A method for mining driving factors of land use change based on random forest and crowd-source geographic information

Abstract

The present invention provides a method for mining driving factors of land use change based on random forest and multi-source geographic information, namely: first, a multi-source geographic data set of potential driving factors affecting land use change is constructed using multi-source geographic data mainly based on POI points, and data spatialization is performed; then, a random forest classifier model is constructed with multi-potential driving factors as characteristic variables and land use types of land use thematic maps as predictive variables, and model training is performed; then, the trained model is used to perform K random permutations of a single variable, thereby calculating the importance score of the variable, and ranking the importance of the driving factors according to the score; finally, the recursive feature elimination principle is used to screen the core driving forces affecting land use change. The present invention can quantitatively evaluate the importance of driving factors affecting urban land use change and screen core factors, thereby revealing the micro-driving mechanism of urban evolution.

Description

Technical Field

[0001] The present invention belongs to the field of geographic big data analysis and mining, and in particular to a method for mining driving factors of land use change based on random forests and crowd-source geographic information. Background Art

[0002] Mining the driving factors of land use change is the basis for revealing the mechanism of land use change, evolution rules and future trend simulation, and has always been an important direction of land use research. Domestic and foreign scholars have carried out a lot of research on the driving forces of land use change. Early studies mainly used empirical analysis to reveal the driving mechanism of a certain land use type in a characteristic region at the macro level. For example, Moran pointed out that forest degradation in the Amazon region of Brazil from 1975 to 1987 was mainly affected by changes in local government policies on livestock farms, rather than population growth factors. Although these qualitative analyses have laid a good foundation for the identification of land use driving forces, it is difficult to assess the impact of different types of human or natural factors on land use change. For this reason, statistical methods have been used in subsequent studies to conduct quantitative research on factor driving forces, and linear equation models have been constructed with multivariate driving factors as independent variables and land use changes as dependent variables, such as correlation analysis, logistic regression, linear regression, principal component analysis and other methods.

[0003] Considering the current situation that urban development is gradually shifting from outward-oriented spatial expansion to small-scale urban space redevelopment such as land function renewal and old town renovation, it is urgent to study the micro-driving mechanism of land use change. However, there are two main deficiencies in existing research that make it difficult to meet the research on the micro-driving force of land use change: first, most of the existing research focuses on the exploration of macro-driving factors of large-scale land use change, which has problems such as too large research scale and inaccurate classification of driving factors, making it difficult to support the discovery of laws and mechanism research on the functional transformation of urban construction land; second, urban land use change is interactively affected by multiple factors such as natural environment and social economy, and land use function and structure are more complicated under the influence of high-intensity human activities. Traditional statistical models are mostly based on linear equations, which simplify the relationship model between multiple driving factors and land use change, and cannot truly and comprehensively reflect the complex and nonlinear mapping relationship between the two.

