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

Abstract

The invention discloses a soil moisture content inversion and determination method based on a remote sensing image, particularly relates to the technical field of surveying and mapping and remote sensing measurement, and is used for solving the problems of space-time inconsistency, system deviation and result non-traceability caused by superposition of a mulching film and a salt shell in spring and tide in an existing saline-alkali area. Under a meter-scale unified grid and a unified radiation and time caliber, three types of interferences of coating, salt crust and phase are extracted as conditional quantities, equivalent bare land processing is completed according to the sequence of phase normalization, coating correction and salt crust correction, then the volumetric moisture content is calculated in a monotonicity controlled model, the stable monotonicity relation between pixel observation and the moisture content is recovered, and the monotonicity controlled model is obtained. The system error caused by the difference between the non-bare ground and the imaging time phase is resolved from the source. Through pre-seasonal and in-seasonal verification, the overall deviation is remarkably reduced, the water content error is kept at a low level, judgment jitter caused by post-rain transient and a high tide window is effectively inhibited by a stabilization strategy, and the engineering precision and consistency requirements of coastal zone spring film covering and salt returning periods can be met.

Description

Technical Field

[0001] This invention relates to the field of surveying and remote sensing measurement technology, specifically to a method for soil moisture inversion and measurement based on remote sensing images. Background Technology

[0002] In existing technologies, soil moisture inversion based on remote sensing mainly relies on optical indices, empirical relationships of microwave backscattering, radiative transfer approximations, or machine learning regression. Common practices include using near-infrared and short-wave infrared sensors to detect water, and radar polarization and incident angle to assess surface water content. These methods are supplemented by cloud cover, shadows, and vegetation masking, and then calculated under the "bare ground assumption." These methods can achieve acceptable accuracy in inland agricultural areas with less mulching and relatively stable surface conditions. However, in saline-alkali agricultural areas, the proportion of mulch film coverage is high in spring, with both black and white films present, and the salt crust is bright and temporally stable. Furthermore, the observation phase changes caused by tides and shallow groundwater result in inconsistent "water content baselines" for the same pixel at different imaging times.

[0003] Existing technologies typically treat mulch and salt crust as "exclusion factors" or coarse-grained occlusions, or only perform overall statistical corrections, rarely quantifying them as pixel-level component conditions. Simultaneously, the coupled effects of tidal level and distance from the coast are often ignored or treated as uniform time shifts, failing to reflect differences in soil texture and coastal zonation. As a result, volumetric water content under non-bare land conditions is prone to systematic bias and temporal fluctuations, particularly noticeable in post-rain transients and high tide windows. Furthermore, removing these pixels entirely creates spatial gaps and sampling biases, affecting the consistency assessment at the plot scale.

[0004] From an engineering implementation perspective, existing solutions still fall short in terms of unifying coordinates, time, and radiation aperture for multi-source heterogeneous data; traceability criteria for film and salt crust; regional parameterization for tidal phase normalization; and quality grading and traceability of pixel-level results. Typical workflows primarily focus on single-scene or limited temporal phase calculations, lacking detailed rules for handling cloud cover and missing strip measurements, and easily amplifying the impact of boundary artifacts and differences in incident angles. Thresholds are mostly set empirically, lacking version management and closed-loop quadrat backtesting based on seasonal and plot differences. Inconsistencies in quadrat-pixel mapping, drying, and electromagnetic methods, as well as the minimum sample size threshold, limit the stability of scale anchoring. At the output level, most are displayed as raster maps or hierarchical maps, lacking pixel-level uncertainty intervals, quality labels, cause codes, and certificate records, as well as chain-like traceability of result versions, parameter table versions, and source data. In application scenarios with strong interference and significant temporal differences, such as saline-alkali areas, these characteristics of existing technologies and data products directly limit the commercial application of soil moisture data and cross-batch consistency assessment during the spring mulching and salt return periods. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for soil moisture inversion and measurement based on remote sensing images. By pre-quantifying three types of strong interferences—covering film, salt crust, and tidal phase—as conditional quantities, and processing them according to the "phase normalization → equivalent bare land conversion → monotonicity controlled measurement" link, and solidifying the results through quadrat anchoring and certificate-based traceability closed loop, the problem mentioned in the background technology is solved.

[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: Methods for soil moisture retrieval and measurement based on remote sensing imagery 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.

[0007] In a preferred embodiment, 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.

[0008] 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-polarization scattering, solar altitude angle, incident angle, near-ground wind speed, surface temperature, tide level, plot boundary and soil texture, and cloud and snow shadows, strip completion, degradation and suspected boundary artifacts are attached to the record. 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.

[0009] In a preferred embodiment, S2 includes: Based on a unified grid and a unified radiation time caliber, the following steps are performed sequentially: receiving registered and consistent optical multi-channel reflectivity and polarimetric radar scattering, as well as incident angle, time, and plot boundaries; performing brightness clipping, edge fidelity filtering, and intra-plot quantization standardization; extracting overlay candidates based on high-brightness and low-chromaticity priors; verifying radar evidence based on scattering differences from adjacent bare land under a consistent incident angle benchmark; combining near-temporal consistency and spatial connectivity correction, while avoiding suspected boundary artifacts, water bodies, and urban construction high-albedo interference, and simultaneously invoking salt crust candidates for exclusion; mapping multiple pieces of evidence to pixel overlay components according to lookup table segmentation rules and generating masks, which, along with quality labels and cause identifiers, are written into the conditional dataset for downstream use.

[0010] In a preferred embodiment, 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.

[0011] In a preferred embodiment, 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.

[0012] In a preferred embodiment, S5 includes: 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.

[0013] In a preferred embodiment, the equivalent bare land transformation employs a segmented lookup table based on geological zoning, coastal distance zones, incident angle groups, film coverage classification, salt crust classification, and phase classification, applying offsets and scale corrections only to the disturbed channels. 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.

[0014] In a preferred embodiment, S6 includes: 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.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention extracts the film component, salt crust component, and phase condition quantity on a 10-meter unified grid and with unified radiation and time aperture. Then, it performs an equivalent bare land transformation in a fixed order of "phase normalization → film correction → salt crust correction". Finally, it calculates the volumetric water content in a monotonic measurement model, so that the pixel observation and water content restore a stable monotonic relationship. This resolves the systematic error caused by the difference between non-bare land and imaging time phase from the source. After pre-season and mid-season verification, the overall deviation is reduced by no less than 30% after phase normalization, the root mean square error of volumetric water content is stabilized within 4%, and the judgment jitter caused by post-rain transients and high tide windows is significantly suppressed by the stabilization strategy. This meets the engineering application accuracy and consistency requirements of spring film covering and salt return period in coastal areas from 0 to 20 kilometers.

