A method for estimating straw coverage degree and mapping conservation tillage based on spatio-temporal fusion of multi-source remote sensing images and ground temperature constraint

Abstract

The present application relates to the field of farmland ecological data estimation, and discloses a straw coverage degree estimation and conservation tillage mapping method based on multi-source remote sensing image space-time fusion and ground temperature constraint, comprising the following steps: step S1, establishing a remote sensing monitoring golden time window based on ground surface temperature constraint; step S2, acquiring multi-source satellite remote sensing data and ground measured auxiliary data; step S3, multi-source remote sensing image consistency preprocessing; step S4, high space-time resolution image reconstruction based on the ESTARFM algorithm; step S5, feature index calculation and regional straw coverage degree CRC inversion; step S6, conservation tillage mode classification determination and mapping; the present application establishes the best dynamic monitoring time window, improves the purity of the straw spectral signal, solves the contradiction that the space-time resolution of a single data source cannot be compatible, and realizes continuous and seamless farmland monitoring.

Description

Technical Field

[0001] This invention relates to the field of farmland ecological data estimation, specifically a method for estimating straw coverage and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints. Background Technology

[0002] Currently, the most basic method for obtaining farmland straw coverage is traditional field measurement, mainly including visual inspection, line transect method, and photographic method. However, these traditional methods have extremely obvious limitations in practical applications: limited spatial range and low efficiency: traditional methods rely heavily on manual labor, are time-consuming and labor-intensive, and are only applicable at a very small scale, making it difficult to meet the needs of large-scale continuous monitoring; strong subjectivity and poor representativeness: measurement results are easily affected by human operation and subjective factors, and the limited discrete sampling points cannot comprehensively reflect the overall condition of farmland with high spatial heterogeneity.

[0003] Limitations of existing single remote sensing monitoring technologies: Although calculating tillage indices (such as the Normalized Difference Tillage Index, NDTI) using satellite remote sensing imagery is currently the mainstream technology, existing single remote sensing data sources face irreconcilable contradictions in practical applications: The temporal resolution bottleneck and weather constraints of high spatial resolution optical remote sensing data: To minimize the impact of green vegetation on the straw index, the optimal time for remote sensing classification of tillage practices is strictly limited to a very short time window between crop sowing and seedling stage. Landsat 8's long 16-day revisit cycle, coupled with the extreme sensitivity of optical sensors to weather conditions such as clouds, rain, and fog, makes it extremely difficult to acquire high-quality, cloud-free effective images during the critical but cloudy and rainy spring sowing period, resulting in data gaps during key agricultural seasons; Lack of spatial detail in high temporal resolution remote sensing data: Although MODIS imagery can provide a daily revisit and has extremely high temporal resolution, its spatial resolution in the shortwave infrared band is extremely coarse (up to 500 meters). At a scale of 500 meters, the problem of mixed pixels is extremely serious. A single pixel often contains multiple types of farmland, roads, woodlands, etc., making it impossible to accurately identify straw coverage and farming patterns at the plot level, thus losing the significance of precision agriculture monitoring.

[0004] The application gaps and uncertainties of existing spatiotemporal fusion models in conservation tillage monitoring: To reconcile the contradiction between spatial and temporal resolution, introducing spatiotemporal fusion algorithms (combining Landsat and MODIS) is a theoretically feasible path. However, existing technologies still have significant blind spots in the application of spatiotemporal fusion models in the specific and complex scenario of "conservation tillage monitoring": Model applicability is unclear: The Spatiotemporal Adaptive Reflectance Fusion Model (STARFM) assumes that land cover type has not changed significantly, and its predictive ability is weak for rapidly changing surfaces and areas with high spatial heterogeneity; the Enhanced Spatiotemporal Adaptive Reflectance Fusion Model (ESTARFM) considers the trend of reflectance changes, but requires more reference images; the Adaptive Data Fusion Algorithm (FSDAF) is optimized for abrupt changes in land cover, but its computation is extremely complex. It is currently unclear which model will achieve the best results in predicting the NDTI index, which determines straw cover; a lack of performance-efficiency balancing mechanism: In large-scale agricultural monitoring, not only is estimation accuracy (R², RMSE) required, but the model also needs reasonable computational efficiency (time cost). Current technologies lack a systematic comparative evaluation of the accuracy and computation time of these three mainstream models, making it difficult to choose the right one for practical engineering applications. They also neglect the mechanistic impact of landscape fragmentation (cultivated land percentage) on fusion accuracy: existing studies often treat fusion models as a black box tool, ignoring the land cover characteristics of the study area itself (such as the proportion of cultivated land to the total area and the degree of land fragmentation). Furthermore, they lack a precise mechanism for dynamically establishing phenological time windows: existing remote sensing monitoring often relies on fixed calendar times to acquire images, ignoring the fact that crop sowing is strictly driven by surface temperature (e.g., corn can only be sown when the temperature reaches above 10℃). Blindly selecting time windows can lead to severe contamination of images by green vegetation after emergence, damaging the spectral signal of straw.

