Intelligent recommendation method of management measures for forest sub-compartment precision management

Abstract

The application discloses a kind of forest small class precision management-oriented management measure intelligent recommendation method, belong to electric digital data processing field, including steps: forest is divided into small class, and portrait of each small class in each time cycle is generated by collecting multi-source data;Measure library is constructed;The candidate measure set of small class is generated;Based on growth prediction and residual error correction, the predicted accumulation of candidate measures is calculated in portrait deduction duration;Corresponding carbon storage, mixed degree, fire risk index, pest risk index, cumulative management cost are predicted;Based on the comprehensive score of each candidate measure of small class, obtain the recommended scheme;The uncertainty of the recommended scheme is judged.The application realizes the quantitative comparison and intelligent recommendation of different forest management measures by multi-source data unified modeling, small class portrait, ecological process model, XGBoost residual error correction and uncertainty supplement mechanism, and provides a new technical path for forest small class precision management, measure dynamic optimization and whole life cycle intelligent decision-making.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to an intelligent recommendation method for management measures for precise management of small forest classes. Background Technology

[0002] With the rapid development of intelligent forest management and decision-making technologies, methods such as multi-source data fusion, ecological process modeling, machine learning correction, and scenario simulation have shown significant potential in forest resource management, forest management, and disaster prevention. As forest management gradually shifts towards refined and differentiated management of small forest stands, how to automatically generate appropriate management measures based on the stand status and management conditions of different small stands has become an important issue in precision forest management.

[0003] In forestry management, the accurate integration of forest resource surveys, remote sensing data on forest growth, meteorological environment, disaster risk, and historical management data is a crucial foundation for precise management operations such as forest tending, thinning, replanting, enclosure, fire prevention, and pest and disease control. However, existing technologies still suffer from problems such as reliance on human experience, the inability of single remote sensing data to interpret management effects, and the lack of ecological process constraints in pure machine learning. In particular, different management measures can simultaneously affect stand growth, stand structure, disaster risk, ecological function, and management costs. Existing methods struggle to uniformly extrapolate and quantitatively compare the future effects of different measures, thus making it difficult to directly generate management recommendations for specific forest stands. For example, patent application number 201910411776.5 focuses on solving the problem of forest biomass estimation, mainly outputting resource or biomass results, but it does not further develop management measure screening, scheme optimization, or scenario extrapolation. Patent application number 202310203619.1 focuses on the dynamic measurement of carbon sink project boundaries, carbon layers, sample plots, and changes in carbon storage, primarily outputting carbon storage and carbon change, but lacks further recommendations for small-compartment management measures and multi-objective scheme selection. Patent application number 202310203751.2 emphasizes the dynamic assessment of the quantity and value of ecosystem service functions, mainly serving ecological value accounting, but lacks mechanisms for predicting and selecting the effects of management measures such as thinning, replanting, enclosure, and fire prevention. Therefore, existing related technologies largely concentrate on forest status monitoring, resource estimation, or ecological function evaluation. For specific forest compartments, questions remain regarding which management measures to adopt and the future effects of different measures. A technical solution combining ecological process prediction, machine learning correction, multi-objective evaluation, and measure recommendation is still lacking. Summary of the Invention

[0004] The purpose of this invention is to provide a method for intelligently recommending management measures for precise forest management, which addresses the problems of traditional forest management relying on human experience, inaccurate judgment from a single data source, lack of ecological process constraints in model predictions, and difficulty in quantifying and comparing different management measures. This method predicts, compares, and optimizes the future implementation effects of candidate management measures based on the state of the forest plot, providing a new technical path for precise forest management, dynamic optimization of management measures, and intelligent decision-making throughout the entire life cycle.

[0005] To achieve the above objectives, the technical solution adopted by this invention is as follows: an intelligent recommendation method for management measures for precise management of small forest plots, comprising the following steps:

[0006] S1, divide the forest into N sub-classes, with a preset number of time periods T, collect data to generate a profile of each sub-class at each time period, and the profile of sub-class i at time period t is x. i,t ;

[0007] ;

[0008] Among them, V i,t For accumulation, C i,t For carbon reserves, ρ i,t For canopy closure, A i,t For forest age, H i,t For tree species mixing degree, For stand density, For fire risk index, This is a pest and disease risk index. To manage the difficulty of the task, Q i,t For data reliability, 1≤i≤N, 1≤t≤T;

[0009] S2, construct a library of measures, including S21~S22;

[0010] S21, preset N1 type of management measures, configure four types of rules for each type of management measure, including applicable conditions, implementation month, budget constraints and policy constraints, and generate a code for each type of management measure;

[0011] S22, store the configured management measures into the measure library, and label the n1th type of management measure in the library as m. n1 , 1≤n1≤N1;

[0012] S3, calculate x i,t Execute m n1 Management costs B i,t (m n1 Based on rule matching, management measures are selected to constitute x. i,t Candidate measure set M i,t and M i,tThe k-th candidate measure is marked as m. k ;

[0013] S4, preset simulation duration T h Based on growth prediction and residual correction, x is calculated. i,t In T h Execute m within the specified time k Predicted accumulation ;

[0014] S5, based on Predict x i,t Execute m k Duration T h The following are the carbon storage, mixing degree, fire risk index, pest and disease risk index, and cumulative management cost;

[0015] S6, based on predicted accumulation, parameters from S5, and data reliability Q. i,t Calculate the candidate measures m for class i. k Overall score S i,k ;

[0016] S7, calculate x according to S4~S6 i,t The candidate measure with the highest comprehensive score among all candidate measures will be the proposed recommended solution.

