Population spatialization method based on building population and height information improvement

Abstract

The invention belongs to the technical field of population spatialization, and discloses a population spatialization method based on building population and height information improvement, and the method comprises the following steps: obtaining multi-source data, and carrying out the data processing; according to the building population information, constructing a population spatialization prediction model based on a random forest algorithm; building height information is introduced, a random forest population threshold regression equation is constructed, and a population estimation value is corrected; and verifying the precision, the fitting degree and the stability of the population spatialization prediction model. According to the population spatialization method based on building population and height information improvement, on the basis of a random forest algorithm, the performance of a population spatialization prediction model is effectively improved, higher spatial resolution is achieved, urban population spatialization precision is effectively improved, a new idea is provided for fine population spatialization, and the population spatialization method is suitable for popularization and application. The method has an important reference value for urban fine management and promotion of coordinated development of regional economy.

Description

Technical Field

[0001] The invention relates to the technical field of population spatialization, and in particular to a population spatialization method based on improved building population and height information. Background Art

[0002] Detailed population spatial distribution data are widely used in disaster risk assessment, urban development planning, public health and other fields, and are of great significance to the refined management of cities and the coordination of regional economic development. At present, population census data is the main source of data for population research, but it has the characteristics of long update cycle, large workload, and administrative district as statistical unit, which leads to low temporal resolution and spatial resolution in practical applications, and does not support spatial operations and analysis.

[0003] At present, the commonly used population spatialization models include spatial interpolation, statistical models, regression models and random forest models. Among them, the random forest model uses the decision tree modeling mechanism and has nonlinear modeling capabilities. It can effectively describe the nonlinear relationship distribution of population in geographic space and is the mainstream model for population spatialization. However, the random forest model constructed based on DEM, night light data and census data has a low resolution of population spatialization results due to the low resolution of the data source, which limits the ability to depict the differences in population distribution details. It is necessary to introduce a data source with finer spatial granularity to optimize the model.

[0004] In addition, some scholars have proposed using various types of building data to spatialize population data, by adding two-dimensional or three-dimensional feature information such as POI and building height as independent variables to the random forest model for training, and improving the resolution of population spatialization results. Among them, two-dimensional data such as points of interest (POI), road network (OSM) and night lights are used to provide information on the horizontal distribution of urban population and activity hotspots, and data such as building height and volume are used to provide information on the vertical distribution of population and space occupancy. However, in the simulation process of the random forest model, the data structures of two-dimensional information and three-dimensional information are different, and there are obvious differences in information characteristics, and the two types of information are incomparable. When added to the model for training at the same time, the three-dimensional information that is highly correlated with the independent variable will be overfitted, so that its training effect in the model is over-enhanced, thereby unreasonably weakening the role of other indicators on population distribution.

[0005] Therefore, the present invention proposes a population spatialization method based on improved building population and height information to solve the above technical problems. Summary of the invention

[0006] The purpose of the present invention is to provide a population spatialization method based on improved building population and height information. Based on the random forest algorithm, it effectively improves the performance of the population spatialization prediction model, has a higher spatial resolution, and effectively improves the accuracy of urban population spatialization. It provides new ideas for fine population spatialization and has important reference value for fine urban management and promoting the coordinated development of regional economy.

[0007] To achieve the above object, the present invention provides a population spatialization method based on improved building population and height information, comprising the following steps:

[0008] Step S1, obtaining multi-source data and performing data processing;

[0009] Step S2: construct a population spatialization prediction model based on the random forest algorithm according to the building population information;

[0010] Step S3, introducing building height information, constructing a random forest population threshold regression equation, and correcting the population estimate;

[0011] Step S4: verify the accuracy, goodness of fit and stability of the population spatialization prediction model.

[0012] Preferably, in step S1, multi-source data including modeling data, comparison data, remote sensing data, and statistical data are obtained; the obtained data are processed and normalized, and used as independent variables of the random forest model for population estimation;

[0013] Among them, the modeling data include POI data, road network data, night light data and building patch data; the comparison data include WorldPop dataset and census data; remote sensing data is used to obtain building patch data; statistical data include the number of households, the number of people per household and the number of people in the urban area.