[0016] 2. This invention establishes a quality and traceability closed loop encompassing coverage acquisition, component measurement, phase normalization, equivalent bare land conversion, measurement calculation, quadrat anchoring, and certification. Through quality tags, versioned lookup tables and anchoring, certificate JSON key sets, and SHA-256 hash and prev_hash chaining, it achieves full-process visibility of the source, parameters, usage boundaries, and uncertainties of pixel-level results. In data gap and anomaly scenarios, it ensures continuous output and consistency through degradation, interpolation, playback, and version rollback. End-to-end latency and concurrency metrics meet the requirements of batch business operations. Furthermore, standardized interfaces support irrigation scheduling, early warning, and regulatory certification, significantly improving the reliability, repeatability, and management usability of measurement results. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for soil moisture inversion and measurement based on remote sensing images according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example: Figure 1 A flowchart illustrating the method for soil moisture inversion and measurement based on remote sensing imagery of the present invention is provided. The method for soil moisture inversion and measurement based on remote sensing imagery includes: 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.

[0020] The technical connections and implementation logic of the six steps are as follows: First, at a 10-meter resolution, multi-source optical and radar images, along with tidal and meteorological conditions, are aligned to the same coordinates, time, and radiation aperture, generating a pixel feature set with quality labels and version numbers. This set serves as the sole input aperture for all subsequent judgments. Then, on this aperture, the coating component is identified. Candidates are initially provided using optical brightness and chromaticity, and then evidence is enhanced by radar polarization and incident angle consistency. The pixel coating mask and component percentages are output, and a reason code is recorded. Next, salt crust contiguity is determined based on the January temporal brightness stability and radar polarization differences, forming salt crust mask and component percentages. These results are mutually exclusive with the coating results while preserving connectivity and artifact identifiers. With both coating and salt crust conditions in place, the water-bearing baseline shift is estimated based on the pixel's distance to the coast, soil texture, and observation time, combined with tide level, wind speed, and temperature grading. Phase lag and amplitude levels are provided, and phase normalization is completed. This process generates indexable phase condition quantities. Then, pixel features, along with the three condition quantities of overlay, salt crust, and phase, are fed into the equivalent bare land transformation. Phase normalization, overlay correction, and salt crust correction are performed in a fixed order, with segmented bias and scale correction applied only to affected channels. After obtaining the equivalent bare land observations, the volumetric water content from 0 to 5 cm is calculated within a monotonicity-controlled measurement model, generating confidence intervals, quality identifiers, and cause codes. Finally, the scale and bias are anchored using quadrat measurements, and the current batch of anchored versions is provided. Uncertainty intervals and certified traceability records are output along with the results. Anchoring summaries are sent back upstream to update lookup tables and parameter versions. Anomalies and deficiencies trigger downgrading, interpolation, or playback verification, thus forming a closed loop of acquisition, component measurement, phase normalization, equivalent bare land transformation, measurement calculation, and value traceability. This ensures continuous output, consistent caliber, and traceability and reverification within the same measurement range.

[0021] S1. Acquire multi-source remote sensing images and tidal meteorological data, complete geometric registration and radiometric consistency, and establish a unified grid observation benchmark and pixel feature set. The specific implementation is as follows: Targeting the spring emergence period in saline-alkali agricultural areas and the coastal zone from 0 to 20 kilometers, a 10-meter unified grid was used to establish an observation benchmark and pixel feature set, unifying observations from different satellites, orbits, and imaging times to the same measurement range. The spatial benchmark used the national 2000-year coordinate system, with the projection being the Gauss-Krüger 3-degree zone (zone number recorded according to the meridian of the image center). The time was unified to Beijing time, and the original acquisition time zone field was retained for traceability. The tide level was calculated using the national elevation benchmark, and the station number and benchmark conversion information were recorded.

[0022] Input elements include optical surface reflectance (dimensionless, ESA and national operational imagery, revisited every three to five days, with a time window of one day before and after the current date, and an allowable deviation of no more than 0.5%) and microwave radar backscattering coefficient (decibels, synchronous revisit frequency and time window, with an allowable deviation of no more than 0.5 decibels), along with solar altitude angle and incident angle (degrees), near-surface wind speed (meters per second), surface temperature (degrees Celsius), tidal height (meters, time resolution of no less than one hour, with an allowable deviation of no more than 0.1 meters), observation time (minute-level timestamps), and land parcel boundaries and soil zoning (annual updated classification codes). Before data is entered into the database, coordinates and time zone standards are standardized, and source numbers and version numbers are recorded. Cross-source data is paired on the same day and aligned with minute-level timestamps, with a tolerance of no more than ten minutes. Excessive grids are not merged in the current batch and are included in the "pending supplementation" queue. They are supplemented with the most recent phase interpolation in the next one to two revisit cycles, and the interpolation time difference and interpolation identifier are marked.

[0023] The processing chain is executed in the following order: "Geometric registration - Radiometric homogenization - Cloud and snow shadow removal - Unified resampling - Feature assembly". First, the images are aligned to a 10-meter grid using ground control points, with a planar error of no more than one meter. Microwave strips undergo terrain-related corrections before registration to avoid geometric stretching. Then, radiometric homogenization is performed while maintaining the same dimensions. Optical images are checked against the reflection baseline of stable targets in the scene, and radar images are checked against stable echo targets. Only small linear corrections to offset and scale are allowed, and acceptance is based on quantitative thresholds: On stable targets and representative plots, the absolute value of the difference between the mean values ​​of each optical channel before and after homogenization does not exceed 0.5%, and the dispersion from the 5th to the 95th percentile is reduced by no less than 20% compared to before homogenization; the absolute value of the difference between the mean values ​​of radar dual-polarization before and after homogenization does not exceed 0.5 dB, and the standard deviation is no more than 80% of that before homogenization. Scenes that do not meet the standards are not included in feature assembly, and a reason code is recorded in the log.

[0024] Subsequently, clouds, snow, and shadows are identified and processed. Fully occluded areas are set as invalid values. Thin clouds and semi-transparent edges are locally interpolated within a 3x3 grid and assigned low-confidence labels. When the cloud cover in a single scene exceeds 70%, the entire scene is removed from the current processing. All channels are then uniformly resampled to a 10-meter grid, using edge-fidelity interpolation (window not exceeding 3 grids). At the same time, boundary consistency is checked: using a 3x3 grid as a window, when the brightness difference across grids of similar ground features exceeds 1% of reflectivity or the radar scattering difference exceeds 0.8 dB, the pixel is marked as a "suspected boundary artifact" so that downstream connections can avoid it during connectivity determination. For missing strip measurements, the strips are handled by length: those with a continuous length of no more than one kilometer are linearly filled with the same diameter from a nearby track and the measurement is recorded; those exceeding one kilometer are marked as invalid. When multiple images exist on the same day, the main image with the incident angle at the median of the image distribution is selected, and the rest are included in the backup queue. If the main and backup images have conflicting conclusions on individual grids, the image with the incident angle closer to the median and lower cloud cover is given priority and marked "Conflict resolved". The other result is retained as supporting evidence.

[0025] The threshold and weight are initially calibrated before the season using no less than 30 quadrats and no less than 10 fixed-angle reflection targets. After the season begins, the threshold and weight are adaptively and slightly adjusted weekly based on the quadrat deviation statistics and the residual of the stable target. The adjustment range of a single adjustment shall not exceed 20% of the original value. All parameters are generated with version numbers and written with each batch.

