Land use data downscaling method and related equipment based on generalized additive model

Abstract

The application discloses a land use data downscaling method based on a generalized additive model and related equipment, and the method comprises the following steps: acquiring a LUH2 data set and downscaling requirements, wherein the LUH2 data set comprises data of various land class types; performing fishing net construction on a preset area based on the LUH2 data set, the downscaling requirements and preset driving factor data to obtain a coarse fishing net and a fine fishing net; inputting the coarse fishing net and the fine fishing net into a preset initial generalized additive model for downscaling processing to obtain an intermediate land class area prediction result for the preset area; optimizing the initial generalized additive model based on the intermediate land class area prediction result to obtain a target generalized additive model; and outputting a target land class area prediction result for the preset area based on the target generalized additive model. The problem that the resolution of the LUH2 data set is low is solved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a land use data downscaling method and related equipment based on a generalized additive model. The related equipment includes a land use data downscaling system based on a generalized additive model, a computing device, and a computer-readable storage medium. Background Technology

[0002] Global-scale land use scenario data (taking the well-known LUH2 dataset as an example) provides an important data foundation and support for many fields such as ecosystem models, climate models, and biodiversity assessment models. For example, the LUH2 (Land Use Harmonization 2) dataset is a standardized dataset of global land use and land cover change developed and widely used by international climate change research institutions. This dataset integrates historical land use change observation data and future land use scenario prediction data, covering continuous time series from historical periods (850 AD) to future scenarios (2100 AD), providing long-term data support for the evolution of global land use patterns. However, because the LUH2 dataset is designed for large-scale global studies, its spatial resolution is relatively low, generally 0.25° × 0.25° (approximately 25 kilometers). While this is sufficiently detailed for global-scale studies, it results in significant spatial scale errors in regional or local-scale studies (e.g., urban or rural scales). Such low spatial resolution land use data is difficult to apply directly to refined regional ecological environment management, land use planning, biodiversity conservation, and urban and rural planning, because land use changes at local scales often exhibit strong spatial heterogeneity, and this heterogeneity has a significant impact on ecosystem services and land resource management. Therefore, there is an urgent need for a LUH2 data downscaling method to improve the resolution of land use data. Summary of the Invention

[0003] To overcome the low resolution of the LUH2 dataset, this application provides a land use data downscaling method and related equipment based on a generalized additive model. The related equipment includes a land use data downscaling system based on a generalized additive model, a computing device, and a computer-readable storage medium.

[0004] Firstly, to address the aforementioned technical problems, this application provides a land use data downscaling method based on a generalized additive model, comprising:

[0005] Obtain the LUH2 dataset and meet the downscaling requirements. The LUH2 dataset includes data from various land cover types.

[0006] Based on the LUH2 dataset, downscaling requirements, and preset driving factor data, fishing nets are constructed in a preset area to obtain coarse and fine fishing nets.

[0007] The coarse and fine fishing nets are input into a preset initial generalized additive model for downscaling to obtain the predicted area of ​​intermediate land types for the preset area.

[0008] The initial generalized additive model is optimized based on the intermediate land area prediction results to obtain the target generalized additive model;

[0009] The target generalized additive model outputs the predicted area of ​​the target land type for the preset area.

[0010] Furthermore, based on the LUH2 dataset, downscaling requirements, and pre-defined driving factor data, fishing nets are constructed in a pre-defined area to obtain coarse and fine fishing nets, including:

[0011] Based on the LUH2 dataset and preset driving element data, a fishing net is constructed in a preset area to obtain a coarse fishing net;

[0012] Based on the downscaling requirements and driving factor data, a fishing net is constructed in the preset area to obtain a fine fishing net.

[0013] Furthermore, based on the LUH2 dataset and preset driving element data, a fishing net is constructed in the preset area to obtain a coarse fishing net, including:

[0014] The LUH2 dataset is cropped based on a preset region to obtain a low-resolution dataset for the preset region;

[0015] A first vector fishing net matching the size of the low-resolution dataset is constructed;

[0016] The low-resolution dataset and the preset driving element data are added to the first vector fishing net to form a coarse fishing net.

[0017] Furthermore, the downscaling requirement is to downscale the LUH2 dataset to the target resolution;

[0018] Based on the downscaling requirements and driving factor data, a fishing net is constructed in the preset area to obtain a fine fishing net, including:

[0019] A second vector fishing net is constructed corresponding to the preset area, and the resolution of the second vector fishing net is the same as the target resolution;

[0020] The driving element data is added to the second vector fishing net to form a fine fishing net.

