Soil moisture content inversion and determination method based on remote sensing image

Abstract

The application discloses a soil moisture content inversion and determination method based on remote sensing images, and particularly relates to the technical field of surveying and remote sensing measurement, and is used for solving the problems of time and space inconsistency and system deviation caused by spring film and salt shell superposition and tide and the problem of untraceable results of the prior art. Through unified grid and unified radiation and time caliber, three types of interference of film, salt shell and phase are first refined as conditional quantities, equivalent bare land is completed in the order of "phase normalization, film correction, salt shell correction", and then the volume moisture content is calculated in a monotonicity controlled model, so that the stable monotonic relationship between pixel observation and moisture content is restored, and the system error caused by non-bare land and imaging phase difference is resolved from the source. After pre-season and mid-season verification, the overall deviation is significantly reduced, the moisture content error is kept at a low level, the determination jitter caused by rain after transient state and high tide window is effectively inhibited by the stabilization strategy, and the engineering precision and consistency requirements of coastal spring film and salt return period can be met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of surveying and remote sensing measurement, in particular to a soil moisture inversion and determination method based on remote sensing images. BACKGROUND

[0002] In the prior art, soil moisture inversion based on remote sensing mainly relies on optical index, microwave backscatter empirical relationship, radiation transmission approximation or machine learning regression. The common practice is to use near-infrared and short-wave infrared which are sensitive to water, radar polarization and incident angle which are sensitive to surface water content as the basis, and to calculate the water content under the assumption of "bare ground" after cloud, shadow and vegetation masking. These methods can obtain acceptable accuracy in inland agricultural areas, where the proportion of mulching is low and the surface conditions are relatively stable. However, in saline-alkali areas, the proportion of mulching is high in spring, black and white mulching coexist, salt crust is high and stable in time series, and the observation phase changes caused by tides and shallow groundwater. The "water content baseline" of the same pixel at different imaging times is not consistent.

[0003] The prior art usually treats mulching and salt crust as "excluded items" or coarse-grained obstructions, or only makes overall correction in statistics, and rarely quantifies them as pixel-level component conditions. At the same time, the coupling effect of tidal level and distance from the coast is often ignored or treated as a uniform time shift, which is difficult to reflect the differences of different soil textures and alongshore zoning. As a result, the volumetric water content under non-bare ground conditions is prone to systematic deviation and phase jitter, especially in the case of rain after transient, high tide window, etc. If these pixels are excluded as a whole, it will also cause spatial gaps and sampling bias, affecting the consistency evaluation at the plot scale.

[0004] From the perspective of engineering implementation, the existing scheme still has deficiencies in the unification of multi-source heterogeneous data coordinates, time, radiation caliber, traceable criteria of mulching and salt crust, regionalized parameters of tidal phase, and quality grading and value tracing of pixel-level results. The typical process is mainly based on single scene or a small number of phase solutions, and the processing rules for thin clouds and strip missing data are not detailed. The influence of boundary artifacts and incident angle differences is easily amplified. The threshold is mostly set by experience, and lacks version management and sample back-checking loop for seasonal and plot differences. The mapping of sample-pixel, consistency of drying and electromagnetic method, and minimum sample size threshold are not uniform, which limits the stability of scale anchoring. On the output level, most of them are displayed in raster or hierarchical graphs, lacking of uncertainty interval, quality label, reason code and certification record at pixel level, and lacking of chain tracing of result version, parameter table version and source data. In the application scenarios of strong interference and strong phase difference in saline-alkali areas, these characteristics of existing technology and data products directly limit the application of soil moisture in spring mulching period and salt return period. SUMMARY

[0005] (I) Technical problems solved

[0006] In view of the deficiencies of the prior art, the present application provides a soil moisture condition inversion and determination method based on remote sensing images, which quantifies three types of strong interference, i.e., mulch, salt crust and tidal phase, as conditional quantities, processes them according to the link of "phase normalization→ equivalent bare land→ monotonicity-controlled determination", and uses sample anchor and certificate to solidify the results in a closed loop, thereby solving the problems mentioned in the background art.

[0007] (II) Technical solutions

[0008] To achieve the above object, the present application provides the following technical solutions:

[0009] The soil moisture condition inversion and determination method based on remote sensing images comprises:

[0010] S1, acquiring multi-source remote sensing images and tidal meteorological condition data, completing geometric registration and radiation unification, and establishing a unified grid observation reference and pixel feature set;

[0011] S2, determining the mulch component of the unified image, combining the spectral morphology and microwave radar scattering response to generate a pixel mulch component mask and record it as a conditional quantity;

[0012] S3, determining the salt crust component of the unified image, generating a pixel salt crust component mask according to the time series brightness stability and microwave radar polarization difference, and recording it as a conditional quantity;

[0013] S4, estimating the water content baseline offset caused by the tidal action according to the coastal distance, soil texture partition and observation time, completing phase normalization and outputting the phase conditional quantity;

[0014] S5, inputting the pixel features, mulch component, salt crust component and phase conditional quantity into the equivalent bare land conversion to obtain the equivalent bare land observation, and then solving the shallow soil moisture condition in the monotonicity-controlled determination model;

[0015] S6, using sample measurement to anchor the scale and bias, outputting the pixel-level moisture condition results and quality identification, forming the measurement record and value traceability chain.

[0016] In a preferred embodiment, S1 comprises:

[0017] Under the unified national coordinate reference, unified projection, unified time caliber and unified tidal level reference, the multi-source optical and microwave images are sequentially subjected to geometric registration, radiation unification, cloud and snow shadow disposal, unified resampling and element assembly;

[0018] The pixel alignment is completed by using ground control points, and the radiation unification is performed under the constraint of maintaining consistent dimensions;

[0019] On the same day, the image with the median incident angle is set as the main scene, and when there is a conflict, the image with the incident angle close to the median and lower cloud cover is determined as the main scene;

[0020] The missing strip marks are processed and output as boundary suspected artifact identification.

[0021] In a preferred embodiment, a pixel feature vector is constructed with grid number and timestamp as the primary key, including optical multi-channel reflectivity, radar dual-polarized scattering, solar elevation angle, incident angle, near-surface wind speed, ground temperature, tidal level, land parcel boundary, and soil texture, and cloud and snow shadow, strip completion, degradation, and boundary suspected artifact identification are attached as records;

[0022] A hierarchical quality label and parameter version management mechanism is established, and when the pixel feature coverage is insufficient, the output is suspended and added to the queue for supplement, and a standardized interface is provided with grid number and timestamp as the key to output to the coating component, salt crust component measurement unit, and phase normalization module.

[0023] In a preferred embodiment, S2 includes:

[0024] Based on a unified grid and a unified radiation time caliber, the following are sequentially performed: receiving registered and consistent optical multi-channel reflectivity and polarized radar scattering and incident angle, time, and land parcel boundary; completing brightness clipping, edge fidelity filtering, and intra-parcel quantile standardization; extracting coating candidates based on high-brightness low-chroma prior; performing radar evidence verification based on adjacent bare ground scattering differences under the consistent incident angle reference; correcting in combination with near-time sequence consistency and spatial connectivity, while avoiding boundary suspected artifacts, water bodies, and urban construction high albedo interference, and simultaneously calling salt crust candidate exclusion; mapping multiple evidences into pixel coating components according to a segmented lookup table rule and generating a mask, and writing into a conditional quantity dataset along with quality labels and reason identifications for downstream calling.

[0025] In a preferred embodiment, S3 includes:

[0026] Salt crust component measurement includes: under the unified grid, unified radiation, and unified time caliber, normalizing across track and incident angle, and calling optical multi-temporal and dual-polarized radar data;

[0027] Sequentially performing time sequence brightness stability candidate, polarization difference verification, coating component exclusion, connectivity repair, and boundary consistency test, outputting pixel salt crust components and generating a mask according to a multi-evidence lookup table;

[0028] The results, along with quality labels, reason codes, and parameter versions, are recorded as conditional quantities, and are called for phase normalization and equivalent bare ground.

[0029] In a preferred embodiment, S4 includes:

[0030] Phase normalization includes: obtaining the shortest distance from the pixel to the shoreline, soil texture zoning, tidal level and fluctuation state, observation time and near-ground wind temperature under unified coordinates, time and radiation aperture;

[0031] Retrieving phase lag and amplitude parameters according to "distance band x texture", combining tidal level extreme window and wind temperature correction to determine water baseline offset;

[0032] When the difference between adjacent phases is too large, implement adjacent and spatial stabilization, and limit the correction strength in the scenarios of bays, estuaries and hydrological discontinuity;

[0033] Output phase condition quantities including distance band and texture encoding, tidal level station and aperture version, lag and amplitude level, fluctuation state and stabilization identification for equivalent bare soil call.

