A land use simulation method for patch generation

Abstract

The present application provides a patch-generated land use simulation method, which relates to the field of geographic information science and technology. The method comprises: obtaining change raster data by acquiring initial land use data; performing random point sampling on driving factor data and change raster data to obtain sampling data; training a random forest model with the preprocessed sampling data; obtaining a development probability and a development potential map with the driving factor data and the random forest model; inputting the development potential map and the initial land use data into a CA model, iteratively scanning each pixel of the initial land use data, and obtaining an overall probability; based on the overall probability, using a roulette mechanism and a threshold descent rule to introduce competition among land use types, combining multiple types of random patch seeds to obtain a land use simulation map, thereby achieving accurate simulation of changes in multiple types of land use patches and landscapes and simultaneously applying it to the prediction of future land use scenarios.

Description

Technical Field

[0001] The present application relates to the field of geographic information science and technology, and in particular to a patch generation land use simulation method. Background Art

[0002] In land-use change simulations, geographic cellular automata (GCAs) can simulate the spatial distribution and evolution of land-use types, enabling a better understanding of the mechanisms and trends of land-use change. They are a mainstream model for studying land-use change simulations. Cellular automata are an important simulation tool that simulates a spatial system composed of many small units (cells), where the state of each cell changes over time and is influenced by its neighbors.

[0003] A key characteristic of cellular automata is their bottom-up operation, where each cell interacts only with a limited number of its surrounding cells, and global patterns are formed through regular local behavior. Compared to traditional mathematical models, cellular automata can simulate irregular and complex phenomena in a more intuitive way, and computers can be used to model them with lossless precision, thereby simulating the behavior of various natural systems. Through cellular automata, we can better understand how individual and local behaviors in complex systems generate globally ordered patterns.

[0004] A geographic cellular automaton is a special form of cellular automaton that decomposes a spatial system into regular geographic units with defined properties and states. Each unit can represent a specific land-use type. Each cell is surrounded by a neighborhood, which consists of the cell itself and its neighboring cells. The size and shape of the neighborhood can be customized; common neighborhood shapes include four-, eight-, and twelve-neighborhoods. The states of cells in a cellular automaton change over time, governed by transition rules. A transition rule is a function that defines how the next state of each cell is calculated based on its current state and the states of its neighbors. Transition rules are typically based on the states of the cell's neighborhood and can be simple local functions or more complex global functions. By defining and designing the states, neighborhoods, and transition rules of a cellular automaton, it is possible to simulate the evolution of land-use change.

[0005] Currently, two transformation rule mining strategies are commonly used in geographic cellular automata: the Transformation Analysis Strategy (TAS) and the Pattern Analysis Strategy (PAS). TAS-based CA models extract all land use transformations between two periods of land use data. However, analyzing all transformation types increases the computational complexity of the CA model structure and reduces the model's flexibility and universality.

[0006] The PAS-based CA model, on the other hand, uses only one period of land-use data to calculate the probability of occurrence of a land-use type in each cell and determines the future state of the cell through competition between land-use types. This avoids the exponential increase in the number of transformation types as the number of land-use types increases during the process of discovering transformation rules. However, the PAS-based CA model itself lacks the ability to reveal how driving factors cause land-use change because the model is not based on historical "change" analysis. Therefore, the PAS-based CA model cannot obtain land-use change rules for specific time intervals, just as the FLUS model lacks the ability to reveal how driving factors cause land-use change.

[0007] Furthermore, existing models are insufficiently capable of simulating the evolution of land-use patches. Vector-based CA cannot simulate the dynamic changes of natural land-use types (such as forests and grasslands). Raster-based CA models struggle to synchronize the evolution of multiple types of patches. For example, the Fore-SCE model struggles with the presence of numerous small patches, and it lacks spatiotemporal dynamics. Summary of the Invention

[0008] The purpose of the present invention is to provide a patch generation land use simulation method to solve the limitations of the traditional land use change simulation model, such as the reduction of simulation accuracy due to error transmission, the inability to explore the potential factors of land use change, and the inability to dynamically simulate the generation and evolution of multiple types of land patches in time and space.