[0005] Therefore, existing technologies urgently need a new method for estimating crop straw coverage that can dynamically lock the optimal monitoring window, collaboratively utilize the advantages of multi-source remote sensing data, reconstruct high-quality images by selecting the most efficient and accurate spatiotemporal fusion algorithm, and clarify the mechanism by which regional features affect accuracy. Summary of the Invention

[0006] The purpose of this invention is to provide a method for estimating straw coverage and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints, comprising the following steps:

[0008] Step S1: Establishing the golden time window for remote sensing monitoring based on surface temperature constraints. Specifically, this involves acquiring high temporal resolution surface temperature remote sensing products for the target study area, setting the minimum physiological ground temperature threshold for crop germination, calculating the date when the temperature threshold is reached, and acquiring the time window.

[0009] Step S2, acquisition of multi-source satellite remote sensing data and ground-based measured auxiliary data, specifically high spatial resolution image acquisition, high temporal resolution image acquisition, and ground-based measured data acquisition;

[0010] Step S3, multi-source remote sensing image consistency preprocessing, specifically radiometric calibration, projection transformation, and pixel unification;

[0011] Step S4, high spatiotemporal resolution image reconstruction based on the ESTARFM algorithm, specifically involves input data configuration, finding similar pixels, calculating weight functions, determining spectral conversion coefficients, and deducing high-resolution pixel values.

[0012] Step S5: Calculation of characteristic index and CRC inversion of regional straw coverage;

[0013] Step S6: Classification and mapping of conservation tillage patterns.

[0014] Preferably, step S1 uses the VNP21A1N nighttime surface temperature product of the VIIRS satellite to obtain the high temporal resolution land surface temperature remote sensing product of the target study area. Its spatial resolution is 1 km, temporal resolution is 1 day, the transit time is consistent with the local daily minimum temperature occurrence time, and the date of reaching the temperature threshold is calculated on a pixel-by-pixel basis in the cloud computing platform, which calculates the date when the spring daily minimum ground temperature first stably reaches or exceeds 10℃.

[0015] Preferably, the acquisition of the time window in step S1 is specifically achieved by combining the revisit characteristics of the selected high spatial resolution satellite, taking the date when 10℃ is reached as the starting point, and extending it forward by one seedling cycle to dynamically determine the optimal remote sensing image acquisition time window for each pixel in the study area.

[0016] Preferably, step S2, high spatial resolution image acquisition, specifically involves acquiring Landsat 8 OLI L1TP level images of the target area within a defined time window and before and after it. Two sets of images for reference dates (t1, t3) need to be acquired to form the basis for building the fusion model. At the same time, the target prediction date (t2) is recorded. High temporal resolution image acquisition specifically involves acquiring MODIS surface reflectance products provided by existing technologies, and it is necessary to acquire images synchronized with the Landsat 8 reference dates (t1, t3) and images for the target prediction date (t2).

[0017] Preferably, step S2, ground measurement data acquisition, specifically involves setting up representative sampling points within the target area within a defined time window and using a standardized transect method for measurement: using a fixed-length transect rope marked with equidistant markers, the rope is stretched along the diagonal of the quadrat, and the number of intersections between the transect rope markers and the crop straw on the ground is counted. The ratio of the number of intersecting markers to the total number of markers is calculated as the measured straw coverage CRC of the quadrat. Based on a spatial resolution of 30 meters, the average value is obtained by using a five-point cross-sampling method within a 30m×30m pixel area, and the center coordinates of the sampling points are recorded using GPS.