[0034] In a preferred embodiment, S5 includes:

[0035] On a unified grid and unified radiation, time aperture, receive pixel features, mulch components, salt crust components and phase condition quantities;

[0036] Perform aperture and version consistency check on input;

[0037] Implement equivalent bare soil conversion according to fixed order "phase normalization-mulch correction-salt crust correction" to obtain equivalent bare soil observation;

[0038] In a monotonicity-controlled measurement model, there is only one turning point, and the monotonic direction is set for each observation channel. Under the above constraints, the volume water content of the pixel is calculated;

[0039] And complete scale and bias correction with sample anchor, output version, quality and reason identification synchronously.

[0040] In a preferred embodiment, the equivalent bare soil conversion uses a segmented lookup table with texture zoning, coastal distance band, incident angle group, mulch sub-range, salt crust sub-range and phase classification as keys, and only applies bias and scale correction to disturbed channels;

[0041] When the input components are in high interference, phase and version inconsistency, and the output difference caused by the exchange of correction order exceeds the preset threshold, automatically degrade to interval estimation and record the reason;

[0042] When monotonicity is unstable, tighten the feasible interval and keep the model direction from reversing.

[0043] In a preferred embodiment, S6 includes:

[0044] Collect samples under unified coordinates and time aperture and register with the grid, establish sample-pixel mapping according to standard sample and neighborhood weighting, and the center pixel has the highest weight;

[0045] Electromagnetic method is replaced after calibration and comparison based on drying method;

[0046] The sample plot is excluded from scale anchoring and bias correction when any of the following conditions is met: the difference in satellite observation time exceeds the predetermined limit and the review fails; the plane positioning deviation exceeds the predetermined limit and cannot be corrected; it is determined to be abnormal through robust statistics and is still abnormal after review;

[0047] Scale anchoring is completed based on sample plot statistics, and bias is refined according to soil texture and management plots;

[0048] The anchor version, uncertainty, and quality identification are attached to the pixel result, and the certificate and traceability record containing machine-readable fields and hash check are generated, and the anchoring information is returned.

[0049] Compared with the prior art, the present application has the following beneficial effects:

[0050] 1. In the present application, the film component, salt shell component and phase condition component are extracted on the ten-meter unified grid and unified radiation and time caliber, and the equivalent bare land transformation is performed in the fixed order of "phase normalization→film correction→salt shell correction", and then the volume moisture content is calculated in the determination model with controlled monotonicity, so that the pixel observation and moisture content restore stable and monotonic relationship, and the system error caused by non-bare land and imaging phase difference is resolved from the source; through pre-season and mid-season verification, the overall deviation after phase normalization is reduced by not less than 30%, the root mean square error of volume moisture content is stable within 4%, the determination jitter caused by rain after transient and high tide window is significantly suppressed by the stabilization strategy, and the engineering application precision and consistency requirements of zero to twenty kilometers coastal belt in spring film and salt return period are met.

[0051] 2. The present application forms a quality and traceability closed loop of coverage acquisition-component determination-phase normalization-equivalent bare land transformation-determination calculation-sample plot anchoring-certificate, realizes the source, parameter, use boundary and uncertainty of pixel-level results through quality label, versioned lookup table and anchoring, certificate JSON key set and SHA-256 hash and prev_hash chain tracing, guarantees continuous output and caliber consistency in data gap and abnormal scene through degradation, interpolation, playback and version rollback, meets batch business operation in end-to-end time delay and concurrency index, and supports irrigation scheduling, early warning and supervision certification through standardized interface, and significantly improves the credibility, retestability and management availability of determination results. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 It is a flowchart of the soil moisture content inversion and determination method based on remote sensing image of the present application. DETAILED DESCRIPTION

[0053] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0054] Embodiments Figure 1 A flowchart of a soil moisture inversion and determination method based on remote sensing images is given, the soil moisture inversion and determination method based on remote sensing images comprises the following steps.

[0055] S1, acquiring multi-source remote sensing images and tidal meteorological condition data, completing geometric registration and radiation unification, establishing a unified grid observation reference and pixel feature set;

[0056] S2, determining the mulching component of the unified image, combining the spectral morphology and microwave radar scattering response, generating a pixel mulching component mask and recording it as a condition quantity;

[0057] S3, determining the salt crust component of the unified image, generating a pixel salt crust component mask according to the time sequence brightness stability and microwave radar polarization difference, and recording it as a condition quantity;

[0058] S4, estimating the water content baseline offset caused by the tide according to the distance from the coast, the soil texture partition and the observation time, completing the phase normalization and outputting the phase condition quantity;

[0059] S5, inputting the pixel features and the mulching component, the salt crust component and the phase condition quantity into the equivalent bare ground conversion to obtain the equivalent bare ground observation, and then solving the shallow soil moisture in the determination model controlled by monotonicity;

[0060] S6, using the quadrat measurement to anchor the scale and bias, outputting the pixel-level moisture result and quality identification, forming the measurement record and the value traceability chain.

[0061] The technical connection relationship and implementation logic of the six steps are as follows:

[0062] Firstly, align multi-source optical and radar images and tidal meteorological conditions to the same coordinate, time and radiation caliber at ten-meter resolution, generate a pixel feature set with quality label and version number as the only input caliber for all subsequent judgments; then identify the mulch component on this caliber, first give the candidate with optical brightness and chrominance, then use radar polarization and incidence angle consistency as evidence to enhance, output pixel mulch mask and component percentage and record reason code; next, determine the salt crust continuous piece with one-month time series brightness stability and radar polarization difference, form the salt crust mask and component percentage, and mutually exclude with the mulch result while retaining connectivity and artifact identification; after having both mulch and salt crust conditions, estimate the water content baseline offset according to the distance from the pixel to the coast, the soil texture and the observation time, combined with tidal level, wind speed and air temperature, and give the phase lag and amplitude level and complete the phase normalization to form the phase condition that can be indexed; then send the pixel features together with the three conditions of mulch, salt crust and phase into the equivalent bare ground transformation, in fixed order first phase normalization, then mulch correction, and then salt crust correction, only to the affected channels to apply segmented bias and scale correction, get the equivalent bare ground observation, and then solve the volume water content from zero to five centimeters in the monotonicity controlled measurement model and output the confidence interval, quality identification and reason code; finally, anchor the scale and bias with sample measurement and give the batch anchor version, the uncertainty interval and the certification traceability record are output with the result, the anchor summary is returned to the upstream to update the lookup table and parameter version, and the abnormal and insufficient are triggered to degrade, interpolate or playback review, so as to form the acquisition, component measurement, phase normalization, equivalent bare ground, measurement and solving and value traceability closed loop, to ensure continuous output, caliber consistency and traceability and verifiability in the same range.

[0063] S1, acquire multi-source remote sensing images and tidal meteorological condition data, complete geometric registration and radiation consistency, establish a unified grid observation reference and pixel feature set, the specific implementation is:

[0064] For the coastal zero to twenty kilometer belt of saline-alkali farmland area before and after spring germination, a ten-meter unified grid is used to establish an observation reference and pixel feature set, and observations of different satellites, different orbits and different imaging times are unified to the same range; the spatial reference uses the national two-thousand-year coordinate system, projected as Gauss-Kruger three-degree zone (zone number recorded according to the central meridian of the image), the time is unified to Beijing time and the original collection time zone field is retained for tracing, and the tidal level height is calculated according to the national height datum and the station number and datum conversion information is recorded.

[0065] The input elements include optical surface reflectance (dimensionless, ESA and NBSI, revisit 3-5 days, time window 1 day before and after, allowable deviation no more than 0.005%) and microwave radar backscatter coefficient (dB, synchronous revisit frequency and time window, allowable deviation no more than 0.5 dB), and are accompanied by solar elevation and incidence angles (degrees), near-surface wind speed (m / s), surface temperature (degrees Celsius), tidal height (m, time resolution no less than 1 hour, allowable deviation no more than 0.1 m), observation time (minute-level timestamp), land parcel boundaries and soil region classification (annual update classification code); all data are unified in terms of coordinates and time zones before being stored in the database, and the source number and version number are recorded; cross-source same-day pairs are aligned according to the minute-level timestamp, with a tolerance of no more than 10 minutes; out-of-limit grids are not fused into the "to be supplemented" queue in the current batch, and are supplemented by interpolation of the latest time phase within the next one to two revisit periods, with the interpolation time difference and interpolation identifier being marked.