[0026] The pixel feature vector uses "grid number + timestamp" as the primary key, and is equipped with optical multi-channel reflectivity, radar dual-polarization scattering, solar altitude angle and incident angle, wind speed, surface temperature, tide height, plot boundary code, and soil zoning code. It is also accompanied by cloud and snow shadow markers, strip completion markers, downgrade markers, quality labels, source number and parameter version number. The quality label is divided into three levels: high, medium and low. The high level requires the registration error to be no more than 0.5 meters and the cloud and snow interpolation ratio to be no more than 5%, and there is no strip completion marker. The medium level requires the registration error to be no more than 1 meter and the interpolation ratio to be no more than 15%, or there is a strip completion with a single segment length of no more than 1,000 meters. All other cases are low level and will be explicitly prompted when the interface returns. When the feature coverage of a single grid is less than 80%, the production of that grid is suspended and it is added to the "pending completion" queue.

[0027] The scheduling module drives the entire link with a task queue. External data enters this module through the access gateway. After processing, the status and exception codes are returned. The caliber versions of the tide level, wind speed, and temperature of this batch are registered to the condition quantity manager to ensure phase normalization and cross-batch traceability of equivalent bare land. The timing and resource objectives are that the end-to-end time for every 100 square kilometers should not exceed 60 minutes, with no less than 20 concurrent tasks, no more than 3 retries for failure with an interval of no less than 5 minutes, and when the queue pressure increases, it is divided into batches according to spatial slices and time slices, prioritizing the production of the main image first and the backup image later.

[0028] Anomaly and fault tolerance strategies include: when the registration error exceeds one meter, downgrading the output to a 30-meter grid and recording the reason for downstream filtering; when tidal data is scarce, temporarily replacing it with the median value of the same phase times in the last three days and lowering the confidence level, while leaving a trace in the conditional quantity manager; abandoning the phase when the cloud cover is too high to avoid false brightness misleading subsequent criteria; uniformly labeling edge artifacts introduced by resampling and strip completion to avoid them in the judgment of the continuity of film and salt crust. Quality verification adopts a two-layer check of geometry and radiation: the geometric layer is verified by independent ground control points, with the average plane error not exceeding 0.5 meters and the maximum error under control; the radiation layer is verified by stable targets and representative plots to check the channel differences before and after consistency, and the mean deviation and dispersion must meet the above hard thresholds; the total number of check plots for the two layers is not less than thirty, and the verification report and parameter version are archived and associated with the batch number for playback audit.

[0029] The results are stored as partitioned datasets, and the raster is permanently stored using GeoTIFF with block-based lossless compression. Metadata and indexes are stored in columnar format. A query interface with "region number + grid number + timestamp" as the primary key and subscription-based incremental push are provided. The returned content includes the unit, precision, source number, version number, and quality tag. The standardized output is simultaneously sent to the film component and salt crust component measurement units, and provides the tide level and time caliber to the phase normalization module.

[0030] The smallest example is as follows: input an optical image covering approximately 200 square kilometers, two radar images from the same day but with different orbits, and the tidal weather records of the same day. The process involves registration, standardization, cloud and snow shadow processing, resampling, strip processing, and feature assembly, and outputs a 10-meter grid feature set with quality and version labels, which is then stored in the database. When high-resolution optical images are severely lacking, radar is used as the primary reference. The optical channels are supplemented with stable statistics from the previous cycle and labeled as low confidence before continuing to flow to ensure business continuity.

[0031] The operation log records the input list, parameter version and threshold weight, anomalies and alarms, degradation and interpolation positions, time consumption statistics and concurrency strategies, and data source numbers; the original images and external condition quantities are stored for no less than one complete growing season, intermediate products are stored for no less than six months, and the pixel feature set is stored long-term. At least one percent of the grid samples are extracted monthly for playback verification and archiving reports, thereby normalizing heterogeneous and spatiotemporally inconsistent multi-source observations into a pixel feature set with the same dimension, resolution, and time caliber. This supports downstream stable identification of film and salt crust components and tidal phase normalization under the same measurement range, meeting the requirements of full disclosure, verifiability, and engineering usability.

[0032] S2. The overlay component of the unified image is determined. Combining spectral morphology and microwave radar scattering response, a pixel overlay component mask is generated and recorded as a conditional quantity. The specific implementation is as follows: Building upon the 10-meter unified grid and time caliber established in the previous unit, the film covering component was measured on the unified imagery. The presence and proportion of plastic film within pixels were quantified and used as conditional quantities for phase normalization and equivalent bare land conversion. The input remained consistent with the upstream data, including five channels of surface reflectance (dimensionless, allowable deviation ≤ 0.5%) for blue, green, red, near-infrared, and short-wave infrared, and radar dual-polarization backscattering coefficient (decibels, allowable deviation ≤ 0.5 dB). Incident angle (degrees), observation time (minute-level timestamp), and plot boundary vector were also input, inheriting quality labels and source versions. To unify the data caliber, the reference center wavelengths for the five optical channels were approximately 490 nm, 560 nm, 660 nm, 840 nm, and 1610 nm, respectively. Reflectance was stored at a percentage scale and the scaling factor was recorded. Radar data was processed in groups by incident angle, prioritizing the 30°–45° range; values ​​outside this range were corrected and labeled using the median incident angle of the adjacent time phase.

[0033] Optically, the coating exhibits overall high brightness and low chromaticity, while on radar, it displays reduced roughness and suppressed cross-polarization energy. Preprocessing first involves pruning the brightness range on a uniform grid to remove extreme noise outside the top and bottom 1% quantiles of channel statistics. Then, a 3x3 detail-preserving filter is applied to the edge regions to suppress isolated high frequencies while maintaining boundary positions. Subsequently, quantile normalization is performed on the multi-channel reflectivity within the same plot to reduce the impact of inter-plot albedo differences on the threshold. All processing actions are written into metadata. The execution sequence is: "Optical prior candidate selection—radar evidence secondary screening—temporal consistency verification—spatial connectivity correction—table lookup and segmented assignment." The quantification criteria for optical candidates are: plots with normalized brightness within the upper quartile to upper decimal range and normalized chromaticity ≤ 0.25 are considered strong candidates; those only reaching the upper quartile and with chromaticity between 0.25 and 0.30 are considered weak candidates, only upgraded when strong radar evidence is present.

[0034] Radar secondary screening uses adjacent bare land at the same incident angle as the comparison benchmark. Strong evidence is defined as having a polarization difference ≥1.0 dB and cross-polarization suppression ≥1.0 dB, or a polarization difference ≥4.0 dB, with at least 5 consecutive pixels within a 3x3 window. When multiple time phases overlap on the same day, results with a time difference ≤6 hours from the main image are prioritized, with the rest used as circumstantial evidence for consistency checks. To prevent confusion with high-brightness objects, a unified exclusion mechanism is implemented: when the near-infrared and short-wave infrared combination exhibits water-like characteristics and its area ratio >20%, overlay output is suspended; urban construction high-reflectivity masks are excluded; when the overlap area with salt crust candidates is >30%, the salt crust unit confirms the result before restoring the current unit's estimation; simultaneously, the "boundary suspected artifact" marker from the previous unit is used to avoid it during connectivity correction. For black films, if visible light is not bright, short-wave infrared is significantly low-reflectivity, and the polarization is consistently higher than the cross-polarization, it enters the conservative channel and its confidence level is lowered.