[0009] The above-mentioned purpose of this application is achieved through the following technical solutions:

[0010] S1: Obtain initial land use data; obtain change raster data through the initial land use data;

[0011] Obtain a preset number of driving factors affecting land use and land cover changes to form driving factor data;

[0012] S2: Perform random point sampling on the driving factor data and the change grid data to obtain sampling data;

[0013] S3: Preprocess the sampled data; build a random forest model based on the random forest classification algorithm; train the random forest model using the preprocessed sampled data;

[0014] S4: Input all driving factor data into the trained random forest model to obtain the development probability of each type of land use and the development potential map of each land use type in the region;

[0015] S5: Construct a CA model based on a multi-type random patch seed mechanism; set the transformation matrix and neighborhood weights of the CA model; set the target number of pixels in the CA model according to the required area of ​​each land use type;

[0016] S6: Input the development potential map and the initial land use data into the CA model, iteratively scan each pixel of the initial land use data, and obtain the land use type contained in the neighborhood, the proportion of the neighborhood, and the adaptive inertia coefficient of each pixel;

[0017] The overall probability of each land use type in each pixel is obtained by combining the land use types contained in the neighborhood, the proportion of the pixel in the neighborhood, the adaptive inertia coefficient, and the development probability of each type of land use.

[0018] S7: Construct a roulette wheel based on the overall probability; and transform various land use types within the pixel area into the land use type that wins the competition through the roulette wheel method.

[0019] When the adjacent effect of a land use type is equal to 0, different types of random seeds are generated by calculating the overall probability; new land use types are generated by random seeds;

[0020] S8: Based on the threshold decreasing rule, determine whether the conversion of the new land use type that wins the competition occurs;

[0021] S9: Repeat steps S6 to S8 until the iteration reaches the target number of pixels, stop the iteration, and output the land use simulation map.

[0022] This application utilizes the aforementioned technical solutions to more accurately explore the relationship between driving factors and land-use change simulations for both two- and multiple-category land-use change. By combining a land expansion analysis strategy with a random forest classification algorithm, this approach allows for more precise exploration of the relationship between driving factors and land-use change. By employing a multi-type random patch seeding mechanism based on threshold descent, this approach enables accurate simulation of changes in multiple land-use patches and landscapes, more realistically reflecting the land-use change process and enabling more accurate simulation of future land-use scenarios.

[0023] Optionally, step S1 includes:

[0024] S11: Acquire an initial land use classification image; convert or reclassify the initial land use classification image to obtain initial land use data; the initial land use data includes: first-period land use data and second-period land use data;

[0025] S12: extracting the changed parts between the two periods of the initial land use data, that is, extracting the grids with changed states from the second period land use data;

[0026] S13: Extracting grids whose states have changed between the two periods, ie, changed grid data, based on the first period land use data and the second period land use data.

[0027] This application adopts the above technical solution, sampling the change results of two periods of land use data instead of the two-period sampling mode or the one-period sampling mode commonly used in cellular automata. This effectively avoids the analysis of all transformation types, thus effectively simplifying the land use change analysis process and providing the ability to describe the characteristics of land use changes within a specific time interval.

[0028] Optionally, step S2 includes:

[0029] Using uniform sampling strategy or proportional sampling strategy, random point sampling is performed on the driving factor data and the change grid data to obtain the sampling data. The sample formula after sampling is expressed as:

[0030]

[0031] in Indicates the The first sampling point is extracted Driver data; Indicates the The data of the change grid data extracted from the sampling points; is transposed.

[0032] Optionally, step S3 includes:

[0033] The sampled data is normalized, and the normalization formula is as follows:

[0034]

[0035] in and Respectively represent The maximum and minimum values ​​of each driver; .

[0036] Optionally, step S3 further includes:

[0037] The random forest classification algorithm is expressed as follows:

[0038]

[0039] in is a constant, either 0 or 1; It is a vector composed of a preset number of driving factors; is the indicator function of the decision tree set; is a vector The prediction type of the nth decision tree; is the total count of the decision tree.

[0040] Optionally, step S6 includes:

[0041] The overall probability synthesis formula is:

[0042]

[0043] in is a random value from 0 to 1; For land use type The threshold for generating new land use patches, which are mined by users; is the neighborhood function;

[0044] A scanning window is constructed through the neighborhood function, and the number of pixels in the scanning window is counted to measure the spatial interaction of various land use types.