[0018] Preferably: In step S3, radiometric calibration is performed on the acquired Landsat 8 OLI image to convert it to atmospheric top reflectance, followed by atmospheric correction to obtain true surface reflectance data; projection conversion uses batch processing tools to perform band extraction and reprojection operations on the MOD09GA data, converting its projection coordinate system to the UTM-WGS84 coordinate system that is completely consistent with the Landsat image; pixels are uniformly extracted from the shortwave infrared bands corresponding to MODIS, and bilinear interpolation is used to resample the 500-meter spatial resolution MODIS image to a 30-meter spatial resolution, ensuring that it achieves sub-pixel-level complete alignment with the Landsat 8 image in terms of spatial range, pixel size, number of rows and columns, and projection type.

[0019] Preferred configuration for step S4: Input data configuration: Input Landsat 8 surface reflectance images and resampled MODIS surface reflectance images for two reference dates (t1, t3), and MODIS image for the predicted date (t2); Find similar pixels: Within a set sliding window (w), using the cross value of the Landsat 8 reflectance data from the two reference dates, identify and determine similar pixels with similar land cover category (c) to the center pixel; Calculate the weight function (w) i ): Determine the number N of similar pixels within the sliding window, and assign weights to each similar pixel by comprehensively considering the spatial distance, spectral differences, and temporal evolution characteristics between similar pixels and the center pixel.

[0020] Preferably: step S4 determines the spectral conversion coefficient (V) i ): A linear regression analysis was performed on MODIS data from the baseline date and Landsat 8 data from the same period to determine the conversion coefficient (V) reflecting the systematic spectral response differences between the two sensors when observing the same ground feature. i High-resolution pixel value extrapolation: Based on the formula, the high spatial resolution surface reflectance prediction value of the central pixel on the predicted date (t2) is calculated. The mathematical expression is:

[0021]

[0022] Where L represents the derived high-resolution reflectance, and M represents the low-resolution reflectance; x i y i t represents the pixel coordinates; p t2 is the predicted date, t0 is the base date; b is the specified band; w i V represents the comprehensive weight of the i-th similar pixel; i is the spectral conversion coefficient; N is the total number of similar pixels within the sliding window.

[0023] Preferably: Step S5 utilizes the high-resolution surface reflectance data of the predicted date (t2) generated in step S4 to extract the first and second shortwave infrared bands. Based on the characteristic absorption valley principle of lignin and cellulose in crop straw around 2100nm, the Normalized Differential Tillage Index (NDT) is calculated using the following formula:

[0024]

[0025] Next, the calculated pixel-by-pixel NDTI data is substituted into the pre-constructed or preferred straw coverage linear estimation model of this patent. The model formula is as follows:

[0026] CRC=754.71NDTI+5.3817

[0027] By substituting each pixel into the calculation, a high spatial resolution spatial distribution map of straw coverage in the entire study area on the key prediction date is generated.

[0028] Preferably, step S6 uses GIS spatial analysis tools to reclassify the continuous CRC numerical map obtained in step S5 into discrete farming pattern type maps, and calculates the area proportion of each farming pattern in the study area, ultimately generating a regional conservation tillage spatial distribution map.

[0029] Compared with the prior art, the beneficial effects of this invention are as follows:

[0030] This invention extracts high-frequency nighttime thermal infrared data through a cloud computing platform, calculates the date on a pixel-by-pixel basis when the lowest daily ground temperature in spring first reaches 10°C, and uses this as a starting point to define an extremely short time window from "after sowing to before emergence", thus eliminating the pollution of the straw index (NDTI) by the green vegetation spectrum from the physical source.

[0031] Through real-world experimental comparisons, it was found that in an environment where the surface reflectance changes rapidly during the sowing period of farmland, the prediction accuracy (R²) of STARFM is only 0.09~0.49. Although the FSDAF algorithm considers surface abrupt changes, due to the flaws in its algorithm principle (the decomposition calculation of a single domain reduces the edge accuracy) and its extremely complex calculation, it takes up to 2 hours and 44 minutes and has an R² of only 0.33~0.48. In contrast, the ESTARFM model of this invention perfectly matches the law of surface abrupt changes, with a prediction accuracy of up to 0.56~0.69 and a single scene calculation only takes 3 minutes and 3 seconds.