[0004] With the development of Web 2.0 technology, crowdsourcing geographic information, a massive amount of data generated actively or passively by the public in their daily lives, has become an important supplement to professional geographic information data. The depth and breadth of geoscientific research have been further enhanced by using the micro characteristics of human activities and socio-economics reflected in crowdsourcing geographic information. Point of Interest (POI) data is the most widely used type of crowdsourcing geographic information. The large amount of dynamic and detailed socio-economic information contained in POI tags is applied to urban land use research, which provides the possibility of exploring the micro-driving mechanism of urban land use. However, in practical applications, the high data redundancy and strong information correlation brought by the rich crowdsourcing geographic information will inevitably cause serious interference to the accurate identification and screening of the core driving force. Therefore, it is necessary to establish a land use change driving force analysis method that is less affected by the correlation between variables, which can not only construct a nonlinear model between driving factors and land use change, but also avoid the interference of data redundancy, accurately identify the driving factors that dominate and contribute most from the multi-factor driving factors, and more deeply and finely explore the core driving factors that affect urban land use change. Random forest has a natural advantage in feature screening. It evaluates the importance of feature variables according to the contribution of the variables to the model as the result of analyzing the driving force factors of land use change. Random forest can measure the importance of feature variables according to the contribution of variables to the prediction results. By applying the random forest model to construct the relationship between land use categories and spatial variables, the core driving factors of land use change can be analyzed by calculating the importance of variables. The random forest model provides a method for evaluating variable importance, such as calculating the variable importance MDI (mean decrease in impurity) index based on Gini impurity. This method calculates the impact of each variable on the heterogeneity of observations at each node of the classification tree, thereby comparing the importance of variables. However, this algorithm may cause serious deviations in the evaluation of variable importance, mainly because the MDI index is a statistical value calculated based on the model training data, which cannot fully reflect the contribution of the variable to the model prediction. Related studies have shown that when the sample data is unevenly distributed, the MDI index will be misjudged as an important factor when the model is overfitted, and the indicators without obvious contribution will be misjudged as important factors. In addition, the variable importance evaluation based on Gini impurity tends to give high values ​​to continuous variables and underestimate the importance of discrete variables. To this end, this study introduced a variable permutation test method to evaluate the importance of feature variables based on the random forest model and crowd-source geographic information. By randomly permuting a single variable, the relationship between the variable and the prediction target that has been established in the model is broken. Then, indicators are calculated for the changes in model errors caused by the variable changes, thereby deriving the importance of a single variable.The advantage of this method is that there is no obvious bias in the importance assessment of discrete and continuous variables, and it can more accurately quantify the importance of the driving factors affecting urban land use changes and screen the core factors. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in order to reveal the micro-driving mechanism of land use change, a method for mining driving factors of land use change based on random forests and crowd-source geographic information is proposed to accurately reflect the importance of characteristic factors affecting urban land use change and realize the screening of core driving factors of land use change.

[0006] The present invention solves the technical problem by adopting the following technical solutions:

[0007] The present invention provides a method for mining driving factors of land use change based on random forest and crowd-source geographic information. The method comprises the following steps: firstly, a multivariate potential driving factor data set affecting land use change is constructed using crowd-source geographic data mainly based on POI points, and data spatialization is performed; then, a random forest classifier model is constructed using the multivariate potential driving factors as characteristic variables and the land use type of the land use thematic map as a predictor variable, and the model is trained; then, a single variable is randomly permuted K times using the trained model to calculate the importance score of the variable, and the importance of the driving factors is ranked according to the score; finally, the core driving force affecting land use change is screened using the recursive feature elimination principle.

[0008] The present invention can use the following methods to perform data spatialization processing: according to the data type characteristics of different factors, multiple methods such as kernel density estimation, buffer zone creation, Euclidean distance calculation, zoning statistics and slope and aspect calculation are used to perform data spatialization processing to generate a spatial variable set with consistent resolution and continuous surface type.

[0009] The present invention can eliminate the difference in the dimension of the spatial variable set data by using the following method: using the Fuzzy tool in Arcmap10.2 software to perform deviation standardization on the variables, normalizing the spatial variable pixel values, and mapping the numerical range of the variables to between 0 and 1.

[0010] The present invention can adopt the following method to construct a random forest classifier model: taking multivariate potential driving factors as characteristic variables and the land use type of the land use thematic map as a predictive variable, and constructing a mapping relationship between the two.

[0011] The random forest classifier model described in the present invention can be constructed by the following method to collect model training samples:

[0012] (1) The land use maps of the study area in two different years were imported into ArcMap software. The spatial classification rate of the data was 30m. The land use types were divided into nine categories: water area, forest land and grassland, cultivated land, unused land, residential land, industrial land, commercial land, public management land and mixed land. The category codes were numbers from 1 to 9.

[0013] (2) Use the Raster calculator in ArcMap software to detect changes in land use types in different years. The algebraic expression is Con("landuse1.tif"!="landuse2.tif",1,0). Generate a new raster dataset landuse_difference.tif. The pixel value of 1 represents the area where the land use type has changed, and the pixel value of 0 represents the area where there has been no change.

[0014] (3) For pixels where land use has changed, the random traversal sampling method is used according to the spatial position index of the pixel. The corresponding traversal step length is set for different types of land use, and the whole area search and sampling are performed to form a training sample set D = [(x 1 ,y 1 ),,...,(x n ,y n )].