[0045] Neighborhood function is defined as follows:

[0046]

[0047] in Indicates the land use type at the last iteration within the n×n window The total number of occupied grid cells, is the weight between different land use types;

[0048] Adaptive function The adaptive method is defined as follows:

[0049]

[0050] in and It is Second and Land use type at the iteration The difference between current quantities and future demand.

[0051] Optionally, step S7 includes:

[0052] The threshold decreasing rule, the formula is as follows:

[0053]

[0054]

[0055] in, Decrease threshold The attenuation coefficient, Random values ​​that follow a normal distribution; is the decreasing threshold; is the number of attenuation times; is the decreasing step size; Defines whether land use type k is allowed to be converted to type The transformation matrix; Represents the random forest probability, that is, the pixel Developed into a type probability.

[0056] This application utilizes the aforementioned technical solution, employing a threshold-decreasing multi-type random patch seeding mechanism. This mechanism uses a Monte Carlo method to generate seeds and proposes a threshold-decreasing rule for a competitive process to limit the organic and spontaneous growth of all land use types. This allows for dynamic spatiotemporal simulation of the generation and evolution of any number of land patches, more realistically reflecting land use change processes and enabling more accurate simulation of future land use scenarios.

[0057] This application utilizes the aforementioned technical solution, employing a threshold-decreasing multi-type random patch seeding mechanism. This mechanism uses a Monte Carlo method to generate seeds and proposes a threshold-decreasing rule for a competitive process to limit the organic and spontaneous growth of all land use types. This allows for dynamic spatiotemporal simulation of the generation and evolution of any number of land patches, more realistically reflecting land use change processes and enabling more accurate simulation of future land use scenarios.

[0058] A computer-readable storage medium stores instructions, which, when executed, perform a patch generation land use simulation method.

[0059] The beneficial effects of the technical solution provided by this application are:

[0060] It integrates the ability of cellular automata (CA) to simulate complex spatiotemporal evolution processes and the advantage of simulating competition among multiple land use types, and overcomes the inherent defects of traditional cellular automata through the integration of advantages; it adopts a rule mining framework based on land expansion analysis strategy to realize the intelligent calculation of the development probability of land use types. The framework adopts the random forest classification (RFC) algorithm, which can reveal the potential driving factors of land use change; in addition, it adopts a multi-type random patch seeding mechanism based on threshold descent to realize the accurate simulation of changes in multiple land use patches and landscapes; and compared with traditional models, it has practical advantages such as higher accuracy, suitability for multiple scales, low data requirements, fewer parameters, simple operation, fast speed, more accurate landscape simulation results, and more rigorous mining of driving factors.

[0061] In summary, the present invention integrates the advantages of the old model and eliminates error propagation; introduces the random forest classification algorithm to calculate the development probability of various types of land use, and combines it with the land expansion analysis strategy to explore the driving factors of land use change; then uses the roulette mechanism and the threshold descent rule to introduce competition among land use types, and combines multiple types of random patch seeds to achieve accurate simulation of changes in multiple types of land use patches and landscapes and simultaneously apply it to future land use scenario prediction; ultimately, it makes up for the shortcomings of the old model and obtains simulation results with higher accuracy than the old model. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present application will be further described below with reference to the accompanying drawings and embodiments, in which:

[0063] Figure 1 is a flow chart in an embodiment of the present application;

[0064] Figure 2 It is the initial land use classification map of 2003 in the embodiment of this application;

[0065] Figure 3 is a schematic diagram of various driving factors of land use changes in the embodiments of the present application;

[0066] Figure 4 Schematic diagram of the development probability of various types of land use output by the random forest in the embodiment of the present application;

[0067] Figure 5 It is a simulation result diagram in the embodiment of the present application;

[0068] Figure 6 This is a schematic diagram showing the comparison of the actual land use change distribution in 2013 in the embodiment of the present application. DETAILED DESCRIPTION

[0069] In order to have a clearer understanding of the technical features, purposes and effects of this application, the specific implementation methods of this application are now described in detail with reference to the accompanying drawings.

[0070] An embodiment of the present application provides a patch generation land use simulation method.