[0017] S8: Perform uncertainty judgment on the proposed solution. If the judgment is correct, output the result; otherwise, execute S9.

[0018] S9, reconstruct the small class profile by collecting supplementary data and execute S3~S8 again or trigger manual review. If it still fails after N2 executions, manual review will be triggered directly. N2 is the preset upper limit for supplementary data collection.

[0019] As a preferred option, generate image x i,t The methods include Sa1~Sa4;

[0020] Sa1, obtain the accumulated amount V corresponding to class i. i,t Carbon reserves C i,t Canopy closure ρ i,t Lin Ling A i,t Tree species mixing degree H i,t forest stand density Management difficulty Data credibility Q i,t ;

[0021] Sa2, obtain the combustible index F of class i. i,t Meteorological dryness index Topographic index Human activity index Historical fire index Calculate the fire risk index ;

[0022]

[0023] In the formula, clip(⋅,0,1) is the cutoff function. ~ These are the weights of the corresponding multiplication terms, with initial values ​​set to 0.30, 0.40, 0.15, 0.10, and 0.05 respectively.

[0024] Sa3, obtain the host tree species susceptibility index of class i. Vegetation anomaly index Pest and Disease Climate Suitability Index Historical Pest and Disease Index Calculate the pest and disease risk index ;

[0025]

[0026] In the formula, ~ These are the weights of the corresponding multiplication terms, with initial values ​​set to 0.30, 0.30, 0.25, and 0.15 respectively.

[0027] Sa4, generated .

[0028] Preferably, the management measures include individual measures and combined measures. The individual measures include no management, light thinning, replanting of mixed tree species, enclosure, clearing of understory combustibles, pest and disease control, and adjustment of tree species structure. The combined measures are composed of multiple compatible individual measures.

[0029] The applicable conditions include tree species, forest age, stand density, canopy closure, fire risk, pest and disease risk, slope, and road accessibility; policy constraints are formed by manually compiling policy texts into rules and updating them regularly.

[0030] As a preferred embodiment, S3 includes S31 to S34;

[0031] S31, for x i,t Generate a candidate measure set M i,t Initially empty;

[0032] S32, calculate x i,t Execute m n1 Management costs B i,t (m n1 ), Area i c is the area of ​​class i. n1 For m n1The cost per unit area;

[0033] S33, x i,t With m n1 If all four types of rules are matched, the resulting measure is stored in M ​​as a candidate measure. i,t ;

[0034] S34, after processing all management measures in the measure library in sequence according to S32~S33, then M i,t The k-th candidate measure is marked as m. k .

[0035] Preferably, S4 includes S41 to S43;

[0036] S41, for x i,t Growth prediction was performed using the Chapman-Richards model, based on the tree species s at t and t+T. h Corresponding forest age A i,t and Output tree species s in forest age A i,t and At that time, the site conditions of the small class i i Theoretical storage capacity per unit area Y s (A i,t Site i ), Calculate t ~ t+T h Basic accumulation increment , Area i It is the area of ​​class i;

[0037] S42, a pre-trained XGBoost model used as input x i,t External environment variables and m k Encode and output t~t+T h The increase in actual accumulation during the period and The difference is used as the residual correction amount. The external environmental variables include annual or quarterly temperature, rainfall, drought index, remote sensing crop growth changes, soil and topographic conditions;

[0038] S43, Calculate the predicted accumulation volume , In the formula, L i (m k ) for t~t+T h Execute m k The resulting immediate reduction in accumulation.

[0039] As a preferred option, in S41, Y s (A i,t Sitei Calculate according to the following formula:

[0040]

[0041] In the formula, , , Tree species s in small class i, site conditions i The maximum potential accumulation, growth rate parameter, and curve shape parameter are given, and exp(⋅) is the exp function.

[0042] As a preferred embodiment, in S42, when pre-training the XGBoost model, the input is x. i,t The external environmental variables and actual management measures are coded, and the output is the residual correction amount. During the reasoning process, the actual management measures are replaced with candidate measures.

[0043] Preferably, S6 includes S61 to S62;

[0044] S61, comparing t and t+T h Calculate the increment of stock volume Δa1, the increment of carbon storage Δa2, the increment of hybridization degree Δa3, and the decrease in fire risk index Δa. 41 Δa, a decrease in the pest and disease risk index 42 Let the cumulative management cost be denoted as Δa5, and calculate Δa1, Δa2, Δa3, and Δa5 respectively. 41 , Δa 42 The normalized value corresponding to Δa5 , , , , , Then , Weighted fusion reduces normalization risk. ;

[0045] S62, calculate the candidate measures m adopted by class i. k Overall score S i,k ;

[0046]

[0047] In the formula, w1~w6 are the weights of the corresponding multiplication terms.

