Land utilization space distribution prediction method considering ground feature space distribution characteristics

Abstract

The invention relates to a land utilization spatial distribution prediction method considering ground feature spatial distribution characteristics, and the method comprises the steps: obtaining key factors and land utilization data of target land class spatial distribution, inputting the key factors and the land utilization data into an artificial neural network model, and obtaining a suitability distribution probability, utilizing a Markov model to predict the future grid number of each land utilization type; the method comprises the following steps: acquiring a land type area variation ratio, determining a neighborhood weight parameter based on the land type area variation ratio, acquiring a land utilization type transfer matrix, setting a cost matrix through the land utilization type transfer matrix, further acquiring spatial distribution of each land utilization type, and calculating a shape control parameter according to the spatial distribution; and inputting the suitability distribution probability, the future grid number, the neighborhood weight parameter, the cost matrix and the shape control parameter into a CA model to obtain a prediction result of target land class spatial distribution. The method can better adapt to and cope with complex scenes in practical application.

Description

Technical Field

[0001] The present invention relates to the technical field of land use spatial distribution prediction, and particularly to a land use spatial distribution prediction method considering the spatial distribution characteristics of ground objects. Background Art

[0002] With the acceleration of the pace of globalization and urbanization, land use / cover change has become a research hotspot in the fields of environmental science and geographical science. Selecting a scientific land use simulation model to predict the future spatial pattern of land use not only helps to optimize the spatial structure of land use, promote the efficient and reasonable use of national land space, but also can promote the healthy and sustainable development of regions. In recent years, the idea of research on the simulation and prediction of future land spatial distribution is to use a top-down method to determine the overall trend and macro demand of land use, and then combine a bottom-up method to simulate and verify the implementation process of these trends and demands at the micro level, achieving the combination of top-down and bottom-up.

[0003] With the in-depth research and the increasing complexity of application scenarios, the Markov-FLUS model has gradually exposed some deficiencies. In the application of the Markov-FLUS model, calculating the conversion probability through the CA model is the core link of the whole model. However, when the CA model considers the probability of the target grid changing at the next moment, it only focuses on the quantity of ground objects within the selected neighborhood range, and fails to fully consider the spatial distribution characteristics of ground objects. Summary of the Invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the object of the present invention is to propose a land use spatial distribution prediction method considering the spatial distribution characteristics of ground objects, improve the Markov-FLUS model, and improve the accuracy of the model for predicting future land use changes, so as to better adapt to and cope with complex scenarios in practical applications.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A land use spatial distribution prediction method considering the spatial distribution characteristics of ground objects, comprising:

[0007] Obtaining key factors of the spatial distribution of the target land type, where the key factors are environmental factors affecting the spatial distribution of land use types;

[0008] Input the key factors and land use data into an artificial neural network model to obtain the suitability distribution probability, and use the Markov model to predict the future grid numbers of each land use type; the artificial neural network model and the Markov model are respectively trained using a first training set and a second training set, the first training set includes: historical key factors and initial period land use data, and the second training set includes: historical period land use data;

[0009] Obtain the proportion of the change in land type area, which is used to quantitatively describe the expansion intensity of different land types to determine the neighborhood weight parameter. Determine the neighborhood weight parameter based on the proportion of the change in land type area, and obtain the land use type transfer matrix. Set the cost matrix through the land use type transfer matrix, where the cost matrix is the change rule between land use types. Further obtain the spatial distribution of each land use type, and calculate the shape control parameter according to the spatial distribution;

[0010] Input the suitability distribution probability, the future grid numbers, the neighborhood weight parameter, the cost matrix, and the shape control parameter into the CA model to obtain the prediction result of the target land type spatial distribution; the CA model is trained using a third training set, and the third training set includes: historical suitability distribution probability, historical future grid numbers, historical neighborhood weight parameter, historical cost matrix, and historical shape control parameter.

[0011] Optionally, the key factors include: elevation, slope, aspect, temperature, precipitation, population, distance from the economic center, distance from the road, and GDP.