[0014] Preferably, the processing method for modeling data is as follows:

[0015] (1) POI data: Perform kernel density estimation on POI data to obtain the corresponding kernel density layer;

[0016] (2) Road network data: Euclidean distance analysis is performed on the road network data, and the road network information after Euclidean distance analysis is added to the model to further improve the population spatial prediction model and learn the spatial distribution of the population;

[0017] (3) Night light data: The night light data is preprocessed by projection transformation, clipping and resampling to obtain 30m resolution night light data to simulate the spatial distribution of population within the region;

[0018] (4) Building patch data: Building patch data are obtained through visual interpretation of historical images.

[0019] Preferably, in step S2, a population spatialization prediction model based on a random forest algorithm is constructed according to the building population information; wherein the building population information is expressed by the building population density, and the specific process is as follows:

[0020] Step S21: Use building population density as the dependent variable of the model training sample to describe the spatial distribution characteristics of the population, as shown below:

[0021] B 人数 =B 户数 *x (1);

[0022]

[0023] Where x is the estimated number of people per household in the census data; B 户数 is the number of households in the building; B 人数 is the population in the building; B 密度 is the building population density; N is the ratio of the building grid unit area to the total area; n is the total number of buildings in the i-th grid unit; j is the j-th building among n buildings;

[0024] The population in the building is obtained by the above formula, and the ratio of the building grid unit area to the total area is used as the weight to resample to a 30*30m grid unit to obtain the building population density;

[0025] Step S22: Based on the building population information, a population spatialization prediction model based on the random forest algorithm is constructed, and the building population density is used as the dependent variable of the model, and the multi-source geospatial data is used as the independent variable of the model to predict the estimated population value of each grid cell, as shown below:

[0026] Y(D) 训 =H(d1,d2,…,d n ) 训 (3);

[0027] Y(T) 预 =H(t1,t2,…,t n ) 训 (4);

[0028] Among them, H 训 is the function obtained by random forest model training; d is the independent variable in the training sample; t is the independent variable in the prediction sample; n is the number of independent variables; Y(D) 训 is the dependent variable of the model, i.e., building population density; Y(T) 预Get population estimates for random forest model predictions.

[0029] Preferably, in step S3, building height information is introduced to construct a random forest population threshold regression equation to correct the population estimate; wherein the building height information is expressed by building height data, and the specific process is as follows:

[0030] Step S31, acquiring building height data;

[0031] First, K-means image segmentation is used to obtain the shadows of buildings in remote sensing images. Then, the ratio of building height to shadow length in remote sensing images at the same time is taken as a fixed value, and a certain building and its corresponding shadow length are measured to obtain the specific relationship between the two, as shown below:

[0032] H = L*λ (5);

[0033] Among them, H is the height of the building to be calculated; λ is the relationship between the shadow length and the building height; L is the extracted shadow length;

[0034] Step S32, constructing a random forest population threshold regression equation;

[0035] The building population density is taken as the dependent variable of the model, and the building height and building area are taken as independent variables to construct the population threshold regression equation, which is the same as formula (3) and formula (4). Based on this equation, the population threshold of each building in the study area is calculated, and the ratio of the building grid unit area to the total area is used as the weight for resampling to obtain a 30m resolution population threshold map.

[0036] Step S33, correcting the population estimate;

[0037] In a grid cell, if the population estimate exceeds the threshold, it is adjusted to the threshold; otherwise, the population estimate is retained, as shown below:

[0038]

[0039] Among them, POP C_i is the estimated population value after correction for the ith grid; POP A_i is the number threshold of the i-th grid; POP B_i is the estimated value of the initial population of the ith grid;

[0040] Step S34, grid population allocation: According to the revised population estimate, the total number of people in the yearbook of the study area is allocated to each grid to obtain the population spatialization result, as shown below:

[0041] POP district_i =POP district*W i (7);

[0042] Among them, POP district_i is the population estimate of the ith grid; POP district is the number of yearbooks in the study area; W i is the ratio of the revised population estimate in the ith grid to its total population.