[0035] Spatial connectivity correction first performs majority decision-making using a 3x3 window to eliminate isolated pixels, then fills holes smaller than one pixel to prevent fragmentation from affecting component estimation. Component assignment uses a lookup table and segmentation rules: optical brightness is divided into four levels, chromaticity into two levels, same polarization difference into two levels, and cross-polarization suppression into two levels, mapped to coating component intervals with a step size of 5%, ranging from 0 to 100%. The results are rounded to the nearest step size and clipped within the 0-100% range. When both strong optical and strong radar evidence are satisfied and connectivity is passed, the upper limit of the interval is shifted up by one step size; when only a single piece of evidence is valid or there is thin cloud interpolation or incident angle correction, the interval is shifted down by one step size and a reason code is written. The initial weight is optical:radar = 6:4. Before the season, a one-time calibration is completed using no fewer than 30 sample plots. After the season begins, weekly adaptive fine-tuning is performed based on sample plot back-checking deviations and stable target residuals, with a single adjustment range ≤20%, and a version number is generated for each adjustment.

[0036] The trigger and stop conditions are as follows: demixing starts when the candidate ratio is ≥10%, and grid processing stops when the results of two consecutive rounds of screening are consistent; if there are insufficient candidates, optical highlight evidence is retained and added to the "to be supplemented" list. The output includes two types of formats: pixel-level overlay component percentage raster and binary overlay mask raster, as well as plot-level statistical reports (plot number, overlay area, component mean and confidence interval); pixel records are stored as percentage integers with a step size of 1, missing measurements are assigned a value of -1, and masks are encoded with 0 / 1. The record also returns the grid number, timestamp, confidence level, reason code, weight version, input quality label, source data batch number and incident angle group number.

[0037] Quality labels are divided into three levels: High-level requires strong optical and radar evidence with connectivity passed, registration error ≤0.5 meters, cloud and snow interpolation ratio ≤5%, and no strip completion; Medium-level allows registration error ≤1 meter, interpolation ratio ≤15%, or the presence of strip completion with a single segment length ≤1000 meters, and strong evidence plus weak circumstantial evidence; the rest are classified as low-level and explicitly indicated in the interface; when the cloud thin interpolation ratio within the plot is >10% or the incident angle correction exceeds the interquartile range of the panoramic incident angle distribution, the quality label must not be higher than medium-level. Processing performance indicators are: end-to-end ≤15 minutes per 100 square kilometers, concurrent ≥10 tasks, ≤2 retry attempts with an interval ≥5 minutes, and batch processing by region slice when queue pressure increases, prioritizing the main image coverage area; output synchronously pushes salt crust component measurement units as rejection references and writes them into the conditional quantity dataset for direct indexing of equivalent bare land conversion and phase normalization.

[0038] Quality verification employs stratified sampling: three zones are established between 0 and 20 kilometers from the sea, with ≥10 quadrats in each zone; each quadrat must cover at least two of the three widths of the mulch film: <1 meter, 1-2 meters, and >2 meters, with coverage for both small and large plots; evaluation indicators are: absolute value of component deviation ≤10%, recall ≥80%, precision ≥75%, and area crossover ratio ≥50%. Verification reports and parameter versions are archived with each batch. The operation log records the input list, threshold and weight versions, anomalies and alarms, locations of downgraded and suspended production awaiting replenishment, time consumption statistics and concurrency strategies, and data source numbers, stored in the same database as the results to ensure traceability of the source of mulch film component conditions, reproducibility of processing, and verifiable quality, and to ensure natural connection with the upstream grid baseline and downstream salt crust components and phase normalization in data and control flow.

[0039] S3. Measure the salt crust component of the unified image. Based on the temporal brightness stability and microwave radar polarization difference, generate a pixel salt crust component mask and record it as a conditional quantity. The specific implementation is as follows: Based on a unified grid and aperture, using multi-temporal optical surface reflectance, dual-polarization microwave radar backscattering, observation time, precipitation records, and plot boundaries and soil geology delineation as inputs for a continuous monthly sequence, salt crust component measurements are performed on each pixel, and salt crust component masking and conditional quantities are output. The input data maintains the same coordinates, time, and radiation aperture as S1. For radar backscattering across different orbits, orbital consistency is first performed to control the comparability between channels of the same polarization within 0.5 dB. If the difference in incident angle exceeds 5 degrees, incident angle normalization is performed first according to the lookup table correction rules before execution. Polarization criteria are applied; the temporal "brightness" adopts a joint brightness caliber of red, near-infrared, and short-wave infrared, with each of the three having a default weight of one-third, and can be slightly adjusted within a range of plus or minus 20% on a weekly basis. The full range of joint brightness is given by statistically analyzing the quantile interval of the cloudless effective pixels in the current batch, which is used as a reference for the stability criterion of "fluctuation amplitude less than 10% of the full brightness range"; the minimum lower limit of the effective temporal phase within the month is no less than four for optical and no less than two for radar. If either does not reach the lower limit, only the interval estimate is output and reduced to medium or low confidence, and "insufficient temporal phase" is written in the reason code.

[0040] The processing is performed in the following order: "temporal stability candidate → polarization evidence verification → coating rejection → connectivity and hole repair → component quantization → quality and version labeling". First, a monthly optical temporal brightness statistic is constructed. For each pixel, the monthly median brightness and dispersion are calculated. Pixels whose median brightness is located in the upper quartile of the scene (default upper quartile to upper decimal) and whose monthly fluctuation is less than 10% of the overall brightness range are retained as "stable high brightness candidates". At the same time, based on the precipitation record, the short-term high brightness one to two days after the rain is reviewed temporally and labeled as "post-rain transient" and cannot be included in the candidate set.

[0041] Secondly, polarization verification is performed in the radar domain. Using a 3x3 grid as the window, it is compared with the adjacent bare ground baseline. If the overall increase in unidirectional polarization (VV) within one month (default increase not less than 1 dB) and the overall decrease in cross-polarization (VH) (default decrease not less than 0.8 dB) are met simultaneously in at least three time phases, then the polarization evidence is considered sufficient. If only some time phases are met, it is recorded as "critical" and downweighted. The overlay component output by S2 is called again as the exclusion condition. When the overlay component is higher than 20%, the "stable high brightness" interpretation of the pixel is preferentially attributed to overlay, and the salt crust evidence is downgraded. When the overlay component is lower than 20% and the polarization evidence is sufficient, it enters the salt crust determination set.