[0032] It overcomes the "data gap" problem of lacking high-quality, high-resolution, and effective remote sensing images during critical agricultural seasons, solves the blind application of existing fusion models, and addresses the engineering dilemma of "efficiency and accuracy being difficult to balance" in large-scale applications. It also solves the problem that traditional measurement methods cannot achieve large-area, high-precision, and continuous spatial mapping and pattern classification, breaks through the technical bottleneck of the incompatibility of spatiotemporal resolution, and realizes true all-weather, large-scale, and refined dynamic monitoring. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention;

[0034] Figure 2 The following is a detailed step diagram of an embodiment of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example

[0037] Please see Figures 1-2 The figure shows a method for estimating straw coverage and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints. The method includes the following steps: Step S1: Establishing the golden time window for remote sensing monitoring based on ground temperature constraints. In order to minimize the interference of green vegetation on the straw spectrum (especially the short-wave infrared band) after crop emergence, the monitoring time must be strictly limited to after sowing and before emergence.

[0038] Specifically, this includes: acquiring high temporal resolution land surface temperature remote sensing products for the target study area: preferably using the VNP21A1N nighttime land surface temperature product from the VIIRS satellite, which has a spatial resolution of 1 km, a temporal resolution of 1 day, and a transit time close to the local daily minimum temperature occurrence time, accurately reflecting the soil baseline temperature; setting the minimum physiological soil temperature threshold for crop germination: taking maize as an example, setting the minimum temperature required for its germination to 10℃; calculating the date of reaching the temperature threshold: in a cloud computing platform (such as Google Earth Engine, GEE), calculating the date on a pixel-by-pixel basis when the spring daily minimum soil temperature first stably reaches or exceeds 10℃; acquiring the time window: combining the revisit characteristics of the selected high spatial resolution satellite (such as Landsat 8, with a revisit period of 16 days and an 8-day overlap between adjacent orbits), starting from the date of reaching 10℃, extending forward by one seedling cycle (such as 10-15 days), thereby dynamically determining the optimal remote sensing image acquisition time window for each pixel in the study area.

[0039] Step S2: Acquisition of multi-source satellite remote sensing data and ground-based auxiliary data, specifically including: High spatial resolution image acquisition: Within the determined time window and before and after it, acquire Landsat 8 OLI L1TP level images of the target area. Two sets of images for the reference date (t1, t3) are required (with extremely low cloud cover) to form the basis for building the fusion model; at the same time, record the target prediction date (t2), which is usually within the time window but Landsat images are missing or of extremely poor quality due to clouds and rain; High temporal resolution image acquisition: Acquire MODIS surface reflectance products (preferably MOD09GA) provided by NASA. Images synchronized with the Landsat 8 reference date (t1, t3) and images for the target prediction date (t2) must be acquired.

[0040] Further, ground-based measured data acquisition (for modeling and verification): Within a defined time window, representative sampling points are set up in the target area, and measurements are taken using the standardized transect method: A transect rope of fixed length (e.g., 20 meters) marked with equidistant marks (e.g., every 0.4 meters) is stretched straight along the diagonal of the quadrat, and the number of times the transect rope marks intersect with the crop straw on the ground is counted. The ratio of the number of intersecting marks to the total number of marks is calculated as the measured straw coverage (CRC) of the quadrat. To match the 30-meter spatial resolution, a five-point cross-sampling method is used within a 30m×30m pixel area to take the average value, and the center coordinates of the sampling points are recorded using GPS.

[0041] Step S3: Strict consistency preprocessing of multi-source remote sensing images. Since spatiotemporal fusion requires a high degree of matching between the two types of data in terms of spatial and radiometric characteristics, strict preprocessing must be performed. Specifically, this includes radiometric calibration: performing radiometric calibration on the acquired Landsat 8OLI image to convert it to atmospheric top reflectance, followed by atmospheric correction to obtain the true surface reflectance data, and extracting the shortwave infrared first band (SWIR1, center wavelength approximately 1610nm) and shortwave infrared second band (SWIR2, center wavelength approximately 2200nm) that are sensitive to straw.

[0042] Projection transformation: Batch processing tools (such as MODIS Reprojection Tool, MRT) are used to perform band extraction and reprojection operations on MOD09GA data, converting its projection coordinate system to the UTM-WGS84 coordinate system that is completely consistent with the Landsat image; Pixel unification: The corresponding shortwave infrared bands (Band 6 and Band 7) of MODIS are extracted, and the 500-meter spatial resolution MODIS image is resampled to 30-meter spatial resolution using bilinear interpolation to ensure that it achieves sub-pixel level complete alignment with the Landsat 8 image in terms of spatial range, pixel size, number of rows and columns, and projection type.