[0043] Preferably, in step S4, the root mean square error RMSE and the coefficient of determination R are used. 2 , mean absolute error (MAE) and mean relative error (MRE) are used to verify the accuracy, fit and stability of the population spatial prediction model. The specific formula is as follows:

[0044]

[0045] Among them, POP i is the population count of the ith street; POP' i is the estimated population of the ith street; is the average population count of all streets in the area; A is the total number of streets within the statistical scope.

[0046] Therefore, the present invention adopts the above-mentioned population spatialization method based on improved building population and height information, based on the random forest algorithm, which effectively improves the performance of the population spatialization prediction model, has higher spatial resolution, and effectively improves the accuracy of urban population spatialization. It provides new ideas for fine population spatialization and has important reference value for fine urban management and promoting the coordinated development of regional economy.

[0047] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a technical roadmap of the present invention for an improved population spatialization method based on building population and height information;

[0049] Figure 2 is the location of the study area in the embodiment of the present invention;

[0050] Figure 3 It is a population change graph of Jixi City from 2003 to 2022 in an embodiment of the present invention;

[0051] Figure 4 2 is a graph showing the relationship between parameter values ​​and model accuracy in 2010 in an embodiment of the present invention; wherein (a) is the number of decision trees; (b) is the depth of the decision tree; (c) is the maximum number of features;

[0052] Figure 5is the importance index of the population distribution influencing factor in the embodiment of the present invention;

[0053] Figure 6 1 is a comparison of the accuracy of population estimation values ​​of different methods in the embodiments of the present invention; wherein (a) is a comparison chart of the accuracy results of model 1, model 2 and model 3; (b) is the importance index of model 1;

[0054] Figure 7 It is a visualization of the correction of population estimates and simulation errors in an embodiment of the present invention;

[0055] Figure 8 2 is a diagram showing a partial comparison of the results of the method of the present invention and WorldPop in 2010 in an embodiment of the present invention; wherein (a) is the result of the present invention; (b) is the result of WorldPop;

[0056] Fig. 9 30m population spatialization result of Jixi City in the embodiment of the present invention; wherein, (a) is the population in 2015; (b) is the population in 2020; (c) is the population in 2010; (d) is the regional enlarged map of the population in 2010; (e) is the remote sensing image of the regional enlarged population in 2010;

[0057] Fig.10 It is the change of population from 2010 to 2020 in the embodiment of the present invention. DETAILED DESCRIPTION

[0058] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0059] The present invention provides a population spatialization method based on improved building population and height information, comprising the following steps:

[0060] Step S1: Acquire multi-source data and perform data processing.

[0061] Step S11, acquiring multi-source data.

[0062] The data types include modeling data, comparison data, remote sensing data, and statistical data. Modeling data includes POI data, road network data, night light data, and building patch data; comparison data includes WorldPop data set and census data; remote sensing data is mainly used to obtain building patch data; statistical data includes the number of households, the number of people per household, and the number of people in the city, as shown in Table 1.

[0063] Table 1 Overview of research data

[0064]

[0065] Step S12: Process the modeling data.

[0066] The processing method for POI data is as follows:

[0067] (1) POI data: POI data can provide richer semantic information than land use data. The present invention performs kernel density estimation on 15 types of POI data and obtains corresponding kernel density layers. The specific classification and quantity are shown in Table 2.

[0068] Table 2 POI data categories and quantities

[0069] Data Category Quantity / pcs Data Category Quantity / pcs Data Category Quantity / pcs government 581 Shopping 1300 Commercial Residence 149 Science and Education 325 transportation 43 finance 284 Lifestyle Services 970 FOOD 771 stay 173 Automobile related 355 physical education 285 Medical 412 Public Facilities 147 company 421 landscape 22

[0070] (2) Road network data: The road network is closely related to population distribution. In order to further improve the model to learn the spatial distribution of population, the road network information after Euclidean distance analysis is added to the model.