[0021] Furthermore, the initial generalized additive model includes a generalized additive model for each land type, and the driving element data includes multiple driving elements.

[0022] The coarse and fine fishing nets are input into a pre-defined initial generalized additive model for downscaling to obtain the predicted area of ​​intermediate land types for the pre-defined region, including:

[0023] The coarse and fine fishing nets are input into the preset initial generalized additive model. The land use generalized additive model is used to make predictions based on the driving elements in the fine fishing net, and the original area prediction results of the corresponding land use type in the fine fishing net are obtained. The original area prediction results include the original land use area value and the original standard error.

[0024] Based on multiple original area prediction results, an initial land category area prediction result is generated for the preset area.

[0025] The initial land area prediction results are corrected based on coarse and fine fishing nets to obtain intermediate land area prediction results for the preset area.

[0026] Furthermore, the initial generalized additive model includes the generalized additive model for each land use type, and the intermediate land use area prediction results include the intermediate land use area value and intermediate standard error of the generalized additive model for each land use type.

[0027] Based on the intermediate land category area prediction results, the initial generalized additive model is optimized to obtain the target generalized additive model, including:

[0028] Using a pre-defined constraint optimization function, constraint values ​​are calculated based on multiple intermediate land area values, multiple intermediate standard errors, and corresponding multiple real land area values.

[0029] The average difference is calculated based on the difference between the predicted value and the corresponding actual land area value for each intermediate land category.

[0030] The network parameters in the initial generalized additive model are adjusted based on the constraint values ​​and average differences to obtain the target generalized additive model.

[0031] Furthermore, the network parameters in the initial generalized additive model are adjusted based on the constraint values ​​and average differences to obtain the target generalized additive model, including:

[0032] When the constraint value is greater than the preset value, and / or the average difference is greater than the preset difference, the network parameters in the initial generalized additive model are adjusted to obtain the intermediate generalized additive model.

[0033] The coarse and fine fishing nets are input into the intermediate generalized summation model for downscaling to obtain the predicted area of ​​transitional land types for the preset area.

[0034] When the constraint value corresponding to the predicted area of ​​transitional land is less than or equal to the preset value, and the corresponding average difference is less than or equal to the preset difference, the corresponding intermediate generalized additive model is determined as the target generalized additive model.

[0035] Secondly, this application also provides a land use data downscaling system based on a generalized additive model, comprising:

[0036] The acquisition module is used to acquire the LUH2 dataset and the downscaling requirements. The LUH2 dataset includes data from various land cover types.

[0037] The fishing net construction module is used to construct fishing nets in a preset area based on the LUH2 dataset, downscaling requirements, and preset driving element data, resulting in coarse and fine fishing nets.

[0038] The downscaling module is used to input coarse and fine fishing nets into a preset initial generalized additive model for downscaling processing, and to obtain the predicted area of ​​intermediate land types for the preset area.

[0039] The model optimization module is used to optimize the initial generalized additive model based on the intermediate land area prediction results to obtain the target generalized additive model;

[0040] The results output module is used to output the predicted area of ​​the target land type for the preset area based on the target generalized additive model.

[0041] Thirdly, this application also provides a computing device, including a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the above-described land use data downscaling method based on a generalized additive model.

[0042] Fourthly, this application also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform steps of a land use data downscaling method based on a generalized additive model.

[0043] The beneficial effects of this application are as follows: First, by constructing coarse and fine nets for a preset area based on the LUH2 dataset, downscaling requirements, and preset driving element data, the coarse and fine nets are input into a preset initial generalized additive model for downscaling. The intermediate land use area prediction results obtained from the downscaling process are then used to optimize the initial generalized additive model, thereby improving the downscaling accuracy of the optimized target generalized additive model. This ensures that the target land use area prediction results output by the target generalized additive model for the preset area can meet the downscaling requirements, thereby improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset. Attached Figure Description

[0044] Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of the present application of a land use data downscaling method based on a generalized additive model;

[0045] Figure 2 This is a schematic diagram of the process of applying the land use data downscaling method based on the generalized additive model provided in an exemplary embodiment of this application;

[0046] Figure 3 This is an example diagram showing the predicted area of ​​intermediate land categories in an exemplary embodiment of this application;

[0047] Figure 4 This is an example diagram showing the predicted area of ​​the target land type in an exemplary embodiment of this application;

[0048] Figure 5 This is a schematic diagram illustrating the structure of a land use data downscaling system based on a generalized additive model, as an exemplary embodiment of this application. Detailed Implementation

[0049] The following embodiments are further explanations and supplements to this application and do not constitute any limitation on this application.