[0066] The processing chain is executed in the order of "geometric registration - radiometric normalization - cloud and snow shadow removal - unified resampling - element assembly": first, the images are aligned to a 10-meter grid using ground control points, with a plane error of no more than 1 meter; before registration, the microwave strip is first corrected for terrain correlation to avoid geometric stretching; then, radiometric normalization is performed without changing the dimension, with the optical being checked by the reflection baseline of the scene stable target and the radar being checked by the stable echo target, only allowing small linear correction of bias and scale, and being accepted by a quantitative threshold: on stable targets and representative land parcels, the absolute value of the mean difference before and after optical channel normalization is no more than 0.005%, and the dispersion from the 5th to 95th percentile is reduced by no less than 20% compared to before normalization; the absolute value of the mean difference before and after radar dual polarization normalization is no more than 0.5 dB, and the standard deviation is no more than 80% of that before normalization, and scenes that do not meet the standard are not entered into element assembly and the reason code is recorded in the log.

[0067] After that, clouds, snow and shadows are identified and processed, the fully shaded area is set to an invalid value, and the thin cloud and semi-transparent edge are locally interpolated in a three-by-three grid and assigned a low trust identifier. If the single-scene cloud cover exceeds 70%, the entire scene is excluded from the current processing. All channels are resampled to a ten-meter grid, and edge-preserving interpolation is used (window size not exceeding three grid cells). At the same time, boundary consistency is checked: using a three-by-three grid as a window, when the boundary of the same type of feature crosses the grid with a brightness difference of more than 1% reflectance or a radar scattering difference of more than 0.8 decibels, the pixel is marked as a "boundary suspected artifact" to avoid it in the downstream connectivity determination. For strip missing, length-based processing is used: if the continuous length is not more than 1 km, it is filled with linear interpolation of the same caliber from the adjacent track and a trace is left. If the length exceeds 1 km, it is marked as invalid. When multiple scenes exist on the same day, the main image with the incident angle in the middle of the scene distribution is selected, and the rest is included in the standby queue. If the main and standby results are different in individual grids, the one with the incident angle closer to the middle and the cloud cover lower is output, and a "conflict resolved" identifier is marked. The other result is kept as a side note.

[0068] Threshold and weight are initially calibrated with not less than thirty sample plots and not less than ten fixed-angle reflective targets before the season. After the season, they are adjusted adaptively and slightly based on sample plot deviation statistics and stable target residual on a weekly basis. The single adjustment amplitude does not exceed 20% of the original value. All parameters generate a version number and are written with the batch.

[0069] The pixel feature vector takes "grid number + timestamp" as the primary key, assembles optical multi-channel reflectivity, radar dual-polarization scattering, solar elevation angle and incident angle, wind speed, ground temperature, tidal height, land boundary code, soil quality regional code, and attaches cloud and snow shadow identifier, strip filling identifier, degradation identifier, quality label, source number and parameter version number. The quality label is divided into high, medium and low levels. The high level requires that the registration error is not more than 0.5 meters, the cloud and snow interpolation ratio is not more than 5%, and there is no strip filling identifier. The medium level requires that the registration error is not more than 1 meter and the interpolation ratio is not more than 15% or there is a single strip filling with a length of not more than 1 km. The rest is low level and is explicitly prompted when returned in the interface. When the single-grid feature coverage rate is less than 80%, the grid output is suspended and put into the "to be filled" queue.

[0070] The scheduling module drives the entire link with a task queue. External data enters the module through the access gateway, and after processing, the state and exception code are returned. The caliber version of the current batch of tidal height, wind speed and air temperature is registered to the conditional quantity manager to ensure phase normalization and equivalent bare land cross-batch traceability. The time sequence and resource target are not more than 60 minutes per 100 square kilometers end-to-end, not less than 20 tasks are executed concurrently, failure retry is not more than three times and interval is not less than five minutes. When the queue pressure rises, it is divided into space slices and time slices in batches, and priority is given to the main image first and the standby image later.

[0071] The abnormality and fault-tolerant strategy includes: when the registration error exceeds one meter, degrade the output of thirty-meter grid and record the reason for downstream filtering; when the tidal data is short, replace it with the median value of the same phase time statistics for nearly three days, and reduce the reliability, while leaving a trace in the conditional quantity manager; when the cloud cover is too high, give up this phase to avoid false brightness misleading subsequent criteria; the edge artifacts introduced by resampling and strip filling are uniformly marked to avoid the continuity of the film and salt crust for the film and salt crust continuity judgment. Quality verification uses geometric and radiation double-layer checking: geometric layer verifies with independent ground control points, average plane error is not more than 0.5 meters, and maximum error is controlled; the radiation layer checks the channel difference before and after the consistency, and the mean deviation and dispersion must meet the above hard threshold; the total number of two-layer checking samples is not less than thirty, and the calibration report and parameter version are archived and associated with batch number for playback audit.

[0072] The results are stored in the form of regional data sets, the grid uses block lossless compression GeoTIFF for long-term storage, and the metadata and index use columnar storage; Provide query interface and subscription incremental push with "region number + grid number + timestamp" as the primary key, the returned content includes unit, precision, source number, version number and quality label; Standardized output is sent to film component and salt crust component measurement unit at the same time, and provides tide and time aperture to phase normalization module.

[0073] The minimum example is: input a scene of optical image covering about 200 square kilometers, two scenes of radar image of the same day and different orbits, and tidal weather records of the day, complete registration, consistency, cloud and snow shadow processing, resampling, strip processing and element assembly according to the above link, output ten-meter grid feature set with quality and version label and enter the database; When high-resolution optical image is seriously missing, establish a reference based on radar, fill in the last period of stable statistics for optical channels and mark low reliability, and continue to flow to ensure business continuity.

[0074] The running log records the input list, parameter version and threshold weight, abnormality and alarm, degradation and interpolation position, time consumption statistics and concurrent strategy, and data source number; The original image and external condition quantity are saved for not less than one complete growing season, the intermediate product is saved for not less than six months, the pixel feature set is saved for a long time, and not less than one percent of the grid samples are extracted monthly for playback verification and archiving report, so that the heterogeneous and time and space inconsistent multi-source observation is regularized into pixel feature set with same dimension, same resolution and same time aperture, which supports the downstream to stably carry out film and salt crust component identification and tidal phase normalization in the same measurement range, and meets the requirements of open, full, verifiable and engineering available.

[0075] S2, film component measurement is carried out on the unified image, combining spectral morphology and microwave radar scattering response, generating pixel film component mask and recording as conditional quantity, the specific implementation is:

[0076] On the basis of the establishment of a ten-meter unified grid and time caliber in the previous unit, the unified image is executed to determine the film component, and the existence and proportion of plastic mulch in the pixel are quantified and provided as a condition for phase normalization and equivalent bare land calling. Input remains consistent with the upstream, including blue, green, red, near-infrared, and short-wave infrared five-channel ground reflectivity (dimensionless, with an allowable deviation of ≤0.5%) and radar dual-polarization backscatter coefficient (decibel, with an allowable deviation of ≤0.5 decibel), while accessing the incident angle (degree), observation time (minute-level timestamp), and land boundary vector and inheriting the quality label and source version. To unify the data caliber, the reference center wavelengths of the optical five channels are approximately 490 nanometers, 560 nanometers, 660 nanometers, 840 nanometers, and 1610 nanometers, respectively, and the reflectivity is stored and recorded in percentage scaling with a scaling factor; Radar is processed by incident angle grouping, with priority given to the 30°-45° interval, and beyond the interval is corrected and marked according to the median incident angle in the adjacent phase.

[0077] The film is optically bright and low in color, and in radar it shows a response characteristic of reduced roughness and suppressed cross-polarization energy. Preprocessing first performs brightness range clipping on the unified grid, removing extreme noise outside the lower and upper 1% quantiles of channel statistics; then a three-by-three detail-preserving filter is applied to the edge area to suppress isolated high frequencies while preserving the boundary position; then the multi-channel reflectivity within the same land is standardized by quantile to weaken the disturbance of inter-land albedo differences on the threshold, and all processing actions are written into the metadata. The execution sequence is "optical prior candidate - radar evidence secondary screening - temporal consistency review - spatial connectivity correction - table lookup segmentation assignment", among which the quantitative caliber of optical candidate is: the standardized brightness within the land is located in the upper quartile to the upper tenth of the scene, and the normalized color is ≤0.25, which is identified as a strong candidate; only reaching the upper quartile and the color is between 0.25-0.30 is identified as a weak candidate, which is upgraded only when there is strong radar evidence.