[0043] Step S4: High spatiotemporal resolution image reconstruction based on the ESTARFM algorithm. In this step, the STARFM model with poor prediction accuracy and the FSDAF model with extremely low computational efficiency are discarded. The enhanced spatiotemporal adaptive reflectivity fusion model (ESTARFM) is specifically selected and spatiotemporal fusion reconstruction of the image is performed. The specific mathematical and physical process is as follows:

[0044] Input data configuration: Input Landsat 8 surface reflectance images and resampled MODIS surface reflectance images for two reference dates (t1, t3), and MODIS image for the predicted date (t2).

[0045] Finding similar pixels: Within a set sliding window (w), using the cross value of Landsat 8 reflectance data from two reference dates, identify and determine similar pixels that have a similar land cover category (c) to the center pixel.

[0046] Calculate the weight function (w) i ): Determine the number (N) of similar pixels within the sliding window, and comprehensively consider the spatial distance, spectral differences, and temporal evolution characteristics between similar pixels and the central pixel. Assign a weight to each similar pixel, and the magnitude of the weight determines the contribution of adjacent pixels to the predicted value of the central pixel.

[0047] Determine the spectral conversion coefficient (V) i ): A linear regression analysis was performed on MODIS data from the baseline date and Landsat 8 data from the same period to determine the conversion coefficient (V) reflecting the systematic spectral response differences between the two sensors when observing the same ground feature. i ).

[0048] High-resolution pixel value extrapolation: Based on the formula, the predicted surface reflectance value of the central pixel on the predicted date (t2) is calculated at a high spatial resolution (30 meters). The core mathematical expression is as follows:

[0049]

[0050] Where L represents the derived high-resolution reflectance, and M represents the low-resolution reflectance; x i y i t represents the pixel coordinates; p t2 is the predicted date, t0 is the base date; b is the specified band; w i V represents the comprehensive weight of the i-th similar pixel; i is the spectral conversion coefficient; N is the total number of similar pixels within the sliding window.

[0051] Step S5: Feature index calculation and regional straw coverage (CRC) inversion. Using the high-resolution surface reflectance data of the predicted date (t2) generated in step S4, extract the first shortwave infrared band (SWIR1) and the second shortwave infrared band (SWIR2).

[0052] Based on the characteristic absorption valley principle of lignin and cellulose in crop straw around 2100 nm, the Normalized Differential Tillage Index (NDTI) is calculated using the following formula:

[0053]

[0054] Next, the calculated pixel-by-pixel NDTI data is substituted into the pre-constructed or present invention's linear estimation model for straw cover. The model formula is as follows:

[0055] CRC=754.71NDTI+5.3817

[0056] By substituting each pixel into the calculation, a high spatial resolution spatial distribution map of straw coverage in the entire study area on the key prediction date is generated.

[0057] Step S6: Classification and mapping of conservation tillage patterns. Based on mainstream international standards and the classification definition of the US Center for Conservation Technology Information (CTIC), set quantitative judgment thresholds for tillage patterns:

[0058] When the pixel At that time, the pixel was classified as Conventional Tillage (CT).

[0059] When the pixel The cell is classified as Reduced Tillage (RT).

[0060] When the pixel The pixel was classified as No-Tillage (NT), which is a typical example of conservation tillage.

[0061] Using GIS spatial analysis tools (such as ArcGIS's zoning statistics function), the continuous CRC numerical map obtained in step S5 is reclassified into discrete tillage pattern type maps, and the area proportion of each tillage pattern in the study area is statistically analyzed, ultimately generating a regional conservation tillage spatial distribution map.

[0062] The effectiveness of the method proposed in this invention will be verified below with reference to specific embodiments. The specific process is as follows:

[0063] 1. Overview of the experimental area and optimization of ground temperature (determining the monitoring window)

[0064] This example selects central Indiana, USA as the study area (this region is a typical global practice area for conservation tillage, with flat terrain, mainly growing corn and soybeans, and soil types of leached soil and black fertile soil).