[0050] Existing global or regional land use datasets (e.g., the LUH2 land use scenario dataset) often have low spatial resolution, making it difficult to meet the needs of refined studies on land use change in local areas. Specifically, such datasets (e.g., LUH2) are important data sources in global-scale ecology, land use change, biodiversity assessment, and environmental policy research, widely used to simulate and predict long-term land use trends, supporting ecological environment management and land planning decisions. However, due to the coarse spatial resolution of LUH2 data (approximately 0.25°, or about 25 kilometers), it is difficult to directly apply to refined-scale land management, environmental risk assessment, and ecological protection planning. It is also insufficient for analyses requiring high-precision spatial data support, such as urban expansion, refined farmland management, and ecosystem service assessment. Therefore, to more effectively leverage the value of such low-resolution data, there is an urgent need for a scientific, efficient, and accurate downscaling method to obtain land use data with higher spatial resolution, meeting the practical needs of regional-scale, especially urban and county-level, refined-scale land use change research and management planning.

[0051] Currently, there are some experimental downscaling methods in academia and industry, which mainly include the following categories:

[0052] (1) Spatial interpolation method

[0053] Commonly used spatial interpolation methods include inverse distance weighted interpolation (IDW) and kriging, which downscale based on the spatial continuity between adjacent grid data. These methods are generally simple to operate and computationally fast, but they assume strong spatial continuity of land use types and perform poorly in areas with clear spatial boundaries and high heterogeneity (such as urban boundaries and agroforestry junctions), potentially producing transitional areas that do not reflect reality.

[0054] (2) Simple area allocation method (proportional area method)

[0055] This method typically uses existing high-resolution reference land cover data (such as MODIS and ESA WorldCover data) to proportionally allocate low-resolution LUH2 land use types to fine-scale rasters. However, this method also has significant drawbacks: it fails to consider the ecological and socio-economic drivers of land use transfer processes, resulting in a static area redistribution that cannot reflect the actual spatial dynamics of land use change.

[0056] (3) Experience-based downscaling methods based on historical statistics or expert knowledge

[0057] This method relies on historical land use statistics, regional planning data, or expert experience, and uses regression models or rule-driven methods for downscaling. Existing studies, such as Giuliani et al. (2022), have used expert knowledge combined with high-resolution remote sensing data to refine the land use data. Although this type of method can reflect the actual land use characteristics at the regional scale to a certain extent, it has high requirements for data quality and expert experience, and is difficult to implement in areas with scarce data or no historical records.

[0058] (4) Model coupling and machine learning downscaling methods

[0059] In recent years, with the rapid development of geographic information technology and machine learning technology, models such as Random Forest and Convolutional Neural Networks (CNN) have been gradually applied to data downscaling processes such as LUH2. For example, Rashidi et al. (2023) used machine learning methods combined with land suitability models to downscale global LUH2 data to the national scale and achieve high accuracy. However, these methods usually require a large amount of sample data for model training, resulting in high computational complexity, and their generalization ability may be insufficient in areas with limited data.

[0060] However, current downscaling methods in academia have some obvious drawbacks when applied:

[0061] (1) Insufficient spatial precision

[0062] Among existing downscaling methods, spatial interpolation and simple area allocation methods, while effective in some cases, suffer from insufficient accuracy. Spatial interpolation methods (such as IDW and Kriging interpolation) assume that land use types are spatially continuous, but this does not reflect reality, especially in areas with clear spatial boundaries such as urban expansion and the boundary between farmland and forest, where inaccurate transition zones are easily created. Simple area allocation methods, by proportionally allocating low-resolution land use data to a high-resolution grid, are simple to implement but ignore the dynamic characteristics of land use change and fail to reflect the spatial heterogeneity of land use transformation.

[0063] (2) Lack of consideration for spatial heterogeneity

[0064] Most existing downscaling methods fail to effectively account for the spatial heterogeneity of land use change. Land use types and ecological environments within a region often exhibit significant differences, while existing methods typically employ simplistic assumptions, neglecting the interactions between different land use types and the influence of environmental factors. For example, while downscaling methods based on historical statistics or expert knowledge can simulate the true distribution of land use types in some cases, they rely heavily on expert experience and historical data, lack consideration for current spatial heterogeneity, and are often difficult to apply in areas with limited data.