[0015] The random forest classifier model constructed in the present invention can be trained by the following method: input the training sample D into the random forest model for model training, and set the maximum number of features to the square root of the number of potential driving factors N; perform model training by iteratively increasing the number of decision trees, and measure the model training effect by the average interval of model error:

[0016]

[0017] Where: MG avg represents the average interval of all samples, n represents the number of samples, mg(x i ,y i ) represents the interval of a single sample. If mg(x i ,y i ) is greater than zero, then the correct category occupies the most votes, and the final classification result is correct under voting; otherwise, the final classification result is wrong.

[0018] The present invention can use the following method to calculate and sort the importance scores of driving factors: for each feature variable j in sample D, randomly replace the value in the variable to generate a new, damaged training sample Calculate the new sample interval MG j, repeat 50 random permutations for each variable, and use the average value as the final result of variable importance:

[0019]

[0020] Where: i j represents the importance score of variable j, MG represents the average model interval before random permutation, K is the number of random permutations, and MG k，j represents the model average interval after the kth random permutation of variable j;

[0021] Using the importance scores of characteristic factors obtained in this step, the characteristic variables are sorted in descending order, which is the importance ranking of the driving factors of land use change.

[0022] The present invention can adopt the following method to screen the core driving forces: using the recursive feature elimination principle, sorting the driving factors according to their importance, starting from the most important driving factor, adding one factor at a time to form a new feature subset, inputting it into the random forest model, and using the cross-validation method to train the model and obtain a new model classification accuracy; repeating the above steps until all driving factors are included in the feature subset.

[0023] The core driving force screening method can determine the number of core driving forces by the following method: draw a curve showing how the model classification accuracy changes as the number of characteristic variables decreases, and find the point in the curve where the classification accuracy converges, which is the number of core driving factors.

[0024] The above-mentioned land use change driving factor mining method based on random forest and crowd-source geographic information provided by the present invention is used to accurately evaluate and screen the micro-driving factors that affect urban land use changes.

[0025] Compared with the prior art, the present invention has the following main technical effects:

[0026] (1) In view of the problems of too large research scale and inaccurate classification of driving factors in existing research on the driving forces of land use change, the present invention introduces multi-source geographic data containing rich socioeconomic information, and transforms the abstract driving mechanism of urban development into a quantitative feature expression in two-dimensional space through spatial statistical methods, thus achieving complete unification of driving factors and land use status in data form. This not only solves the problem of inconsistency between multi-source driving factors and the spatial scale of land use change, but also greatly improves the refinement of land use driving force analysis, laying a good data foundation for revealing the micro-driving mechanism of urban evolution.

[0027] (2) Considering that traditional statistical models oversimplify the complex relationship between multivariate driving factors and land use change, and it is difficult to truly reflect the micro-driving mechanism, the present invention designs a land use change driving factor importance assessment method based on random forests when mining land use micro-driving factors. This method reconstructs and tests the changes of the random forest model through random permutation of a single variable, and independently examines the influence of each driving factor on the model's prediction ability, thereby effectively avoiding the adverse effects of information redundancy between multivariate factors on the identification of core driving factors. Compared with the traditional variable importance MDI index in the random forest model, the advantage of the present invention is that there is no obvious bias when evaluating the importance of discrete and continuous variables, and the core driving factors can be screened out more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of the method of the present invention.

[0029] Figure 2 It is a spatialized result map of potential driving factors based on crowd-source geographic information.

[0030] Figure 3 It is the iterative training result of the random forest model.

[0031] Figure 4 It is a ranking chart of driving factors importance.

[0032] Figure 5 This is the core driving factor screening result chart. DETAILED DESCRIPTION

[0033] Aiming at the problem that traditional statistical models oversimplify the complex relationship between multivariate driving factors and land use change and are difficult to truly reflect the micro-driving mechanism, the present invention proposes a land use change driving factor importance assessment method based on random forest model and multi-source geographic information, uses multi-source geographic information containing rich socio-economic factors to construct a multivariate potential driving factor dataset of land use change, designs a classification interval index to measure the generalization error of the reconstructed model, and identifies the importance of characteristic factors based on the difference in interval sequences before and after variable replacement. This method independently examines the degree of influence of each driving factor on the model's predictive ability, thereby effectively avoiding the adverse effects of information redundancy between multivariate factors on the identification of core driving factors, and solves the problem that the traditional random forest algorithm is not sensitive enough to small variable disturbances, thereby improving the accuracy of identifying micro-driving factors of land use change.