[0071] (3) Night light data: Since the DN value of night light data can well simulate the spatial distribution of the population within a region, the night light data is preprocessed by projection transformation, clipping and resampling to obtain night light data with a resolution of 30m.

[0072] (4) Building patch data: Using the 2020 building patch data as a reference, the building patch data for this time period was obtained by visually interpreting Google historical images from 2010 and 2015.

[0073] Step S13: normalize the four types of data obtained above and use them as independent variables of the random forest model for population estimation.

[0074] Step S2: construct a population spatialization prediction model based on the random forest algorithm according to the building population information.

[0075] The present invention aims to improve the resolution and accuracy of population spatialization. According to the building population information, a population spatialization prediction model based on the random forest algorithm is constructed. Among them, the building population information is expressed by the building population density, such as Figure 1 shown.

[0076] Step S21: Use building population density as the dependent variable of the model training sample to finely describe the spatial distribution characteristics of the population, as shown below:

[0077] B 人数 =B 户数 *x (1);

[0078]

[0079] Where x is the estimated number of people per household in the census data; B 户数is the number of households in the building; B 人数 is the population in the building; B 密度 is the building population density; N is the ratio of the building grid unit area to the total area; n is the total number of buildings in the i-th grid unit; j is the j-th building among n buildings.

[0080] The above formula is used to obtain the number of people in the building, and the ratio of the building grid unit area to the total area is used as the weight to resample to a 30*30m grid unit to obtain the building population density.

[0081] Step S22: Based on the building population information, a population spatialization prediction model based on the random forest algorithm is constructed, and the building population density is used as the dependent variable of the model, and multi-source geospatial data such as POI kernel density is used as the independent variable of the model to predict the estimated population value of each grid cell, as shown below:

[0082] Y(D) 训 =H(d1,d2,…,d n ) 训 (3);

[0083] Y(T) 预 =H(t1,t2,…,t n ) 训 (4);

[0084] Among them, H 训 is the function obtained by random forest model training; d is the independent variable in the training sample; t is the independent variable in the prediction sample; n is the number of independent variables; Y(D) 训 is the dependent variable of the model, i.e., building population density; Y(T) 预 Get population estimates for random forest model predictions.

[0085] Step S3: Introduce building height information, construct a random forest population threshold regression equation, and correct the population estimate. The building height information is expressed by building height data, such as Figure 1 shown.

[0086] Although multi-source data such as POI kernel density can represent information on population distribution and activity hotspots, it is difficult to describe the potential residential capacity in buildings. To this end, the present invention proposes to use building height, building area and population data to construct a random forest population threshold regression equation to correct the population estimate, while also solving the problem of overfitting of three-dimensional information. The specific process is as follows:

[0087] Step S31, acquiring building height data.

[0088] First, K-means image segmentation is used to obtain the shadows of buildings in remote sensing images. Then, the ratio of building height to shadow length in remote sensing images at the same time is taken as a fixed value, and a certain building and its corresponding shadow length are measured to obtain the specific relationship between the two, as shown below:

[0089] H = L*λ (5);

[0090] Among them, H is the height of the building to be determined; λ is the relationship between the shadow length and the building height; L is the extracted shadow length.

[0091] Step S32: construct a random forest population threshold regression equation.

[0092] The building population density is used as the dependent variable of the model, and the building height and building area are used as independent variables to construct the population threshold regression equation, which is the same as formula (3) and formula (4). Based on this equation, the population threshold of each building in the study area is calculated, and the ratio of the building grid unit area to the total area is used as the weight for resampling to obtain a 30m resolution population threshold map.

[0093] Step S33: Correct the population estimate.

[0094] In a grid cell, if the population estimate exceeds the threshold, it is adjusted to the threshold; otherwise, the population estimate is retained, as shown below:

[0095]

[0096] Among them, POP C_i is the estimated population value after correction for the ith grid; POP A_i is the number threshold of the i-th grid; POP B_i is the estimated initial population of the ith grid.