[0048] Preferably, in step S8, an uncertainty judgment is made on the proposed recommended solution, specifically as follows:

[0049] S81, Obtain the data uncertainty of class i over time period t. Model uncertainty and data conflict uncertainty All three are normalized to [0,1].

[0050] in Calculated from missing rate, cloud cover, and survey interval. The discreteness of multiple residual correction values ​​obtained from Bootstrap resampling training of the pre-trained XGBoost model is calculated. Calculated from the differences in results from multiple sources;

[0051] S82, calculate the comprehensive uncertainty U according to the following formula. i,t :

[0052]

[0053] In the formula ω data ω model ω conflict These are the weights of the corresponding multiplication terms;

[0054] S83, preset uncertainty threshold, if U i,t If the value is below the uncertainty threshold, it passes.

[0055] As a preferred option, S81 Calculate according to the following formula:

[0056]

[0057] p miss To determine the missing percentage of data needed to construct a small class profile, p cloud The proportion of cloud-covered pixels in the small-scale remote sensing image. T represents the time interval since the most recent forest resource survey. max To preset the maximum survey interval, γ1, γ2, and γ3 are the weights of the corresponding multiplicative terms, and the sum of the three is 1;

[0058] S81 Based on the differences in results from multiple sources for the same indicator within the same small class and the same time period, if the result of the j-th indicator obtained from the q-th data source is z... j,q ,but

[0059]

[0060] Where J represents the number of indicators with multiple data sources, and ε is a very small positive number to prevent the denominator from being zero. , z j,q The maximum and minimum values.

[0061] Regarding S1: This invention requires the collection of multi-source data when constructing small class profiles. The multi-source data is divided into three categories:

[0062] The first category is essential data, including plot boundaries, forest resource survey data, remote sensing images, meteorological data, and topographic data.

[0063] A forest stand is the basic unit of forestry survey planning and design. It refers to the smallest regional unit in which a forest is divided according to different ownership, land category, origin, forest category, forest type, dominant tree species group, age group, canopy closure level, site type, economic forest production period, and forestry project category. The boundaries of the forest stand are used to unify data within the corresponding regional unit.

[0064] Forest resource survey data is used to describe the stand basis of a forest compartment. It includes tree species composition, stand age, average diameter at breast height (DBH), average tree height, number of trees per unit area, stock volume, canopy closure, site conditions, stand origin, and forest type. If the original data is a forest compartment survey form, it is directly associated with the forest compartment number; if the original data is a sample plot survey form, it is first assigned to the corresponding forest compartment based on the sample plot coordinates or the forest compartment to which the sample plot belongs, and then weighted and summarized.

[0065] Remote sensing imagery is used to supplement the recent growth and canopy changes of vegetation plots, such as NDVI (Normalized Difference Vegetation Index), EVI (Enhanced Vegetation Index), canopy cover, and vegetation anomalies. Remote sensing imagery is originally pixel-level data; the system aggregates pixel values ​​within a plot according to its boundaries to form plot-level remote sensing features. For example, averaging all NDVI pixels within a plot yields the average growth index for that plot at the current time; subtracting NDVI values ​​from two consecutive periods reveals the growth trend.

[0066] Meteorological data refers to data obtained from meteorological stations and interpolated to sub-classes using the nearest station method or inverse distance weighted interpolation.

[0067] Topographic data is statistically analyzed for slope, aspect, and elevation at small plot boundaries; GIS (Geographic Information System) data such as roads, villages, water sources, and power transmission lines are converted into distance or density features.

[0068] The second category is critical data, including records of historical disasters such as fires, pests and diseases, windbreaks, and droughts, as well as historical management records such as thinning, replanting, enclosure, pest and disease control, and clearing of understory combustibles. This type of data is used to improve risk assessment and model calibration capabilities. Historical disaster and management records are assigned to corresponding work plots according to their location, work unit, and time.

[0069] The third category is augmented data, including drone lidar data, individual tree identification data, and special surveys on forest understory combustibles. This type of data can improve accuracy, but its absence does not affect the basic processes.

[0070] The profile of class i over time period t is x. i,t ;

[0071]

[0072] x i,t It contains 10 data points, each with its own source, and the accumulated amount V. i,t Estimated from forest resource survey data or timber volume tables; carbon storage C i,t Calculated from volume and tree species carbon conversion coefficient; canopy closure ρ i,t Lin Ling A i,t Data from forest resource surveys; mixing degree H i,t Calculated from the proportion of different tree species in forest resource survey data; stand density The difficulty of management operations is measured by indicators such as tree density and canopy closure from forest resource survey data. Calculated based on slope, road distance, and access conditions; data reliability is determined by data completeness, age, and consistency. Additionally, the fire risk index... Pest and disease risk index The calculations are performed and backfilled in steps Sa2 and Sa3 respectively.

[0073] Regarding the fire risk index The index consists of five sub-items: combustible material index, meteorological aridity index, topographic index, human activity index, and historical fire index. The combustible material index is obtained from forest stand density, canopy closure, understory combustible material surveys, and remote sensing vegetation indices. The meteorological aridity index is obtained from temperature, humidity, rainfall over the past 7 to 30 days, wind speed, and soil moisture. The topographic index is obtained from slope, aspect, and altitude. The human activity index is obtained from distances to roads, villages, power lines, and tourist attractions. The historical fire index is obtained from the number of historical fires, their distances, and the number of years since the fires occurred.