[0034] The present invention is further described below in conjunction with application examples and drawings, but is not limited to the following contents.

[0035] The present invention provides a method for mining driving factors of land use change based on random forest and multi-source geographic information, specifically: first, using multi-source geographic information mainly based on POI point data (such as education, public services and transportation) to construct a multivariate potential driving factor data set affecting land use change, and generate data spatial processing; then, using multivariate potential driving factors as feature variables and land use types as predictive variables, a random forest classifier model is constructed, and model iterative training is performed; then, using the trained model, K random permutations of a single variable are performed to calculate the importance score of the variable and rank the importance of the driving factors; finally, using the recursive feature elimination principle, the core driving force affecting land use change is screened out. The present invention can accurately perform quantitative importance evaluation and core factor screening on the driving factors affecting urban land use change, thereby mining the micro-evolution mechanism of land use change.

[0036] In the above methods, the following methods can be used for data spatialization processing: for the multivariate potential driving factors represented by POI point data, linear data, surface data and raster data, the data are spatialized using kernel density estimation, buffer creation, Euclidean distance calculation, zonal statistics and slope and aspect calculation spatialization methods (the spatialization methods corresponding to different driving factors are shown in Table 1) to generate a spatial variable set with a resolution of 30m and continuous surface type. The specific spatialization method is implemented using the kernel density, Multiple Ring Buffer, euclidean allocation, zonal statistics, slope and aspect tools in Arcmap10.2 software. The original feature data are imported into the corresponding tools respectively, and the output is uniformly set to a result map with a resolution of 30m and tif format to form a spatial variable set, which is the multivariate potential driving factor dataset.

[0037] In the above method, the following method can be used to eliminate the difference in the dimension of the spatial variable set data: use the Fuzzy membership tool in Arcmap10.2 software to standardize the variables, import all the tif format data of the spatial variables into the tool in turn, keep the default settings, export the result map in tif format, and convert the pixel value into a floating point decimal between 0 and 1 to achieve the normalization of the spatial variable pixel value and eliminate the impact of the dimension and data level differences between different variables.

[0038] In the above method, the following method can be used to construct a random forest classifier model: 20 normalized spatial variables are used as feature variables (independent variables), and 9 land use types of land use thematic maps are used as predictive variables (dependent variables), and a random forest model is used to construct a mapping relationship between the two.

[0039] In the above method, the following methods can be used to collect model training samples:

[0040] (1) The land use maps of the study area in two different years were imported into ArcMap software. The data spatial classification rate was 30m. The land use types were divided into nine categories: water area, forest land and grassland, cultivated land, unused land, residential land, industrial land, commercial land, public management land and mixed land. The category codes were numbers from 1 to 9.

[0041] (2) The Raster calculator in ArcMap software was used to detect changes in land use types in different years. The algebraic expression was Con("landuse1.tif"!="landuse2.tif",1,0). A new raster dataset landuse_difference.tif was generated. The pixel value of 1 represented the area where the land use type changed, and the pixel value of 0 represented no change.

[0042] (3) Write a Python language program. For pixels with changed land use, use the random traversal sampling method according to the spatial position index of the pixel, set the corresponding traversal step size for different types of land use, perform full-domain search and sampling, and form a training sample set D = [(x 1 ,y 1 ),,...,(x n ,y n )], where: x 1 ...x n Represents the independent variable of n samples in the random forest model, y 1 ...y n Represents the dependent variable of n samples.

[0043] In the above method, the following method can be used for iterative training of the random forest model: write a Python language program, input the training sample D into the random forest model for model training, set the maximum number of features to the square root of the number of potential driving factors N; perform model training by iteratively increasing the number of decision trees, and measure the model training effect by the average interval of model error:

[0044]

[0045] Where: MG avg represents the average interval of all samples, n represents the number of samples, mg(x i ,y i ) represents the interval of a single sample. If mg(x i ,y i) is greater than zero, then the correct category occupies the most votes, and the final classification result is correct under voting; otherwise, the final classification result is wrong.