[0012] Optionally, obtaining the suitability distribution probability includes:

[0013] p(p,k,t)=∑ j w j,k ×signoid(net j (p,t))

[0014]

[0015] net j (p,t)=∑ i w i,j ×x i (p,t)

[0016] where p(p,k,t) represents the probability that grid p appears as land type k at time t; w j，k is the adaptive weight between the hidden layer and the output layer; sigmoid(net j (p,t)) is the S-shaped activation function or S-shaped growth curve, which is the association function between the hidden layer and the output layer; netj (p, t) represents the signal sent by grid p on the input layer to neuron j at time t, that is, the intensity of the change of grid p in land type j at time t; w i，j Same as w j，k Both are adaptive weights, x i (p, t) is the relationship function between variable i at time t and grid p in input layer neuron i.

[0017] Optionally, determining the neighborhood weight parameter based on the proportion of the change amount of the land type area includes:

[0018] Performing dimensionless processing on the proportion of the change amount of the land type area, and determining the neighborhood weight parameter based on the processed proportion of the change amount of the land type area;

[0019] Performing dimensionless processing on the proportion of the change amount of the land type area includes:

[0020]

[0021] Among them, X * Represents the standardized value, X represents the proportion of the change amount of the land type area, max is the maximum value of the data, and min is the minimum value of the data.

[0022] Optionally, setting the cost matrix through the land use type transfer matrix includes:

[0023] Obtaining the change rule in the land use transfer matrix of historical time periods with equal annual intervals, and determining the conversion rule between land use types. The conversion rule is set to 1 when a certain land use type is allowed to be converted into other land use types, and set to 0 when a certain land use type is not allowed to be converted into other land use types;

[0024] Setting the cost matrix according to the change rule.

[0025] Optionally, the shape control parameters include: common point parameter, common edge parameter, distance parameter, and aggregation parameter; the common point parameter is the common point between grid cells of the same land use type, the common edge parameter is the common edge between grid cells of the same land use type, the distance parameter is used to judge the continuity and aggregation of the spatial distribution of land use types, and the aggregation parameter is the aggregation degree of areas with the same land use type.

[0026] Optionally, calculating the shape control parameter according to the spatial distribution includes:

[0027] Calculating the common point parameter includes:

[0028]

[0029] Among them, Ot p，k It represents the common point parameter of grid p for land use type k at time t. b represents the number of common points, and (N + 3)(N - 1) represents the total number of common points when all grids in the N×N Moore neighborhood are of the same land use type;

[0030] Calculating the common edge parameter includes:

[0031]

[0032] Among them, S t p,k It represents the common edge parameter of grid p for land use type k at time t. a represents the number of common edges, and 2N(N - 1) - 4 represents the total number of common edges when all grids in the N×N Moore neighborhood are of the same land use type;

[0033] Calculating the distance parameter includes:

[0034]

[0035] Among them, D t p,k It represents the distance parameter of grid p for land use type k at time t, It represents the Euclidean distance from the grid where the i-th land use type k is located to the central grid p;

[0036] Calculating the aggregation parameter includes:

[0037]

[0038] In the formula, G t p,k It represents the aggregation parameter of grid p for land use type k at time t, g t p,k It represents the number of grids where land use type k forms an island within the N×N Moore neighborhood at time t.

[0039] Optionally, obtaining the prediction result of the land use spatial distribution includes:

[0040] Inputting the suitability distribution probability, the number of future grids, the neighborhood weight parameter, the cost matrix, and the shape control parameter into the CA model, and through the roulette wheel competition mechanism, obtaining the prediction result of the land use spatial distribution.

[0041] The beneficial effects of the present invention are:

[0042] The present invention determines the key influencing factors of the spatial distribution of land use types. Based on the selected influencing factor data, it calculates the suitability distribution probability of each land use type through an artificial neural network, predicts the number of grids of each future land use type using the Markov model, determines the neighborhood weight parameters based on the change amount of land use type area, sets the cost matrix based on the land use type transfer matrix, and calculates the shape control parameters based on the spatial distribution of land use types. It substitutes the suitability distribution probability, simulation coefficient, and shape control parameters into the CA model to predict the spatial distribution of land use types during the verification period. It verifies the simulation accuracy according to the existing data and predicts the future spatial distribution of land use. The technical solution provided by the present invention can simulate the spatial distribution of different future land use types, improve the accuracy of the model's prediction of future land use changes, and better adapt to and respond to complex scenarios in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0044] Figure 1 It is a flowchart of a method for predicting the spatial distribution of land use considering the spatial distribution characteristics of ground objects according to an embodiment of the present invention;

[0045] Figure 2 It is a distribution map of the land use types in the Yellow River Basin in 2030 simulated according to an embodiment of the present invention;

