtain the spatial distribution of the crop under study in the current year and the true value of ground observation phenology, and generate a shape model. S2. Determine N pixels within the local spatial window of the historical year that have the same crop type as the target pixel, and obtain the complete VI time series curves of these N pixels; then perform translation matching only on the finite VI time series curves of the target pixel and the N complete VI time series curves in the local time window, and record the highest correlation coefficient as the optimal translation step size, and then select the top 10 optimal translation step sizes and complete VI time series curves in the correlation coefficient ranking. S3. Use the SMF-S method to perform post-season phenological detection on the 10 complete VI time series curves obtained in S2 to obtain 10 first phenological estimates. S4. Add the 10 optimal translation step sizes obtained in S2 to the corresponding first phenological estimates obtained in S3 to obtain 10 second phenological estimates for the target pixel in the current year. For stability considerations, select the median of the 10 second phenological estimates as the near-real-time phenological detection result for the target pixel in the current year. In step S2, for the current year target pixel phenological period In other words, take the finite VI time series curve A section Simultaneously, it extracts pixels of the same crop within a local spatial window of a historical year. Complete VI timing curve A section , where subscript Then, stable real-time curve matching is achieved using only linear translation: wherein, is the optimal translation step size, denotes the Pearson correlation coefficient; the optimal translation step size with the highest correlation coefficient is selected, and then the optimal translation step sizes with the top 10 correlation coefficients are selected; Let the phenophase For the current year, the end date of the VI timing curve is The local matching segment of the target timing curve and the historical year timing curve is expressed as follows: , with in, It is a predefined time window; Then, the top 10 complete VI time series curves ranked by Pearson correlation coefficient were selected.

2. The near-real-time remote sensing monitoring method for crop phenology as described in claim 1, characterized in that, In step S3, the SMF-S method is used to perform post-phenological detection on the 10 complete VI time-series curves obtained in S2; assuming the target phenological period in the shape model is... The SMF-S method was used to obtain the time series curves for historical years. The first estimate of the phenological period : in, These are the optimal linear translation parameters obtained through curve matching using the SMF-S method; finally, 10 first phenological estimates are obtained.

3. The near-real-time remote sensing monitoring method for crop phenology as described in claim 2, characterized in that, In step S4, the results of steps S2 and S3 are combined, and then... ( )right The formula for phenological estimation is as follows: Obtain 10 second phenological estimates for the target pixel in the current year. The median of the 10 second phenological estimates was taken as the target pixel. In the current year The final phenological estimation result is given by the following formula: 。

Images