[0046] In the above method, the following method can be used to calculate and rank the importance scores of the driving factors: Write a Python language program to randomly replace the value of each feature variable j in sample D to generate a new, damaged training sample Calculate the new sample interval MG j , repeat the random permutation for each variable K times (50 times in this case), and take the average value as the final result of variable importance:

[0047]

[0048] Where: i j represents the importance score of variable j, MG represents the average model interval before random permutation, K is the number of random permutations, and MG k，j represents the model average interval after the kth random permutation of variable j.

[0049] Using the importance scores of characteristic factors obtained in this step, the characteristic variables are sorted in descending order, which is the importance ranking of the driving factors of land use change.

[0050] In the above method, the following method can be used to screen the core driving forces: use the recursive feature elimination principle to sort the driving factors according to their importance, starting from the most important driving factors, adding one factor at a time to form a new feature subset, input it into the random forest model, and use the cross-validation method to train the model and obtain the new model classification accuracy; repeat the above steps until all driving factors are included in the feature subset.

[0051] In the above method, the following method can be used to determine the number of core driving forces: draw a curve showing how the model classification accuracy changes as the number of feature variables decreases, and find the point in the curve where the classification accuracy tends to converge, which is the number of core driving factors.

[0052] The method for mining driving factors of land use change based on random forests and crowd-source geographic information provided by the present invention is used to reveal the micro-driving mechanism of urban evolution.

[0053] Application examples:

[0054] This case takes a central area of ​​a city as the research area. The area of ​​the area is 2724.228 square kilometers, accounting for 31.79% of the total area of ​​a city, and is the area with the highest degree of urbanization. Taking the driving force analysis of land use change in the area from 2015 to 2020 as an example, the present invention is further explained in combination with the attached drawings and tables.

[0055] Specific processing steps: Figure 1 )as follows:

[0056] Step 1: From the perspectives of socio-economic and natural ecology, a dataset of multiple potential driving factors affecting land use change is constructed based on multi-source geographic information, and the data is spatialized, including:

[0057] (1) The factors affecting urban land use change mainly include natural ecology and social economy, totaling 20 factors. Natural ecological factors include three topographic factors: elevation, slope and aspect, and three ecological factors: soil and water conservation function, soil organic matter and water system. Social economic factors come from multi-source geographic information such as POI, including 14 factors including population, economy, education, public service and transportation (Table 1).

[0058] (2) Obtain raster (tif format) or vector data (shapefile format) representing all natural ecological and socio-economic factors in the study area, and import them into ArcMap10.2, a professional geographic information data processing and analysis software, in turn. Use the Project function in the toolbox to perform coordinate projection transformation on data from multiple sources to keep the coordinate projection of the data consistent, which is: WGS_1984_UTM_Zone_49N. Then use the Clip function to clip all data into a uniform shape with the boundary of the study area as the clipping range.

[0059] (3) Different spatial processing methods are adopted for different types of factors, including kernel density estimation, buffer creation, Euclidean distance calculation and zonal statistics, which are implemented using the Kernel Density, Multiple Ring Buffer, Euclidean Distance and Zonal Statistics tools in Arcmap10.2 software respectively. The spatial processing methods corresponding to the 20 factors are detailed in Table 1. After processing, a spatial variable dataset with a resolution of 30m and a pixel number of 3366990 (2090*1611) will be generated, and the data format is tif type ( Figure 2 ).

[0060] (4) In the first step, in order to eliminate the differences in data dimensions between different factors, the Fuzzy tool in Arcmap10.2 software was used to standardize the variables and normalize the pixel values ​​of spatial variables. The numerical range of the variables was mapped to between 0 and 1.

[0061] In the second step, the multivariate potential driving factors were used as characteristic variables, and the land use type of the land use thematic map was used as the predictor variable to construct a random forest classifier model and conduct model training, which included:

[0062] (1) The land use thematic maps of a city center in 2015 and 2020 were imported into ArcMap software. The data spatial classification rate was 30m. The land use types were divided into nine categories: water area, forest land and grassland, cultivated land, unused land, residential land, industrial land, commercial land, public management land and mixed land. The category codes were numbers from 1 to 9.