[0046] Figure 3 It is a partial map of the spatial distribution of the land use types in the Yellow River Basin simulated in 2015 according to an embodiment of the present invention;

[0047] Figure 4 It is a partial map of the actual spatial distribution of the land use types in the Yellow River Basin in 2015 according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0049] To make the above objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] As Figure 1 shown, this embodiment discloses a method for predicting the spatial distribution of land use considering the spatial distribution characteristics of ground objects, including: obtaining the key factors of the spatial distribution of the target land type, where the key factors are environmental factors that affect the spatial distribution of land use types; inputting the key factors and land use data into an artificial neural network model to obtain the suitability distribution probability, and using the Markov model to predict the future grid numbers of each land use type; the artificial neural network model and the Markov model are respectively trained using the first training set and the second training set, the first training set includes: historical key factors and land use data in the initial period, and the second training set includes: historical land use data; obtaining the proportion of the change in land type area, which is used to quantitatively describe the expansion intensity of different land types to determine the neighborhood weight parameter, determining the neighborhood weight parameter based on the proportion of the change in land type area, and obtaining the land use type transfer matrix, setting the cost matrix through the land use type transfer matrix, where the cost matrix is the change rule between land use types, further obtaining the spatial distribution of each land use type, and calculating the shape control parameter according to the spatial distribution; inputting the suitability distribution probability, future grid numbers, neighborhood weight parameters, cost matrix, and shape control parameter into the CA model to obtain the prediction result of the spatial distribution of the target land type; the CA model is trained using the third training set, and the third training set includes: historical suitability distribution probability, historical future grid numbers, historical neighborhood weight parameters, historical cost matrix, and historical shape control parameter.

[0051] Specifically: This embodiment discloses a method for predicting the spatial distribution of land use considering the spatial distribution characteristics of ground objects, including:

[0052] Step (1), determining the key factors that affect the spatial distribution of land use types: Selecting environmental factors that affect the spatial distribution of land use through historical research and expert experience.

[0053] Step (2), calculating the suitability distribution probability, and calculating the suitability distribution probability through an artificial neural network based on the selected influence factor data.

[0054] Step (3), predicting the future grid numbers of different land types: Using the Markov model to predict the future grid numbers of each land type.

[0055] Step (4), determining the neighborhood weight parameter: Determining the neighborhood weight parameter based on the proportion of the change in the area of each land type, where the neighborhood weight parameter is the expansion intensity of each land type, that is, the ability to expand itself under the drive of external factors, with a threshold range of 0 to 1, and the value closer to 1 represents the stronger expansion ability of this type.

[0056] Step (5), set the cost matrix: Set the cost matrix based on the land use type transfer matrix, where the cost matrix refers to the change rules between land use types and is used to indicate whether the conversion between land types is allowed. Set it to 1 when the conversion is allowed and 0 when the conversion is not allowed.

[0057] Step (6), calculate the shape control parameters: Calculate the common point parameters, common edge parameters, distance parameters, and aggregation parameters based on the spatial distribution of land use types.

[0058] Step (7), predict the land use spatial distribution during the verification period: Substitute the suitability distribution probability, the number of future grids, the neighborhood weight parameter, the cost matrix, and the shape control parameters into the CA model to predict the land use spatial distribution during the verification period.

[0059] Step (8), verify the simulation accuracy based on existing data: Verify the overall accuracy through the Kappa coefficient, OA coefficient, and FoM coefficient.

[0060] Step (9), predict the future land use spatial distribution: Predict the future land use spatial distribution based on the improved Markov-FLUS model.

[0061] Furthermore, the key factors include: elevation, slope, aspect, temperature, precipitation, population, distance from the economic center, distance from the road, and GDP.

[0062] Specifically: In step (1), summarize the previous studies and select multiple natural factors and human factors as the factors affecting the spatial distribution of the target land type.