[0078] Radar secondary screening uses adjacent bare land with the same incident angle as the comparison reference, and meets the same polarization difference ≥1.0 decibel and cross-polarization suppression ≥1.0 decibel, or the same and cross-polarization difference value ≥4.0 decibel, and at least 5 pixels are continuous within a three-by-three window, which is recorded as strong evidence; When there are multiple time phases overlapping on the same day, the results within 6 hours of the main image time are preferred, and the rest are used as collateral evidence for consistency check. To prevent confusion with high-brightness bodies, the following are excluded: near-infrared and short-wave infrared combination showing water body characteristics and area proportion >20% temporarily suspend mulch output; city construction high albedo mask hit; and overlapping area >30% with salt crust candidate, first confirmed by the salt crust unit and then restored to the estimated value of this unit; At the same time, reference the "boundary suspected artifact" identifier from the previous unit, and avoid during connectivity correction. For black film, if the visible light is not high-brightness, the short-wave infrared is significantly low-albedo, and the same polarization is stable and higher than the cross-polarization, then enter the conservative channel and lower the confidence level.

[0079] Spatial connectivity correction performs majority vote in 3x3 window to eliminate isolated pixels, and fills holes smaller than one pixel to prevent fragmentation from affecting component estimation. Component assignment adopts look-up table and piecewise rules: optical brightness is divided into four bins, chrominance into two bins, co-polarized difference into two bins, cross-polarized suppression into two bins, mapped to the interval of coating component, interval step 5%, value 0-100%, result rounded to the nearest step and clipped within 0-100%; when both optical and radar strong evidence are satisfied and connectivity passes, the upper limit of the interval is moved up by one step; only single evidence is established or there is thin cloud interpolation, incidence angle correction, the interval is moved down by one step and the reason code is written. The initial weight is optical: radar = 6:4, pre-season calibration is completed at least 30 sample points once, and after the season, the deviation and residual error of the stable target are checked according to the sample points to adjust adaptively, the single adjustment amplitude is ≤20%, and a version number is generated each time the adjustment is made.

[0080] The trigger and stop conditions are to start demixing when the candidate ratio is ≥10%, and to stop the grid processing when the results of the last two rounds of screening are consistent; when the candidate is insufficient, the optical high-brightness collateral evidence is retained and entered into the "to be supplemented" list. The output includes two types of forms: pixel-level coating component percentage grid and binary coating mask grid, as well as block-level statistical report (block number, coating area, component mean and confidence interval); pixel records are stored as integer percentages, with a step size of 1 and a missing value of -1, the mask is encoded as 0 / 1, and the grid number, timestamp, confidence level, reason code, weight version, input quality label, source data batch number, and incidence angle grouping number are returned with the record.

[0081] The quality label is divided into three levels: high level requires strong evidence from both optical and radar, connectivity passes, registration error ≤0.5 meters, cloud snow interpolation ratio ≤5%, and no strip filling; medium level allows registration error ≤1 meter, interpolation ratio ≤15% or single segment length ≤1000 meters of strip filling and single strong evidence plus weak collateral evidence; the rest are classified as low level and explicitly prompted in the interface; when the cloud thin interpolation occupies >10% in the block or the incidence angle correction amplitude exceeds the four quantile width of the full view incidence angle distribution, the quality label cannot be higher than the medium level. The processing performance index is ≤15 minutes per 100 square kilometers end-to-end, ≥10 concurrent tasks, ≤2 failed retries and ≥5 minutes interval, when the queue pressure rises, slice by region, priority to the main image coverage area; the output is pushed to the salt crust component measurement unit as a reference for exclusion, and the condition quantity dataset is written for direct indexing of equivalent bare land and phase normalization.

[0082] Quality verification adopts stratified sampling: 0-20 km from the sea is divided into three zones, ≥10 sample plots in each zone; Each sample plot covers at least two of the three categories: <1m, 1-2m, >2m, and both small and large plots are covered; The evaluation index is absolute value of component deviation ≤10%, recall rate ≥80%, precision rate ≥75%, area intersection ratio ≥50%, verification report and parameter version are archived with batch; The running log records the input list, threshold and weight version, abnormality and alarm, degradation and stop production position to be repaired, time consumption statistics and concurrent strategy, and data source number, which is saved in the same library as the results, to ensure that the coverage component condition quantity source is traceable, the processing is reproducible, and the quality is checkable, and it is naturally connected with the upstream grid reference and downstream salt shell component and phase normalization on data and control flow.

[0083] S3, salt shell component measurement of unified image, according to the time sequence brightness stability and microwave radar polarization difference, generate pixel salt shell component mask and record as condition quantity, the specific implementation is:

[0084] On the basis of unified grid and caliber, input the multi-temporal optical ground reflectivity, dual-polarized microwave radar backscatter, observation time, precipitation record, plot boundary and soil quality zoning of nearly one month continuous sequence, measure the salt shell component of each pixel and output the salt shell component mask and condition quantity; The input data maintains the same coordinate, time and radiation caliber as S1, the radar backscatter is first made consistent in orbit, so that the comparability between the same polarization channels is controlled within 0.5dB, if the difference in incident angle exceeds 5 degrees, the incident angle is first normalized according to the lookup table correction rule, and then the polarization criterion is executed; The time sequence "brightness" uses the joint brightness caliber of red, near-infrared and short-wave infrared, the default weight of each is one third, and it can be adjusted slightly within the range of plus and minus two tenths of each week, the global range of joint brightness is given by the quantile interval statistics of the current batch of cloud-free valid pixels, which is used for the stability criterion of "fluctuation amplitude less than 10% of the global brightness range"; The minimum lower limit of valid phase within a month is optical not less than four times and radar not less than two times, if any does not meet the lower limit, only the interval estimate is output and the confidence is reduced to medium or low, and "insufficient time phase" is written in the reason code.

[0085] The processing is executed in the order of "time sequence stability candidate → polarization evidence verification → coverage exclusion → connectivity and hole repair → component quantization → quality and version labeling": First, construct a one-month optical time sequence brightness statistics, calculate the monthly brightness median value and dispersion for each pixel, retain the pixels with brightness median value in the upper quantile of the scene (default upper quartile to upper decile interval) and monthly fluctuation amplitude less than 10% of the global brightness range as "stable high brightness candidate", and at the same time, according to the precipitation record, look back at the short-term high brightness one to two days after the rain and label it as "rain after temporary" which cannot enter the candidate set.

[0086] Secondly, polarization verification is performed in the radar domain. A three-by-three grid is used as a window, and compared with the adjacent bare ground baseline. If the co-polarization (VV) is generally enhanced (the default increase is not less than one decibel) in one month, the cross-polarization (VH) is generally weakened (the default decrease is not less than zero point eight decibels), and at least three time phases are satisfied at the same time, it is determined that the polarization evidence is sufficient. If only part of the time phase is satisfied, it is recorded as "critical" and the weight is reduced; thirdly, the mulch component output by S2 is called as a rejection condition. When the mulch component is higher than 20%, the "stable highlight" of the pixel is explained to be preferentially attributed to mulch, and the salt crust evidence is degraded. When the mulch component is lower than 20% and the polarization evidence is sufficient, it enters the salt crust determination set.

[0087] Subsequently, connectivity and hole repair are performed. The connected components are calculated within a five-by-five grid. The minimum four ten-meter pixels above the connected area are retained. The holes with an area less than two pixels are filled. A three-by-three structural element is used to perform a closed operation once to improve the boundary stability. At the same time, boundary consistency test is performed. If the cross-grid brightness difference exceeds one percent reflectivity or the radar scattering difference exceeds zero point eight decibels, it is marked as "boundary suspected artifact", and the downstream connectivity determination needs to be avoided; component quantization uses "multi-evidence lookup table" rule. Three indexes of optical stable highlight degree (relative quantile value), polarization difference amplitude and connectivity scale are used as independent variables. According to the segmented lookup table established by the pre-season sample and the patrol annotation, the pixel is mapped to the salt crust component percentage (between zero and one). The default weight of optical and radar is five percent each, and the weekly adaptive fine-tuning within two percent amplitude is allowed according to the ground feedback; when the effective time phase in the month is less than three times or the polarization evidence is "critical", only the interval estimate is output and the confidence is reduced to medium or low.