[0042] Subsequently, connectivity and hole repair were performed. Connectivity components were calculated within a 5x5 grid, retaining contiguous areas with a minimum of four 10-meter pixels. Holes smaller than two pixels were filled, and a closure operation was performed using a 3x3 structuring element to improve boundary stability. Boundary consistency was also checked. If the cross-grid brightness difference exceeded 1% reflectivity or the radar scattering difference exceeded 0.8 dB, it was marked as a "potential boundary artifact" and should be avoided in downstream connectivity assessments. Component quantization adopted a "multi-evidence lookup table" rule, using three indicators—optical stable high brightness (relative quantile), polarization difference amplitude, and connectivity scale—as independent variables. A segmented lookup table was established based on pre-season quadrats and patrol markings to map pixels to salt crust component percentages (between zero and one). The default weights were 50% for optical and 50% for radar, and adaptive fine-tuning was allowed weekly based on ground feedback within a range not exceeding 20%. When there were fewer than three valid time phases within a month or the polarization evidence was "critical," only interval estimates were output and downgraded to medium or low confidence.

[0043] The triggering and stopping conditions are as follows: at the plot level, if the proportion of stable bright candidates does not reach 20%, the output will be postponed and merged into the next revisit cycle; at the pixel level, if two consecutive judgments are consistent, the result will be solidified and the update of the pixel in this cycle will be stopped; the post-rain transient time window is limited to 24 to 48 hours after the rainfall, and the wet transient time window is limited to 6 hours before the high tide imaging. If the time window is exceeded, the transient mark will be automatically removed. The output consists of the percentage of salt crust component per pixel, salt crust component mask (zero-one grid), quality label (high / medium / low) and reason code ("post-rain transient", "high coverage", "critical polarization", "potential boundary artifact", "insufficient temporal phase", etc.), and the parameter version number and data source number are written in. The quality label quantification threshold is as follows: high level requires simultaneous satisfaction of stable high brightness and polarization dual criteria, effective temporal phase within the month reaching the lower limit, no post-rain transient and wet transient markers, and connected component area of ​​not less than four pixels and no suspected boundary artifact annotation; medium level requires any core criterion to be "critical" or use interval estimation but the temporal phase reaches the lower limit, or a single threshold is slightly lower than the threshold; low level requires triggering any reason code of "radar priority", "insufficient temporal phase", "post-rain transient", "wet transient" or "high coverage", or the proportion of suspected boundary artifacts exceeds five percent of the plot.

[0044] The results are stored in the conditional dataset. The key-value interface returns the grid number, timestamp, salt crust component percentage, salt crust mask, quality label, reason code, parameter version number, and data source number, using "grid number + timestamp" as the primary key. It also carries the unit, precision, and version information, allowing direct reading by S4 phase normalization and S5 equivalent bare land transformation. The timing and resource metrics are: end-to-end processing time per 100 square kilometers should not exceed 20 minutes, with a minimum of ten concurrent tasks. When queue pressure increases, processing is done in batches by spatial slices to ensure priority for the main scene.

[0045] Anomaly and fault-tolerance strategies include: when a one-third time-series gap in January occurs, a "radar-first" process is activated, generating results based solely on polarization evidence and connectivity, and uniformly downgrading them to low confidence; missing rainfall records are supplemented by nearby stations with their sources recorded; when rapid vegetation regreening disrupts optical stability, VV / VH differences and connectivity scales are prioritized, avoiding pixels with NDVI higher than 0.2; and wet highlights appearing within six hours of high tide imaging are marked as "wet transients" and their determination is delayed. Quality verification involves cross-checking at no fewer than thirty sample plots and ground inspection points, with salt crust identification accuracy no less than 80%, absolute deviation of salt crust components no greater than 10%, spatial consistency of contiguous areas evaluated by the intersection-union ratio of connected components with a threshold no lower than 0.6, and repeatability stability characterized by the median absolute deviation of the monthly revisit difference with a threshold no higher than 5%.

[0046] For example, the minimum sample size is: inputting the optical time series of a certain plot of land for one month (no less than four times) and two radar observations and rainfall records, and after stability screening, polarization verification, overlay rejection, connectivity processing and table lookup quantization, the output pixel salt shell component is 30%, the mask is 1, the quality label is medium, and the reason code is empty; in the case of severe incompleteness of multiple time phases, the consistency of optical data from multiple angles on the same day and the consistency of radar data from adjacent orbits are used to replace the time series stability criteria, the confidence level is reduced simultaneously, and the "rule replacement" reason code is written into the result to ensure business continuity and traceability.

[0047] S4. Based on the coastal distance, 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. The specific implementation is as follows: Based on the aforementioned unified coordinate, time, and dimensional benchmarks, a phase normalization process is established with coastline distance, soil texture, and observation time as independent variables. This process pushes back the systematic offset of the 0-5 cm shallow water content measurement caused by the coupling of tides and shallow groundwater to the unified benchmark. The inputs include the shortest horizontal distance from the pixel to the coastline (meters, the coastline is obtained by smoothing and denoising the official coastline vector of the recent year, and the distance from the coastline is defined as the shortest projected distance to the mean high tide line), soil zoning code (updated annually, with the minimum mapping unit after connectivity revision being no less than four 10-meter grids), tide height and fluctuation state (meters, derived from national and local tide stations and tide gauges, unified to the national elevation benchmark and recording the station number and benchmark conversion information, with a time resolution of no less than one hour), observation time (minute-level timestamps, unified to Beijing time and retaining the original acquisition time zone field for traceability), and near-surface wind speed and air temperature (meters per second and degrees Celsius, using the same caliber and version as the aforementioned conditional quantities).

[0048] Preprocessing first establishes a fixed-distance zonation table, encoding four distance zones—0 to 2 km, 2 to 5 km, 5 to 10 km, and 10 to 20 km—into the pixel attributes. Simultaneously, connectivity revision is performed on the texture zones to remove scattered small patches. The tidal level sequence undergoes a daily stationarity check, and abrupt changes are smoothly replaced using a sliding window between adjacent times. Cross-source time alignment is performed using minute-level timestamps with a tolerance of no more than ten minutes. Records exceeding this limit are marked as pending and interpolated with the most recent phase within one to two subsequent revisit cycles, with the interpolation time difference and interpolation identifier labeled. In bay and estuary scenarios, shore distance is measured along the shoreline. When the straight-line distance across waterways exceeds 3,000 meters or is blocked by seawalls or reclamation, it is judged as hydrological disconnection, and the upper limit of amplitude correction intensity is lowered by one level. At the same time, a disconnection mark is written. For tide level stations, priority is given to those with the closest shoreline and a shoreline distance of no more than 30 kilometers. When the distance exceeds 30 kilometers, the shoreline segments of the two stations are weighted, with the weights inversely proportional to the shoreline distance, and the station pair list is recorded. If the tide level phase difference between the two stations exceeds 15 minutes, the inconsistency between the stations is marked and the confidence level is lowered.