[0074] Regarding the risk index of pests and diseases The index consists of four sub-items: host tree species susceptibility, vegetation anomaly index, pest and disease climate suitability index, and historical pest and disease index. Host tree species susceptibility is derived from the pest and disease host tree species table; vegetation anomaly index is derived from NDVI decline, EVI changes, and canopy color anomalies; pest and disease climate suitability index is derived from temperature, humidity, and rainfall conditions; and historical pest and disease index is derived from historical pest and disease locations and spread distances.

[0075] Regarding the weights of each sub-item in the fire risk index and pest and disease risk index: when historical samples are insufficient, a fixed initial weight is used. ~ The values ​​are 0.30, 0.40, 0.15, 0.10, and 0.05, respectively. ~ The weights are set to 0.30, 0.30, 0.25, and 0.15 respectively to ensure system reproducibility. When there are sufficient historical fire or pest samples, the weights can be calibrated using logistic regression. However, the calibrated weights still need to meet the non-negativity and normalization constraints to avoid the model becoming uninterpretable.

[0076] Regarding management measures, each type of management measure needs to be configured with four types of rules, including applicable conditions, implementation months, budget constraints, and policy constraints. Applicable conditions include indicators and corresponding thresholds such as tree species, forest age, stand density, canopy closure, fire risk, pest and disease risk, slope, and road accessibility. Whether the applicable conditions are met is determined by comparing the corresponding indicators and thresholds. Implementation months set time constraints for each management measure; for example, "light thinning" can only be implemented in certain months. Budget constraints set a cost cap for each management measure; exceeding the cap disqualifies the measure. Policy constraints are manually compiled and derived from current policy documents such as those concerning public welfare forests, ecological red lines, logging quotas, and local management regulations.

[0077] Compared with the prior art, the advantages of the present invention are as follows:

[0078] (1) Introducing a small-scale forest profiling and management prescription generation mechanism. This invention uses the forest small-scale forest as the smallest decision-making unit, unifying forest resource surveys, remote sensing images, meteorological environment, topographic conditions, historical disasters and management records to the small-scale forest scale, forming a status profile that includes stock volume, carbon storage, mixed forest ratio, fire risk, pest and disease risk, operational difficulty and data reliability, and generating comparable and executable management plans based on this profile, thereby improving the precision of forest management plan formulation.

[0079] (2) An ecological process model and the XGBoost residual correction method are introduced. The ecological model (Chapman-Richards model) is used to express the basic influence of tree species, forest age, site conditions and stand density on growth. The basic growth trend of the small plot is predicted. The XGBoost model is then pre-trained using local forest plot data. After learning the deviation between ecological prediction and actual growth, it is only responsible for predicting the residual of the ecological model based on the local actual situation. This solves the defects of traditional ecological models, which are interpretable but difficult to adapt to local differences, and pure machine learning models, which can fit the data but lack ecological constraints. While maintaining ecological rationality, the accuracy of forest growth simulation and prediction of the effects of different management measures is improved, providing a reliable basis for the quantitative comparison of management measures in small plots.

[0080] (3) This invention introduces an uncertainty-based supplementary sampling closed loop. It compares scenarios such as no management, immediate management, delayed management, and alternative measures, and triggers UAV supplementary shooting, sample plot re-survey, or manual survey when there is data missing, model divergence, or significant multi-source conflict, thereby improving the interpretability and reliability of the management plan.

[0081] In summary, this solution improves the quality and application value of intelligent forest management decisions through unified multi-source data modeling, stand state profiling, ecological process modeling, XGBoost residual correction, counterfactual scenario extrapolation, and uncertainty re-collection mechanisms. This method addresses the problems of traditional forest management relying on human experience, inaccurate judgments from single data sources, lack of ecological constraints in model predictions, and difficulty in quantifying and comparing management measures. It provides a new technical path for precise management of forest stands, dynamic optimization of measures, and intelligent decision-making throughout the entire life cycle, and is particularly suitable for forest tending, stand structure optimization, disaster risk prevention and control, comparison of management measures, and refined management of forest resources. Attached Figure Description

[0082] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0083] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0084] Example 1: See Figure 1 A smart recommendation method for management measures aimed at precise management of small forest plots includes the following steps:

[0085] S1, divide the forest into N sub-classes, with a preset number of time periods T, collect data to generate a profile of each sub-class at each time period, and the profile of sub-class i at time period t is x. i,t ;

[0086] ;

[0087] Among them, V i,t For accumulation, C i,t For carbon reserves, ρ i,t For canopy closure, A i,t For forest age, H i,t For tree species mixing degree, For stand density, For fire risk index, This is a pest and disease risk index. To manage the difficulty of the task, Q i,t For data reliability, 1≤i≤N, 1≤t≤T;

[0088] S2, construct a library of measures, including S21~S22;