[0071] Please refer to Figure 1 , Figure 1 Flowchart of a land use simulation method for patch generation in an embodiment of the present application, including:

[0072] S1: Obtain initial land use data; obtain change raster data through the initial land use data;

[0073] Obtain a preset number of driving factors affecting land use and land cover changes to form driving factor data;

[0074] S2: Perform random point sampling on the driving factor data and the change grid data to obtain sampling data;

[0075] S3: Preprocess the sampled data; build a random forest model based on the random forest classification algorithm; train the random forest model using the preprocessed sampled data;

[0076] S4: Input all driving factor data into the trained random forest model to obtain the development probability of each type of land use and the development potential map of each land use type in the region;

[0077] Specifically, the output of the random forest in this example is Figure 4 Development potential of the seven land use types shown.

[0078] S5: Construct a CA model based on a multi-type random patch seed mechanism; set the transformation matrix and neighborhood weights of the CA model; set the target number of pixels in the CA model according to the required area of ​​each land use type;

[0079] S6: Input the development potential map and the initial land use data into the CA model, iteratively scan each pixel of the initial land use data, and obtain the land use type contained in the neighborhood, the proportion of the neighborhood, and the adaptive inertia coefficient of each pixel;

[0080] The overall probability of each land use type in each pixel is obtained by combining the land use types contained in the neighborhood, the proportion of the pixel in the neighborhood, the adaptive inertia coefficient, and the development probability of each type of land use.

[0081] S7: Construct a roulette wheel based on the overall probability; and transform various land use types within the pixel area into the land use type that wins the competition through the roulette wheel method.

[0082] When the adjacent effect of a land use type is equal to 0, different types of random seeds are generated by calculating the overall probability; new land use types are generated by random seeds;

[0083] S8: Based on the threshold decreasing rule, determine whether the conversion of the new land use type that wins the competition occurs;

[0084] S9: Repeat steps S6 to S8 until the iteration reaches the target number of pixels, stop the iteration, and output the land use simulation map.

[0085] Specifically, Figure 1 The models in this paper include a transformation rule mining framework (LEAS) module based on the land expansion analysis strategy and a CA model (CARS) module based on the multi-type random patch seed mechanism.

[0086] Step S1 includes:

[0087] S11: Acquire an initial land use classification image; convert or reclassify the initial land use classification image to obtain initial land use data; the initial land use data includes: first-period land use data and second-period land use data;

[0088] S12: extracting the changed parts between the two periods of the initial land use data, that is, extracting the grids with changed states from the second period land use data;

[0089] S13: Extracting grids whose states have changed between the two periods, ie, changed grid data, based on the first period land use data and the second period land use data.

[0090] Specifically, Figure 2 To collect the initial land use change data, Landsat images were classified into seven land use types, including built-up areas, cultivated land, deciduous broad-leaved forest, evergreen broad-leaved forest, grassland, water bodies, and bare land. Figure 3 The raster maps of various driving factors used in this example include: Wuhan City elevation, Wuhan City slope, Wuhan City soil type, Wuhan City average annual precipitation, Wuhan City average annual temperature, Wuhan City adjacent open water, Wuhan City expressway linear vector, Wuhan City railway linear vector, Wuhan City main traffic arterial road vector, Wuhan City secondary traffic road vector, Wuhan City tertiary traffic road vector, Wuhan City adjacent high-speed rail stations, Wuhan City adjacent government, Wuhan City GDP data, Wuhan City population data.

[0091] Step S2 includes:

[0092] Using uniform sampling strategy or proportional sampling strategy, random point sampling is performed on the driving factor data and the change grid data to obtain the sampling data. The sample formula after sampling is expressed as:

[0093]

[0094] in Indicates the The first sampling point is extracted Driver data; Indicates the The data of the change grid data extracted from the sampling points; is transposed.

[0095] Specifically, the uniform sampling strategy makes the sampling points of each land use type consistent, which is suitable for areas with large differences in the areas of different land use types; the proportional sampling method makes the proportion of sampling points of each land use type in the total sampling points consistent with the ratio of each initial land use type to the regional area, with less calculation amount, and is suitable for sampling in larger-scale areas.

[0096] Step S3 includes:

[0097] The sampled data is normalized, and the normalization formula is as follows:

[0098]

[0099] in and Respectively represent The maximum and minimum values ​​of each driver; .

