Drought adaptability assessment system for helan mountain coniferous forest based on multi-source data

Abstract

The application discloses a Helan Mountain coniferous forest drought adaptability evaluation system based on multi-source data and concretely relates to the field of forestry information processing and ecological monitoring, and comprises a same shooting controller and an observation terminal, writes root zone moisture content and crown layer images after collecting air temperature and humidity according to the same shooting number, supplements growth state codes by the investigation terminal, positions according to the stand number, sequences according to the shooting number, divides sources according to the source number, and outputs observation records; comprises a communication gateway and a storage device, is used for correcting the shooting number, writes the stand number according to the position, extracts the crown layer coverage value from the crown layer image, writes according to the shooting number after collecting according to the stand number, forms stand time series; forms the stand time series through the same shooting collection and the shooting number correction, constructs drought receiving relationship chain and recovery relationship chain according to the phenological change, generates dislocation time series source by source, verifies the relationship chain and the sequence adjacent position by using the backtracking maximum common subgraph, and identifies and cancels the drought adaptability evaluation result formed due to the distortion of the time sequence relationship.

Description

Technical Field

[0001] This invention relates to the field of forestry information processing and ecological monitoring technology, and more specifically, to a drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data. Background Technology

[0002] Currently, the assessment of drought adaptability of coniferous forests is mostly based on remote sensing images, meteorological observations, soil moisture and stand survey data. Through time registration, phenological correction, index fusion or grade determination, the degree of damage and recovery performance of the stand during drought are analyzed.

[0003] The coniferous forests of Helan Mountain are affected by altitude, slope aspect and local hydrothermal conditions. Different forest stands may be in the growth initiation, vigorous growth or growth decline stage on the same date. The satellite transit time, meteorological sampling time and ground survey time are independent of each other. Therefore, all kinds of data can only jointly characterize the drought response of the forest stands when they conform to the actual growth stage and the sequence of drought evolution.

[0004] Existing methods typically generate evaluation results directly after time registration or phenological correction without further examining the dependence of these results on temporal relationships. Consequently, it is possible that the evaluation grade remains unchanged even after replacing data from adjacent growth stages, data from the same period in non-drought years, or data from other drought events. This phenomenon indicates that the evaluation results are mainly affected by stand age, slope aspect, canopy density, or seasonal background, and fail to truly reflect the correspondence between drought exposure, stand response, and subsequent recovery.

[0005] The technical problem this application aims to solve is: how to identify and eliminate misjudgments of coniferous forest drought adaptability caused by distortion of temporal relationships during the multi-source data knowledge fusion process. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data. This system forms forest stand time series through simultaneous data collection and time series number correction, constructs drought-affected relationship chains and recovery relationship chains based on phenological changes, generates misaligned time series source by source, and verifies relationship chains and sorted neighbors using a retrospective maximum common subgraph. It identifies and revokes drought adaptability assessment results caused by distorted time series relationships, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data, comprising:

[0008] The multi-source acquisition module includes a simultaneous capture controller and an observation terminal. After acquiring air temperature and humidity according to the simultaneous capture number, it writes the root zone water content and canopy images. The growth status code is added by the survey terminal. After locating according to the stand number, it is ordered according to the capture number and then sorted by source number to output the observation record.

[0009] The aggregation module, including a communication gateway and a storage device, is used to correct the time stamp number, write the stand number according to the location, extract the canopy cover value from the canopy image, and then aggregate it according to the stand number and write it according to the time stamp number to form the stand time series.

[0010] The knowledge fusion module is used to perform information processing and knowledge fusion in the forestry field. It divides phenological segments based on the reversal of canopy cover difference and changes in growth status codes, merges the same-direction differences into change segments, first connects meteorological aridity to root zone water loss, then connects root zone water loss to canopy descent, and finally connects root zone recovery to canopy return, forming an original knowledge graph. It generates the number of withstand beats and the number of recoverable beats by connecting the beat number differences at both ends, and forms an initial sorting table by descending order of the number of withstand beats and ascending order of the number of recoverable beats.

[0011] The misalignment construction module is used to execute the iterative amplitude adjustment Fourier method, shift the time sequence source by source, replace the shifted amplitude with the original Fourier amplitude and retain the phase, and after the inverse transformation, fill the original value according to the rank of the updated value. The same value is ordered according to the beat number. Iterate until the rank sequence first coincides to form a misalignment time sequence, and generate a misalignment knowledge graph and misalignment sorting table according to the knowledge fusion rules.

[0012] The evaluation module is used to perform operations research or analysis using a backtracking maximum common subgraph algorithm. Among nodes with the same fragment type and source number, they are paired according to the difference sign and the connection direction. After backtracking and pruning of the remaining nodes, a common subgraph is obtained. When the common subgraph retains the dehydration to descending connection and the recovery to return connection and the adjacent stand number has not changed, the initial sorting is canceled and the time-series distortion record is output. Otherwise, the drought adaptability evaluation record is output.

[0013] In a preferred embodiment, the multi-source acquisition module includes:

[0014] The synchronous controller generates an incremental time number and sends the sampling edge to the meteorological observation terminal. After the meteorological observation terminal reads the air temperature and air humidity values, it writes back the completion bit. The synchronous controller triggers the root zone observation terminal and the canopy observation terminal with the same time number based on the completion bit to obtain the root zone water content and canopy image.

[0015] The simultaneous sampling controller refers to the hardware controller that uniformly controls the sampling sequence of each observation end. It is equipped with a sampling number register and a trigger signal output terminal. Each time a sampling trigger is completed, the sampling number is incremented by one bit. After reading the completion bit returned by the meteorological observation end, the current sampling number is sent to the root zone observation end and the canopy observation end to make the data corresponding to the same sampling number belong to the same forest stand observation.

[0016] The sampling edge refers to the moment when the trigger signal output by the synchronous controller changes from low to high. This moment serves as the unified start time for a sampling and is used to determine the data acquisition order and survey record reading time point of each observation end under the current time number.

[0017] The root zone observation terminal refers to the soil moisture monitoring device buried in the soil layer where the main root system of the target forest stand is distributed. It consists of a soil moisture probe, a data acquisition circuit, and a data transmission interface. After receiving the trigger signal corresponding to the current sequence number, it reads the root zone moisture value, binds the root zone moisture value with the current sequence number, and then transmits it back. The canopy observation terminal refers to the image acquisition device installed at a position that can cover the canopy of the target forest stand and maintains a fixed imaging direction. It consists of a camera, imaging range calibration parameters, and a data transmission interface. After receiving the trigger signal corresponding to the current sequence number, it acquires the canopy image, binds the canopy image with the current sequence number, and then transmits it back.

[0018] After obtaining the root zone water content and canopy image, the survey terminal reads the last growth status code at a time no later than the current sampling edge and binds the air temperature value, air humidity value, root zone water content, canopy image and growth status code into a co-shot data group;

[0019] The survey terminal refers to a mobile data terminal used by forestry surveyors to input forest stand growth survey results. It internally stores the survey time, forest stand number and growth status code, and stores survey records in the order of survey time. When forming the current data group, it reads the survey record whose survey time is no later than the current sampling time and whose time order is later.

[0020] A growth status code is a fixed-length code used to indicate the growth stage of a target forest stand at a corresponding survey time. The code value corresponds to each item in a pre-established forest stand growth stage table. Only one growth status code is written for the same forest stand at the same survey time, and subsequent status changes are indicated by writing new survey records.

[0021] In a preferred embodiment, the multi-source acquisition module further includes:

[0022] Based on the installation coordinates of each observation terminal, the corresponding forest compartment is determined. When the canopy image crosses the forest compartment boundary, the canopy image is segmented according to the intersection of the imaging range and the forest compartment boundary. The segmented canopy images are written into the corresponding forest stand number so that all data in the same data group point to the same forest stand.

[0023] Among them, a forest compartment refers to a forestry management spatial unit delineated based on the results of forest resource surveys. Each forest compartment corresponds to a unique forest stand number and a closed boundary. The closed boundary is formed by connecting the boundary points in a unified coordinate system in order, and is used to determine the target forest stand to which the observation terminal's installation coordinates and the canopy image imaging range belong.

[0024] For the same batch of data for which the stand number has been written, record checks are performed according to the batch number and source number. If multiple records appear from the same source under the same batch number, the first record in the receiving order is retained. Data that is not returned before the next batch number is generated is marked as missing. Late data is bound to the original batch number and marked as supplementary. Then, observation records are formed according to the stand number, batch number, and source number.

[0025] In a preferred embodiment, the aggregation module includes:

[0026] After receiving the observation record and the time number table written by the time controller, the communication gateway reads the local order difference of adjacent observation records according to the source number, and determines the current corrected time number by the sum of the previous corrected time number and the local order difference; if the current corrected time number is inconsistent with the time number carried in the observation record, the time number conflict bit is written and the current corrected time number is retained to form a correction record.

[0027] Based on the collection location in the calibration record, read the forest compartment boundary. When the collection location falls inside the forest compartment boundary, write the corresponding forest stand number. When the collection location falls on the forest compartment boundary line, write the first forest stand number in ascending order of the forest compartment numbers on both sides of the boundary to form a forest stand record.