[0089] S21, preset N1 type of management measures, configure four types of rules for each type of management measure, including applicable conditions, implementation month, budget constraints and policy constraints, and generate a code for each type of management measure;

[0090] S22, store the configured management measures into the measure library, and label the n1th type of management measure in the library as m. n1 , 1≤n1≤N1;

[0091] S3, calculate x i,t Execute m n1 Management costs B i,t (m n1 Based on rule matching, management measures are selected to constitute x. i,t Candidate measure set M i,t and M i,t The k-th candidate measure is marked as m. k ;

[0092] S4, preset simulation duration T h Based on growth prediction and residual correction, x is calculated. i,t In T h Execute m within the specified time k Predicted accumulation ;

[0093] S5, based on Predict x i,t Execute m k Duration T h The following are the carbon storage, mixing degree, fire risk index, pest and disease risk index, and cumulative management cost;

[0094] S6, based on predicted accumulation, parameters from S5, and data reliability Q. i,t Calculate the candidate measures m for class i. k Overall score S i,k ;

[0095] S7, calculate x according to S4~S6 i,t The candidate measure with the highest comprehensive score among all candidate measures will be the proposed recommended solution.

[0096] S8: Perform uncertainty judgment on the proposed solution. If the judgment is correct, output the result; otherwise, execute S9.

[0097] S9, reconstruct the small class profile by collecting supplementary data and execute S3~S8 again or trigger manual review. If it still fails after N2 executions, manual review will be triggered directly. N2 is the preset upper limit for supplementary data collection.

[0098] In this embodiment, image x is generated. i,t The methods include Sa1~Sa4;

[0099] Sa1, obtain the accumulated amount V corresponding to class i. i,t Carbon reserves C i,t Canopy closure ρ i,t Lin Ling A i,t Tree species mixing degree H i,t forest stand density Management difficulty Data credibility Q i,t ;

[0100] Sa2, obtain the combustible index F of class i. i,t Meteorological dryness index Topographic index Human activity index Historical fire index Calculate the fire risk index ;

[0101] ;

[0102] In the formula, clip(⋅,0,1) is the cutoff function. ~ These are the weights of the corresponding multiplication terms, with initial values ​​set to 0.30, 0.40, 0.15, 0.10, and 0.05 respectively.

[0103] Sa3, obtain the host tree species susceptibility index of class i. Vegetation anomaly index Pest and Disease Climate Suitability Index Historical Pest and Disease Index Calculate the pest and disease risk index ;

[0104] ;

[0105] In the formula, ~ These are the weights of the corresponding multiplication terms, with initial values ​​set to 0.30, 0.30, 0.25, and 0.15 respectively.

[0106] Sa4, generated .

[0107] The management measures include individual measures and combined measures. Individual measures include no management, light thinning, replanting mixed tree species, enclosure, clearing of understory combustibles, pest and disease control, and adjustment of tree species structure. The combined measures are composed of multiple compatible individual measures.

[0108] The applicable conditions include tree species, forest age, stand density, canopy closure, fire risk, pest and disease risk, slope, and road accessibility; policy constraints are formed by manually compiling policy texts into rules and updating them regularly.

[0109] S3 includes S31~S34;

[0110] S31, for x i,t Generate a candidate measure set M i,t Initially empty;

[0111] S32, calculate x i,t Execute m n1 Management costs B i,t (m n1 ), Area i c is the area of ​​class i. n1 For m n1 The cost per unit area;

[0112] S33, x i,t With m n1 If all four types of rules are matched, the resulting measure is stored in M ​​as a candidate measure. i,t ;

[0113] S34, after processing all management measures in the measure library in sequence according to S32~S33, then M i,t The k-th candidate measure is marked as m. k .

[0114] S4 includes S41~S43;

[0115] S41, for x i,t Growth prediction was performed using the Chapman-Richards model, based on the tree species s at t and t+T. h Corresponding forest age A i,t and Output tree species s in forest age A i,t and At that time, the site conditions of the small class i i Theoretical storage capacity per unit area Y s (A i,t Site i ), Calculate t ~ t+T h Basic accumulation increment , Area i It is the area of ​​class i;

[0116] S42, a pre-trained XGBoost model used as input x i,t External environment variables and mk Encode and output t~t+T h The increase in actual accumulation during the period and The difference is used as the residual correction amount. The external environmental variables include annual or quarterly temperature, rainfall, drought index, remote sensing crop growth changes, soil and topographic conditions;

[0117] S43, Calculate the predicted accumulation volume , In the formula, L i (m k ) for t~t+T h Execute m k The resulting immediate reduction in accumulation.

[0118] In S41, Y s (A i,t Site i Calculate according to the following formula:

[0119] ;

[0120] In the formula, , , Tree species s in small class i, site conditions i The maximum potential accumulation, growth rate parameter, and curve shape parameter are given, and exp(⋅) is the exp function.

[0121] In S42, when pre-training the XGBoost model, the input is x. i,t The external environmental variables and actual management measures are coded, and the output is the residual correction amount. During the reasoning process, the actual management measures are replaced with candidate measures.