[0063] Furthermore, obtaining the suitability distribution probability includes:

[0064] p(p,k,t) = ∑ j w j,k × signoid(net j (p,t))

[0065]

[0066] net j (p,t) = ∑ i w i,j × x i (p,t)

[0067] where p(p,k,t) represents the probability that grid p appears as land type k at time t; w j，k is the adaptive weight between the hidden layer and the output layer; sigmoid(net j(p, t)) is an S-shaped activation function or S-shaped growth curve, which is the correlation function between the hidden layer and the output layer; net j (p, t) represents the signal sent by grid p on the input layer to neuron j at time t, that is, the intensity of change of grid p in the j-th type of land at time t; w i，j Same as w j，k Both are adaptive weights, x i (p, t) is the relationship function between variable i at time t and grid p in input layer neuron i

[0068] Specifically:

[0069] In step (2), the artificial neural network is used to calculate the suitability distribution probability based on the impact factor data and land use data, and establish the relationship between the probability of occurrence of each land type and the impact factors

[0070] p(p, k, t) = ∑ j w j,k × signoid(net j (p, t))

[0071]

[0072] net j (p, t) = ∑ i w i,j × x i (p, t)

[0073] In the formula, p(p, k, t) represents the probability of grid p appearing in the k-th land type at time t; w j，k is the adaptive weight between the hidden layer and the output layer; sigmoid(net j (p, t)) is also called the S-shaped activation function, or S-shaped growth curve, which is the correlation function between the hidden layer and the output layer; net j (p, t) represents the signal sent by grid p on the j-th input layer to neuron j at time t, that is, the intensity of change of grid p in the j-th type of land at time t; w i，j Same as w j，k Both are adaptive weights, and the difference is that w i，j represents the adaptive weight relationship between the input layer and the hidden layer; x i (p, t) is the relationship function between variable i at time t and grid p in input layer neuron i

[0074] In step (3), the Markov model is used to predict the total number of pixels of each land use type during the verification period through the land use type transfer probability matrix of the two periods before the verification period. Thus, the number of pixels of various future lands can be predicted based on the historical change trend of the land. After passing the accuracy verification, the total number of pixels of various future lands is predicted again.

[0075] The formula of the Markov model is:

[0076] S (t+1) = P ij S (t)

[0077] In the formula, S(t) and S(t + 1) are the system states at times t and t + 1 respectively; P ij is the state transition probability matrix and satisfies: 0 ≤ P ij < 1.

[0078] Furthermore, determining the neighborhood weight parameter based on the proportion of the change in land use type area includes:

[0079] Performing dimensionless processing on the proportion of the change in land use type area, and determining the neighborhood weight parameter based on the processed proportion of the change in land use type area;

[0080] Performing dimensionless processing on the proportion of the change in land use type area includes:

[0081]

[0082] Among them, X * represents the standardized value, X represents the proportion of the change in land use type area, max is the maximum value of the data, and min is the minimum value of the data.

[0083] Specifically: In step (4), the change in the total land area (TotalArea, TA) of each land use type in the landscape pattern index can quantitatively describe the expansion intensity of different land use types, thereby determining the neighborhood weight parameter. Perform dimensionless processing on the proportion of the change in TA, so that the change values of TA of different land use types are at the same order of magnitude for comprehensive evaluation and analysis.

[0084]

[0085] In the formula, X * represents the standardized value; max is the maximum value of the data; min is the minimum value of the data.

[0086] Furthermore, setting the cost matrix through the land use type transfer matrix includes:

[0087] Obtain the variation law in the land use transfer matrix for historical time periods with equal annual intervals, determine the conversion rules between different land use types. The conversion rule is set to 1 when a certain land use type is allowed to be converted into other land use types, and set to 0 when a certain land use type is not allowed to be converted into other land use types; according to the variation law, set up the cost matrix.

[0088] Specifically: In step (5), summarize the rules from the land use transfer matrix of the time period equal to the prediction year before the initial period, summarize the conversion rules between land use types, and set up the cost matrix.

[0089] Furthermore, the shape control parameters include: common point parameter, common edge parameter, distance parameter, and aggregation parameter; the common point parameter is the common point between grid cells of the same land use type, the common edge parameter is the common edge between grid cells of the same land use type, the distance parameter is used to judge the continuity and agglomeration of the spatial distribution of land use types, and the aggregation parameter is the degree of aggregation of areas with the same land use type.