[0100] Step S3 further includes:

[0101] The random forest classification algorithm is expressed as follows:

[0102]

[0103] in is a constant, either 0 or 1; It is a vector composed of a preset number of driving factors; is the indicator function of the decision tree set; is a vector The prediction type of the nth decision tree; is the total count of the decision tree.

[0104] Specifically, this example uses the random forest algorithm to explore the mapping relationship between the development and changes of urban functional land type k and the driving factors . It is a binary classifier, so the value of d is 0 or 1.

[0105] Step S6 includes:

[0106] The overall probability synthesis formula is:

[0107]

[0108] in is a random value from 0 to 1; For land use type The threshold for generating new land use patches, which are mined by users; is the neighborhood function;

[0109] A scanning window is constructed through the neighborhood function, and the number of pixels in the scanning window is counted to measure the spatial interaction of various land use types.

[0110] Neighborhood function is defined as follows:

[0111]

[0112] in Indicates the land use type at the last iteration within the n×n window The total number of occupied grid cells, is the weight between different land use types;

[0113] Adaptive function The adaptive method is defined as follows:

[0114]

[0115] in and It is Second and Land use type at the iteration The difference between current quantities and future demand.

[0116] The threshold decreasing rule, the formula is as follows:

[0117]

[0118]

[0119] in, Decrease threshold The attenuation coefficient, Random values ​​that follow a normal distribution; is the decreasing threshold; is the number of attenuation times; is the decreasing step size; Defines whether land use type k is allowed to be converted to type The transformation matrix; Represents the random forest probability, that is, the pixel Developed into a type probability.

[0120] This application provides an embodiment as follows:

[0121] The study target is Wuhan, the capital of Hubei Province. Located in central Hubei Province, Wuhan covers an area of ​​approximately 8,494 square kilometers, making it a megacity in central China. The test data used in this study area include land use data from Wuhan in 2003 and 2013, which served as the initial simulation data. A decision tree was used to classify the corresponding Landsat imagery from 2003 and 2013. Land use change data from Wuhan interpreted from Landsat imagery in 2013 served as validation data. The 2003 and 2013 data share the same land use classification system. All regional data files were unified into raster images of the same scale to facilitate pixel-level homogeneity analysis. Based on research experience in land use change simulation and combining historical and current data from Wuhan, fifteen drivers of land use / land cover change were selected for this example. The driving factors are: Wuhan City elevation, Wuhan City slope, Wuhan City soil type, Wuhan City average annual precipitation, Wuhan City average annual temperature, Wuhan City adjacent to open water, Wuhan City highway linear vector, Wuhan City railway linear vector, Wuhan City traffic main road vector, Wuhan City traffic secondary road vector, Wuhan City traffic tertiary road vector, Wuhan City adjacent to high-speed rail stations, Wuhan City adjacent to government, Wuhan City GDP data, Wuhan City population data.

[0122] The main reasons for selecting Wuhan as the research subject in this paper are: Wuhan is located in the middle reaches of the Yangtze River. Its gross domestic product (GDP) reached US$91.3 billion in 2011, ranking 13th among all Chinese cities. It is a major transportation hub and economic center. Wuhan enjoys advantageous geographical conditions, being located at the confluence of the Yangtze and Han Rivers and endowed with abundant natural resources such as forests, minerals, and water resources. Water bodies occupy a large area of ​​Wuhan, and forests are mainly distributed in hilly areas, with natural vegetation mainly consisting of deciduous broad-leaved trees. Wuhan has experienced rapid urban development over the past decade, resulting in extensive expansion of built-up land (from 4.19×104 hectares in 1988 to 49.39×104 hectares in 2011), encroaching on surrounding land with ecological value (i.e., cultivated land, forest land, grassland, and water surface). Studying the potential causes of land use change in Wuhan and identifying sustainable land use options for the city are of great significance to its urban development. The study area includes 11 of Wuhan's 13 districts, encompassing the central metropolitan area.

[0123] Step 1: Convert or reclassify the initial land use classification image to obtain initial land use data; select several driving factors that affect land use / land cover changes to form driving force data; overlay the land use data of the two periods and extract the changed parts between the two periods, that is, extract the grids with changed states from the second period of land use data.