[0049] During the execution phase, within each "distance zone × texture zone" combination, a default phase lag and amplitude parameter set is generated before the season using playback statistics of no less than thirty sample plots and short baseline field sampling. Two maintenance correction tables for wind speed and temperature are also compiled. Wind speed is divided into 0 to 3, 3 to 6, 6 to 10, and greater than 10 meters per second, while temperature is divided into below 10 degrees Celsius, 10 to 20 degrees Celsius, and above 20 degrees Celsius. The lag range is selected from the discrete range of 2 to 24 hours, and the amplitude correction scale is limited to 5% to 20%. The selection of the default range follows the principle of minimizing the deviation of the verification sample plots. When there are fewer than eight valid sample plots in a certain "distance zone × texture zone", the default range of the adjacent zone is borrowed and a neighboring zone borrowing mark is written.

[0050] Upon entering the current time, the system locates the default parameters according to the pixel distance zone and texture code, and then retrieves the corresponding entries in the correction table based on the tide height and fluctuation state, wind speed and temperature to obtain the water content baseline offset magnitude of the pixel at the current time (expressed jointly as an amplitude percentage and offset level number, with the offset level being a 5% amplitude level) and correction intensity; when the observation time falls within 30 minutes before or after the extreme value of high tide or low tide, it switches to a more stringent entry to reduce instantaneous phase error; when the extreme value half-hour window and extreme wind speed are triggered simultaneously, the stringent entry of the extreme value window is executed first, and then the intensity convergence is performed according to the extreme wind speed, and the decision order is written to the processing log.

[0051] To suppress time jitter, a stability threshold is set. When the difference in the level between two adjacent normalization results exceeds two levels (i.e., the amplitude difference exceeds 10%), the adjacent time three-point median strategy is activated and a stabilization intervention flag is written. At the same time, spatial consistency constraints are applied to isolated pixels using the regional median of a three-by-three spatial window. Under extreme conditions such as wind speed greater than 10 meters per second or temperature below zero degrees Celsius, an amplitude upper limit convergence strategy is activated to suppress the correction intensity to the preset upper limit and reduce the confidence level. When the tide level caliber version is inconsistent with the current batch, execution is rejected and the process is rolled back to the time of the most recent consistent version, and the reason is recorded.

[0052] The output is written into the conditional quantity dataset as phase conditional quantities. Fields include distance zone code, soil texture code, tide station number, tide caliber version, wind speed and air temperature caliber versions, phase lag level, amplitude percentage, offset level number, fluctuation state, stabilization indicator, hydrological disconnection and inter-station inconsistency markers, confidence level and processing timestamp. The confidence level is divided into three levels: high, medium and low. The high level requires complete tide and meteorological sources, no interpolation and stabilization triggered, and wind speed not exceeding six meters per second and not within the extreme value 30-minute window. The medium level allows one stabilization or one interpolation with an interpolation time difference not exceeding two hours. The low level corresponds to tide interpolation, extreme wind speed and low temperature scenarios or inter-station inconsistencies, and is explicitly prompted in the interface. This result is directly indexed by the subsequent equivalent bare land transformation through the interface with grid number and timestamp as the key. The quality control module synchronously registers the version of this correction table for traceability.

[0053] The timing and resource objectives are: end-to-end processing time of no more than ten minutes per 100 square kilometers, with no fewer than ten concurrent tasks, no more than two retries with an interval of no less than five minutes between failures, and outputting pending record entries to ensure uninterrupted pipeline operation when the tidal source is delayed; anomalies and fault tolerance include: when tidal data is missing, temporarily replacing it with the median value of the phase times of the same tidal level in the last three days and reducing it to low confidence; triggering a distance too far alarm and limiting the upper limit of correction intensity when the distance to the nearest station is more than 50 kilometers away from the shore; temporarily replacing the misjudgment of the distance zone boundary due to shoreline geometric errors with the parameters of the adjacent zone and recording the inter-zone replacement code; and triggering a phase instability alarm for manual review when three consecutive phase outputs are stabilized.

[0054] Validation employs stratified replay assessments before and during the season, using an independent validation set of no fewer than fifty quadrats. The deviations in shallow water content measurements are compared between unnormalized and normalized samples. The overall deviation reduction is required to be no less than 30%, no less than 25% for the 10-20 km zone, and no less than 35% for the 0-2 km zone. Pairwise nonparametric significance tests are performed, with a 5% significance threshold used to record pass / fail status. If the primary threshold is not met, the system rolls back to the previous stable version, and the reason for the rollback and its impact are recorded in the log. Output and storage adhere to the aforementioned data and version specifications. The interface returns units, accuracy, version, and quality tags, and supports incremental push subscriptions by region and time window.

[0055] The smallest sample is a scenario where a pixel within a 10-kilometer distance zone is in the rising tide phase, with a wind speed of 6 meters per second, a temperature of 18 degrees Celsius, and aligned with the station's time. The system provides the phase lag level and amplitude percentage based on the parameter set of "10 to 20 kilometers and soil" and the wind and temperature correction table. If it is 20 minutes before the high tide, strict entries are enabled, and the final output fields include offset level plus 2, amplitude of 10%, medium confidence, and no stabilization intervention. When soil texture zoning is missing, three alternative zoning levels are established using topographic slope and surface roughness statistics: flat and smooth, flat and rough, and slightly undulating. Parameters are selected according to the closest level, and the confidence level is reduced by one level while writing the texture substitution mark. This ensures that the input caliber, processing rules, parameter range, anomaly backoff, and acceptance criteria for phase normalization are all traceable and verifiable, and seamlessly connected with the subsequent equivalent bare land transformation.

[0056] S5. Input the pixel features, covering film components, salt crust components, and phase condition quantities into the equivalent bare land transformation to obtain the equivalent bare land observation. Then, solve the shallow soil moisture in the monotonicity-controlled measurement model. The specific implementation is as follows: For the measurement range of 0 to 5 cm volumetric water content, the upstream pixel features, covering film components, salt crust components, and phase condition quantities are integrated into the same processing chain. At the pixel level, an equivalent bare land transformation is first performed to remove the systematic influence of covering film, salt crust, and tidal phase on the observation. Then, the shallow soil moisture is solved in a monotonicity-controlled measurement model, so that the observed values ​​and water content restore a consistent, stable, and traceable monotonic relationship. The input includes the characteristics of each channel of the unified grid, the coating component and the salt crust component, the phase condition quantity and the sample plot anchoring information. Before entering, a consistency check is performed: if the combined coating component and the salt crust component exceed 80% or the coating is ≥70% and the salt crust is ≥60% respectively, it will be converted to a downgraded channel and only the interval estimate will be output; the version number of the phase condition quantity must be consistent with the version registered in the condition quantity manager. If they are inconsistent, the fusion will be stopped and a reason code and a low confidence mark will be output; the optical reflectivity is clipped to 0–1, the radar backscattering is clipped to the upper and lower limits of the operable decibels, the feasible water content is limited to 0–45% volume percentage, and out-of-bounds records are potential anomalies.