[0090] Furthermore, calculating the shape control parameters according to the spatial distribution includes:

[0091] Calculating the common point parameter includes:

[0092]

[0093] Among them, O t p，k represents the common point parameter of grid p for land use type k at time t, b represents the number of common points, and (N + 3)(N - 1) represents the total number of common points when all grid cells in the N×N Moore neighborhood are of the same land use type;

[0094] Calculating the common edge parameter includes:

[0095]

[0096] Among them, S t p,k represents the common edge parameter of grid p for land use type k at time t, a represents the number of common edges, and 2N(N - 1) - 4 represents the total number of common edges when all grid cells in the N×N Moore neighborhood are of the same land use type;

[0097] Calculating the distance parameter includes:

[0098]

[0099] Among them, D t p,k represents the distance parameter of grid p for land use type k at time t, Denote the Euclidean distance from the grid where the \(i\)-th land use type \(k\) is located to the central grid \(p\);

[0100] The calculation of the aggregation parameter includes:

[0101]

[0102] In the formula, \(G\) t p,k Denote the aggregation parameter of land use type \(k\) at grid \(p\) at time \(t\), and \(g\) t p,k Denote the number of grids where land use type \(k\) forms an island within the \(N\times N\) Moore neighborhood at time \(t\).

[0103] Specifically: In step (6), based on the spatial distribution of land use types, calculate the common point parameter, common edge parameter, distance parameter, and aggregation parameter of different land use types at a certain moment in the iteration. A common point refers to the common point of grid cells of the same land use type. Since grid data is discrete, by examining the attribute values of common edges, the continuity, change rate, etc. of spatial data can be analyzed.

[0104]

[0105] In the formula, \(O\) t p，k Denote the common point parameter of land use type \(k\) at grid \(p\) at time \(t\), \(b\) represents the number of common points, and \((N + 3)(N - 1)\) represents the total number of common points when all grids in the \(N\times N\) Moore neighborhood are of the same land use type.

[0106] A common edge refers to the common edge of grid cells of the same land use type. By examining the changes in the attribute values of these common points, spatial distribution characteristics such as gradients and trends can be revealed. The formula for setting the common edge parameter is as follows:

[0107]

[0108] In the formula, \(S\) t p,k Denote the common edge parameter of land use type \(k\) at grid \(p\) at time \(t\), \(a\) represents the number of common edges, and \(2N(N - 1)-4\) represents the total number of common edges when all grids in the \(N\times N\) Moore neighborhood are of the same land use type.

[0109] The distance between the same land use types will affect the continuity and aggregation of their distribution. When the distance between the same land use types is relatively close, they tend to form continuous and large-scale distribution areas. When the distance is relatively far, they tend to decrease.

[0110]

[0111] In the formula, \(D\)t p,k Denote the distance parameter of grid p for land use type k at time t. Denote the Euclidean distance from the grid where the i-th land use type k is located to the central grid p.

[0112] Considering the four directions of up, down, left, and right of the grid, the number of grids in the largest aggregation area with the same land use type is defined as the aggregation degree.

[0113]

[0114] In the formula, G t p,k Denote the aggregation parameter of grid p for land use type k at time t, and g t p,k Denote the number of grids where land use type k forms an island at time t within the N×N Moore neighborhood range.

[0115] Furthermore, obtaining the prediction result of the land use spatial distribution includes:

[0116] Input the suitability distribution probability, the number of future grids, the neighborhood weight parameter, the cost matrix, and the shape control parameter into the CA model. Through the roulette wheel competition mechanism, obtain the prediction result of the land use spatial distribution for the overall transition probability of cells.

[0117] Specifically: In step (7), combine the development probability with the domain factor, the adaptive inertia coefficient, and the conversion cost to obtain the overall transition probability of cells. Through the roulette wheel competition mechanism, finally obtain the land use simulation result during the verification period.

[0118] After calculating the total probability of land use change, use the roulette wheel mechanism to determine which future land use type will appear on the pixel. The random feature of roulette wheel selection enables the model to reflect the uncertainty of land use change in the real world and the alternating development of land use, and better considers the competition between land types.

[0119] In step (8), compare the verification period distribution map simulated by the Markov-FLUS model with the real verification period land use distribution map, and conduct an overall spatial accuracy test of the simulation results through the Kappa coefficient, the FoM coefficient, and the OA coefficient. When the Kappa coefficient is higher than 0.80, it indicates that the model reaches a good state in terms of statistical significance. The FoM index evaluates the performance of the grids where transfer occurs. The larger the FoM parameter value, the better the simulation effect and the higher the accuracy. However, practical verification shows that most of its results are within 0.3, and the results between 0.1 and 0.2 are the most common. A high OA value (close to 1) indicates that the simulation result is very consistent with the actual situation, while a low OA value indicates low classification accuracy.