[0028] In a preferred embodiment, the aggregation module further includes:

[0029] For canopy images in the forest stand record, canopy reference pixels and ground reference pixels saved during camera calibration are read in the corresponding forest compartment projection area. A pixel adjacency map is established according to the channel difference between adjacent pixels and random walk segmentation is performed. Pixels with a higher probability of reaching canopy reference pixels than those with a higher probability of reaching ground reference pixels are recorded as canopy pixels. If the probabilities are the same, they are recorded as ground pixels in ascending order of pixel row number and column number. The canopy cover value is obtained by dividing the number of canopy pixels by the number of pixels in the forest compartment projection area.

[0030] Among them, the canopy reference pixels refer to the set of pixels in the canopy image that are confirmed to belong to the coniferous forest canopy during the camera calibration and the forest stand survey. The ground reference pixels refer to the set of pixels that are confirmed to belong to the bare ground, understory vegetation or shaded background during the camera calibration and the forest stand survey. The two types of reference pixels respectively store the pixel row number, pixel column number, source number and calibration version, and are used as fixed seeds for random walk segmentation of canopy images.

[0031] The random walk segmentation process is as follows: each pixel in the forest compartment projection area is established as a node, and a connection is established between nodes whose row or column numbers differ by one bit. The connection weight is obtained by dividing by one and adding the sum of the absolute values ​​of the differences in the color channels of the pixels at both ends of the connection. The canopy arrival probability of the canopy reference pixel is fixed at 1 and the ground arrival probability is fixed at zero. The canopy arrival probability of the ground reference pixel is fixed at zero and the ground arrival probability is fixed at 1. For the remaining nodes, an equation is established in which the node probability is equal to the weighted average of the probabilities of adjacent nodes according to the connection weight. After arranging the unknown nodes in ascending order of pixel row and column numbers, the system of equations is solved to obtain the probability of each pixel reaching the canopy reference pixel and the ground reference pixel.

[0032] After the canopy cover value is calculated, the storage device uses the stand number, correction sequence number and source number to form a record key. The correction record, stand number and canopy cover value are written to the corresponding record key. The supplementary record is written back to the original record key without changing the order of subsequent records. The missing record is left empty. Then, the records are collected according to the stand number and the corrected sequence number is used to form the stand time sequence.

[0033] The supplementary collection record refers to a record formed when a missing collection record is first formed under the corresponding auction number for a certain source, and then the communication gateway receives data carrying the same source number and original auction number. The received data is filled into the original record key and a supplementary collection mark is written. The supplementary collection record uses the original stand number, corrected auction number, and source number, and no new record order is added. The missing collection record refers to a placeholder record formed by creating a record key with the current stand number, corrected auction number, and source number before the next auction number is generated, when a certain source number has not returned the current auction number data. The data fields are left blank and a missing collection mark is written to maintain the positional correspondence of the time sequence of each source under the same auction number.

[0034] In a preferred embodiment, the knowledge fusion module includes:

[0035] Based on the forest stand time-series execution of information processing and knowledge fusion in the forestry field, the canopy cover difference reversal beat and growth status code change beat are read as candidate segment boundaries. The candidate segment boundaries are used as nodes, and the beat number intervals between adjacent candidate segment boundaries are used as edges. The edge cost is formed by the difference reversal number and the number of missing beats within the interval. The growth status code change beat is set as a necessary node. Through interval dynamic programming, the first path is read according to the cumulative edge cost increasing and the segment boundary beat number ascending order, and the phenological segment is output.

[0036] Within each phenological segment, the adjacent beat differences of air temperature and humidity, root zone water content, and canopy cover are calculated. Differences with the same sign and adjacent beat numbers are merged into change segments. Segment nodes are generated by segment type, difference sign, start and end beat numbers, and source number. Candidate edges are established only when the meteorological dryness segment precedes the root zone water loss segment, the root zone water loss segment precedes the canopy decline segment, and the root zone recovery segment precedes the canopy return segment. A candidate knowledge graph is then output.

[0037] In a preferred embodiment, the knowledge fusion module further includes:

[0038] For the candidate knowledge graph, each segment node enters only one similar relation chain, the candidate edge direction increases along the beat number and the relation chain does not cross phenological segments as constraints, the difference between the beat numbers at the two ends of the candidate edge, the difference in the amplitude rank of the segment and the number of missing beats are used as the edge cost, the network simplex algorithm is used to perform basis replacement with increasing reduced cost until the reduced cost of all non-basis edges is not negative, then the number of complete relation chain paths is calculated according to the node beat number in the forward and reverse directions, candidate edges that are not in complete relation chains are deleted and basis replacement is performed again until the edge sets of two adjacent rounds are consistent, forming the original knowledge graph;

[0039] The number of withstandable beats is generated based on the beat number difference between the starting point of the root zone water loss segment and the starting point of the canopy descent segment in the original knowledge graph. The number of recoverable beats is generated based on the beat number difference between the starting point of the root zone recovery segment and the starting point of the canopy return segment. The stands are first sorted in descending order of withstandable beats, and then the stands with the same withstandable beats are sorted in ascending order of recoverable beats. If the recoverable beats are still the same, they are sorted in ascending order of stand number. The initial sorting table is then output.

[0040] In a preferred embodiment, the misalignment construction module includes:

[0041] Read the phenological segment boundaries, original knowledge graph, and stand time series of the target forest stand. Fix one source time series in sequence and freeze the remaining source time series. Develop forward and backward shifts based on the beat number difference between the starting point of the target phenological segment and the starting points of the adjacent phenological segments. Use the two shifts to cyclically move the fixed source time series. Delete the corresponding result when the shift is zero. Output the source-by-source shifted time series.

[0042] The phenological segment boundary refers to the segmented beat number selected by the knowledge fusion module from the canopy cover difference reversal beat and the growth status code change beat. The first valid beat number and the last valid beat number serve as the starting segment boundary and the ending segment boundary, respectively. Two adjacent phenological segment boundaries define one phenological segment. The beat number difference between the starting point of the target phenological segment and the starting point of the previous phenological segment is used to form the forward shift amount, and the beat number difference between the starting point of the subsequent phenological segment and the starting point of the target phenological segment is used to form the backward shift amount.

[0043] For each source-by-source shift time series, perform iterative amplitude adjustment Fourier method, read the Fourier amplitude and original value ascending table of the corresponding original source time series, replace the transformed amplitude of the source-by-source shift time series with the original Fourier amplitude and retain the transformed phase, obtain the updated value through inverse transformation, and then fill the original value ascending table according to the rank of the updated value. For the same value, determine the rank according to the step number ascending order and save the rank sequence. Repeat amplitude replacement and rank backfilling until the rank sequence of this round coincides with the saved rank sequence, and output the candidate misaligned time series that retain the original value set and the original Fourier amplitude.

[0044] The process of performing the iterative amplitude adjustment Fourier method is as follows: using the source-by-source shift timing sequence as the first round timing sequence, performing a discrete Fourier transform on the first round timing sequence, reading the phase of the complex result at each frequency bit, multiplying the Fourier amplitude of the original source timing sequence at the same frequency bit by the corresponding phase to form a replacement spectrum, and then performing an inverse discrete Fourier transform to obtain the updated value; arranging each time number according to the updated value from low to high, and arranging them in ascending order when the updated values ​​are the same, then writing the first original value in the ascending list of original values ​​into the first time number, the second original value into the second time number, until all time numbers are filled back to form the next round timing sequence; saving the rank sequence after filling back to each round, when the rank sequence of the current round is consistent with any saved rank sequence, reading the timing sequence corresponding to the first round of the repeated rank sequence as the candidate misaligned timing sequence;

[0045] Fourier amplitude refers to the magnitude of the complex result of each frequency bit after performing a discrete Fourier transform on the original source time series; the zero frequency bit records the sum of the values ​​of the original source time series; the positive and negative frequency bits corresponding to the real time series adopt conjugate symmetric amplitudes; when the number of samples is even, the Nyquist frequency bits are stored separately to maintain the frequency composition of the original source time series in each round of spectrum replacement.

[0046] The original value ascending order table refers to the value table formed by arranging all valid sampled values ​​in the original source time series in ascending order of numerical value. When the values ​​are the same, the order is determined by the original beat number in ascending order. Missing sampling positions are not included in the value table and are retained after the rank backfilling is completed, so that the candidate misaligned time series retains the set of valid values ​​and the distribution of missing sampling from the original source time series.

[0047] The rank sequence refers to the sequence of rank and position formed by writing each valid rank number into its corresponding arrangement position according to the ascending order of the update values ​​in the current round. When the update values ​​are the same, the rank is assigned according to the ascending order of the rank numbers. In the same round, each valid rank number corresponds to one rank number, and each rank number corresponds to one valid rank number. This sequence is used to determine whether adjacent iterations have entered a repeated arrangement state.