[0063] (2) Use the Raster calculator in ArcMap software to detect changes in land use types from 2015 to 2020. The algebraic expression is Con("landuse2015.tif"!="landuse2020.tif",1,0) to generate a new raster dataset landuse_difference.tif. A pixel value of 1 represents a change in land use type from 2015 to 2020, and a pixel value of 0 represents no change. Pixels with a pixel value of 1 are selected, which are the areas where land use types have changed from 2015 to 2020, totaling 823,636 pixels.

[0064] (3) For the 823,636 pixels with land use changes, the random traversal sampling method was used according to the spatial position index of the pixels. The corresponding traversal step size was set for different types of land use, and the whole area search and sampling were performed. The 20 potential driving factors were used as characteristic variables and the land use type in 2020 was used as the predictor variable to form a training sample set D = [(x 1 ,y 1 ),,...,(x n ,y n )], including 5717 water samples, 3283 forest and grassland samples, 10583 cultivated land samples, 6509 unused land samples, 6852 residential land samples, 6791 industrial land samples, 4985 commercial land samples, 2416 public management land samples and 2570 mixed land samples.

[0065] The specific implementation code is as follows:

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072] (4) Input the training sample D into the random forest model for model training. Set the maximum number of features to the square root of the number of potential driving factors N. In this case, 20 driving factors are used, so the maximum number of features is The model is trained by iteratively increasing the number of decision trees, and the average interval of model error is used to analyze the model training effect:

[0073]

[0074] Where: MG avg represents the average interval of all samples, n represents the number of samples, mg(x i ,y i ) represents the interval of a single sample. If mg(x i ,y i ) is greater than zero, then the correct category has the most votes, and the final classification result is correct under voting; otherwise, the final classification result is wrong. The results show that when the number of decision trees is 60, the average interval of the model tends to be stable ( Figure 3 ). The specific implementation code is as follows:

[0075]

[0076]

[0077]

[0078] The third step is to evaluate the importance of variables. For each feature variable j in sample D, the value of the variable is randomly replaced to generate a new, corrupted training sample. Calculate the new sample interval MG j Considering the instability of random permutation, random permutation was repeated 50 times for each variable, and the average value was taken as the final result of variable importance:

[0079]

[0080] Where: i j represents the importance score of variable j, MG represents the average model interval before random permutation, K is the number of random permutations, and MG k，j represents the model average interval after the kth random permutation of variable j.

[0081] Using the importance scores of the characteristic factors obtained in this step, the characteristic variables are sorted in descending order, which is the ranking of the driving forces of land use change ( Figure 4), the importance of the 20 driving factors in this case is ranked from high to low: population, K12 schools, industrial facilities, cultural and leisure venues, parks and green spaces, bus stations, colleges and universities, stadiums, hospitals, water systems, housing prices, subway stations, soil and water conservation, commercial facilities, main roads, elevation, soil organic matter, long-distance bus stations, slope and slope aspect. The specific implementation code is as follows:

[0082]

[0083]

[0084] The fourth step is to use the principle of recursive feature elimination to screen the core driving forces. The driving factors are sorted according to their importance, starting from the most important driving factor, and adding one factor at a time to form a new feature subset, which is input into the random forest model. The model is trained using the cross-validation method to obtain a new model classification accuracy. Repeat the above steps until all driving factors are included in the feature subset. Draw a curve of the model classification accuracy as the number of feature variables decreases, and find the point in the curve where the classification accuracy converges at the 15th factor ( Figure 5 ), after which the classification accuracy of the model remains basically unchanged, and the 15 core driving factors screened out are: population, K12 schools, industrial facilities, cultural and leisure venues, parks and green spaces, bus stops, colleges and universities, stadiums, hospitals, water systems, housing prices, subway stations, soil and water conservation, commercial facilities and main roads.