[0097] Step S34, grid population allocation: According to the revised population estimate, the total number of people in the yearbook of the study area is allocated to each grid to obtain the population spatialization result, as shown below:

[0098] POP district_i =POP district *W i (7);

[0099] Among them, POP district_i is the population estimate of the ith grid; POP district is the number of yearbooks in the study area; W i is the ratio of the revised population estimate in the ith grid to its total population.

[0100] Step S4: verify the accuracy, goodness of fit and stability of the population spatialization prediction model.

[0101] The population of the study area after the allocation is summarized in the streets, and the accuracy is compared with the street census population. At the same time, the spatial population product of WorldPop is compared to prove the feasibility of the experiment through model accuracy and model fit.

[0102] The root mean square error (RMSE) and the coefficient of determination (R 2 ), mean absolute error (MAE) and mean relative error (MRE) are used to verify the accuracy, fit and stability of the model proposed in this invention. 2 It is used to evaluate the degree of fit of the model. The closer it is to 1, the higher the degree of fit of the model. MAE and MRE are used to evaluate the accuracy of the model as a whole and in each partition, respectively. The lower the value, the higher the model accuracy. RMSE reflects the stability of the estimation results, as shown below:

[0103]

[0104] Among them, POP i is the statistical population of the ith street; POP' i is the estimated population of the ith street; is the average population count of all streets in the area; A is the total number of streets within the statistical scope.

[0105] Example

[0106] This embodiment takes Jixi City, Heilongjiang Province as an example. Jixi City has a district such as Figure 2 As shown, it includes 6 districts, 29 sub-districts, 8 townships, and a total area of ​​2535.86km 2 .like Figure 3 As shown in the figure, from 2003 to 2021, the population of Jixi City decreased from 913,200 to 738,500, and the rate of population decline has gradually accelerated since 2015. As of 2022, the permanent population of Jixi City is about 728,800, with a population density of 343.48 people / km 2 As the first of the four coal cities in Heilongjiang Province, Jixi City is experiencing a continuous decline in population. This phenomenon has not only had a profound impact on the social development and economic structure of the region, but also brought new challenges to urban planning and resource allocation. This example deeply explores the spatial distribution of population in Jixi City, aiming to provide scientific decision-making support for future urban planning in the region.

[0107] 1. Correlation analysis and model construction of factors affecting population distribution.

[0108] (1) Correlation analysis of factors affecting population distribution.

[0109] SPSS software was used to perform Pearson correlation analysis on population data and influencing factors, as shown in Table 3. The results show that the factors related to distance from roads, scenery, and cars are significant at the 0.05 level, and the other factors are significantly correlated at the 0.01 level; the maximum correlation coefficient value of the night light factor is 0.503, indicating that it has a strong correlation with population distribution. Therefore, the feature library of the random forest model in this embodiment contains 15 types of POI data, night light data, distance from roads, and building patch area, a total of 18 influencing factors.

[0110] Table 3 Pearson correlation analysis results

[0111] Factor Name Correlation coefficient Factor Name Correlation coefficient government 0.107** Shopping 0.070** Science and Education 0.326** transportation 0.297** Night Lights 0.503** FOOD 0.265** Lifestyle Services 0.202** physical education 0.279** Automobile related -0.052* company 0.236** Public Facilities 0.279** finance 0.101** Commercial Residence 0.358** stay 0.158** Building area 0.380** Medical 0.226** Distance from road 0.050* landscape -0.054*

[0112] (2) Construct a spatial population prediction model.

[0113] The population density of buildings was used as the dependent variable, and 18 influencing factors were used as independent variables to construct the training samples of the random forest model. The training samples were sampled with replacement (Bootstrap), of which 70% of the training samples were used for the training data of the random forest model, and 30% of the training samples were used for the accuracy verification of the out-of-bag error rate (OOB). The model improves the accuracy of the model by adjusting the three parameters of the number of decision trees, the maximum depth of the decision tree, and the maximum number of features, and the grid search method is used to select the optimal parameter values, as shown in Table 4. Figure 4 As shown in Figure 2, the relationship between parameter values ​​and model accuracy in 2010 is shown. 2 The closer it is to 1, the higher the accuracy of the model. According to the results in the figure, the optimal values ​​of the number of decision trees, the maximum depth of decision trees, and the maximum number of features are 80, 17, and 11, respectively.