[0088] The trigger and stop conditions are: at the plot level, if the stable highlight candidate proportion does not reach 20%, the output is delayed and the next revisit cycle is combined; at the pixel level, if the determination is consistent for two consecutive times, the result is solidified and the update of the pixel in this cycle is stopped; the rain after transient time window is limited to 24 to 48 hours after rainfall, and the wet transient time window is limited to within 6 hours of high tide imaging, and the time window is automatically removed if it exceeds the time window. The output is the percentage of salt shell component per pixel, salt shell component mask (zero-one grid), quality label (high / medium / low) and reason code ("rain after transient" "high mulch" "polarization critical" "boundary suspected artifact" "insufficient time phase" etc.), and the parameter version number and data source number are written; the quality label quantization threshold is: high level requires stable highlight and polarization dual criteria, effective time phase reaches the lower limit within a month, no rain after transient and wet transient mark, connected component area is not less than four pixels and no boundary suspected artifact mark; medium level is any core criterion is "critical" or interval estimation is used but time phase reaches the lower limit, or single threshold is slightly lower than the threshold; low level is any one of the reason codes "radar priority" "insufficient time phase" "rain after transient" "wet transient" "high mulch", or the proportion of boundary suspected artifacts exceeds 5% of the plot.

[0089] The results are stored in the conditional quantity data set, and the "grid number + timestamp" is the primary key to return the grid number, timestamp, salt shell component percentage, salt shell mask, quality label, reason code, parameter version number and data source number, and carry unit, precision and version information, for S4 phase normalization and S5 equivalent bare land conversion direct reading. The time sequence and resource index is that each 100 square kilometers end-to-end processing does not exceed 20 minutes, and the concurrency is not less than 10 tasks, and when the queue pressure rises, the space is sliced and processed in batches to ensure that the main scene is given priority.

[0090] The abnormality and fault tolerance strategy includes: when the one-month time sequence gap exceeds one-third, the "radar priority" process is enabled, the result is generated only with polarization evidence and connectivity, and the credibility is uniformly reduced to low; the rainfall record is missing, which is filled with adjacent stations and the source is recorded; when the optical stability is destroyed due to rapid vegetation recovery, the VV / VH difference and connected scale are used as the main and the pixel with NDVI higher than 0.2 is avoided; the wet highlight label appearing within 6 hours of high tide imaging is "wet transient" and the determination is delayed. The quality verification is cross-checked at not less than 30 sample sites and ground patrol points, the salt shell identification accuracy is not less than 80%, the salt shell component absolute deviation is not more than 10%, the spatial consistency of the connected region is evaluated by the intersection ratio of the connected component, the threshold is not less than 0.6, and the repeated phase stability is represented by the median absolute deviation of the revisit difference within a month, and the threshold is not higher than 5%.

[0091] The minimum sample is, for example: input the optical time sequence of a certain plot in January (not less than four times) and two radar observations and rainfall records, after stability screening-polarization verification-coating exclusion-connectivity processing and table quantification, output the pixel salt shell component percentage 30, mask 1, quality label medium, reason code empty; In the case of serious incomplete multi-temporal, use the same day multi-angle optical consistency and adjacent orbit radar consistency instead of time sequence stability criterion, simultaneously reduce the reliable level and write "rule replacement" reason code in the result, to ensure business continuity and traceability.

[0092] S4, according to the distance from the coast, soil texture zoning and observation time, estimate the baseline offset caused by tidal action, complete the phase normalization and output the phase condition quantity, the specific implementation is:

[0093] On the basis of the aforementioned unified coordinate, time and dimension reference, the phase normalization process is established with the distance from the coast, soil texture and observation time as independent variables, and the imaging phase difference caused by the coupling of tide and shallow groundwater is pressed back to the unified reference of zero to five centimeters shallow water measurement system offset; Input includes the shortest horizontal distance from the pixel to the coastline (meters, the coastline is obtained after smoothing and denoising of the official coastline vector in the near year, the distance from the coast is defined as the shortest projection distance to the mean high tide line), soil texture zoning code (annual update, after connectivity revision, the minimum mapping unit is not less than four ten-meter grids), tidal height and fluctuation state (meters, from national and local tide stations and tide gauges, unified to national elevation reference and record station number and reference conversion information, time resolution not less than one hour), observation time (minute level time stamp, unified to Beijing time and keep the original collection time zone field for traceability), near ground wind speed and air temperature (meters per second and degrees Celsius, consistent with the same caliber and version as the aforementioned condition quantity).

[0094] The preprocessing first establishes a fixed distance zoning table, writes four distance bands of zero to two, two to five, five to ten and ten to twenty kilometers into the pixel attribute, at the same time, performs connectivity revision on the texture zoning to remove scattered small patches, performs daily smoothness check on the tidal level sequence and replaces the mutation points with sliding window smoothing, cross-source time alignment pairs by minute level time stamp, tolerance not more than ten minutes, out of limit record is set to be supplemented and filled in the next one to two revisit period with the latest phase interpolation and labeled with interpolation time difference and interpolation identifier; The distance from the coast is measured along the coastline in the bay and estuary scene, when the straight-line distance across the water area is greater than three kilometers or is blocked by seawalls and reclamation, it is determined as hydrological discontinuity and the upper limit of the amplitude modification strength is lowered by one level, at the same time, the discontinuity marker is written; The tidal level station is preferentially selected to be the nearest along the coastline and the distance from the coast is not more than thirty kilometers, when it exceeds thirty kilometers, the two stations are segmented and weighted along the coastline, the weight is inversely proportional to the distance from the coast and the station pair list is recorded, if the tidal level phase difference of the two stations exceeds fifteen minutes, it is marked as inconsistent between stations and the reliable level is lowered.

[0095] The execution stage gives the parameter set of default phase lag and amplitude with no less than thirty playback statistics of sample plots and short baseline field sampling in each combination of "distance band x texture partition" before the season, and compiles two-dimensional correction table of wind speed and air temperature, wind speed is divided into zero to three, three to six, six to ten and more than ten meters per second, air temperature is divided into less than ten degrees Celsius, ten to twenty degrees Celsius and more than twenty degrees Celsius, lag is selected in two to twenty-four hour discrete range, amplitude correction scale is limited to five to twenty percent; the default gear landing selection follows the principle of minimum deviation of verification sample plots, when the effective sample plots of a "distance band x texture partition" are less than eight, the default gear of adjacent partition is borrowed and the adjacent partition borrowing mark is written.

[0096] When the current time enters, the system locates the default parameters according to the pixel distance band and texture code, and then retrieves the corresponding items in the correction table according to the tidal height and fluctuation state, wind speed and air temperature, to get the water content baseline offset level of the pixel at the current time (expressed in amplitude percentage and offset level number jointly, and the offset level is one level with five percent amplitude); when the observation time is within thirty minutes before and after the high or low tide value, switch to stricter items to reduce the instantaneous phase error; when the extreme value half hour window and extreme wind speed are triggered at the same time, first execute the strict item of the extreme value window, then perform intensity convergence according to the extreme wind speed, and write the decision order to the processing log.

[0097] To suppress time jitter, set a stability threshold, when the level difference of adjacent two normalized results exceeds two levels (i.e. the amplitude difference exceeds ten percent), enable the three-point median strategy of adjacent time and write the stability intervention mark, at the same time, apply spatial consistency constraint to isolated pixels with three by three spatial window median; under the extreme condition of wind speed greater than ten meters per second or air temperature less than zero degrees Celsius, enable the amplitude upper limit convergence strategy, press the correction intensity to the preset upper limit and reduce the confidence level; when the tidal level version is inconsistent with the current batch, refuse to execute and fall back to the time consistent with the latest version and record the reason.

[0098] The output is written into a condition amount dataset with phase conditions, fields including distance band encoding, soil texture encoding, tide station number, tide gauge version, wind speed and air temperature version, phase lag level, amplitude percentage, offset level number, rise and fall state, stabilization identifier, hydrological disconnection and station inconsistency marker, trust level and processing timestamp, the trust level is divided into high, medium and low levels, the high level requires complete tide and weather sources, no interpolation and stabilization triggered, wind speed not higher than 6 meters per second and not in the extreme value 30-minute window, the medium level allows one stabilization or one interpolation with an interpolation time difference of no more than 2 hours, and the low level corresponds to tide interpolation, extreme wind speed and low temperature scene or station inconsistency and explicit prompt in the interface; the result is directly indexed by the subsequent equivalent bare land transformation through the grid number and timestamp as the key, and the quality control module simultaneously records the current revision table version for traceability.