[0120] In step (9), based on the land use distribution map and the impact factor normalization map during the verification period, and according to the land use type transfer matrix and the proportion of the change in the area of different land classes in the time period equal to the interval from before the verification period to the prediction year, the cost matrix and the neighborhood factor parameters are set to predict the future spatial distribution of land use.

[0121] As Figure 1 shown, this embodiment discloses a method for predicting the spatial distribution of land use considering the spatial distribution characteristics of ground objects, and the steps are as follows:

[0122] Step (1): Summarize the previous studies and select elevation, slope, aspect, temperature, precipitation, population, distance from the economic center, distance from the road, and GDP as the factors affecting the spatial distribution of land use types.

[0123] Step (2): Calculate the suitability distribution probability of each land class through an artificial neural network: First, use the initial data, after normalizing the driving force factors, extract 70% of the pixels of its raster data using a 10% random sampling strategy as training samples, and use the ANN model to calculate the suitability probability of each land class represented by each pixel.

[0124] Step (3): Predict the future raster numbers of each land class: Use the Markov model to predict the total number of pixels of each land class in 2015 and 2030 respectively through the land use data from 1995 to 2005 and from 2010 to 2020. Among them, the simulated value in 2015 is used to test the prediction accuracy of the Markov model. After passing the accuracy verification, the simulated value of the raster number of each land class in 2030 is used as the future pixel total parameter of the model.

[0125] Step (4): Determine the neighborhood weight parameter based on the proportion of the change in land class area: The neighborhood weight parameter value is obtained by dimensionless quantization of the proportion of the change in the TA value of each land class in the Yellow River Basin from 1980 to 2000.

[0126] Step (5): Set the cost matrix based on the land use transfer matrix: The present invention summarizes the rules using the land use type transfer matrix in the Yellow River Basin from 1980 to 2000 to determine whether different land classes can be converted into each other.

[0127] Step (6): Calculate the shape control parameter: Based on the spatial distribution of land use at each iteration, calculate the common point parameter, common edge parameter, distance parameter, and aggregation parameter of each land class.

[0128] Step (7), predict the spatial distribution of land use during the verification period: Combine the suitability probability with domain factors, adaptive inertia coefficient, conversion cost, and shape control parameters to obtain the overall conversion probability of the cells. Through the roulette competition mechanism, finally obtain the simulation results of the spatial distribution of land use types in the Yellow River Basin in 2015.

[0129] Step (8), verify the simulation accuracy based on existing data: Compare the spatial distribution map of land use types in the Yellow River Basin in 2015 simulated by the Markov-FLUS model with the real spatial distribution map of land use types in 2015, and conduct an overall spatial accuracy test of the simulation results through the Kappa coefficient and FoM coefficient. The results show that: the Kappa coefficient is 0.93, the FoM coefficient is 0.09, and the OA coefficient is 0.95.

[0130] Step (7), predict the future land use distribution: Use the spatial distribution map of land use types in the Yellow River Basin in 2015 and the normalized map of environmental factors, and set relevant parameters according to the data from 2000 to 2015, such as Figures 3 - 4 , simulate the spatial distribution of land use types in the Yellow River Basin in 2030, such as Figure 2 .

[0131] The present invention adopts the method of an improved Markov-FLUS model to complete the prediction of the spatial distribution of land use types in the Yellow River Basin in 2030, thereby providing a decision-making and management basis for the sustainable development of the Yellow River Basin.

[0132] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting the spatial distribution of land use taking into account the spatial distribution characteristics of land objects, characterized in that: include: Obtain key factors for the spatial distribution of target land use types, wherein the key factors are environmental factors that affect the spatial distribution of land use types; The key factors and land use data are input into an artificial neural network model to obtain a suitability distribution probability, and a Markov model is used to predict the number of future grids for each land use type; the artificial neural network model and the Markov model are trained using a first training set and a second training set, respectively, the first training set includes: historical key factors and initial land use data, and the second training set includes: historical land use data; Obtaining the proportion of land area change, which is used to quantitatively describe the expansion intensity of different land types to determine the neighborhood weight parameter, determining the neighborhood weight parameter based on the proportion of land area change, and obtaining the land use type transfer matrix, setting the cost matrix through the land use type transfer matrix, the cost matrix is ​​the change rule between each land use type, further obtaining the spatial distribution of each land use type, and calculating the shape control parameter according to the spatial distribution; The suitability distribution probability, the future grid number, the neighborhood weight parameter, the cost matrix and the shape control parameter are input into the CA model to obtain the prediction result of the spatial distribution of the target land type; the CA model is trained using the third training set, and the third training set includes: historical suitability distribution probability, historical future grid number, historical neighborhood weight parameter, historical cost matrix and historical shape control parameter.