[0048] In a preferred embodiment, the misalignment construction module further includes:

[0049] Based on phenological segmentation, fragment generation, and node connection rules, a candidate misaligned time series is reconstructed into a candidate misaligned knowledge graph and a candidate misaligned sorting table. The number of edges retained in the drought-affected relationship chain and the recovery relationship chain in the original knowledge graph are counted. The number of neighboring trees of the target forest stand that maintain the same predecessor and successor numbers in the initial sorting table and the candidate misaligned sorting table are counted. The first candidate misaligned time series is read in descending order of the number of retained edges, descending order of the number of neighboring trees, descending order of the number of shifted steps, and ascending order of the source number.

[0050] Among them, drought-affected relationship chains and recovery relationship chains refer to directed paths within the same phenological segment, from meteorological dryness segment to root zone water loss segment, and then from root zone water loss segment to canopy descent segment, used to represent the sequence of time when changes in air temperature and humidity are transmitted to root zone water loss and cause canopy descent; recovery relationship chains refer to directed paths within the same phenological segment, from root zone recovery segment to canopy return segment, used to represent the sequence of time when canopy recovery occurs after root zone water content increases; the number of retained edges in the candidate misaligned knowledge graph is counted one by one according to the edges in the above two types of relationship chains that have the same start point type, end point type, and connection direction as the original knowledge graph;

[0051] The first candidate misaligned time series corresponding to each source is replaced with the original source time series, while the time series of the remaining sources remain unchanged. Then, a misaligned knowledge graph and a misaligned sorting table are formed according to the processing rules of the knowledge fusion module. The misaligned knowledge graph and the misaligned sorting table are used as the time series counter-evidence input for the drought adaptability results of the target forest stand.

[0052] In a preferred embodiment, the evaluation module includes:

[0053] Based on the original knowledge graph and each misaligned knowledge graph, information processing and operational analysis in the forestry field are performed. Candidate node pairs are established between nodes with the same fragment type, source number and difference sign. Candidate edge pairs are established when both ends of the candidate node pairs are paired and the connection directions are the same. Then, the pairing order is generated according to the ascending order of the original knowledge graph node sequence number, the ascending order of the candidate misaligned knowledge graph node sequence number and the ascending order of the source number.

[0054] The process of information processing and operations research in the forestry field is as follows: The original knowledge graph and its corresponding misaligned knowledge graph are read according to the stand number. Segment nodes representing meteorological aridity, root zone water loss, canopy decline, root zone recovery, and canopy return are taken as forestry status objects. The segment type, source number, difference sign, and starting time number are written into the node field. Between the two knowledge graphs, only nodes with the same segment type, source number, and difference sign are allowed to form candidate node pairs, and each node is required to enter only one candidate node pair. When the two ends of a connection in the original knowledge graph and the two ends of a connection in the misaligned knowledge graph are paired, and the starting point type, ending point type, and direction of the two connections are consistent, candidate edge pairs are established. The order of reading candidate node pairs is determined according to the starting time number of the nodes in the original knowledge graph, the starting time number of the nodes in the misaligned knowledge graph, and the source number. A common subgraph is obtained through a combination search of candidate node pairs. The retention status of drought-affected relationships, recovery relationships, and target stand ranking neighbors in the common subgraph is written into the relationship retention record, serving as the operations research result for withdrawing or retaining the drought adaptability evaluation results.

[0055] The backtracking maximum common subgraph algorithm is executed along the pairing order. At each level, the search continues from two branches: adding candidate node pairs and skipping candidate node pairs. After adding a candidate node pair, the successor node pair that is duplicated with the already paired node or disrupts the direction of the candidate edge is deleted. The current branch ends when the sum of the current number of paired nodes and the remaining number of candidate nodes does not exceed the number of nodes already saved. The first common subgraph is retained in descending order of the number of node pairs, descending order of the number of candidate edge pairs, and ascending order of the pairing order.

[0056] The process of executing the backtracking maximum common subgraph algorithm is as follows: A search hierarchy is established based on the order in which candidate node pairs are read. At each level, two branches are generated: one for adding the current candidate node pair and one for skipping the current candidate node pair. The adding branch first checks whether the current candidate node pair shares original knowledge graph nodes or misaligned knowledge graph nodes with already added node pairs. If they do, the adding branch is deleted. If they do not, it checks whether the candidate edge pairs between the current candidate node pair and already added node pairs maintain their original connection direction. If the directions are inconsistent, the adding branch is deleted. If the directions are consistent, the current candidate node pair and already added node pairs are added together. The corresponding candidate edge pairs are written into the current common subgraph. For the retained branches, the sum of the number of node pairs in the current common subgraph and the number of candidate node pairs that have not yet been processed is calculated. If the sum is not greater than the number of node pairs in the saved common subgraph, the branch is terminated; otherwise, the next layer is read. After reaching the last candidate node pair, the current common subgraph is saved, and all saved results are sorted in descending order of the number of node pairs, descending order of the number of candidate edge pairs, and ascending order of the order in which candidate node pairs are read. The first result is read as the first common subgraph, thereby ensuring that the output result is unique and the search stops after all candidate node pairs are processed or the branch is pruned.

[0057] For the first common subgraph, read the connection from the root zone water loss segment to the canopy descent segment and the connection from the root zone recovery segment to the canopy return segment. Read the predecessor number and successor number of the target forest stand from the initial sorting table and the corresponding misaligned sorting table. Replace the predecessor number with the table head code at the beginning of the table and replace the successor number with the table tail code at the end of the table to form a relationship retention record.

[0058] When any relationship retention record contains two connections and the predecessor and successor numbers remain unchanged, the target forest stand's ranking in the initial ranking table is revoked and a time-series distortion record is output. When all relationship retention records are missing any connection or the predecessor or successor number has changed, the target forest stand's ranking in the initial ranking table is retained and a drought adaptability evaluation record is output.

[0059] The technical effects and advantages of this invention are as follows:

[0060] 1. This scheme constructs misaligned time series from source to source and compares two types of knowledge graphs to examine the dependence of ranking on time series relationships, thereby eliminating adaptive misjudgments caused by time series distortion.

[0061] 2. By using simultaneous sampling, time stamp correction, and unified multi-source sequence for stand attribution, mismatches caused by independent sampling and delayed data transmission are reduced, thus improving the temporal consistency of stand data.

[0062] 3. Perform random walk segmentation based on canopy reference pixels and ground surface reference pixels to extract canopy cover values ​​and reduce the interference of ground surface and shadow pixels on canopy characterization;

[0063] 4. Generate change segments according to phenological boundaries and connect drought-affected chains and recovery chains to make meteorological, root zone and canopy changes correspond in evolutionary order, thereby improving the pertinence of the evaluation;

[0064] 5. The network simplex is used to filter out duplicate assignments and incomplete connections, so that nodes enter a unique relation chain, reducing the interference of connection conflicts on the number of cycles to be sustained and the number of cycles to be recovered. Attached Figure Description

[0065] Figure 1 This is a flowchart of the drought adaptability assessment of Helan Mountain coniferous forests according to the present invention. Detailed Implementation

[0066] 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.

[0067] Refer to the instruction manual appendix Figure 1 The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data of the present invention includes:

[0068] The multi-source acquisition module includes a simultaneous capture controller and an observation terminal. After acquiring air temperature and humidity according to the simultaneous capture number, it writes the root zone water content and canopy images. The growth status code is added by the survey terminal. After locating according to the stand number, it is ordered according to the capture number and then sorted by source number to output the observation record.

[0069] Multi-source observation data need to be recorded in a unified manner under the same sampling order, the same spatial object, and the same survey conditions in order to be used for subsequent forest stand time series generation and drought adaptability knowledge integration;

[0070] In specific processing, the simultaneous shooting controller establishes a task number and shooting number, sequentially triggering the meteorological observation terminal, root zone observation terminal, and canopy observation terminal to obtain observation data. Then, the survey terminal supplements the growth status code corresponding to the sampling time, and subsequently completes the writing of forest compartment affiliation, forest stand number, and anomaly record processing. This implementation process includes the following steps:

[0071] The uniformity of sampling timing is accomplished by the synchronous controller; when the evaluation task starts, the synchronous controller generates a task number and sets the time number register to zero. Each time a sampling edge arrives, the time number is incremented, and the task number, the current time number, and the time of sampling edge occurrence are written into the time number table. At the same time, a trigger signal is sent to the meteorological observation terminal.

[0072] The meteorological observation terminal reads the air temperature and air humidity values, binds the two observation values ​​with the task number, time number, source number and local sequence number respectively, and returns the completion position. The time controller only receives the first completion position that is consistent with the current task number and the current time number, and triggers the root zone observation terminal and the canopy observation terminal accordingly.

[0073] If the completion bit is not received before the next sampling edge arrives, a missing sampling mark is written to the corresponding field under the current time number; a late completion bit only allows the meteorological record to be written back to the original time number, and will no longer trigger the root zone observation end and the canopy observation end.

[0074] After the time register reaches the last bit of the encoding, the next sampling edge increments the task number and returns the time to zero; when the same time controller restarts, it regenerates the task number so that the same time number before and after the restart can be distinguished.

[0075] The attribution of root zone data and canopy data is completed using task number and time stamp number. The root zone observation terminal sets up soil moisture probes based on the main root distribution soil layers determined by the forest stand survey. After receiving the current task number and time stamp number, it reads the root zone water content and combines the root zone water content, task number, time stamp number, source number, local sequence number and installation coordinates to form the root zone record.