[0057] The equivalent bare land transformation is implemented using a segmented lookup table method. The lookup key is fixed as "geological zoning × coastal distance zone × incident angle group × film classification × salt crust classification × phase classification". The step size for film and salt crust classification is 10% each, and the phase classification is divided into three levels: low, medium, and high. Each key value produces two types of correction coefficients: bias and scale, as well as their validity period and version number. The number of segments does not exceed 3 (2 segments by default). Conditional corrections are applied only to the disturbed channels while maintaining the dimensionless nature. The phase conditional quantity is first mapped to the baseline offset level, and the application order is fixed as "phase normalization → film correction → salt crust correction". If both film and salt crust exceed their respective high thresholds and the two orders are interchanged, resulting in an output difference > 1 volume percentage, the result is directly downgraded to an interval estimate with a "component conflict" reason code. The initial values ​​for the lookup table are jointly calibrated by pre-season quadrats and stable targets. At the beginning of the season, the values ​​are updated slightly on a weekly basis (single amplitude ≤ 20% of the original value). Each update generates an independent version number, and the table version and classification position are recorded synchronously when the pixel is applied.

[0058] Monotonicity controlled measurements are conducted on equivalent bare land observations. Monotonic directions are set according to physical feasibility and empirical order, with the maximum number of inflection points limited to one, and direction reversal strictly prohibited. Channel directions are constrained through enumeration: the water-sensitive optical channel (a combination of near-infrared and short-wave infrared channels) decreases with increasing water content; at a specified incident angle, horizontal polarization backscattering from the radar tends to increase with increasing water content, while vertical polarization tends to decrease; local directional instability narrows the feasible range of that interval and outputs a "monotonicity tightening" reason code. The location of the unique inflection point is determined by minimizing the error of the sample set. When a new sample triggers an update, only minor adjustments are made within a predetermined neighborhood, and the "inflection point version number" is written.

[0059] Scale and bias anchoring are implemented according to the strategy of "unified before the season and fine-tuned in batches": no less than 3 quadrats per administrative township, with a small number of quadrats used for incremental fine-tuning within each batch (≤20% of the original value per batch), the time difference between anchored quadrats and satellite observation is ≤1 hour, and the sampling depth is 0–5 cm. Quadrats exceeding the limit are only used for trend verification and do not participate in scale updates; the anchoring version and the table lookup version used are written back together. The output provides volumetric water content, upper and lower confidence intervals, confidence level, quality identifier, cause code, table lookup version, and anchoring version in 10-meter grid units; volumetric water content is retained to 0.1 volume percentage, 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: the high level requires complete input conditions and the absolute value of the residual is ≤3%, the medium level allows adjacent phase interpolation of a single condition or a residual of ≤4%, and the low level corresponds to inconsistent versions, strong components, or phase instability that have triggered degradation.

[0060] The upper and lower confidence intervals are derived from the residual distribution quantile statistics of the most recent four weeks under the same lookup key. When a sample is labeled as low confidence or has been downgraded, the interval width is increased by one level, not exceeding 50% of the original width, to cover uncertainty. The reason codes are limited to seven categories: "high coating," "high salt crust," "phase instability," "component conflict," "default table," "version inconsistency," and "monotonicity tightening." The interface only returns enumerated values ​​for downstream parsing. Results are stored in both raster and vector formats. Fields include region ID, grid ID, timestamp, volumetric water content, upper and lower confidence limits, confidence level, quality identifier, reason code, table version, and anchor version. Querying and subscription are provided through the service interface, supporting traceability and certification of measurement values, drought and flood early warning, and irrigation applications. The quality identifier is also sent back for scheduling to drive data collection and retesting.

[0061] The timing and resource metrics are: end-to-end processing time ≤ 20 minutes per 100 square kilometers, concurrent tasks ≥ 10; if the queue exceeds the threshold, it will be run in batches according to spatial slices, ensuring that the table version and anchor version are consistent within the same batch; the task-level availability requirement is that the effective cell coverage of the current batch is ≥ 90%; if the spatial connectivity verification pass rate is < 85%, the entire batch will be marked as "requires review" and external release will be suspended. Anomalies and fault tolerance include: if the condition quantity is missing, interpolation of adjacent time phases will be performed and the reliability will be reduced; if the monotonicity check fails, the feasible interval and the number of segments will be automatically tightened; if the component is too high, only the interval estimate will be output and recorded; if the version is inconsistent, fusion will be skipped and an alarm will be written; the operation log synchronously records the input list, the table version used, the anchor version, anomalies and alarms, the cause code distribution, processing time and concurrency strategy.

[0062] The verification is performed on an independent quadrat set and covers ≥50 quadrats. The system bias and root mean square error are required to meet the aforementioned targets, and the correlation coefficient of repeated measurements is ≥0.7. At the same time, consistency verification is carried out on the results of different orbits or different sources but within the same time window. If the difference exceeds the limit, the system will automatically roll back to the previous stable table version and trigger a review. Minimum Sample: A pixel has complete channel characteristics, 35% overlay, 20% salt crust, moderate phase shift, and consistent version. After applying two corrections to the affected channel through equivalent bare ground lookup table, it enters monotonicity controlled measurement, outputting a volumetric moisture content of 13.0%, upper and lower confidence intervals of 11.0%–15.0%, medium confidence, normal quality label, and empty reason code. When the overlay of the same pixel increases to 75%, a downgrade is triggered, outputting only the interval estimate of 12%–16% with a "high overlay" reason code. When the latest anchored quadrats are missing in the area for a short period, the "default table" is used for temporary calculation and marked as "default table". In the next batch, the results are automatically corrected and resent with the feedback of the newly added quadrats, thus continuously maintaining physical consistency, repeatability, and management availability under real business conditions.

[0063] S6. Scale and bias anchoring are performed using quadrat measurements, outputting pixel-level soil moisture results and quality labels to form measurement records and a traceability chain. The specific implementation is as follows: A closed loop of "sample plot - remote sensing - certification" is established to correspond pixel-level remote sensing soil moisture with ground standards and form a verifiable data chain. The unified standards are as follows: sample plot locations use the latitude and longitude of the national 2000 coordinate system, time is unified with Beijing time and the original collection time zone is retained, sampling depth is limited to 0 to 5 cm, volumetric water content is expressed as a volume percentage, and soil texture is updated annually with zone coding records. Surface conditions (covering color and integrity, salt crust visibility, vegetation density) and photo numbers are supplemented. The sampling frequency is once a week, completed within two hours before and after the satellite transit day. The allowable deviation is that the sampling-transit time difference does not exceed one hour and the sampling depth deviation does not exceed one centimeter. It is recommended that the same point be retested at least twice, and if the duplication difference exceeds one percentage point, the on-site retest should be conducted and the reason recorded.

[0064] The quadrat-pixel mapping adopts a unified rule: the standard quadrat is a square with a side length of 10 meters, and the center point falls within 1 meter of the center of the pixel to be verified; when the heterogeneity of the plot is significant or the film and salt crust patches are obvious, three points are set up at the center of the pixel and 5 meters away from the east and west, and then measured after mixing; when the quadrat boundary crosses the pixel boundary, the anchoring reference value is generated by weighting the nine neighborhoods, with the weights fixed as 1.0 for the center pixel, 0.5 for the four side neighbors, and 0.25 for the four corner neighbors, and written with each batch.