[0085] Table 1 Classification and spatial processing of potential driving factors of land use

[0086]

Claims

1. A method for mining driving factors of land use change based on random forest and crowd-source geographic information, characterized by: Firstly, a dataset of multivariate potential driving factors affecting land use change is constructed using multivariate potential driving factors as feature variables and land use types of land use thematic maps as predictor variables. A random forest classifier model is constructed and trained. Then, the trained model is used to perform K random permutations of a single variable to calculate the importance score of the variable and rank the importance of the driving factors according to the score. Finally, the recursive feature elimination principle is used to screen the core driving forces affecting land use change. The following method was used to construct the random forest classifier model: taking the multivariate potential driving factors as the characteristic variables and the land use type of the land use thematic map as the predictor variable, a mapping relationship between the two was constructed; The following method is used to build the random forest classifier model for random forest model training: the training sample D is input into the random forest model for model training, and the maximum number of features is set to the square root of the number of potential driving factors N; the model training is performed by iteratively increasing the number of decision trees, and the average interval of model error is used to measure the model training effect. Where: MG avg represents the average interval of all samples, n represents the number of samples, mg(x i ,y i ) represents the interval of a single sample, if mg(x i ,y i ) is greater than zero, then the correct category occupies the most votes, and the final classification result is correct under voting; otherwise, the final classification result is wrong; The following method is used to calculate and rank the importance scores of driving factors: for each feature variable j in sample D, the value of the variable is randomly replaced to generate a new, damaged training sample Calculate the new sample interval MG j , repeat 50 random permutations for each variable, and use the average value as the final result of variable importance: Where: i j represents the importance score of variable j, MG represents the average model interval before random permutation, K is the number of random permutations, and MG k，j represents the model average interval after the kth random permutation of variable j; Using the importance scores of characteristic factors obtained in this step, the characteristic variables are sorted in descending order, which is the importance ranking of the driving factors of land use change.

2. The method for mining driving factors of land use change based on random forest and crowd-source geographic information according to claim 1, characterized in that: The following methods are used for data spatialization processing: according to the data type characteristics of different factors, multiple methods such as kernel density estimation, buffer zone creation, Euclidean distance calculation, zoning statistics, and slope and aspect calculation are used to spatially process the data and generate a set of spatial variables with consistent resolution and continuous surface shape.

3. The method for mining driving factors of land use change based on random forest and crowd-source geographic information according to claim 1, characterized in that: The following method is used to eliminate the differences in the dimensions of the spatial variable set data: the Fuzzy tool in Arcmap10.2 software is used to standardize the deviation of the variables, normalize the pixel values ​​of the spatial variables, and map the numerical range of the variables to between 0 and 1.

4. The method for mining driving factors of land use change based on random forest and crowd-source geographic information according to claim 1, characterized in that: The following method is used to collect random forest model training samples: (1) The land use maps of the study area in two different years were imported into ArcMap software. The spatial classification rate of the data was 30m. The land use types were divided into nine categories: water area, forest land and grassland, cultivated land, unused land, residential land, industrial land, commercial land, public management land and mixed land. The category codes were numbers from 1 to 9. (2) Use the Raster calculator in ArcMap software to detect changes in land use types in different years. The algebraic expression is Con("landuse1.tif"!="landuse2.tif",1,0). Generate a new raster dataset landuse_difference.tif. The pixel value of 1 represents the area where the land use type has changed, and the pixel value of 0 represents the area where there has been no change. (3) For pixels where land use has changed, the random traversal sampling method is used according to the spatial position index of the pixel. The corresponding traversal step size is set for different types of land use, and the whole area search and sampling are performed to form a training sample set D = [(x1, y1),…, (x n ,y n )].

5. The method for mining driving factors of land use change based on random forest and crowd-source geographic information according to claim 1, characterized in that: The following method is used to screen the core driving forces: sort the driving factors according to their importance, starting from the most important driving factor, adding one factor at a time to form a new feature subset, input it into the random forest model, use the cross-validation method to train the model and obtain a new model classification accuracy; repeat the above steps until all driving factors are included in the feature subset.

6. The method for mining driving factors of land use change based on random forest and crowd-source geographic information according to claim 5, characterized in that: The core driving force screening method uses the following method to determine the number of core driving forces: draw a curve showing how the model classification accuracy changes as the number of characteristic variables decreases, and find the point in the curve where the classification accuracy tends to converge, which is the number of core driving factors.

7. The method for mining driving factors of land use change based on random forest and crowd-source geographic information according to any one of claims 1 to 6, characterized in that: Used to accurately evaluate and screen micro-driving factors that affect urban land use changes.

Images