[0099] The timing and resource target is to process not more than ten minutes per hundred square kilometers end-to-end, not less than ten concurrent tasks, no more than two times of failure retry with an interval of not less than five minutes, and the tide source delay first outputs a placeholder record to ensure the pipeline is not interrupted; exceptions and fault tolerance include temporarily replacing the missing tide data with the median value of the same tide phase moment in the past three days and reducing the trust level to low, triggering a distance too far warning when the distance from the nearest station to the coast exceeds 50 km and limiting the upper limit of the correction strength, temporarily replacing the misjudgment caused by the geometric error of the coastline at the boundary of the distance band with the parameters of the adjacent band and recording the inter-band replacement code, and triggering a phase instability alarm for manual review when the output of three consecutive phases is intervened by stabilization.

[0100] Verification uses pre-season and mid-season stratified playback evaluation, uses an independent validation set of not less than fifty sample plots, compares the deviation of the non-normalized and normalized shallow water content determination, requires a reduction amplitude of not less than 30%, not less than 25% in the 10-20 km band, and not less than 35% in the 0-2 km band, simultaneously performs paired non-parametric significance test and records pass and fail flags with a 5% significance threshold, and when the main threshold is not met, rolls back to the last stable version and records the rollback reason and impact range in the log; the output and storage comply with the aforementioned data and version specifications, the interface returns units, precision, version and quality label and supports incremental push subscription by region and time window.

[0101] The minimum sample is a scenario where a pixel in a ten-kilometer distance band is in the rising tide, the wind speed is six meters per second, the air temperature is eighteen degrees Celsius, and the time is aligned with the station. The system gives the phase lag level and amplitude percentage according to the "ten to twenty kilometers and loam" parameter set and the wind temperature correction table. If it is within twenty minutes before the high tide level, the strict item is enabled. The final output is the offset level plus two, the amplitude percentage is ten, the reliability is medium, and the stabilization is not involved. When the soil texture partition is missing, three partitions of flat smooth, flat rough, and micro-relief are established based on the terrain slope and the surface roughness statistics. The parameters are selected according to the closest partition and the reliability level is reduced by one. The texture replacement marker is written, which ensures that the input range, processing rules, parameter range, abnormal rollback, and acceptance standards are all traceable, verifiable, and seamlessly connected with the subsequent equivalent bare land conversion.

[0102] S5, input the pixel characteristics, mulch component, salt crust component, and phase condition quantity into the equivalent bare land conversion to obtain equivalent bare land observations, and then calculate the shallow soil moisture content in a monotonicity-controlled measurement model. The specific implementation is as follows:

[0103] For zero to five centimeter volume water content measurement range, the upstream pixel characteristics, mulch component, salt crust component, and phase condition quantity are integrated into the same processing chain. The equivalent bare land conversion is first performed at the pixel level to strip the system influence of mulch, salt crust, and tidal phase on observations. Then, the shallow soil moisture content is calculated in a monotonicity-controlled measurement model to restore a consistent, stable, and traceable monotonic relationship between the observation quantity and the moisture content. The input includes uniform grid characteristics, mulch component, salt crust component, phase condition quantity, and sample anchor information. Before entering, consistency check is performed: if the total of mulch component and salt crust component exceeds 80% or reaches mulch ≥ 70% and salt crust ≥ 60% respectively, it is transferred to the degraded channel to output only interval estimation; the version number of the phase condition quantity must be consistent with the condition quantity manager registration version. If it is not consistent, the fusion is aborted and the reason code and low reliability identifier are output; the optical reflectance is clipped between 0-1, the radar backscatter is clipped between the upper and lower limits of the operable decibels, the water content feasible region is limited to 0-45 volume percent, and the out-of-range record is a potential anomaly.

[0104] The equivalent bare ground conversion is implemented by piecewise lookup table method, and the lookup table key is fixed as "soil region division x coastal distance zone x incident angle group x mulch subinterval x salt crust subinterval x phase classification". The subinterval step of mulch and salt crust is 10%, and the phase classification is divided into low, medium and high three grades. Each key value outputs bias and scale correction coefficients, their effective period and version number. The piecewise is not more than 3 (default 2), and only the disturbed channel is subjected to conditional correction and the dimension is kept unchanged. The phase condition quantity is first mapped to the baseline offset magnitude, and the application order is fixed as "phase normalization→mulch correction→salt crust correction". If both mulch and salt crust exceed their respective high thresholds at the same time, and the output difference is greater than 1% by volume when the order is interchanged, the output is directly degraded to interval estimation and the "component conflict" reason code is attached. The initial value of the lookup table is calibrated by the pre-season sample and the stable target, and it is updated weekly in season with a small amplitude (single amplitude ≤20% of the original value). Each update generates an independent version number, and the table version and subinterval position used are recorded synchronously when the pixel is applied.

[0105] The monotonicity controlled measurement is performed on the equivalent bare ground observation, and the monotonic direction is set according to the physical feasible range and the empirical order, and the number of turning points is limited to 1, and the direction is strictly prohibited to reverse. The channel direction is constrained in an enumerated manner: the water-sensitive optical channel (combined channel of near-infrared and short-wave infrared) decreases with increasing water content, and the radar horizontal polarization backscattering tends to increase with increasing water content at a specified incident angle, and the vertical polarization tends to weaken. If the local direction is unstable, the feasible range of this interval is narrowed and the "monotonicity tightening" reason code is output. The position of the only turning point is determined by the error minimization criterion of the sample set, and the new sample triggers the update only within the specified neighborhood, and the "turning point version number" is written.

[0106] The scale and bias anchoring is performed according to the strategy of "pre-season unification and batch fine-tuning": not less than 3 sample points per administrative township, and a small number of sample points are used for incremental fine-tuning (single ≤20% of the original value) within the batch. The anchoring sample and the satellite observation time difference is ≤1 hour, and the sampling depth is 0-5 cm. The sample is only used for trend checking and does not participate in scale updating; the anchoring version and the used lookup table version are written together. The output is given in 10-meter grid units, including volume water content, upper and lower confidence intervals, confidence level, quality identifier, reason code, lookup table version and anchoring version. The volume water content is retained to 0.1% by volume, the system bias target is ±3%, and the root mean square error target is ≤4%. The confidence level is divided into high, medium and low: high level requires that the input condition quantity is complete and the residual absolute value is ≤3%, medium level allows single condition quantity interpolation or residual ≤4% in adjacent phase, and low level corresponds to version inconsistency, strong component or phase instability which has triggered degradation.

[0107] The upper and lower confidence intervals are derived from the residual distribution quantile statistics of the last 4 weeks under the same lookup key; when the sample is labeled as low confidence or has been degraded, the interval width is expanded by no more than 50% of the original width to cover the uncertainty. The reason code is limited to "high coating", "high salt shell", "phase instability", "component conflict", "default table", "version inconsistency", and "monotonicity tightening", and the interface only returns the enumeration value for downstream parsing. The results are stored in both raster and vector formats, with fields including region number, grid number, timestamp, volume moisture content, upper and lower confidence intervals, confidence level, quality identifier, reason code, table version, and anchor version. The service interface provides query and subscription, supports value traceability and certificate issuance, drought and flood warning, and irrigation applications, and returns the quality identifier to the dispatch to drive resampling and retesting.

[0108] The timing and resource indicators are ≤20 minutes per 100 square kilometers of end-to-end processing and ≥10 concurrent tasks, and the queue threshold is batched by spatial slicing and ensures that the table version and anchor version are consistent within the same batch; the task-level availability requirement is that the current batch effective pixel coverage rate is ≥90%, and when the spatial connectivity verification pass rate is <85%, the entire batch is marked as "needs review" and external publication is suspended. Abnormalities and fault tolerance include: conditional quantity missing, adjacent phase interpolation, and confidence degradation; monotonicity check failure automatically tightens the feasible interval and segment number; high component only outputs interval estimates and records; version inconsistency skips fusion and writes an alert; running logs synchronously record input lists, table versions used, anchor versions, exceptions and alerts, reason code distribution, processing time consumption, and concurrency strategy.