[0065] Ground benchmarks are based on the drying method, with operating conditions of a constant temperature of 105 degrees Celsius ± 2 degrees Celsius for 24 hours. The constant weight criterion is that the difference between two weighings two hours apart does not exceed 0.1 grams. Container numbers and tare weights are entered into the table, and the resolution of the electronic balance is not less than 0.01 grams. Electromagnetic methods are limited to equipment that has passed legal metrological verification and is within its validity period. On-site, zero-point and span calibrations are performed daily using at least two standard soil samples or dried-back-wet soil samples with different moisture levels, and the instrument serial number and time are entered into the database. A "magnetic-drying" comparison is organized monthly. If the correlation coefficient is lower than 0.7 or the absolute value of the system deviation exceeds 3% of the volumetric moisture content, electromagnetic readings are suspended from participating in scale anchoring and a review is initiated.

[0066] Before sample plots are stored, consistency verification and registration are performed. For inconsistencies in time, space, and statistical anomalies, weighting is first reduced before exclusion. Sample plot weights range from zero to one, with a default of one; weighting to 0.2 and 0.5 are fixed values. Predefined limits are derived from the "Anchoring Strategy Parameter Table," managed by version and written with each batch: the default time difference limit is one hour, the default planar positioning deviation limit is five meters, and statistical anomalies are defined by the interquartile range and interquartile distance; plots falling outside the lower quartile minus one interquartile distance or the upper quartile plus one interquartile distance are considered anomalies. Weighting rules: Sample plots with time differences exceeding the limit but passing verification and capable of being resampled within two hours are weighted to 0.2; sample plots with planar deviations exceeding the limit but capable of being re-registered and snapped to adjacent grids with consistent field indicator points are weighted to 0.2; sample plots with statistical anomalies supported by written field explanations and converged after retesting are weighted to 0.5. The exclusion rules are rigidly triggered; if any of the following conditions are met, the grid is excluded and does not participate in scale anchoring and offset correction: the time difference with satellite observation exceeds a predetermined limit and the review fails; the planar positioning deviation exceeds a predetermined limit and cannot be corrected to the target grid; or the grid is identified as abnormal by robust statistics and remains abnormal after review. The review process includes on-site verification, resampling, coordinate remeasurement, and photo verification. The criteria for failing the review are that the volumetric water content difference of the resampling is greater than one volume percentage point, or the coordinate remeasurement is inconsistent with the positioning of the sample plot photo; if either criterion is met, the grid is deemed to have failed. The adsorption adsorption decision order for adjacent grids is: minimum center distance, then higher input quality label, then lower cloud / snow interpolation ratio, and finally, priority given to the main image.

[0067] Anchoring and correction follow the order of "anchoring the scale first, then refining the bias": First, the median value of all qualified sample plots in the current period is used to complete the global scale anchoring, so that the remote sensing range is consistent with the ground caliber; then, the bias is refined according to soil texture zoning and management plots, giving priority to covering plots with high film coverage and significant salt crust; each batch of new sample plots is updated incrementally, and the single correction range is capped at 1% of absolute volumetric water content and must not exceed 20% of the previous version's correction value. If the threshold is exceeded, it will be automatically transferred to manual review; sample plots with low weight in time and space alignment participate in scale anchoring with a weight of 0.2, and are only used to stabilize trends and do not dominate the scale.

[0068] When outputting pixel-level results, add the current batch anchoring version number, uncertainty range, and quality identifier to the upstream remote sensing measurements (including volumetric water content, confidence level, and cause code). The uncertainty is given as the fifth to ninety-fifth percentile of the independent verification residuals. Each soil texture zone should have no fewer than three valid quadrats participating in the scale anchoring and no fewer than two independent verification quadrats. When there are fewer than thirty independent samples, supplement the cross-batch residual library to thirty and indicate the source in the certificate.

[0069] The finished product is output as "test certificate + traceability record": The certificate specifies the region, time period, model version, lookup table version, anchoring version, sample plot summary (quantity, distribution, time difference statistics, method composition), error statistics (systematic deviation and root mean square error), uncertainty interval, quality identification and reason code; The traceability record lists the sample plot details (anonymous number, coordinates, time, depth, method, original readings, retest information, photo number), processing parameters (threshold and weight, rejection rules, reasons for low weight) and batch operation log summary. The certificate and service interface synchronously provide machine-readable fields. The minimum key set of JSON 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, and hash. The SHA-256 hash is calculated and echoed in the certificate footer and response header. When the electromagnetic method is used as a temporary baseline, basis is set to "em" and the validity period is given. After the drying data is replenished, the anchoring and re-issuance are automatically replayed. The anchor_version is incremented by one. The old and new certificates are linked by prev_hash for traceability. At the same time, a "replacement report" is generated, which specifies the replacement ratio, error changes, and the list of affected certificates.

[0070] Maintain a closed-loop connection with upstream and downstream processes: Upstream pixel-level results enter this unit, and after anchoring, certificates and records are generated and written to the archiving system. The "anchoring version number + correction summary" is then sent back to the configuration center for subsequent batch scale and initial bias values. The phase and condition quantity manager synchronously records the environmental caliber version for each batch, ensuring consistency across batches. Performance and resource targets are: single batch certificate generation time not exceeding ten minutes, concurrency not less than ten certificates, and no more than two retries with an interval of not less than five minutes between failures; when there is a large queue for certificate issuance, processing is done in batches according to administrative regions, with priority given to problematic plots.

[0071] Anomalies and tolerances include: if the number of sample plots in a given week is less than the minimum threshold (no less than three plots per township), the anchoring from the previous week will be used and marked "weak anchoring" and the applicable period in a prominent position on the certificate; if the absolute value of the deviation between individual sample plots and remote sensing results exceeds 8% of the volumetric water content, a review and on-site revisit will be triggered and recorded on the certificate; if archiving fails, the sample will be queued for three retryes and a local encrypted copy will be retained and an alarm will be issued; when electromagnetic and drying coexist, drying will be used as the final benchmark, and electromagnetic will only be included in the scale anchoring after the offline comparison in the current season passes.

[0072] Quality verification employs a dual-track approach of independent sample cross-checking and repeatability consistency: no fewer than fifty samples are independently checked, with the absolute value of systematic deviation not exceeding 3% of the volumetric water content and the root mean square error not exceeding 4%. The correlation coefficient of repeated sampling points is not less than 0.7. Verification results and a list of abnormal samples are archived together. Storage and playback follow the same standards as the base: certificates and traceability records are permanently stored; original sample plot records and photos are kept for at least one complete growing season; parameters and logs are kept for at least six months. Each month, at least 1% of grid samples are randomly selected for cross-batch playback verification and a comparison report is generated. If minor adjustments are needed, a new version is generated according to the constraint that "a single adjustment shall not exceed 20% of the original value." Hash values ​​and version numbers are generated throughout the entire process to ensure that any certificate issuance can be traced back to the sample plot, processing parameters, and operating environment. Ultimately, the measurement chain of "remote sensing—ground—certification" is closed to meet the requirements of G01C metrology, which is reproducible, traceable, and engineering-usable.

[0073] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0074] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0075] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as 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, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0076] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0078] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0079] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0080] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0082] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

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