[0114] Table 4 Optimal values ​​of random forest model parameters

[0115]

[0116]

[0117] Taking the population training data sets of 2010, 2015 and 2020 as an example, the importance index of each influencing factor is obtained using the optimal parameters of the random forest model. Figure 5 As shown, in these three years, the influence of building area is the most important, which is 0.356, 0.416 and 0.358 respectively, indicating that the spatial dependence between population distribution and buildings is very strong; followed by the brightness of night lights; while scenery, catering and life services have less impact on population distribution.

[0118] 2. Verification of the validity of the revised population estimates based on building height data.

[0119] In order to verify the effectiveness of using building height data to modify population estimates, the results of the random forest model with building height introduced (model 1), the random forest model without building height (model 2), and the improved model using building height data (model 3) were compared. Figure 6 shown.

[0120] like Figure 6 As shown in (a) in Figure 2, in 2010, at the street level, the R of the population predicted by Model 1 is 2 The lowest is 0.481, and the R of model 2 predicting population size 2 is 0.543, while the improved model predicts the population size R 2 The highest is 0.702, which shows that the improved method can effectively improve the performance of the model. The reason why the population prediction accuracy of model 1 is low is that the random forest model over-emphasizes the training effect of building height data, such as Figure 6 As shown in (b), its importance index is as high as 0.854, which results in more population being allocated to high-building areas, resulting in lower population spatialization accuracy. 2 It is still not high because the random forest model does not include building height information.

[0121] like Figure 7 As shown, the remote sensing image directly reveals the locations where the initial population estimate simulation errors occur, mainly in the locations of low buildings adjacent to high-rise buildings, resulting in the simulated population estimates of the low buildings being too high.

[0122] In summary, after the random forest population threshold regression equation was constructed using building height, building area and population data to correct the population estimate, MAE, MRE and RMSE all decreased, and R 2 It rose to 0.702, and the overall accuracy was slightly improved, indicating that the use of building height data correction for population estimation is effective to a certain extent.

[0123] 3. Accuracy test

[0124] In this example, the spatialization results of the 30m population in 2010 and 2020 are counted in 34 towns / streets in Jixi City for accuracy verification and compared with the WorldPop dataset. As shown in Table 5, during the two study periods, R 2The results show that the improved model has higher accuracy than the WorldPop dataset in terms of overall and regional precision, and can effectively improve the spatial precision of urban population.

[0125] Table 5 Comparison of model accuracy between this embodiment and WorldPop at street scale

[0126]

[0127] Since there is no census data for townships / streets nationwide in 2015, the accuracy verification was conducted on the county scale, and the obtained MAE was 11490.333, which is also better than the MAE of 17209 of the WorldPop dataset, as shown in Table 6. The above indicators show that the accuracy of the improved model is higher than that of the WorldPop dataset in both county and street scale verification.

[0128] Table 6 Comparison of the accuracy of this embodiment and the WorldPop model at the county level in 2015

[0129] Dataset <![CDATA[R 2 ]]> MRE MAE RMSE WorldPop Dataset 0.975 0.141 17209 22890.897 Results 0.991 0.106 11490.333 13420.350

[0130] In order to more intuitively demonstrate the difference between the results of this embodiment and the WorldPop dataset, the WorldPop dataset of some regions is compared with the results of this embodiment, such as Figure 8 As shown, (a) is the result of this embodiment, and (b) is the WorldPop result. Figure 8 It can be seen that the population spatialization results of this embodiment not only show similar population spatial distribution characteristics to the WorldPop dataset, but also have a higher spatial resolution, revealing more detailed information on population distribution. In addition, the results can not only accurately identify densely populated areas, but also clearly simulate the trend of population changes with building height. From the above, it can be seen that the present invention can not only accurately simulate the population spatialization results of Jixi City using the improved model of building height, but also further highlight the texture information of population spatial distribution.

[0131] 4. Analysis of population spatialization results.