[0109] Verification is performed on an independent sample set covering ≥50 sample points, with system bias and root mean square error meeting the above targets, and repeated measurement correlation coefficient ≥0.7; consistency verification is also performed on different orbits or different sources but the same time window, and differences exceeding the limit are automatically rolled back to the previous stable table version and trigger a review. The minimum sample: a pixel has complete channel characteristics, with a coating of 35%, a salt shell of 20%, a phase of moderate offset, and a consistent version. After applying two segment corrections to the affected channels through equivalent bare land lookup, it enters the monotonicity-controlled measurement, outputs a volume moisture content of 13.0%, upper and lower confidence intervals of 11.0%-15.0%, a medium confidence level, a normal quality identifier, and an empty reason code; when the same pixel coating rises to 75%, it triggers degradation, only outputs the interval estimate of 12%-16%, and attaches the "high coating" reason code; when the region lacks the latest anchor sample for a short time, it uses the "default table" for temporary calculation and labels it as "default table". In the next batch, the new sample feedback is automatically corrected and the results are reissued, thereby continuously maintaining physical consistency, verifiability, and management availability under real business conditions.

[0110] S6、Use sample measurement to anchor scale and bias, output pixel-level soil moisture results and quality identifier, form measurement records and value traceability chain, and implement as follows:

[0111] Establish a "quadrat-remote sensing-evidence" closed loop to one-to-one correspondence between pixel-level remote sensing and ground standard moisture and form a verifiable evidence chain, unified caliber: quadrat position using the national two thousand coordinate system longitude and latitude, time unified Beijing time and keep the original collection time zone, sampling depth limited to zero to five centimeters, volume water value in volume percentage, soil texture according to annual update partition code record, supplemented by surface conditions (color and integrity of mulch, salt shell visibility, vegetation sparseness) and photo number; Sampling frequency once a week, within two hours before and after satellite overpass, allowed deviation is sampling-overpass time difference not more than one hour, sampling depth deviation not more than one centimeter, recommended same point retest not less than two times and repeat difference more than one percentage point, that is, on-site retest and record the reason.

[0112] Quadrat-pixel mapping uses uniform rules: standard quadrat is a square with a side length of ten meters, the center point falls within one meter of the center of the pixel to be verified; When the heterogeneity of the plot is significant or the mulch and salt shell patches are obvious, three points are set up at the center and five meters east and west of the pixel center after mixing; When the quadrat boundary crosses the pixel boundary, the anchor reference value is generated according to the nine-neighbor weighted method, the weight is fixed as one point zero for the center pixel, zero point five for the four side neighbors, and zero point two five for the four corner neighbors, and is written with the batch.

[0113] The ground reference is based on the drying method, the execution condition is 105 degrees Celsius plus or minus 2 degrees Celsius constant temperature for 24 hours, the constant weight criterion is that the difference between two times of weighing within two hours is not more than 0.1 grams; The container number and tare weight are entered into the table, the electronic balance resolution is not less than 0.01 grams; The electromagnetic method is limited to equipment that has been legally measured and is within the valid period, on-site daily calibration of zero and span is completed with at least two standard soil samples with different water levels or oven-dried and rewetted soil samples, and the instrument serial number and time are entered into the database; Organize a "electromagnetic-drying" comparison once a month, if the correlation coefficient is less than zero point seven or the absolute value of the system deviation is more than three percent of the volume water content, the electromagnetic reading is suspended for scale anchoring and the review is started.

[0114] Consistency verification and registration are performed before the sample plot is put into the warehouse. For time inconsistency, spatial inconsistency and statistical anomalies, the weight is first reduced and then determined to be excluded. The sample plot weight value range is zero to one, and the default is one. The weight is reduced to zero point two and zero point five, which is a fixed value. The pre-set limit value comes from the "anchoring strategy parameter table", which is managed according to the version and written with each batch: the time difference limit value is one hour by default, the plane positioning deviation limit value is five meters by default, and the statistical anomaly is based on the quartile and interquartile range. The lower quartile minus one interquartile range or the upper quartile plus one interquartile range is outside the range and is determined to be abnormal. The weight reduction rule: the time difference exceeds the limit but the review passes and the supplementary sampling can be completed within two hours, the weight is reduced to zero point two; the plane deviation exceeds the limit but can be re-registered to the adjacent grid and the on-site indication point is consistent, the weight is reduced to zero point two; the statistical anomaly has on-site written explanation and converges after re-measurement, the weight is reduced to zero point five. The exclusion rule is rigidly triggered, and any of the following conditions will be excluded and will not participate in scale anchoring and bias correction: the time difference with satellite observation exceeds the predetermined limit and the review does not pass; the plane positioning deviation exceeds the predetermined limit and cannot be corrected to the target grid; it is still abnormal after being determined to be abnormal by robust statistics and completing the review. The review process includes on-site review, repeated sampling, coordinate re-measurement and photo check; the criterion for failing the review is that the volume water content difference of repeated sampling is greater than one volume percentage point, or the coordinate re-measurement and sample number photo positioning are inconsistent, and any of the above conditions will be determined to be failed. The adjacent grid adsorption decision-making order is: the center distance is the smallest, followed by the input quality label, followed by the cloud snow interpolation proportion, and finally the main image is given priority.

[0115] The anchoring and correction follow the order of "scale anchoring first, bias refinement second": first, the global scale is anchored with the median value of all qualified sample plots in the current period, so that the remote sensing measurement is consistent with the ground caliber; then the bias is refined according to the soil texture partition and management area, and the area with high film covering proportion and significant salt crust is given priority; the newly added sample plots in each batch are updated incrementally, the single correction amplitude is limited to one percent of the absolute volume water content and cannot exceed twenty percent of the correction value of the last version, and the threshold is automatically converted to manual review; the time and space alignment of the low weight sample plot is zero point two in the scale anchoring, which is only used for stable trend and does not dominate the scale.

[0116] When outputting the pixel-level result, the anchoring version number, uncertainty interval and quality identification of this batch are added to the upstream remote sensing measurement (including volume water content, confidence level, reason code); the effective sample plots participating in scale anchoring in each soil texture partition are not less than three, and the independent verification sample plots are not less than two; when the number of independent samples is less than thirty, the cross-batch residual library is supplemented to thirty and the source is marked in the certificate.

[0117] The finished product is output with "certificate + traceability record": the certificate indicates the region, period, model version, table version, anchor version, sample summary (number, distribution, time difference statistics, method composition), error statistics (systematic bias and root mean square error), uncertainty interval, quality identification and reason code; the traceability record lists the sample details (anonymous number, coordinates, time, depth, method, original reading, retest information, photo number), processing parameters (threshold and weight, exclusion rule, low weight reason) and batch operation log summary. The certificate and service interface provide machine-readable fields simultaneously, and the JSON minimum key set is region_id, cell_id, ts, vwc, vwc_ci_p05, vwc_ci_p95, quality, anchor_version, model_version, table_version, basis (dry|em), basis_valid_until, reasons, hash, and SHA-256 hash is calculated and echoed at the end of the certificate page and the response header; when using electromagnetic method as temporary reference, basis is set to "em" and the valid period is given, and after supplementing the drying data, the anchor is automatically re-played and the certificate is re-issued, anchor_version is incremented by one, and the old and new certificates are linked through prev_hash for traceability, and a "replacement report" is generated indicating the replacement ratio, error change and affected list.

[0118] The connection with the upstream and downstream is kept closed loop: the upstream pixel-level results enter this unit, the certificate and record are generated after anchoring and written to the archive system, and "anchor version number + correction summary" is returned to the configuration center for subsequent batch scale and bias initial value; the phase and condition quantity manager records the batch environment caliber version synchronously, ensuring consistent across batches. The performance and resource target is that the single batch certificate generation time does not exceed ten minutes, the concurrency is not less than ten, the failure retry does not exceed twice and the interval is not less than five minutes; when the certificate queue is too long, it is processed by administrative division and the problem plots are given priority.

[0119] Abnormalities and fault tolerance include: when the number of sample points per week is less than the minimum threshold (not less than three in each township), the anchor of the previous week is used, and "weak anchor" is marked in the certificate and the applicable period; the absolute value of the deviation of individual sample points and remote sensing results exceeds 8% of the volume moisture content, which triggers review and on-site visit and records in the certificate; if the archiving fails, it will be re-tried three times and a local encrypted copy will be kept and an alarm will be given; when electromagnetic and drying coexist, drying is the final reference, and electromagnetic is only counted in scale anchoring after passing the offline comparison in the current season.