[0065] To determine the optimal time for acquiring remote sensing images: the nighttime surface temperature product of VNP21A1N from the VIIRS satellite (representing the lowest daily ground temperature, with a resolution of 1km) was accessed on the Google Earth Engine (GEE) platform. Threshold screening was performed using the GEE's filter() function to calculate the date when each pixel first reached 10℃ in spring (10℃ is the physiological baseline temperature for safe planting of maize).

[0066] Analysis revealed that most areas within the study region reached 10℃ between April 8 and May 2, 2015. This period was designated as the golden monitoring window from "post-sowing to pre-emergence," and satellite imagery was retrieved and ground sampling was arranged based on this.

[0067] 2. Standardized acquisition of ground-based sampling data (used to verify the reliability of this invention)

[0068] Based on the time window defined by ground temperature, 115 sampling points were strictly set up in the study area from April 8 to May 26, 2015.

[0069] Sampling method: The rope transect method was used. A 20-meter long transect rope marked with red cloth tape (marked every 0.4 meters) was used. Spatial matching operation: To accurately match Landsat's 30-meter resolution, the transect rope was stretched taut at the target center point in a "plum blossom" pattern (i.e., the center and four opposite corners). The number of intersections between the marked points on the transect rope and the surface waste corn / soybean straw was recorded. The average intersection ratio of these 5 quadrats was calculated as the true ground CRC value of the 30m×30m pixel. The center latitude and longitude were accurately recorded using a Garmin GPSMAP 62s handheld GPS.

[0070] 3. Acquisition and preprocessing of multi-source remote sensing images: Data acquisition: Based on the time window, two sets of cloudless Landsat 8 OLI images (imaged on April 5, 2015 and May 7, 2015, respectively, were downloaded from USGS as baseline images. At the same time, the MOD09GA product of the corresponding date was downloaded, as well as the MODIS image of April 12, 2015, which was in the core period of the monitoring window but had poor Landsat data quality (as the input image to be predicted).

[0071] Collaborative preprocessing: Radiometric calibration and atmospheric correction were performed on Landsat data. The MRT tool was used to extract short-wave infrared Band 6 (1628-1652nm) and Band 7 (2105-2155nm) from MODIS data. The data were converted from the Sinusoidal projection to the UTM-WGS84 projection. Finally, bilinear interpolation was applied to resample the data from 500m to 30m, achieving spatiotemporal registration of the geometric and radiometric layers of the multi-source data. In order to assess the impact of cultivated land share, the study area was divided into Area A (80% cultivated land share), Area B (70% cultivated land share), and Area C (85% cultivated land share) using the USDA Cultivated Land Data Layer (CDL).

[0072] 4. Run the ESTARFM spatiotemporal fusion algorithm to extrapolate high-resolution imagery. Input the preprocessed Landsat and MODIS data pairs from April 5 (t1) and May 7 (t3), as well as the MODIS data from April 12 (t2), into the ESTARFM algorithm model. The algorithm searches for pixels with similar ground cover categories to the central pixel within a sliding window, and then uses neighboring pixels in t... 1， The dynamic weight W is calculated based on the spectral similarity and spatial distance between t3 and t2. i Then, linear regression was performed to obtain the conversion coefficient V between MODIS and Landsat at each pixel. i， After the algorithm was run, the system successfully synthesized a high spatiotemporal resolution (30 meters) surface reflectance image of April 12, 2015 in a very short time (3 minutes and 3 seconds).

[0073] 5. Estimating Crop Straw Cover (CRC) and Conservation Tillage Classification: Using SWIR1 and SWIR2 band data from the fused high-resolution imagery of April 12th, the tillage index map for the entire region was calculated pixel-by-pixel using the formula NDTI=(SWIR1-SWIR2) / (SWIR1+SWIR2). Applying the validated quantitative estimation model CRC=754.71×NDTI+5.3817, the NDTI layer was transformed into a quantitative spatial distribution map of crop straw cover (CRC). Based on ArcGIS zoning statistical tools, classification criteria were set: CRC <15% for conventional tillage (CT), 15%-30% for reduced tillage (RT), and >30% for no-tillage (NT). The system automatically output the following statistics: In the study area of ​​this embodiment, CT accounted for 22%, RT for 11%, and NT for 67% (conservation tillage totaled as high as 78%).