[0076] The canopy observation terminal obtains canopy images based on the camera's intrinsic and extrinsic parameters and imaging orientation stored in the calibration version, and combines the canopy images, mission number, photo number, source number, local sequence number, and imaging range into a canopy record;

[0077] When the root zone observation terminal or the canopy observation terminal returns data after the next time number is generated, the returned data is still bound to the original task number and the original time number and a supplementary sampling mark is written in it;

[0078] When multiple records appear for the same task number, time stamp number, and source number, the first record in the receiving order is retained, and subsequent records are marked as duplicates. When the observation terminal returns a fault code but does not return any observation values, a missing data mark is written to the corresponding data field and the fault code is saved.

[0079] The time attribution of survey information is determined according to the survey time and survey record number; the survey terminal saves survey records according to the stand number and survey time, and each survey record is written with the survey record number, growth status code and growth stage table version;

[0080] When forming the current batch of data, first filter the survey records with the same stand number and the survey time is no later than the current sampling time, then read the last survey record in ascending order of survey time and survey record number, and write the growth status code and growth stage table version into the current batch number.

[0081] When there are no valid survey records before the current sampling line, the growth status code field is written with a status missing marker; when there are multiple survey records in the same stand at the same survey time, the record with the later survey record number is read and the replaced record number is saved;

[0082] After the reading is completed, the air temperature value, air humidity value, root zone water content value, canopy image and growth status code are bound into the same data group according to the task number and the photo number. The air temperature value and air humidity value are respectively assigned independent source numbers, and the growth status code is used for subsequent phenological segment division reading.

[0083] Spatial attribution and record verification are performed after the same-shot data set is formed; the system reads the observation terminal installation coordinates, canopy image imaging range, forest compartment closure boundary, and the forest compartment number and stand number mapping table that are effective at the time of acquisition, and converts the forest compartment number to which the installation coordinates fall into the stand number;

[0084] When the canopy image covers multiple forest compartments, the imaging range is projected onto the forest compartment coordinate system according to the camera calibration parameters. The canopy image is segmented along the intersection of the forest compartment boundary and the projection range. When the boundary passes through a pixel, the forest compartment to which the pixel center belongs is determined. When the pixel center is on the boundary line, it is assigned to the first forest compartment in ascending order of forest compartment number.

[0085] Each segmented image is paired only with air temperature, air humidity, and root zone water content values ​​for the same task number, same photo number, and installation coordinates corresponding to the same forest stand number; when the corresponding source is missing, a missing data mark is written into the same photo data group of the target forest stand;

[0086] After pairing is completed, the record key is composed of stand number, task number, shot number and source number. Late data is written back to the original record key and marked as re-collected. Finally, the observation record is formed according to the order of stand number and shot number.

[0087] Taking a collection process spanning two forest compartments as an example: the simultaneous shooting controller generates task number 2026072601 and sampling number 156. After the meteorological observation terminal returns the air temperature value, air humidity value and completion position, both the root zone observation terminal and the canopy observation terminal perform sampling according to sampling number 156. The survey terminal reads the last growth status code 02 at the survey time no later than the sampling edge and writes it into the simultaneous shooting data group.

[0088] When the canopy image projection covers both forest compartment 031 and forest compartment 032, the system segments the canopy image along the projection intersection line of the two forest compartment boundaries and reads the forest stand number corresponding to each of the two forest compartments. When forest compartment 032 lacks root zone water content value in photo number 156, a missing collection mark is written. When the root zone observation end returns the data of photo number 156 after photo number 157 is generated, the system writes the returned data back to the record key corresponding to photo number 156 and writes a supplementary collection mark, thus forming an observation record that can be traced in time, space and source.

[0089] The aggregation module, including a communication gateway and a storage device, is used to correct the time stamp number, write the stand number according to the location, extract the canopy cover value from the canopy image, and then aggregate it according to the stand number and write it according to the time stamp number to form the stand time series.

[0090] The aggregation processing is used to restore the observation records returned by each observation terminal to a unified sampling order, and to complete the calculation of stand attribution, canopy cover value and stand time sequence writing based on the corrected time number;

[0091] The communication gateway first reconstructs the corrected time number corresponding to the observation record using the time number table and local sequence number, then determines the stand number based on the acquisition location and the active forest compartment-stand mapping relationship, subsequently performs random walk segmentation on the canopy image, and finally the storage device saves the data from each source according to the recording key; this implementation process includes the following steps:

[0092] The time stamp reconstruction begins with the source number, task number, local priority number, and carried time stamp number of the observation record; after each observation end completes one sampling, the local priority number is incremented by one bit, and the incremented local priority number is written into the observation record.

[0093] The communication gateway collects observation records according to the source number and task number, reads the sampling edge occurrence time, step number and step number wrapping number of the same source from the step number table, and combines the task number, step number wrapping number and step number register value into an expanded step number.

[0094] The first observation record from the same source uses the expanded sequence number that matches the task number, source number, and carried sequence number in the sequence number table as the first corrected sequence number. Subsequent records calculate the modulus difference between the current local sequence number and the previous confirmed local sequence number according to the local sequence number coding capacity, and add the local sequence difference to the previous corrected sequence number to obtain the current corrected sequence number.

[0095] When the local sequence number already exists in the source sequence mapping table, the communication gateway reads the original correction sequence number and treats the current record as a late or duplicate record; after the observation end restarts, the new task number is used to re-establish the anchor point of the first record to avoid the local sequence number returning to zero being interpreted as a missed sample;

[0096] When the current corrected time number is inconsistent with the unfolded time number corresponding to the carried time number, the time number conflict bit is written with the source number, carried time number and corrected time number, and subsequent processing reads the corrected time number uniformly.

[0097] Spatial attribution is performed based on the collection location and collection time in the calibration record; the installation coordinates of the point observation terminal are first converted to the coordinate system used by the forest compartment boundary, and then the forest compartment number is read through the inclusion relationship between the point and the closed boundary; when the collection location is on the forest compartment boundary line, the first forest compartment number is read in ascending order according to the forest compartment numbers on both sides of the boundary.

[0098] The communication gateway then reads the mapping version that is within the effective range at the time of collection from the forest compartment and stand mapping table, and converts the forest compartment number into the stand number; when the same forest compartment corresponds to different stand numbers in different survey periods, the mapping version corresponding to the time of collection shall prevail.

[0099] When the mapping version is missing or the collection location does not fall within the boundary of any forest stand, the correction record is written with the missing attribution marker, and the writing of forest stand time series is paused;

[0100] When the canopy image spans multiple forest compartments, the image pixels are projected onto the forest compartment coordinate system based on the camera's internal and external parameters and the topographic surface corresponding to the canopy observation end. Each forest compartment projection area is formed according to the forest compartment to which the pixel center belongs. When the pixel center is located on the forest compartment boundary line, it is still assigned to the first forest compartment in ascending order of forest compartment number, thus forming a forest stand record with a unique forest stand number.

[0101] The canopy cover value is calculated separately for each forest compartment projection area; the system reads the reference pixel version whose acquisition time is no later than the canopy image acquisition time and whose version number is later, and converts the canopy reference pixels and ground reference pixels saved in the version to the current canopy image coordinates;

[0102] Canopy reference pixels represent coniferous forest canopy pixels confirmed by the stand survey, while ground reference pixels represent bare ground, understory vegetation, or shaded background pixels confirmed by the stand survey.

[0103] A node is established for each effective pixel in the forest projection area. Nodes with row or column numbers that differ by one bit are connected. The connection weight is taken as one divided by one plus the sum of the absolute values ​​of the differences between the normalized values ​​of the red, green, and blue channels of the two pixels.

[0104] For each pixel connected region, the canopy arrival probability of the canopy reference pixel is fixed at 1, and the canopy arrival probability of the ground reference pixel is fixed at zero. For the remaining nodes, an equation is established in which the node probability is equal to the weighted average of the probabilities of the adjacent nodes according to the connection weight. The unknown nodes are arranged in ascending order of pixel row number and column number to form a system of Laplace equations, which are then solved using sparse elimination.

[0105] When a connected region contains two types of reference pixels, the pixel type is determined based on the solution results; when it contains only canopy reference pixels, the connected region is written as canopy pixels; when it contains only surface reference pixels, the connected region is written as surface pixels; when it does not contain reference pixels or zero principal components appear in the solution, a segmentation missing marker is written.

[0106] Pixels with a higher probability of canopy arrival than the probability of ground arrival are denoted as canopy pixels. When the two probabilities are equal, they are denoted as ground pixels in ascending order of row and column number. The canopy cover value is obtained by dividing the number of canopy pixels by the number of effective pixels in the segmentation. When the number of effective pixels is zero, the canopy cover value is written into the missing sampling mark.

[0107] Before writing to the storage device, the communication gateway first establishes a record key based on the stand number, correction time number and source number, where the correction time number already includes the task number and unfolded time number information;

[0108] When there is no valid data for the same record key, the correction record, stand number and canopy cover value are written to the corresponding fields; if no pre-registered source data is received before the next correction sequence number is formed, the storage device establishes a missing record and retains the empty value position.