[0132] like Fig. 9As shown, the population spatialization results of Jixi City in 2010, 2015 and 2020 with a resolution of 30m are shown based on the improved model. It can be observed from the figure that the population of Jixi City is mainly concentrated in Jiguan District, with a small part of the population concentrated in Didao District, Hengshan District, Chengzihe District and Lishu District, while the population density in Mashan District is relatively low; the population distribution mainly radiates from Jiguan District to Didao District, Hengshan District and Chengzihe District; this is consistent with the population distribution characteristics of Jixi City and the urban spatial layout of "one core, multiple centers, and point-axis development". Fig. 9 From the observation of the enlarged figures (d) and (e), it can be clearly seen that the research method of the present invention can accurately simulate the spatial variation of small-scale population numbers, that is, the population number is relatively large in the high-rise building area, and the population number is relatively small in the low-rise building area; in addition, in Jiguan District, the population is mainly concentrated in the city center. Even for buildings of the same height, the population number on the edge of the city is less than that in the city center, which is consistent with the population distribution in the real world, further verifying the accuracy and reliability of the method of the present invention in simulating the distribution of urban population numbers.

[0133] In order to observe the changes in population more intuitively, the population changes are obtained based on the population spatialization results in 2010 and 2020, such as Fig.10 As shown in the figure, we can see that the population increase mainly occurs in the centers of each district and county, and the farther away from the district and county center, the more serious the population decrease. This is because the district and county center is the main area of ​​population concentration and population inflow, and the area far away from the district and county center is the main area of ​​population outflow, which is in line with the actual situation of urban development.

[0134] Therefore, in this embodiment, multi-source data such as night light data and POI data are combined, and the population estimate of each grid unit is estimated based on the random forest model; then, the random forest population threshold regression equation is constructed using building height data and population density data to correct the population estimate, and the number of yearbook people in the study area is redistributed according to the corrected estimate; finally, the population spatialization results of Jixi City with a spatial resolution of 30m in 2010, 2015 and 2020 are obtained.

[0135] The results show that in the street scale accuracy verification, the improved model fit R 2 The R 2 The model performance was effectively improved by reducing the spatial resolution by 3 / 10. The improved model has a higher spatial resolution than the WorldPop dataset, and the population distribution shows similar spatial characteristics to the building clusters in the remote sensing images, which is more consistent with the population distribution pattern. At the same time, through comparative analysis with the WorldPop dataset, the improved model has smaller average absolute error and average relative error, higher fitting degree and accuracy, which can effectively improve the accuracy of urban population spatialization, and thus provide new ideas for fine population spatialization.

[0136] Therefore, the present invention adopts the above-mentioned population spatialization method based on improved building population and height information, based on the random forest algorithm, which effectively improves the performance of the population spatialization prediction model, has higher spatial resolution, and effectively improves the accuracy of urban population spatialization. It provides new ideas for fine population spatialization and has important reference value for fine urban management and promoting the coordinated development of regional economy.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A population spatialization method based on improved building population and height information, characterized in that: The following steps are involved: Step S1, obtaining multi-source data and performing data processing; Step S2: construct a population spatialization prediction model based on the random forest algorithm according to the building population information; Step S3, introducing building height information, constructing a random forest population threshold regression equation, and correcting the population estimate; Step S4: verify the accuracy, goodness of fit and stability of the population spatialization prediction model.

2. A population spatialization method based on improved building population and height information according to claim 1, characterized in that: In step S1, multi-source data including modeling data, comparison data, remote sensing data, and statistical data are obtained; the obtained data are processed and normalized, and used as independent variables of the random forest model for population estimation; Among them, the modeling data include POI data, road network data, night light data and building patch data; the comparison data include WorldPop dataset and census data; remote sensing data is used to obtain building patch data; statistical data include the number of households, the number of people per household and the number of people in the urban area.