[0122] S6 includes S61~S62;

[0123] S61, comparing t and t+T h Calculate the increment of stock volume Δa1, the increment of carbon storage Δa2, the increment of hybridization degree Δa3, and the decrease in fire risk index Δa. 41 Δa, a decrease in the pest and disease risk index 42 Let the cumulative management cost be denoted as Δa5, and calculate Δa1, Δa2, Δa3, and Δa5 respectively. 41 , Δa 42 The normalized value corresponding to Δa5 , , , , , Then , Weighted fusion reduces normalization risk. ;

[0124] S62, calculate the candidate measures m adopted by class i. k Overall score S i,k ;

[0125] ;

[0126] In the formula, w1~w6 are the weights of the corresponding multiplication terms.

[0127] In step S8, an uncertainty judgment is made on the proposed recommended solution, specifically as follows:

[0128] S81, Obtain the data uncertainty of class i over time period t. Model uncertainty and data conflict uncertainty All three are normalized to [0,1].

[0129] in Calculated from missing rate, cloud cover, and survey interval. The discreteness of multiple residual correction values ​​obtained from Bootstrap resampling training of the pre-trained XGBoost model is calculated. Calculated from the differences in results from multiple sources;

[0130] S82, calculate the comprehensive uncertainty U according to the following formula. i,t :

[0131] ;

[0132] In the formula ω data ω model ω conflict These are the weights of the corresponding multiplication terms;

[0133] S83, preset uncertainty threshold, if U i,t If the value is below the uncertainty threshold, it passes.

[0134] S81 Calculate according to the following formula:

[0135] ;

[0136] p miss To determine the missing percentage of data needed to construct a small class profile, p cloud The proportion of cloud-covered pixels in the small-scale remote sensing image. T represents the time interval since the most recent forest resource survey. maxTo preset the maximum survey interval, γ1, γ2, and γ3 are the weights of the corresponding multiplicative terms, and the sum of the three is 1;

[0137] S81 Based on the differences in results from multiple sources for the same indicator within the same small class and the same time period, if the result of the j-th indicator obtained from the q-th data source is z... j,q ,but

[0138] ;

[0139] Where J represents the number of indicators with multiple data sources, and ε is a very small positive number to prevent the denominator from being zero. , z j,q The maximum and minimum values.

[0140] The core of this invention is not simply to create a forest monitoring platform, but to form a "closed-loop management decision-making process": first, assess the current status of the forest plots; then, predict future changes; next, compare the effects of different measures; and finally, output actionable management measures. If the data is unreliable, no decision is forced; instead, supplementary data collection or manual verification is triggered.

[0141] Example 2: See Figure 1 Based on Example 1, regarding the candidate measures for S3: for example, for an overly dense coniferous forest sub-compartment, if the fire risk is high, the road is accessible, and the slope is not steep, then light thinning and clearing of understory combustibles will be included as candidate measures; if the sub-compartment is located in a strictly protected area, then logging measures will be excluded by policy constraints.

[0142] Regarding the XGBoost model in S4: Since this invention relies on growth prediction and residual correction for accurate prediction, the XGBoost model is pre-trained. During inference, if a region lacks continuous sample plot surveys or historical management records, making it impossible to train a local XGBoost model, the residual correction is set to 0, the system degenerates into an ecological model prediction, and is marked as "low prediction reliability." This ensures the process is runnable while avoiding the fabrication of unreliable machine learning results.

[0143] Regarding S5 based Predict x i,t Execute m k Duration T h The following are the subsequent carbon storage, mixing degree, fire risk index, pest and disease risk index, and cumulative management cost:

[0144] All of the above are predicted values, among which the carbon storage is the predicted accumulation obtained from S4. Based on the preset carbon conversion coefficient, the degree of mixing is calculated according to the candidate measure m. kThe calculation of tree species composition after implementation, and the cumulative management cost based on step S31, are t~t+T. h During period x i,t Execute m k The sum of management costs, while the fire risk index and pest and disease risk index are still calculated based on steps Sa2 and Sa3 respectively, specifically:

[0145] The above indicators are used together to evaluate the comprehensive impact of candidate measures on stand growth, structural status, disaster risk and management inputs. Among them, carbon storage is included as one of the ecological function evaluation indicators in the comprehensive score.

[0146] t+T h Fire risk index at that time This involves five indices related to class i at this time: combustible material index, meteorological dryness index, topographic index, human activity index, and historical fire index. The combustible material index and meteorological dryness index are expressed as class i in the time period t+T. h The values ​​for time, terrain index, human activity index, and historical fire index all follow the time period t.

[0147] t+T h Pest and disease risk index at that time The study involves four indices for class i: host tree species susceptibility index, vegetation abnormality index, pest and disease climate suitability index, and historical pest and disease index. The host tree species susceptibility index is calculated as m... k The host tree species susceptibility index corresponding to the tree species composition after implementation is updated; the vegetation anomaly index uses the value of time period t; and the pest and disease climate suitability index is based on t+T. h The calculation uses the predicted values ​​of temperature, humidity, and rainfall at the time of the event, while the historical pest and disease index uses the value of the time period t.