[0120] Quality verification adopts independent sample cross-checking and double-line repeatability: independent checking sample is not less than fifty, absolute value of system deviation is not more than three percent of volume water content, root mean square error is not more than four percent, correlation coefficient of repeated sampling point is not less than zero point seven, and verification result is combined with abnormal sample list and archived. Storage and playback follow base caliber: certificate and traceability record are permanently stored, sample plot original record and photo are stored for not less than one complete growing season, and parameter and log are stored for not less than six months; not less than one percent of grid samples are extracted every month for cross-batch playback verification and comparison report is formed, if fine tuning is needed, new version is generated according to the constraint of "single adjustment is not more than twenty percent of original value"; hash and version number are generated in the whole process, so that any certificate can be traced back to sample plot, processing parameter and running environment, and finally the value chain of "remote sensing-ground-certificate" is closed to the requirements of G0 1C determination science which is verifiable, traceable and engineering usable.

[0121] The calculation involved in the embodiments is the de-dimensioned numerical calculation, and the preset parameters and threshold values in the calculation are set by the person skilled in the art according to the actual situation.

[0122] It should be noted that the application can be deployed in the device itself to realize embedded application, or run on a PC terminal or other terminal with a user interface, thereby meeting various hardware environments and use requirements.

[0123] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another through wireless or wired transmission. The wired transmission includes optical fiber, twisted pair, coaxial cable and the like; the wireless transmission includes infrared, microwave and the like. The computer-readable storage medium can be any available medium accessible by the computer or a data storage device such as a server, data center and the like containing one or more available medium collections. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), optical medium (for example, DVD) or semiconductor medium. The semiconductor medium can be a solid state disk.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and module described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0125] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed modules can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0126] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules, which can be located in one place or distributed on a plurality of network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0127] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically, or two or more modules can be integrated in one module.

[0128] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0129] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0130] Finally, the above merely provides the preferred embodiments of the present application, but is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for soil moisture inversion and measurement based on remote sensing imagery, characterized in that, include: S1. Acquire multi-source remote sensing images and tidal meteorological condition data, complete geometric registration and radiometric consistency, and establish a unified grid observation benchmark and pixel feature set; S2. Determine the overlay component of the unified image, and generate a pixel overlay component mask by combining spectral morphology and microwave radar scattering response, and record it as a conditional quantity. S3. Measure the salt crust component of the unified image, generate a pixel salt crust component mask based on the temporal brightness stability and microwave radar polarization difference, and record it as a conditional quantity. S4. Based on the distance from the coast, soil texture zoning and observation time, estimate the water-bearing baseline shift caused by tidal action, complete phase normalization and output the phase condition quantity; S5. Input the pixel features, film components, salt crust components and phase condition quantities into the equivalent bare land transformation to obtain the equivalent bare land observation, and then solve the shallow soil moisture in the monotonicity controlled measurement model. S6. Use quadrat measurements for scale and bias anchoring, output pixel-level soil moisture results and quality labels, and form measurement records and traceability chains.

2. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 1, characterized in that, S1 includes: Under the unified national coordinate benchmark, unified projection, unified time caliber and unified tide level benchmark, geometric registration, radiometric uniformity, cloud and snow shadow processing, unified resampling and element assembly are carried out sequentially on multi-source optical and microwave images. Pixel alignment is achieved using ground control points, and radiometric uniformity is performed under the constraint of maintaining dimensional consistency. On the same day, the main scene is set according to the image with the incident angle in the middle of the distribution. In case of conflict, the rule is made according to the rule that the incident angle is close to the middle and the cloud cover is lower. The missing measurement traces of the strip are processed by diversion, and suspected artifacts at the boundary are identified.

3. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 2, characterized in that: Construct a pixel feature vector with grid number and timestamp as the primary key, including optical multi-channel reflectivity, radar dual-polarization scattering, solar altitude angle, incident angle, near-ground wind speed, surface temperature, tide level, plot boundary and soil texture, and attach cloud and snow shadow, strip completion, downgrade and suspected boundary artifacts to the records. Establish a hierarchical quality label and parameter version management mechanism. When the pixel feature coverage is insufficient, the output is paused and added to the queue to be supplemented. A standardized interface is provided to output to the coating component, salt crust component measurement unit and phase normalization module using grid number plus timestamp as key.

4. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 1, characterized in that, S2 include: Based on a unified grid and a unified radiation time caliber, the following steps are performed sequentially: receiving the registered and consistent optical multi-channel reflectivity and polarimetric radar scattering, as well as the incident angle, time, and plot boundary; completing brightness clipping, edge fidelity filtering, and plot-specific standardization. Candidate overlays are extracted based on high brightness and low chromaticity priors; radar evidence is verified by the scattering difference of adjacent bare land under the consistent incident angle benchmark; near-temporal consistency and spatial connectivity correction are combined, and suspected boundary artifacts, water bodies and urban construction high albedo interference are avoided, while salt crust candidates are called for exclusion; multiple evidences are mapped to pixel overlay components according to the lookup table segmentation rule and a mask is generated, which, along with quality labels and cause identifiers, is written into the conditional quantity dataset for downstream use.

5. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 1, characterized in that, S3 includes: Salt crust component determination includes: normalizing cross-field orbit and incident angle under a unified grid, unified radiation and time aperture, and calling up optical multi-temporal and dual-polarization radar data; The temporal brightness stability candidate, polarization difference verification, overlay component rejection, connectivity repair and boundary consistency test are executed sequentially. The pixel salt shell component is output according to the multi-evidence lookup table and a mask is generated. The results, along with the quality label, reason code, and parameter version, are recorded as conditional quantities and used for phase normalization and equivalent bare ground conversion.

6. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 1, characterized in that, S4 includes: Phase normalization includes: obtaining the shortest distance from the pixel to the shoreline, soil geological zoning, tidal level and fluctuation state, observation time and near-surface wind temperature under unified coordinates, time and radiation aperture; The phase lag and amplitude parameters are retrieved by "distance zone × texture", and the water content baseline offset is determined by combining the extreme value window of tidal level and wind temperature correction. When the difference between adjacent time phases is too large, temporal and spatial stabilization is implemented, and the correction intensity is limited in bay, estuary and hydrological non-connected scenarios. The output includes phase condition quantities such as distance band and texture code, tide level station and caliber version, hysteresis and amplitude level, fluctuation state and stabilization identifier, which are used for equivalent bare land conversion.

7. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 1, characterized in that, S5 include: Under a unified grid and unified radiometric and temporal aperture, the characteristics of the receiving pixel, the coating component, the salt crust component, and the phase condition quantity are analyzed. Verify the consistency of input execution criteria and version; The equivalent bare land transformation is performed in a fixed sequence of "phase normalization - film correction - salt crust correction" to obtain equivalent bare land observations; In the monotonicity-controlled measurement model, only one inflection point is allowed, and a monotonic direction is set for each observation channel. The pixel volume water content is calculated under the above constraints. The system uses quadrat anchoring to complete sizing and offset correction, and simultaneously outputs version, quality, and cause identification.

8. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 7, characterized in that: The equivalent bare land transformation adopts a segmented lookup table based on geological zoning, coastal distance zone, incident angle group, film classification, salt crust classification and phase classification, and only applies offset and scale correction to the disturbed channel. When the input component is under high interference, the phase is inconsistent with the version, or the correction order is swapped, causing the output difference to exceed the preset threshold, it will automatically be downgraded to interval estimation and the reason will be recorded. When monotonicity is unstable, tighten the feasible interval and keep the model direction from reversing.

9. The method for soil moisture inversion and measurement based on remote sensing imagery according to claim 1, characterized in that, S6 include: Sample plots were collected under a unified coordinate and time caliber and registered with the grid. Sample plot-cell mapping was established according to standard sample plots and neighborhood weighting, with the central cell having the highest weight. The drying method was used as the standard, and the electromagnetic method was replaced after verification and comparison. For temporally and spatially inconsistent and anomalous quadrats, the weighting will be reduced. If any of the following conditions are met, the quadrat will be excluded from scale anchoring and bias correction: the time difference with satellite observation exceeds a predetermined limit and the verification fails; the planar positioning deviation exceeds a predetermined limit and cannot be corrected; or the quadrat is determined to be anomalous by robust statistics and remains anomalous after verification. The scale anchoring was completed using quadrat statistics, and the bias was refined according to soil texture and management plots; The pixel results include anchored version, uncertainty, and quality identifier, generate a certificate and traceability record containing machine-readable fields and hash verification, and send back anchored information.

Images