3. A population spatialization method based on improved building population and height information according to claim 2, characterized in that: The processing method for modeling data is as follows: (1) POI data: perform kernel density estimation on POI data to obtain the corresponding kernel density layer; (2) Road network data: Euclidean distance analysis is performed on the road network data, and the road network information after Euclidean distance analysis is added to the model to further improve the population spatial prediction model and learn the spatial distribution of the population; (3) Night light data: The night light data is preprocessed by projection transformation, clipping and resampling to obtain 30m resolution night light data to simulate the spatial distribution of population within the region; (4) Building patch data: Building patch data are obtained through visual interpretation of historical images.

4. The population spatialization method based on improved building population and height information according to claim 1, characterized in that: In step S2, a population spatialization prediction model based on the random forest algorithm is constructed according to the building population information; wherein the building population information is expressed by the building population density, and the specific process is as follows: Step S21: Use building population density as the dependent variable of the model training sample to describe the spatial distribution characteristics of the population, as shown below: B 人数 =B 户数 *x (1); Where x is the estimated number of people per household in the census data; B 户数 is the number of households in the building; B 人数 is the population in the building; B 密度 is the building population density; N is the ratio of the building grid unit area to the total area; n is the total number of buildings in the i-th grid unit; j is the j-th building among n buildings; The population in the building is obtained by the above formula, and the ratio of the building grid unit area to the total area is used as the weight to resample to a 30*30m grid unit to obtain the building population density; Step S22: Based on the building population information, a population spatialization prediction model based on the random forest algorithm is constructed, and the building population density is used as the dependent variable of the model, and the multi-source geospatial data is used as the independent variable of the model to predict the estimated population value of each grid cell, as shown below: Y(D) 训 =H(d1,d2,…,d n ) 训 (3); Y(T) 预 =H(t1,t2,…,t n ) 训 (4); Among them, H 训 is the function obtained by random forest model training; d is the independent variable in the training sample; t is the independent variable in the prediction sample; n is the number of independent variables; Y(D) 训 is the dependent variable of the model, i.e., building population density; Y(T) 预 Get population estimates for random forest model predictions.

5. The population spatialization method based on improved building population and height information according to claim 1, characterized in that: In step S3, the building height information is introduced to construct a random forest population threshold regression equation to correct the population estimate; wherein the building height information is expressed by building height data, and the specific process is as follows: Step S31, acquiring building height data; First, K-means image segmentation is used to obtain the shadows of buildings in remote sensing images. Then, the ratio of building height to shadow length in remote sensing images at the same time is taken as a fixed value, and a certain building and its corresponding shadow length are measured to obtain the specific relationship between the two, as shown below: H = L*λ(5); Among them, H is the height of the building to be calculated; λ is the relationship between the shadow length and the building height; L is the extracted shadow length; Step S32, constructing a random forest population threshold regression equation; The building population density is taken as the dependent variable of the model, and the building height and building area are taken as independent variables to construct the population threshold regression equation, which is the same as formula (3) and formula (4). Based on this equation, the population threshold of each building in the study area is calculated, and the ratio of the building grid unit area to the total area is used as the weight for resampling to obtain a 30m resolution population threshold map. Step S33, correcting the population estimate; In a grid cell, if the population estimate exceeds the threshold, it is adjusted to the threshold; otherwise, the population estimate is retained, as shown below: Among them, POP C_i is the estimated population value after correction for the ith grid; POP A_i is the number threshold of the i-th grid; POP B_i is the estimated initial population of the ith grid; Step S34, grid population allocation: According to the revised population estimate, the total number of people in the yearbook of the study area is allocated to each grid to obtain the population spatialization result, as shown below: POP district_i =POP district *W i (7); Among them, POP district_i is the population estimate of the ith grid; POP district is the number of yearbooks in the study area; W i is the ratio of the revised population estimate in the ith grid to its total population.

6. A population spatialization method based on improved building population and height information according to claim 1, characterized in that: In step S4, the root mean square error RMSE and the coefficient of determination R 2 , mean absolute error MAE and mean relative error MRE, to verify the accuracy, fit and stability of the population spatial prediction model. The specific formula is as follows: Among them, POP i is the statistical population of the ith street; POP' i is the estimated population of the ith street; is the average population count of all streets in the area; A is the total number of streets within the statistical scope.

Images