[0124] Step 2: Random point sampling is performed on the driving force data and the change grid data output by S1. The sampling method is selected according to the characteristics of the initial land use data. The model provides two sampling methods: 1. Uniform sampling strategy; 2. Proportional sampling strategy. In this example, the uniform sampling strategy is used to make the sampling points of various land use types consistent. Figure 2 Land expansion results data and Figure 3 The driving force data in , the sampling rate is set to 0.01.

[0125] Step 3: Use the random forest algorithm to train the sampled data. Set the training parameters and sampling rate, and then train the regression tree. In this example, the number of decision trees is set to 20, and the maximum number of features in the random forest model is set to 16, which is equal to the number of driving factors.

[0126] Step 4: Input all driving force data into the trained random forest and output the development potential of each land use type in the region.

[0127] Step 5: Input the development potential map and initial land use data into the CA model module of the multi-type random patch seed mechanism, set the conversion matrix and neighborhood weights, and set the target number of pixels based on the required area of ​​each land use type, that is, the number of pixels that each type of land use needs to reach in the future;

[0128] In this example, the image pixel size for the Wuhan region is 100 meters, so a 3×3 moorish neighborhood is selected as the neighborhood range. The simulation period for this example runs from 2003 to 2013, so the target pixel number matches the 2013 land use classification image for each category. The simulation automatically stops when the area of ​​all land use types reaches their corresponding required area. The conversion matrix restricts deciduous forests and towns from being converted to other land uses, and evergreen forests from being converted to other land uses except for cultivated land. The neighborhood weight for each land use type is determined based on expert knowledge and a series of model tests, ranging from 0 to 1.

[0129] Step 6: Iteratively scan the pixels of the initial land use data, and synthesize the overall probability of various land use types in the pixel by calculating the land use types contained in the neighborhood of the pixel and the proportion of the land use types in the neighborhood, the adaptive inertia coefficient and the development probability of each type of land use; in addition, when the pixel neighborhood probability is 0, the overall probability of this type of land use type is composed of the development probability, the adaptive inertia coefficient, the random number, and the growth threshold of this type of land use type.

[0130] In this example, a scanning window is constructed using a neighborhood function to count the number of pixels of each type within the window and measure the spatial interaction between different land use types. The neighborhood rule is defined as follows:

[0131]

[0132] in represents the total number of grid cells occupied by land use type k at the last iteration within the n×n window, is the weight between different land use types, because different land use types have different neighborhood effects. In this example, n=3 is used to construct the Moore neighborhood.

[0133] According to an embodiment of the present invention, the overall probability synthesis formula for roulette is:

[0134]

[0135] in is a random value from 0 to 1; is the threshold for generating new land use patches for land use type k, which is mined by users, is the domain function, is an adaptive function. In this example, the maximum threshold for the number of new seeds generated is set to 0.0001, which means that only 0.01% of the cells can generate new patches. The definition is as follows:

[0136]

[0137] in and It is Second and Land use type at the iteration The difference between current quantities and future demand.

[0138] Step 7: The overall probability of various land use types on the pixel forms a roulette wheel. By using the roulette wheel method, various land use types in the region compete on the pixel. This method proposes a threshold decreasing mechanism: when a new land use type wins in a round of competition, a decreasing threshold is used to determine whether the conversion occurs. The decreasing threshold for generating new patches set in this example is is 0.5, and the attenuation coefficient is set to 0.1.

[0139] Step 8: After iterating over all valid pixels in an image, return to step 6 to refresh the land use change image and enter the next iteration. Calculate the difference to the target number of pixels. When the target number of pixels is reached, stop the iteration and output the results.

[0140] Figure 5 The final output land use simulation map is: Figure 6To compare the simulation results of the new model with the previous old model, the models involved in the comparison are: the degraded PLUS model (I) (using PAS), the degraded PLUS model (II) (not using the multi-type random patch seed mechanism) and the FLUS model (CA model based on artificial neural network).

[0141] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned patch generation land use simulation method.

[0142] The above are merely exemplary embodiments of the present disclosure and are not intended to limit the scope of the present disclosure. In other words, any equivalent variations and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Those skilled in the art will readily conceive of other embodiments of the present disclosure after considering the disclosure and the practical implications thereof.