[0074] Comparing the automatically extracted results with the statistical report released by the official authoritative agency CTIC that year not only confirms the historical fact that Indiana has a high percentage of conservation tillage (which was widely promoted in the early years), but also shows that the automatically classified values ​​are highly consistent with the officially published values ​​(NT64.4%), verifying the extremely high application reliability of the entire implementation plan.

[0075] 6. Control group, control group 1 (traditional mainstream algorithm STARFM): Using the same April 5th image as in Example 1 as the baseline, the STARFM formula was used to predict the NDTI on April 12th. The results showed that although the calculation only took 30 seconds, the generated NDTI deviated drastically from the true value. The R² of area C was only 0.49, and that of area B was even as low as 0.09, which was completely unusable for agricultural quantitative inversion. Using this erroneous prediction value to estimate CRC, its correlation R² with 115 ground measurement points plummeted to 0.46 (area C), 0.37 (area A), and even 0.03 (area B).

[0076] Control group 2 (complex nonlinear algorithm FSDAF): Using the same input set, the FSDAF algorithm, which is touted as being able to handle ground feature mutations, was introduced. Experiments showed that the algorithm was not suitable for the agricultural straw background. The highest R² of the predicted NDTI was only 0.48. Not only was the accuracy lower than the ESTARFM of this invention, but its complex domain decomposition mechanism also caused the single operation time to soar to a desperate 2 hours and 44 minutes.

[0077] Conclusion: This invention unequivocally confirms ESTARFM's dominant position in handling gradual changes in farmland surface features (R² reaches 0.69, R² for CRC inversion is as high as 0.76, and the root mean square error is reduced to a minimum). Furthermore, the data from the examples further confirm another major innovative finding of this invention—the dependence of spatial fusion accuracy on landscape fragmentation: regardless of the fusion algorithm used, the prediction accuracy in areas C (85% arable land), A (80% arable land), and B (70% arable land) always strictly follows the order C>A>B. This clearly indicates that in practical engineering applications, the more contiguous and concentrated the arable land in a region (the higher the proportion), the closer the accuracy of the straw coverage remote sensing mapping output by the method provided by this invention will be to the absolute true value. This scientific conclusion provides a solid theoretical foundation and technical feasibility endorsement for the routine monitoring of large-scale farmland conservation tillage using satellite remote sensing.

[0078] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints, characterized in that, Includes the following steps: Step S1: Establishing the golden time window for remote sensing monitoring based on surface temperature constraints. Specifically, this involves acquiring high temporal resolution surface temperature remote sensing products for the target study area, setting the minimum physiological ground temperature threshold for crop germination, calculating the date when the temperature threshold is reached, and acquiring the time window. Step S2, acquisition of multi-source satellite remote sensing data and ground-based measured auxiliary data, specifically high spatial resolution image acquisition, high temporal resolution image acquisition, and ground-based measured data acquisition; Step S3, multi-source remote sensing image consistency preprocessing, specifically radiometric calibration, projection transformation, and pixel unification; Step S4, high spatiotemporal resolution image reconstruction based on the ESTARFM algorithm, specifically involves input data configuration, finding similar pixels, calculating weight functions, determining spectral conversion coefficients, and deducing high-resolution pixel values. Step S5: Calculation of characteristic index and CRC inversion of regional straw coverage; Step S6: Classification and mapping of conservation tillage patterns.

2. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 1, characterized in that: Step S1 acquires high temporal resolution land surface temperature remote sensing products for the target study area using the VNP21A1N nighttime land surface temperature product from the VIIRS satellite. Its spatial resolution is 1 km, temporal resolution is 1 day, and the transit time is consistent with the local daily minimum temperature occurrence time. The date of reaching the temperature threshold is calculated on a pixel-by-pixel basis in the cloud computing platform, calculating the date when the spring daily minimum ground temperature first stably reaches or exceeds 10°C.

3. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 2, characterized in that: The acquisition of the time window in step S1 specifically involves dynamically determining the optimal remote sensing image acquisition time window for each pixel in the study area by taking into account the revisit characteristics of the selected high spatial resolution satellite, starting from the date when 10℃ is reached, and extending forward by one seedling cycle.

4. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 3, characterized in that: The high spatial resolution image acquisition step S2 specifically involves acquiring Landsat 8 OLI L1TP level images of the target area within a defined time window and before and after it. Two sets of images for reference dates (t1, t3) need to be acquired to form the basis for building the fusion model. At the same time, the target prediction date (t2) is recorded. The high temporal resolution image acquisition specifically involves acquiring MODIS surface reflectance products provided by existing technologies, and it is necessary to acquire images synchronized with the Landsat 8 reference dates (t1, t3) and images for the target prediction date (t2).

5. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 4, characterized in that: The specific steps for obtaining ground measurement data in step S2 are as follows: within a defined time window, representative sampling points are set up in the target area, and the standardized transect method is used for measurement: a fixed-length transect rope marked with equidistant marks is used and stretched along the diagonal of the quadrat. The number of times the transect rope marks intersect with the crop straw on the ground is counted, and the ratio of the number of intersecting marks to the total number of marks is calculated as the measured straw coverage CRC of the quadrat. Based on a spatial resolution of 30 meters, the average value is obtained by using a five-point cross-sampling method within a 30m×30m pixel range, and the center coordinates of the sampling points are recorded using GPS.

6. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 5, characterized in that: In step S3, radiometric calibration is performed on the acquired Landsat 8 OLI image to convert it to atmospheric top reflectance, followed by atmospheric correction to obtain true surface reflectance data. For projection transformation, a batch processing tool is used to perform band extraction and reprojection operations on the MOD09GA data, converting its projection coordinate system to the UTM-WGS84 coordinate system, which is completely consistent with the Landsat image. For pixel-level shortwave infrared bands corresponding to MODIS, bilinear interpolation is used to resample the 500-meter spatial resolution MODIS image to a 30-meter spatial resolution, ensuring complete sub-pixel alignment with the Landsat 8 image in terms of spatial range, pixel size, number of rows and columns, and projection type.

7. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 6, characterized in that: Step S4 involves configuring the input data: inputting Landsat 8 surface reflectance images and resampled MODIS surface reflectance images for two reference dates (t1, t3), and a MODIS image for the predicted date (t2); finding similar pixels: within a set sliding window (w), using the cross value of the Landsat 8 reflectance data from the two reference dates, identifying and determining similar pixels with a similar land cover category (c) to the center pixel; calculating the weight function (w). i ): Determine the number N of similar pixels within the sliding window, and assign weights to each similar pixel by comprehensively considering the spatial distance, spectral differences, and temporal evolution characteristics between similar pixels and the center pixel.

8. The method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints as described in claim 7, characterized in that: Step S4 determines the spectral conversion coefficient (V) i ): A linear regression analysis was performed on MODIS data from the baseline date and Landsat 8 data from the same period to determine the conversion coefficient (V) reflecting the systematic spectral response differences between the two sensors when observing the same ground feature. i High-resolution pixel value extrapolation: Based on the formula, the high spatial resolution surface reflectance prediction value of the central pixel on the predicted date (t2) is calculated. The mathematical expression is: Where L represents the derived high-resolution reflectance, and M represents the low-resolution reflectance; x i y i t represents the pixel coordinates; p t2 is the predicted date, t0 is the base date; b is the specified band; w i V represents the comprehensive weight of the i-th similar pixel; i is the spectral conversion coefficient; N is the total number of similar pixels within the sliding window.

9. A method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints, as described in claim 8, is characterized in that: Step S5 utilizes the high-resolution surface reflectance data for the predicted date (t2) generated in step S4 to extract the first and second shortwave infrared bands. Based on the characteristic absorption valley principle of lignin and cellulose in crop straw around 2100nm, the Normalized Differential Tillage Index (NDT) is calculated using the following formula: Next, the calculated pixel-by-pixel NDTI data is substituted into the pre-constructed or preferred straw coverage linear estimation model of this patent. The model formula is as follows: CRC=754.71NDTI+5.3817 By substituting each pixel into the calculation, a high spatial resolution spatial distribution map of straw coverage in the entire study area on the key prediction date is generated.

10. A method for estimating straw cover and mapping conservation tillage based on spatiotemporal fusion of multi-source remote sensing images and ground temperature constraints, as described in claim 9, is characterized in that: Step S6 uses GIS spatial analysis tools to reclassify the continuous CRC numerical map obtained in step S5 into discrete farming pattern type maps, and calculates the area proportion of each farming pattern in the study area, ultimately generating a regional conservation tillage spatial distribution map.

Images