2. The method for predicting the spatial distribution of land use taking into account the spatial distribution characteristics of land objects according to claim 1, characterized in that: The key factors include: elevation, slope, aspect, temperature, precipitation, population, distance to economic center, distance to road and GDP.

3. The method for predicting spatial distribution of land use taking into account spatial distribution characteristics of land objects according to claim 1, characterized in that: Obtaining the suitability distribution probability includes: p(p,k,t)=∑ j w j,k ×signoid(net j (p,t)) net j (p,t)=∑ i w i,j ×x i (p,t) Where p(p,k,t) represents the probability of the occurrence of land type k in grid p at time t; w j，k is the adaptive weight between the hidden layer and the output layer; sigmoid(net j (p, t)) is the S-type activation function or S-type growth curve, which is the correlation function between the hidden layer and the output layer; net j (p, t) represents the signal sent by the grid p on the input layer to the neuron j at time t, that is, the intensity of the change in the grid p in the j-th type of land at time t; w i，j Same as w j，k are all adaptive weights, x i (p,t) is the relationship function between time variable i and grid p in neuron i of the input layer.

4. The method for predicting spatial distribution of land use taking into account spatial distribution characteristics of land objects according to claim 1, characterized in that: Determining the neighborhood weight parameter based on the land area change ratio includes: Performing dimensionless processing on the land area change ratio, and determining a neighborhood weight parameter based on the processed land area change ratio; The dimensionless processing of the land area change ratio includes: Among them, X * represents the standardized value, X represents the percentage of land area change, max represents the maximum value of the data, and min represents the minimum value of the data.

5. The method for predicting spatial distribution of land use taking into account the spatial distribution characteristics of land objects according to claim 1, characterized in that: Setting the cost matrix through the land use type transfer matrix includes: Obtain the change rules in the land use transfer matrix of historical time periods with equal annual intervals, and determine the conversion rules between various land use types. The conversion rule is set to 1 when a certain land use type is allowed to be converted into other land use types, and set to 0 when a certain land use type is not allowed to be converted into other land use types; According to the changing rules, a cost matrix is ​​set.

6. The method for predicting spatial distribution of land use taking into account spatial distribution characteristics of land objects according to claim 1, characterized in that: The shape control parameters include: common point parameters, common edge parameters, distance parameters and aggregation parameters; the common point parameters are: common points between grid units of the same land use type, the common edge parameters are the common edges between grid units of the same land use type, the distance parameters are used to judge the continuity and agglomeration of the spatial distribution of land use types, and the aggregation parameters are the degree of aggregation of areas with the same land use type.

7. The method for predicting spatial distribution of land use taking into account spatial distribution characteristics of land objects according to claim 6, characterized in that: Calculating the shape control parameters according to the spatial distribution includes: Calculating the common point parameters includes: in, represents the common point parameter of grid p at time t for land use type k, b represents the number of common points, and (N+3)(N-1) represents the total number of common points when all grids in the N×N Moore neighborhood are of the same land type; Calculating the common edge parameters includes: in, represents the common edge parameter of the land use type k of grid p at time t, a represents the number of common edges, and 2N(N-1)-4 represents the total number of common edges when all grids in the N×N Moore neighborhood are of the same land type; Calculating the distance parameter includes: in, represents the distance parameter of land use type k at grid p at time t, represents the Euclidean distance from the grid where the i-th land class k is located to the central grid p; Calculating the aggregation parameter includes: In the formula, represents the aggregation parameter of land use type k at grid p at time t, Represents the number of cells forming islands of land use type k at time t within an N×N Moore neighborhood.

8. The method for predicting spatial distribution of land use taking into account spatial distribution characteristics of land objects according to claim 1, characterized in that: Obtaining the prediction result of the land use spatial distribution includes: The suitability distribution probability, the future grid number, the neighborhood weight parameter, the cost matrix and the shape control parameter are input into the CA model, and the overall conversion probability of the cell is subjected to a roulette competition mechanism to obtain the prediction result of the land use spatial distribution.

Images