[0065] (3) High computational complexity

[0066] Machine learning and model-coupled methods (such as random forests and convolutional neural networks) have advantages in improving downscaling accuracy, but these methods often require large amounts of training data and high computational resources, and their computational complexity is relatively high. For example, when using a random forest model for downscaling, it is necessary to train on a large amount of sample data and optimize multiple hyperparameters, which may lead to low computational efficiency when processing large-scale data, especially when processing high-resolution data, where both training and computation time increase significantly.

[0067] (4) Poor universality of the model

[0068] Most existing downscaling methods are developed for specific regions or types of land use data, thus their universality is poor. Many methods rely on high-quality ancillary data (such as remote sensing imagery, climate data, etc.), and in some regions, especially those with scarce data, the effectiveness of existing methods is difficult to guarantee. This leads to significant limitations in the application of these methods on a global scale, especially when ancillary data is unavailable or incomplete, in which case existing methods cannot provide reliable downscaling results.

[0069] To address the aforementioned issues, embodiments of this application provide a land use data downscaling method and related equipment based on a generalized additive model. The related equipment includes a land use data downscaling system based on a generalized additive model, a computing device, and a computer-readable storage medium. These embodiments will be described in detail below.

[0070] The land use data downscaling method based on a generalized additive model provided in this application can be executed by a server. It should be noted that the server can be a standalone server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms; no limitation is imposed here. The automated control logic of the land use data downscaling method based on a generalized additive model can be implemented using code written in languages ​​such as R or Python. The downscaling in this application refers to the process of converting information from a low-resolution LUH2 dataset into higher-resolution, more localized data.

[0071] Please see Figure 1 , Figure 1 An exemplary embodiment of this application illustrates a land use data downscaling method based on a generalized additive model, such as... Figure 1 As shown, this application provides a land use data downscaling method based on a generalized additive model, including:

[0072] S11, Obtain the LUH2 dataset and the downscaling requirements. The LUH2 dataset includes data from various land cover types.

[0073] S12, Based on the LUH2 dataset, downscaling requirements, and preset driving factor data, construct fishing nets in the preset area to obtain coarse and fine fishing nets;

[0074] S13, input the coarse and fine fishing nets into the preset initial generalized additive model for downscaling to obtain the predicted area of ​​intermediate land types for the preset area;

[0075] S14. Based on the predicted area of ​​intermediate land types, optimize the initial generalized additive model to obtain the target generalized additive model;

[0076] S15, based on the target generalized additive model, outputs the target land area prediction results for the preset area.

[0077] The land use data downscaling method based on a generalized additive model provided in this application firstly constructs a coarse and fine net for a preset area based on the LUH2 dataset, downscaling requirements, and preset driving element data. The coarse and fine nets are then input into a preset initial generalized additive model for downscaling. The intermediate land use area prediction results obtained from the downscaling process are then used to optimize the initial generalized additive model, thereby improving the downscaling accuracy of the optimized target generalized additive model. This ensures that the target land use area prediction results output by the target generalized additive model for the preset area meet the downscaling requirements, thus improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset. The target land use area prediction results are downscaled land use scenario data displayed in a planar net format, which is then converted from vector to raster data to obtain the corresponding raster data.

[0078] In an exemplary embodiment provided in this application, the LUH2 dataset may contain multiple land cover types (such as 17 types). Theoretically, data of these different land cover types can be downscaled simultaneously. However, in actual research, in most cases, so many land cover types are not needed. Therefore, based on previous research, the land cover types of the LUH2 dataset obtained in this application are classified and integrated into 5 types, namely PRI (primitive forest), SEC (secondary forest), CROP (arable land), PAS (pasture), and URB (urban).

[0079] Before downscaling, relevant driving factor data needs to be prepared, including but not limited to climate, soil, topography, land use, and human activity intensity. The resolution of these driving factors needs to be consistent with the target resolution of the predicted land area after downscaling to ensure the accuracy of the downscaling method. Downscaling requirements should be set based on the target resolution. Simultaneously, the driving factor data for different selected layers need to be preprocessed. Preprocessing methods include logistic regression, normalization, and logarithmic processing to ensure that the value ranges of different driving factor data do not differ too much, and that the value distribution follows a normal distribution as much as possible.

[0080] The initial Generalized Additive Model (GAM) construction method: The downscaling approach used in this method transforms the relationship between coarse-grained response data and (multiple) fine-grained covariate data into fine-grained response prediction. Therefore, a generalized additive model is used to construct the regression relationship between the original low-resolution dataset and the driving element layer: the area value of each land use type is used as the response variable, and each driving element data is used as the driving variable, resulting in a generalized additive model for each land use type using a quasi-binomial distribution and a logistic join function (see code for details). Subsequently, VIF collinearity detection, comparison of AIC values ​​during stepwise regression, assessment of the need for a random slope, and stepwise simplification of the generalized additive model can be performed to determine the optimal prediction model structure for each land use type, forming the target generalized additive model.