[0109] When late data arrives, the communication gateway locates the original record key through the source order mapping table. It fills in the observation value and writes the supplementary record mark only when the original record key is a missing record, and the record order remains unchanged. When the original record key already has valid data, the late data is written to the duplicate record area without overwriting the first data in the receiving order.

[0110] After completing the record check of each source, the storage device collects the records according to the stand number, and then forms the stand time series according to the ascending order of the correction time number and the ascending order of the source number; when the supplementary collection of records causes the original null value to change, only the data field of the corresponding source under the original correction time number is updated, and subsequent knowledge fusion is still read with the original correction time number;

[0111] Taking a single time number loop and delayed feedback process of a certain forest stand as an example: the local sequence number of the root zone observation end returns to zero from 65,535, the local sequence number encoding capacity is 65,536, the local sequence difference calculated by the communication gateway is one, so the current corrected time number is incremented by one bit based on the previous corrected time number.

[0112] Subsequently, the root record of local sequence number 104 arrived later than local sequence number 105. The source sequence mapping table has saved the original correction sequence number for 104, and the communication gateway writes the late data back to the original missing record accordingly.

[0113] In a concurrent canopy image, a pixel connected region that does not contain any reference pixels is written with a segmentation missing marker and deducted from the denominator of the canopy cover value, ultimately forming a stand time series that can be verified in terms of photo order, stand affiliation, and canopy cover value.

[0114] The knowledge fusion module is used to perform information processing and knowledge fusion in the forestry field. It divides phenological segments based on the reversal of canopy cover difference and changes in growth status codes, merges the same-direction differences into change segments, first connects meteorological aridity to root zone water loss, then connects root zone water loss to canopy descent, and finally connects root zone recovery to canopy return, forming an original knowledge graph. It generates the number of withstand beats and the number of recoverable beats by connecting the beat number differences at both ends, and forms an initial sorting table by descending order of the number of withstand beats and ascending order of the number of recoverable beats.

[0115] After the forest stand time series enters the knowledge integration stage, it is necessary to convert multi-source observation changes into phenological segments, change fragments and directed relational chains, and eliminate duplicate fragment assignments through cost flow to form an original knowledge graph that can calculate the number of sustained beats and the number of recovered beats;

[0116] Air temperature, air humidity, root zone water content, canopy cover, and growth status codes are all collected according to stand numbering, and subsequent catalog numbers will uniformly refer to and correct the catalog numbers; this implementation process includes the following steps:

[0117] Phenological boundaries are obtained based on canopy cover difference and growth status code; within the same forest stand, the difference between adjacent canopy cover values ​​is calculated according to the corrected time number; when the difference is zero, the previous non-zero difference sign is inherited; when the non-zero difference sign is reversed, the subsequent time number is recorded as the reversed time number; when a missing record occurs, the sign inheritance is interrupted.

[0118] When the current valid growth status code is different from the previous valid growth status code, the current beat number is recorded as a changed beat; the first valid beat number, the last valid beat number, the reversal beat, and the changed beat are merged into a candidate segment boundary, only one duplicate beat number is retained, and the changed beat is written into the mandatory marker;

[0119] Candidate segment boundaries are formed into phenological segments through interval dynamic programming; each candidate segment boundary is arranged in ascending order of the beat number, and the previous candidate segment boundary can connect to any subsequent candidate segment boundary, but cannot cross the change beat with the mandatory mark; each edge reads the number of reversals and the number of missing beats in the interval, and the edge cost is formed by the product of the two plus one.

[0120] Starting from the first valid bid number, the cumulative edge cost is calculated node by node. When there are multiple paths, the first path is read in the order of cumulative edge cost, number of segment boundaries, and segment boundary bid number. Then, the closed and open intervals between adjacent segment boundaries in the path are written as phenological segments. If the first or last valid bid number is missing, a phenological segment shortage mark is written.

[0121] The difference between adjacent images from multiple sources is then converted into change segments; when the air temperature difference is positive and the air humidity difference is negative, a meteorological dryness difference is formed; when the root zone water content difference is negative or positive, a root zone water loss difference and a root zone recovery difference are formed respectively; when the canopy cover difference is negative or positive, a canopy descent difference and a canopy return difference are formed respectively.

[0122] When the difference types and signs are consistent and share the same correction time number, the segments are merged. The starting point of the segment is taken from the time number before the first difference, and the ending point is taken from the time number after the last difference. The current segment ends when the difference is zero, the sign changes, the segment crosses the phenological segment boundary, or the relevant source is missing.

[0123] The meteorological dry segment uses a combined source number composed of the air temperature source number and the air humidity source number;

[0124] After the change fragments are generated, a candidate knowledge graph is established according to the forestry drought process; each fragment node is written with the forest stand number, phenological segment number, fragment type, source number, starting time number and ending time number, and the absolute value of the difference within the fragment is accumulated as the fragment amplitude;

[0125] For the arid meteorological segment, the rank of temperature amplitude and the rank of humidity amplitude are calculated separately, and then the sum of the two is used to form the amplitude rank; for other segments within the same forest stand, the same phenological segment, and the same segment type, ranks are assigned in ascending order of amplitude, and if the values ​​are the same, they are determined in ascending order of the segment's starting time number.

[0126] When the starting point of the meteorological dryness segment is earlier than the starting point of the root zone water loss segment, a first-class candidate edge is established; when the starting point of the root zone water loss segment is earlier than the starting point of the canopy descent segment, a second-class candidate edge is established; and when the starting point of the root zone recovery segment is earlier than the starting point of the canopy return segment, a third-class candidate edge is established.

[0127] The unique ownership of a segment is determined through a cost flow network; each segment node is split into an ingress node and an egress node, and connected by nodes with a capacity of one; the source node connects the arid meteorological segment and the root zone rising segment, the canopy descending segment and the canopy returning segment connect the sink node, and the three types of candidate edges are written between the corresponding nodes;

[0128] The candidate edge cost is the sum of the difference in the starting point number at both ends, the absolute value of the difference in amplitude rank, and the corresponding source missing number of samples. Among them, the first type of candidate edge is calculated based on air temperature, air humidity, and root zone water content, while the second and third types of candidate edges are calculated based on root zone water content and canopy cover.

[0129] The supply of the source node is the sum of the number of dry meteorological segments and the number of rising segments in the root region, and the demand of the sink node is the same; a discard edge is added from each starting segment to the sink node, and the cost of the discard edge is the sum of the costs of all candidate edges in the current phenological segment plus one.

[0130] The network simplex algorithm forms an initial basis tree by discarding edges that carry unit flow, and fixes the source node potential to zero; it calculates the node potential based on the edge cost of the basis tree, calculates the reduced cost of non-basis edges that are below the capacity in the original direction, and calculates the reduced cost of non-basis edges that are above the capacity in the reverse residual direction.

[0131] When there is a negative reduction cost, the first non-base edge is read into the base in ascending order of reduction cost and edge number, and the flow is increased or decreased along the formed loop until a loop edge reaches the upper or lower limit of capacity. Then the first edge to the limit is read out of the base in ascending order of edge number.

[0132] Zero augmentation is performed with the same order of the base edges until all residual non-base edges are reduced to a non-negative cost. The flow path from the source node to the sink node without discarding edges is read as the relation chain.

[0133] After solving the cost flow problem, forward reachability markers are calculated in ascending order of node time number, and then reverse reachability markers are calculated in descending order of node time number. Only candidate edges that have both forward and reverse reachability markers are retained.

[0134] After deleting the remaining candidate edges, re-execute the cost flow solution. Stop when the set of edge numbers retained in this round is consistent with that of the previous round, and output the original knowledge graph; if the set of retained edges is empty, write a sign indicating insufficient relational chains, and stop generating the initial sorting record for this forest stand.

[0135] After the original knowledge graph stabilizes, the meteorological aridity segment, the root zone water loss segment, and the canopy decline segment are connected in sequence to form a drought-related chain, and the root zone recovery segment and the canopy return segment are connected to form a recovery chain.

[0136] Each drought-affected relationship chain and its subsequent recovery relationship chain, which has not yet been included in other drought processes, constitute the drought process record; the difference in beat number between the starting point of the water loss segment in the root zone and the starting point of the canopy descent segment, and the difference in beat number between the starting point of the recovery segment in the root zone and the starting point of the canopy return segment.

[0137] When there are multiple drought process records in the same forest stand, the first record is read in ascending order of the number of drought-affected shots, descending order of the number of recovery shots, and ascending order of the starting shot number of the drought-affected relationship chain; all forest stands are then sorted in descending order of the number of drought-affected shots, ascending order of the number of recovery shots, and ascending order of the forest stand number to form an initial sorting table;

[0138] Taking forest stand number HL031 as an example, the candidate segment boundaries are the corrected beat numbers 20, 28, 35 and 47, among which beat number 35 is a necessary change beat. Interval dynamic programming reads 20, 35 and 47 to form two phenological segments;

[0139] The preceding phenological phase forms a drought-affected relationship chain, and the following phenological phase forms a recovery relationship chain. The starting point of the root zone water loss segment is 24, the starting point of the canopy decline segment is 29, and the number of beats is 5. The starting point of the root zone recovery segment is 38, the starting point of the canopy return segment is 41, and the number of recovery beats is 3.