[0143] This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not described herein. The description and examples are to be considered as exemplary only, and the scope and spirit of the present disclosure are to be defined by the claims.

Claims

1. A patch generation land use simulation method, characterized in that: The method comprises the following steps: S1: Obtain initial land use data; obtain change raster data through the initial land use data; Obtain a preset number of driving factors affecting land use and land cover changes to form driving factor data; S2: Perform random point sampling on the driving factor data and the change grid data to obtain sampling data; S3: Preprocess the sampled data; build a random forest model based on the random forest classification algorithm; train the random forest model using the preprocessed sampled data; S4: Input all driving factor data into the trained random forest model to obtain the development probability of each type of land use and the development potential map of each land use type in the region; S5: Construct a CA model based on a multi-type random patch seed mechanism; set the transformation matrix and neighborhood weights of the CA model; set the target number of pixels in the CA model according to the required area of ​​each land use type; S6: Input the development potential map and the initial land use data into the CA model, iteratively scan each pixel of the initial land use data, and obtain the land use type contained in the neighborhood, the proportion of the neighborhood, and the adaptive inertia coefficient of each pixel; The overall probability of each land use type in each pixel is obtained by combining the land use types contained in the neighborhood, the proportion of the pixel in the neighborhood, the adaptive inertia coefficient, and the development probability of each type of land use. S7: Construct a roulette wheel based on the overall probability; and transform various land use types within the pixel area into the land use type that wins the competition through the roulette wheel method. When the adjacent effect of a land use type is equal to 0, different types of random seeds are generated by calculating the overall probability; new land use types are generated by random seeds; S8: Based on the threshold decreasing rule, determine whether the conversion of the new land use type that wins the competition occurs; S9: Repeat steps S6 to S8 until the iteration reaches the target number of pixels, stop the iteration, and output the land use simulation map.

2. The land use simulation method for patch generation according to claim 1, characterized in that: Step S1 includes: S11: Acquire an initial land use classification image; convert or reclassify the initial land use classification image to obtain initial land use data; the initial land use data includes: first-period land use data and second-period land use data; S12: extracting the changed parts between the two periods of the initial land use data, that is, extracting the grids with changed states from the second period land use data; S13: Extracting grids whose states have changed between the two periods, ie, changed grid data, based on the first period land use data and the second period land use data.

3. The land use simulation method for patch generation according to claim 1, characterized in that: Step S2 includes: Using uniform sampling strategy or proportional sampling strategy, random point sampling is performed on the driving factor data and the change grid data to obtain the sampling data. The sample formula after sampling is expressed as: in Indicates the The first sampling point is extracted Driver data; Indicates the The data of the change grid data extracted from the sampling points; is transposed.

4. The patch generation land use simulation method according to claim 1, wherein: Step S3 includes: The sampled data is normalized, and the normalization formula is as follows: in and Respectively represent The maximum and minimum values ​​of each driver; .

5. The land use simulation method for patch generation according to claim 1, wherein: Step S3 further includes: The random forest classification algorithm is expressed as follows: in is a constant, either 0 or 1; It is a vector composed of a preset number of driving factors; is the indicator function of the decision tree set; is a vector The prediction type of the nth decision tree; is the total count of the decision tree.

6. A land use simulation method for patch generation according to claim 5, characterized in that: Step S6 includes: The overall probability synthesis formula is: in is a random value from 0 to 1; For land use type The threshold for generating new land use patches, which are mined by users; is the neighborhood function; A scanning window is constructed through the neighborhood function, and the number of pixels in the scanning window is counted to measure the spatial interaction of various land use types. Neighborhood function is defined as follows: in Indicates the land use type at the last iteration within the n×n window The total number of occupied grid cells, is the weight between different land use types; Adaptive function The adaptive method is defined as follows: in and It is Second and Land use type at the iteration The difference between current quantities and future demand.

7. The land use simulation method for patch generation according to claim 1, wherein: Step S7 includes: The threshold decreasing rule, the formula is as follows: in, Decrease threshold The attenuation coefficient, Random values ​​that follow a normal distribution; is the decreasing threshold; is the number of attenuation times; is the decreasing step size; Defines whether land use type k is allowed to be converted to type The transformation matrix; Represents the random forest probability, that is, the pixel Developed into a type probability.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 7 is executed.

Images