[0148] Regarding the generation of proposed solutions in S7: The system compares the comprehensive scores of multiple candidate measures, and the candidate measure with the highest score is selected as the proposed solution. The comprehensive score includes multiple influencing factors with different weights. For example, if a forest plot is currently classified as "overly dense coniferous forest" with a fire risk index of 0.76, the candidate measures generated in step S3 include "no management," "light thinning," "understory combustible material clearing," and "light thinning + understory combustible material clearing." The system will then perform simulations for each of these measures. Simulation time T. h=5 years. If the fire risk remains high after 5 years under the "no management" measures, while the fire risk is significantly reduced under the "light thinning + understory combustible material clearing" measures, and the stock volume gradually recovers after a short-term decline, then the measures will receive a higher comprehensive score. In practice, the weights of each component in the comprehensive score are determined by the management based on the operational objectives. If the annual operational objective is primarily fire prevention, the weight of risk reduction is increased; if the primary objective is to promote stand growth and enhance ecological functions, the weights of normalized stock volume gain and normalized carbon storage gain are increased; if the operational budget is tight, the weight of the normalized operational cost is increased, and so on. The final proposed recommended plan can also be designed as a prescription card, including: stand number, recommended measures, operational area, operational time window, target density or target canopy closure, operational intensity, estimated cost, changes in stock volume, carbon storage, fire risk, pest and disease risk over the next 3-5 years, data reliability, and re-collection recommendations. For example, the recommended result for a small plot could be written as follows: Light thinning and clearing of understory combustibles will be carried out in plot No. 12, with the operation window from August to October. The target density will be reduced from 1200 trees / ha to 800 trees / ha, while retaining the dominant trees with good growth. It is expected that the stock volume will decrease in the short term, but will recover and grow again in 3 to 5 years. The fire risk will be reduced from high risk to medium risk. Since the sample plot survey is more than 5 years old, it is recommended to conduct a sample plot re-survey before the operation.

[0149] Regarding S8's assessment of uncertainty in the proposed recommended solution, this includes data uncertainty. Model uncertainty and data conflict uncertainty Users can design specific rules as needed, such as in this embodiment: triggering drone re-photographing when remote sensing cloud cover exceeds 60%; triggering plot re-surveying when plot surveys have not been updated for more than 5 years; triggering manual surveys when pest and disease risk is high but vegetation abnormality evidence is insufficient; triggering combustible material re-survey when fire risk is high but understory combustible material data is missing; and marking management prescriptions as low confidence and requiring manual review when model uncertainty is too high. The newly collected data re-enters the sub-plot comprehensive characteristics, updating the sub-plot profile, risk index, growth prediction, scheme score, and management prescription, forming a closed loop of "prediction—recommendation—re-collection—correction—re-prediction".

[0150] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligently recommending management measures for precise management of small forest plots, characterized in that, Includes the following steps: S1, divide the forest into N sub-classes, with a preset number of time periods T, collect data to generate a profile of each sub-class at each time period, and the profile of sub-class i at time period t is x. i,t ; ； Among them, V i,t For accumulation, C i,t For carbon reserves, ρ i,t For canopy closure, A i,t For forest age, H i,t For tree species mixing degree, For stand density, For fire risk index, This is a pest and disease risk index. To manage the difficulty of the task, Q i,t For data reliability, 1≤i≤N, 1≤t≤T; S2, construct a library of measures, including S21~S22; S21, pre-sets N1 type of management measures, configures four types of rules for each type of management measure, including applicable conditions, implementation month, budget constraints and policy constraints, and generates a code for each type of management measure; S22, store the configured management measures into the measure library, and label the n1th type of management measure in the library as m. n1 , 1≤n1≤N1; S3, calculate x i,t Execute m n1 Management costs B i,t (m n1 Based on rule matching, management measures are selected to constitute x. i,t Candidate measure set M i,t and M i,t The k-th candidate measure is marked as m. k ; S4, preset simulation duration T h Based on growth prediction and residual correction, x is calculated. i,t In T h Execute m within the specified time k Predicted accumulation ; S5, based on Predict x i,t Execute m k Duration T h The following are the carbon storage, mixing degree, fire risk index, pest and disease risk index, and cumulative management cost; S6, based on predicted accumulation, parameters from S5, and data reliability Q. i,t Calculate the candidate measures m for class i. k Overall score S i,k ; S7, calculate x according to S4~S6 i,t The candidate measure with the highest comprehensive score among all candidate measures will be the proposed recommended solution. S8: Perform uncertainty judgment on the proposed solution. If the judgment is correct, output the result; otherwise, execute S9. S9, reconstruct the small class profile by collecting supplementary data and execute S3~S8 again or trigger manual review. If it still fails after N2 executions, manual review will be triggered directly. N2 is the preset upper limit for supplementary data collection.

2. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 1, characterized in that, Generate portrait x i,t The methods include Sa1~Sa4; Sa1, obtain the accumulated amount V corresponding to class i. i,t Carbon reserves C i,t Canopy closure ρ i,t Lin Ling A i,t Tree species mixing degree H i,t forest stand density Management difficulty Data credibility Q i,t ; Sa2, obtain the combustible index F of class i. i,t Meteorological dryness index Topographic index Human activity index Historical fire index Calculate the fire risk index ; ； In the formula, clip(⋅,0,1) is the cutoff function. ~ These are the weights of the corresponding multiplication terms, with initial values ​​set to 0.30, 0.40, 0.15, 0.10, and 0.05 respectively. Sa3, obtain the host tree species susceptibility index of class i. Vegetation anomaly index Pest and Disease Climate Suitability Index Historical Pest and Disease Index Calculate the pest and disease risk index ; ； In the formula, ~ These are the weights of the corresponding multiplication terms, with initial values ​​set to 0.30, 0.30, 0.25, and 0.15 respectively. Sa4, generated .

3. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 1, characterized in that, The management measures include individual measures and combined measures. Individual measures include no management, light thinning, replanting mixed tree species, enclosure, clearing of understory combustibles, pest and disease control, and adjustment of tree species structure. The combined measures are composed of multiple compatible individual measures. The applicable conditions include tree species, forest age, stand density, canopy closure, fire risk, pest and disease risk, slope, and road accessibility; Policy constraints are formed by manually compiling policy texts into rules and updating them regularly.

4. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 1, characterized in that, S3 includes S31~S34; S31, for x i,t Generate a candidate measure set M i,t Initially empty; S32, calculate x i,t Execute m n1 Management costs B i,t (m n1 ), Area i c is the area of ​​class i. n1 For m n1 The cost per unit area; S33, x i,t With m n1 If all four types of rules are matched, the resulting measure is stored in M ​​as a candidate measure. i,t ; S34, after processing all management measures in the measure library in sequence according to S32~S33, then M i,t The k-th candidate measure is marked as m. k .

5. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 1, characterized in that, S4 includes S41~S43; S41, for x i,t Growth prediction was performed using the Chapman-Richards model, based on the tree species s at t and t+T. h Corresponding forest age A i,t and Output tree species s in forest age A i,t and At that time, the site conditions of the small class i i Theoretical storage capacity per unit area Y s (A i,t Site i ), Calculate t ~ t+T h Basic accumulation increment , Area i It is the area of ​​class i; S42, a pre-trained XGBoost model used as input x i,t External environment variables and m k Encode and output t~t+T h The increase in actual accumulation during the period and The difference is used as the residual correction amount. The external environmental variables include annual or quarterly temperature, rainfall, drought index, remote sensing crop growth changes, soil and topographic conditions; S43, Calculate the predicted accumulation volume , In the formula, L i (m k ) for t~t+T h Execute m k The resulting immediate reduction in accumulation.

6. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 5, characterized in that, In S41, Y s (A i,t Site i Calculate according to the following formula: ； In the formula, , , Tree species s in small class i, site conditions i The maximum potential accumulation, growth rate parameter, and curve shape parameter are given, and exp(⋅) is the exp function.

7. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 5, characterized in that, In S42, when pre-training the XGBoost model, the input is x. i,t The external environmental variables and actual management measures are coded, and the output is the residual correction amount. During the reasoning process, the actual management measures are replaced with candidate measures.

8. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 1, characterized in that, S6 includes S61~S62; S61, comparing t and t+T h Calculate the increment of stock volume Δa1, the increment of carbon storage Δa2, the increment of hybridization degree Δa3, and the decrease in fire risk index Δa. 41 Δa, a decrease in the pest and disease risk index 42 Let the cumulative management cost be denoted as Δa5, and calculate Δa1, Δa2, Δa3, and Δa5 respectively. 41 , Δa 42 The normalized value corresponding to Δa5 , , , , , Then , Weighted fusion reduces normalization risk. ; S62, calculate the candidate measures m adopted by class i. k Overall score S i,k ; ； In the formula, w1~w6 are the weights of the corresponding multiplication terms.

9. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 1, characterized in that, In step S8, an uncertainty judgment is made on the proposed recommended solution, specifically as follows: S81, Obtain the data uncertainty of class i over time period t. Model uncertainty and data conflict uncertainty All three are normalized to [0,1]. in Calculated from missing rate, cloud cover, and survey interval. The discreteness of multiple residual corrections obtained from Bootstrap resampling training of the pre-trained XGBoost model is calculated. Calculated from the differences in results from multiple sources; S82, calculate the comprehensive uncertainty U according to the following formula. i,t : ； In the formula ω data ω model ω conflict These are the weights of the corresponding multiplication terms; S83, preset uncertainty threshold, if U i,t If the value is below the uncertainty threshold, it passes.

10. The intelligent recommendation method for management measures for precise management of small forest plots according to claim 9, characterized in that, S81 Calculate according to the following formula: ； p miss To construct the missing data ratio required for small class profiles, p cloud The proportion of cloud-covered pixels in the small-scale remote sensing image. T represents the time interval since the most recent forest resource survey. max To preset the maximum survey interval, γ1, γ2, and γ3 are the weights of the corresponding multiplication terms, and the sum of the three is 1; S81 Based on the differences in results from multiple sources for the same indicator within the same small class and the same time period, if the result of the j-th indicator obtained from the q-th data source is z... j,q ,but ； Where J represents the number of indicators with multiple data sources, and ε is a very small positive number to prevent the denominator from being zero. , z j,q The maximum and minimum values.

Images