[0140] This record, after being sorted together with other forest stands, is written into the initial sorting table, providing the original knowledge graph for subsequent misalignment construction and temporal proof by contradiction.

[0141] The misalignment construction module is used to execute the iterative amplitude adjustment Fourier method, shift the time sequence source by source, replace the shifted amplitude with the original Fourier amplitude and retain the phase, and after the inverse transformation, fill the original value according to the rank of the updated value. The same value is ordered according to the beat number. Iterate until the rank sequence first coincides to form a misalignment time sequence, and generate a misalignment knowledge graph and misalignment sorting table according to the knowledge fusion rules.

[0142] Temporal proof by contradiction requires changing the temporal attribution while maintaining the set of observations from a single source and the distribution of missing data, and then examining whether the drought relationship of the forest stand and the ordinal neighbor can continue to hold.

[0143] The misalignment construction module takes the target phenological segment that forms the initial sorting record as its processing scope, and sequentially changes the air temperature time series, air humidity time series, root zone water content time series, and canopy cover time series. The growth status code is only used to preserve the phenological segment boundaries and is not included in the spectrum calculation. This implementation process includes the following steps:

[0144] The target misalignment range is first determined by the phenological segment boundary; the system reads the drought process number from the original knowledge graph to generate the target forest stand sorting record and locates the target phenological segment where the drought process is located; the forward shift is obtained by subtracting the starting point of the previous phenological segment from the starting point of the target phenological segment, and the backward shift is obtained by subtracting the starting point of the target phenological segment from the starting point of the next phenological segment.

[0145] When the target phenological segment is at the beginning, only the shift amount is generated backward; when it is at the end, only the shift amount is generated forward; when the shift amount is zero, the corresponding result is deleted.

[0146] Each time a source time series is fixed, the data position of that source is cyclically moved within the time number interval defined by the start of the previous phenological segment to the end of the next phenological segment, while data outside the interval, other source time series and other stand records remain unchanged;

[0147] Missing sampling locations are used as zero-value placeholders in spectrum calculations, and a missing sampling location table is saved simultaneously. When the shift result is consistent with the original source time series step by step, no candidate result is formed.

[0148] Frequency domain rearrangement is constrained by the statistical structure of the original source time series; the system performs discrete Fourier transform on the original source time series with zeros filled in the missing sampling positions, saves the modulus of the complex result of each frequency bit as the original Fourier amplitude, and forms an ascending table of original values ​​by sorting all valid sampled values ​​in ascending order of numerical value and original time number.

[0149] Subsequently, the source-by-source shift timing is used as the first round timing, the discrete Fourier transform is performed and the phase of each frequency bit is read, the original Fourier amplitude is multiplied by the corresponding phase to form the replacement spectrum, and then the updated value is obtained by the inverse discrete Fourier transform.

[0150] The zero-frequency bit is kept as a real number, and the positive and negative frequency bits are written simultaneously according to the conjugate relationship. When the number of samples is even, the Nyquist frequency bit is saved separately, and the frequency bit with the original Fourier amplitude of zero is set to zero.

[0151] After the rank backfill is completed, the next round of timing is formed; the system only sorts the updated values ​​outside the missing sampling position table in ascending order of numerical value, and determines the rank of the same value in ascending order of correction beat number. Then, each original value in the original value ascending order table is written into the beat number position of the corresponding rank position in sequence. The missing sampling position is restored with the missing sampling mark and continues to occupy the position with zero value in the next round of spectrum calculation.

[0152] In each round, the rank sequence is saved according to the corrected time number order, and an accessed table is built using the rank sequence encoding. When the encoding in this round is the same as the saved encoding, the complete rank sequence is checked bit by bit. After the check is consistent, the iteration stops, and the time sequence corresponding to the first round of the repeated rank sequence is read as the candidate misaligned time sequence.

[0153] The candidate misaligned time series strictly preserves the original value set and missing sampling positions. The original Fourier amplitude is saved as a spectrum replacement constraint for each round along with the candidate results. At the same time, the spectrum residual is recorded as the sum of the absolute values ​​of the differences between the amplitude of each frequency bit of the candidate misaligned time series and the original Fourier amplitude.

[0154] After the candidate misalignment time series is generated, the target forest stand is recalculated according to the phenological segmentation, change segment generation, candidate edge establishment and cost flow screening rules used by the knowledge fusion module to form a candidate misalignment knowledge graph; other forest stands use the sorting records in the original knowledge graph, and then the re-obtained bearing rate and recovery rate of the target forest stand are used for unified sorting to form a candidate misalignment sorting table.

[0155] When the target forest stand has not formed a complete drought process, write the unlisted code in the candidate misaligned sorting table; the system uses the three connections generated in the original knowledge graph from meteorological dryness to root zone water loss, root zone water loss to canopy descent, and root zone recovery to canopy return as the verification object, and counts the number of edges retained in the candidate misaligned knowledge graph where the start point type, end point type, and connection direction are all consistent.

[0156] Simultaneously compare the predecessor and successor numbers of the initial sorting table with the candidate misaligned sorting table. If the predecessors are the same, record one digit; if the successors are the same, record one digit. The first and last codes of the table are compared according to their corresponding positions. The adjacent digits of the codes not included in the table are recorded as zero.

[0157] After candidate verification is completed, the first candidate misalignment sequence is read in descending order of the number of retained edges, descending order of the number of adjacent bits, descending order of the number of shifts, and ascending order of the source number. If all the above fields are the same, the first result is determined in ascending order of the candidate number, in the order of the first shift, the last shift, and the candidate number.

[0158] The first candidate misaligned time series of each source is used as an independent proof of contradiction input. Each time, only the original source time series of that source is replaced, while the time series of other sources and other forest stand records remain unchanged. Then, the corresponding misaligned knowledge graph and misaligned sorting table are formed according to the knowledge fusion rules.

[0159] If a source fails to form a candidate due to insufficient effective sampled values, zero shift amount, or all shift results being consistent with the original source timing, a timing insufficiency flag is written to that source, and it is not shifted simultaneously with other sources.

[0160] Taking the target forest stand numbered HL031 as an example, the starting points of its adjacent phenological segments are sequentially numbered 20, 35 and 47, and the starting point of the target phenological segment is 35. Thus, the forward shift is 15 and the backward shift is 12. After the air temperature time series is shifted forward by 15 beats, after the spectrum replacement and rank backfilling iterations to the sixth round, the rank sequence is consistent with the third round. The system reads the time series of the third round as the candidate misaligned time series and saves the original ascending list, the original Fourier amplitude and the spectrum residual.

[0161] After the candidate results are regenerated into a candidate misalignment knowledge graph and a candidate misalignment ranking table, they are sorted together with other candidate results from the same source. The first and second results are read to form a set of independent misalignment knowledge graphs and misalignment ranking tables, thus providing a verifiable input for the evaluation module to examine the degree of dependence of drought adaptability ranking on time misalignment.

[0162] The evaluation module is used to perform operations research or analysis using a backtracking maximum common subgraph algorithm. Among nodes with the same fragment type and source number, they are paired according to the difference sign and the connection direction. After backtracking and pruning of remaining nodes, a common subgraph is obtained. When the common subgraph retains the dehydration to descending connection and the recovery to return connection and the adjacent stand number has not changed, the initial sorting is canceled and the time-series distortion record is output. Otherwise, the drought adaptability evaluation record is output.

[0163] The evaluation phase is responsible for the structural verification and ranking stability judgment of the time-series proof results. By comparing the forestry state nodes, relationship directions and stand ranking neighbors in the original knowledge graph and each misaligned knowledge graph, it is possible to identify whether the initial ranking depends on the original observation time series.

[0164] The processing objects include the original knowledge graph, each misaligned knowledge graph, the initial ordination table, each misaligned ordination table, and the drought process number for generating the target forest stand ordination record. The processing results are written into the relation retention record, and based on this, time-series distortion records, drought adaptability evaluation records, or time-series evidence of insufficiency records are output. The implementation process includes the following steps:

[0165] Comparable objects between knowledge graphs are first determined by node fields; the system reads the original knowledge graph and a misaligned knowledge graph according to the stand number, and converts the meteorological aridity segment, root zone water loss segment, canopy decline segment, root zone recovery segment and canopy return segment into forestry status nodes respectively, and writes the node number, segment type, source number, difference symbol, starting time number and phenological segment number;

[0166] Nodes in two knowledge graphs form candidate node pairs only when the fragment type, source number, difference sign, and phenological segment number are consistent.

[0167] During the candidate phase, the same node can correspond to multiple candidate nodes. After a node pair is added to a common subgraph, the same original knowledge graph node and the same misaligned knowledge graph node are only paired once. Nodes with missing fields or different phenological segment numbers are not included in the candidate node pair set.

[0168] Around the target drought process that generates the initial sorting record, the evaluation module reads three target connections from the original knowledge graph: meteorological aridity to root zone water loss, root zone water loss to canopy descent, and root zone recovery to canopy return. It then searches for candidate node pairs in the misaligned knowledge graph that match the fragment type, source number, and difference sign at both ends of each connection.

[0169] When both ends of the original connection and both ends of the misaligned connection have been paired, and the start point type, end point type, and connection direction are consistent, a candidate edge pair is established; the candidate node pair combinations corresponding to the three target connections are written into the target pairing table for backtracking search as mandatory pairings to be read.

[0170] When no corresponding candidate edge pair exists for any target connection, a target relation missing marker is written into the current misaligned knowledge graph;

[0171] The combined search of candidate node pairs is completed using a backtracking maximum common subgraph algorithm. The system generates a pairing order in ascending order of the starting point number of the original knowledge graph node, the starting point number of the misaligned knowledge graph node, the source number, the original node number, and the misaligned node number, and records a set of target pairing tables as the initial common subgraph. Then, the remaining candidate node pairs are read layer by layer, and the addition branch and skip branch are established respectively.

[0172] If duplicate node pairings occur when adding a branch, the process ends directly. If no duplicates occur, check if there are candidate edge pairs with the same direction between the current node pair and the already added node pair. Node pairs and candidate edge pairs that meet the connection relationship are written into the current common subgraph.

[0173] The common subgraph uses a non-induced approach; if there is an unselected connection between two paired nodes, it does not affect the current branch; skipping branches preserves existing pairings and reads the next level.

[0174] The search scale is controlled by pruning the remaining nodes; each branch calculates the sum of the number of currently paired nodes and the number of subsequent unprocessed candidate nodes. The current branch ends when the sum does not exceed the number of nodes in the saved common subgraph; if the sums are the same, the current number of candidate edge pairs plus the number of subsequent candidate edge pairs that can be formed is used as the second upper bound. The current branch also ends when the upper bound does not exceed the number of saved candidate edge pairs.

[0175] After all candidate node pairs are processed, the current common subgraph is saved, and the search results are sorted in descending order of the number of node pairs, descending order of the number of candidate edge pairs, and ascending order of the order of complete pairings. Each target pairing table is searched once, and the search stops after all target combinations have been processed or have been pruned.

[0176] To eliminate the influence of common subgraphs of the same size on the judgment of target relationships, the system only reads the first common subgraph in the search results containing three target connections in the target pairing table; when there are multiple results with the same number of node pairs and candidate edge pairs, the system compares the order of complete pairings one by one and reads the result with the first order.

[0177] When the target pairing table is empty, or when all target combinations are deleted during the search due to node duplication or direction conflict, other common subgraphs of the same size are not used to replace the target results. Instead, the corresponding misaligned knowledge graph is written as missing target relationship, thereby maintaining a consistent correspondence between the common subgraph and the target drought process.

[0178] After the common submap is determined, the system forms a relationship retention record. The record first writes the pre-retention position of the connection from meteorological aridity to root zone water loss, then writes the evaluation retention positions of the connection from root zone water loss to canopy descent and the connection from root zone recovery to canopy return, and reads the predecessor number and successor number of the target forest stand from the initial sorting table and the corresponding misaligned sorting table.

[0179] When the target forest stand is at the beginning of the table, the preceding number is replaced by the first number of the table; when it is at the end of the table, the following number is replaced by the last number of the table. If it is not included in the misaligned sorting table, the not-in-table number is written.

[0180] The relationship retention record also saves the stand number, target drought process number, source number of the displaced area, displacement direction, number of displacement steps, number of common subgraph node pairs, and number of candidate edge pairs for result verification;

[0181] After summarizing the relationship retention records from all sources, an evaluation judgment is performed; if any valid relationship retention record simultaneously satisfies the following conditions: the preceding retention position is valid, both evaluation retention positions are valid, the initial predecessor number is consistent with the misplaced predecessor number, and the initial successor number is consistent with the misplaced successor number, the target forest stand is revoked from the initial sorting table, and the time-series distortion record is output.

[0182] If all valid relation retention records show missing target connections or any ordering neighbor changes, the initial ordering is retained and drought adaptability evaluation records are output.

[0183] If a source fails to form a candidate misaligned time series, the target pairing table is empty, or the common subgraph search cannot obtain the target result, a time series proof of insufficiency flag is written for that source; if all sources are time series proof of insufficiency, a time series proof of insufficiency record is output, and the initial sorting is not confirmed for the time being.

[0184] Based on a misalignment result of stand number HL031, all three target connections in the original knowledge graph are paired in the first common subgraph. The predecessor number in both the initial sorting table and the misalignment sorting table is HL026 and the successor number is HL044. Therefore, the corresponding relationship retention record satisfies the connection and adjacent position retention conditions. The system cancels the initial sorting of HL031 and writes the source number of the shift, the shift direction, and the number of shift steps. If the root zone loses water after the canopy misalignment, leading to the canopy descent and the connection disappearing, or if the successor number of HL031 becomes HL052, then the relationship retention record cannot be canceled.

[0185] By using target relationship constraints, unique node pairing, backtracking pruning, and sorting neighbor verification, the evaluation results can be traced back to the specific source time sequence and the corresponding forestry drought process.

[0186] Overall Implementation and Explanation: This scheme first triggers the meteorological observation terminal, root zone observation terminal, and canopy observation terminal in a unified manner by the synchronous controller, and the survey terminal adds the growth status code to form a synchronous observation record; the communication gateway then corrects the sequence number, determines the stand affiliation, and extracts the canopy cover value from the canopy image to form a stand time sequence arranged by stand number and corrected sequence number;

[0187] The knowledge fusion module divides phenological segments based on changes in canopy cover and growth status, converting meteorological aridity, root zone water loss, canopy decline, root zone recovery, and canopy return into segment nodes. It then uses cost flow screening to form drought-affected relationship chains and recovery relationship chains, and calculates the number of withstand beats, the number of recovery beats, and the initial ranking accordingly.

[0188] Subsequently, the time series of individual sources are shifted one by one. Under the condition of maintaining the original value set and missing sampling position, a misaligned knowledge graph and misaligned sorting table are generated. Then, the original relationship and the misaligned relationship are compared by the backtracking maximum common subgraph algorithm. If the relationship chain and sorted adjacent position are still maintained after the misalignment, it is determined that the initial sorting is affected by the time misalignment and is cancelled. Otherwise, the drought adaptability evaluation results are retained.

[0189] For example, in a coniferous forest monitoring area in Helan Mountain, the same-shot controller collects air temperature and humidity, root zone water content and canopy images under the same shot number, and the survey terminal simultaneously writes the current growth status code; after the system completes the shot number correction and forest stand classification, it identifies the sequential relationship between continuous air dryness, decreased root zone water content, decreased canopy cover and subsequent root zone recovery and canopy return, and calculates the number of shots that the forest stand can withstand and the number of shots that it can recover.

[0190] The system then moves the time series of air temperature, air humidity, root zone water content, and canopy cover respectively, and regenerates the knowledge graph and ordination table. If the original drought-affected relationship chain, the recovery relationship chain, and the previous and next ordination positions of the forest stand remain unchanged after moving a certain source, the system records that the ordination is not dependent on the original observation time series and withdraws the result. If these relationships or ordination positions change, the drought adaptability evaluation record of the forest stand is retained.

[0191] The above description is merely 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 drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data, characterized in that, include: The multi-source acquisition module includes a simultaneous capture controller and an observation terminal. After acquiring air temperature and humidity according to the simultaneous capture number, it writes the root zone water content and canopy images. The growth status code is added by the survey terminal. After locating according to the stand number, it is ordered according to the capture number and then sorted by source number to output the observation record. The aggregation module, including a communication gateway and a storage device, is used to correct the time stamp number, write the stand number according to the location, extract the canopy cover value from the canopy image, and then aggregate it according to the stand number and write it according to the time stamp number to form the stand time series. The knowledge fusion module is used to perform information processing and knowledge fusion in the forestry field. It divides phenological segments based on the reversal of canopy cover difference and changes in growth status codes, merges the same-direction differences into change segments, first connects meteorological aridity to root zone water loss, then connects root zone water loss to canopy descent, and finally connects root zone recovery to canopy return, forming an original knowledge graph. It generates the number of withstand beats and the number of recoverable beats by connecting the beat number differences at both ends, and forms an initial sorting table by descending order of the number of withstand beats and ascending order of the number of recoverable beats. The misalignment construction module is used to execute the iterative amplitude adjustment Fourier method, shift the time sequence source by source, replace the shifted amplitude with the original Fourier amplitude and retain the phase, and after the inverse transformation, fill the original value according to the rank of the updated value. The same value is ordered according to the beat number. Iterate until the rank sequence first coincides to form a misalignment time sequence, and generate a misalignment knowledge graph and misalignment sorting table according to the knowledge fusion rules. The evaluation module is used to perform operations research or analysis using a backtracking maximum common subgraph algorithm. Among nodes with the same fragment type and source number, they are paired according to the difference sign and the connection direction. After backtracking and pruning of the remaining nodes, a common subgraph is obtained. When the common subgraph retains the dehydration to descending connection and the recovery to return connection and the adjacent stand number has not changed, the initial sorting is canceled and the time-series distortion record is output. Otherwise, the drought adaptability evaluation record is output.

2. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 1, characterized in that: The multi-source acquisition module includes: The synchronous controller generates an incremental time number and sends the sampling edge to the meteorological observation terminal. After the meteorological observation terminal reads the air temperature and air humidity values, it writes back the completion bit. The synchronous controller triggers the root zone observation terminal and the canopy observation terminal with the same time number based on the completion bit to obtain the root zone water content and canopy image. After obtaining the root zone water content and canopy image, the survey terminal reads the last growth status code at a time no later than the current sampling edge and binds the air temperature value, air humidity value, root zone water content, canopy image, and growth status code into a co-shot data group.

3. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 2, characterized in that: The multi-source acquisition module also includes: Based on the installation coordinates of each observation terminal, the corresponding forest compartment is determined. When the canopy image crosses the forest compartment boundary, the canopy image is segmented according to the intersection of the imaging range and the forest compartment boundary. The segmented canopy images are written into the corresponding forest stand number so that all data in the same data group point to the same forest stand. For the same batch of data for which the stand number has been written, record checks are performed according to the batch number and source number. If multiple records appear from the same source under the same batch number, the first record in the receiving order is retained. Data that is not returned before the next batch number is generated is marked as missing. Late data is bound to the original batch number and marked as supplementary. Then, observation records are formed according to the stand number, batch number, and source number.

4. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 3, characterized in that: The aggregation module includes: After receiving the observation record and the time number table written by the time controller, the communication gateway reads the local order difference of adjacent observation records according to the source number, and determines the current corrected time number by the sum of the previous corrected time number and the local order difference; if the current corrected time number is inconsistent with the time number carried in the observation record, the time number conflict bit is written and the current corrected time number is retained to form a correction record. Based on the collection location in the calibration record, read the forest compartment boundary. When the collection location falls inside the forest compartment boundary, write the corresponding forest stand number. When the collection location falls on the forest compartment boundary line, write the first forest stand number in ascending order of the forest compartment numbers on both sides of the boundary to form a forest stand record.

5. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 4, characterized in that: The aggregation module also includes: For canopy images in the forest stand record, canopy reference pixels and ground reference pixels saved during camera calibration are read in the corresponding forest compartment projection area. A pixel adjacency map is established according to the channel difference between adjacent pixels and random walk segmentation is performed. Pixels with a higher probability of reaching canopy reference pixels than those with a higher probability of reaching ground reference pixels are recorded as canopy pixels. If the probabilities are the same, they are recorded as ground pixels in ascending order of pixel row number and column number. The canopy cover value is obtained by dividing the number of canopy pixels by the number of pixels in the forest compartment projection area. After the canopy cover value is calculated, the storage device uses the stand number, correction sequence number and source number to form a record key. The correction record, stand number and canopy cover value are written to the corresponding record key. The supplementary record is written back to the original record key without changing the order of subsequent records. The missing record is left empty. Then, the records are collected according to the stand number and the corrected sequence number is used to form the stand time sequence.

6. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 5, characterized in that: The knowledge fusion module includes: Based on the forest stand time-series execution of information processing and knowledge fusion in the forestry field, the canopy cover difference reversal beat and growth status code change beat are read as candidate segment boundaries. The candidate segment boundaries are used as nodes, and the beat number intervals between adjacent candidate segment boundaries are used as edges. The edge cost is formed by the difference reversal number and the number of missing beats within the interval. The growth status code change beat is set as a necessary node. Through interval dynamic programming, the first path is read according to the cumulative edge cost increasing and the segment boundary beat number ascending order, and the phenological segment is output. Within each phenological segment, the adjacent beat differences of air temperature and humidity, root zone water content, and canopy cover are calculated. Differences with the same sign and adjacent beat numbers are merged into change segments. Segment nodes are generated by segment type, difference sign, start and end beat numbers, and source number. Candidate edges are established only when the meteorological dryness segment precedes the root zone water loss segment, the root zone water loss segment precedes the canopy decline segment, and the root zone recovery segment precedes the canopy return segment. A candidate knowledge graph is then output.

7. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 6, characterized in that: The knowledge fusion module also includes: For the candidate knowledge graph, each segment node enters only one similar relation chain, the candidate edge direction increases along the beat number and the relation chain does not cross phenological segments as constraints, the difference between the beat numbers at the two ends of the candidate edge, the difference in the amplitude rank of the segment and the number of missing beats are used as the edge cost, the network simplex algorithm is used to perform basis replacement with increasing reduced cost until the reduced cost of all non-basis edges is not negative, then the number of complete relation chain paths is calculated according to the node beat number in the forward and reverse directions, candidate edges that are not in complete relation chains are deleted and basis replacement is performed again until the edge sets of two adjacent rounds are consistent, forming the original knowledge graph; The number of withstandable beats is generated based on the beat number difference between the starting point of the root zone water loss segment and the starting point of the canopy descent segment in the original knowledge graph. The number of recoverable beats is generated based on the beat number difference between the starting point of the root zone recovery segment and the starting point of the canopy return segment. The stands are first sorted in descending order of withstandable beats, and then the stands with the same withstandable beats are sorted in ascending order of recoverable beats. If the recoverable beats are still the same, they are sorted in ascending order of stand number. The initial sorting table is then output.

8. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 7, characterized in that: The misalignment construction module includes: Read the phenological segment boundaries, original knowledge graph, and stand time series of the target forest stand. Fix one source time series in sequence and freeze the remaining source time series. Develop forward and backward shifts based on the beat number difference between the starting point of the target phenological segment and the starting points of the adjacent phenological segments. Use the two shifts to cyclically move the fixed source time series. Delete the corresponding result when the shift is zero. Output the source-by-source shifted time series. For each source-by-source shift time series, perform an iterative amplitude adjustment Fourier method. Read the Fourier amplitude and the original value ascending list of the corresponding original source time series. Replace the transformed amplitude of the source-by-source shift time series with the original Fourier amplitude and retain the transformed phase. Obtain the updated value through inverse transformation. Then, fill the original value ascending list according to the rank of the updated value. For the same value, determine the rank according to the time number ascending order and save the rank sequence. Repeat the amplitude replacement and rank backfilling until the rank sequence of the current round coincides with the saved rank sequence. Output the candidate misaligned time series that retains the original value set and the original Fourier amplitude.

9. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 8, characterized in that: The misalignment construction module also includes: Based on phenological segmentation, fragment generation, and node connection rules, a candidate misaligned time series is reconstructed into a candidate misaligned knowledge graph and a candidate misaligned sorting table. The number of edges retained in the drought-affected relationship chain and the recovery relationship chain in the original knowledge graph are counted. The number of neighboring trees of the target forest stand that maintain the same predecessor and successor numbers in the initial sorting table and the candidate misaligned sorting table are counted. The first candidate misaligned time series is read in descending order of the number of retained edges, descending order of the number of neighboring trees, descending order of the number of shifted steps, and ascending order of the source number. Replace the original source time sequence with the first candidate misaligned time sequence corresponding to each source, while keeping the time sequences of the other sources unchanged, and then form a misaligned knowledge graph and a misaligned sorting table according to the processing rules of the knowledge fusion module.

10. The drought adaptability assessment system for Helan Mountain coniferous forests based on multi-source data according to claim 9, characterized in that: The evaluation module includes: Based on the original knowledge graph and each misaligned knowledge graph, information processing and operational analysis in the forestry field are performed. Candidate node pairs are established between nodes with the same fragment type, source number and difference sign. Candidate edge pairs are established when both ends of the candidate node pairs are paired and the connection directions are the same. Then, the pairing order is generated according to the ascending order of the original knowledge graph node sequence number, the ascending order of the candidate misaligned knowledge graph node sequence number and the ascending order of the source number. The backtracking maximum common subgraph algorithm is executed along the pairing order. At each level, the search continues from two branches: adding candidate node pairs and skipping candidate node pairs. After adding a candidate node pair, the successor node pair that is duplicated with the already paired node or disrupts the direction of the candidate edge is deleted. The current branch ends when the sum of the current number of paired nodes and the remaining number of candidate nodes does not exceed the number of nodes already saved. The first common subgraph is retained in descending order of the number of node pairs, descending order of the number of candidate edge pairs, and ascending order of the pairing order. For the first common subgraph, read the connection from the root zone water loss segment to the canopy descent segment and the connection from the root zone recovery segment to the canopy return segment. Read the predecessor number and successor number of the target forest stand from the initial sorting table and the corresponding misaligned sorting table. Replace the predecessor number with the table head code at the beginning of the table and replace the successor number with the table tail code at the end of the table to form a relationship retention record. When any relationship retention record contains two connections and the predecessor and successor numbers remain unchanged, the target forest stand's ranking in the initial ranking table is revoked and a time-series distortion record is output. When all relationship retention records are missing any connection or the predecessor or successor number has changed, the target forest stand's ranking in the initial ranking table is retained and a drought adaptability evaluation record is output.

Images