[0081] Optionally, based on the LUH2 dataset, downscaling requirements, and preset driving factor data, fishing nets are constructed in a preset area to obtain coarse and fine fishing nets, including:

[0082] Based on the LUH2 dataset and preset driving element data, a fishing net is constructed in a preset area to obtain a coarse fishing net;

[0083] Based on the downscaling requirements and driving factor data, a fishing net is constructed in the preset area to obtain a fine fishing net.

[0084] In the embodiment provided in this application, a fishing net is constructed for a preset area based on the LUH2 dataset and preset driving element data. This results in a low-resolution coarse fishing net containing both the LUH2 dataset and driving element data. Then, based on downscaling requirements and the driving element data, a fine fishing net is constructed for the preset area, resulting in a fine fishing net with the target resolution that contains driving element data and meets the downscaling requirements. This facilitates subsequent downscaling of the coarse fishing net based on the fine fishing net, ensuring that the target land use area prediction results for the preset area meet the downscaling requirements, thereby improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset. The preset area can be the territory of a country.

[0085] Optionally, a fishing net is constructed in a preset area based on the LUH2 dataset and preset driving element data to obtain a coarse fishing net, including:

[0086] The LUH2 dataset is cropped based on a preset region to obtain a low-resolution dataset for the preset region;

[0087] A first vector fishing net matching the size of the low-resolution dataset is constructed;

[0088] The low-resolution dataset and the preset driving element data are added to the first vector fishing net to form a coarse fishing net.

[0089] In the embodiment provided in this application, firstly, the LUH2 dataset is cropped to obtain a low-resolution dataset within a predetermined land area, and a first vector fishing net matching the size of the low-resolution dataset is constructed so that the first vector fishing net can cover the pixels of the low-resolution dataset. Secondly, the low-resolution dataset and predetermined driving element data are added to the first vector fishing net to form a coarse fishing net, which facilitates subsequent downscaling of the coarse fishing net using a fine fishing net based on the fishing net form, thereby achieving downscaling of the land use scenario data corresponding to the predetermined area in the LUH2 dataset, and thus improving the resolution of the land use data.

[0090] In an exemplary embodiment provided in this application, the low-resolution LUH2 dataset typically covers a global or large region. Therefore, it is necessary to crop out the required portion, i.e., the low-resolution dataset within the land area of ​​the preset region. This allows for downscaling of the datasets from different regions within the LUH2 dataset in stages, reducing the amount of data processing required for each downscaling and thus improving the model's downscaling performance. When dividing the preset region for each downscaling operation, it can be done based on different conditions such as the economic zone, climate zone, and ecosystem type of the study area. The dataset regions corresponding to the required preset regions in the LUH2 dataset are pre-cropped using QGIS 3.40.1 software (an open-source geographic information system GIS software). Simultaneously, a first vector fishing net with a pixel size of 0.25° covering the preset region is created using this software. Various driving element information and the data area of ​​each land type in the low-resolution dataset are then statistically analyzed and mapped onto this fishing net to form a coarse fishing net. After statistical analysis, the coarse fishing net layer is exported as a CSV file (plain text file).

[0091] Optionally, the downscaling requirement is the requirement to downscale the LUH2 dataset to the target resolution;

[0092] Based on the downscaling requirements and driving factor data, a fishing net is constructed in the preset area to obtain a fine fishing net, including:

[0093] A second vector fishing net is constructed corresponding to the preset area, and the resolution of the second vector fishing net is the same as the target resolution;

[0094] The driving element data is added to the second vector fishing net to form a fine fishing net.

[0095] In the embodiment provided in this application, a second vector fishing net corresponding to a preset area is constructed so that the resolution of the second vector fishing net is the same as the target resolution corresponding to the downscaling requirement. The driving element data is added to the second vector fishing net to form a fine fishing net. This makes it easier for the subsequent target land use area prediction results for the preset area obtained by using the fine fishing net to downscale the coarse fishing net based on the fishing net form to meet the downscaling requirement, thereby improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset.

[0096] In an exemplary embodiment provided in this application, the final downscaling target and research scope are determined. Here, a 1km area of ​​a country's territory is used as a preset area. In QGIS 3.40.1 software, a 1km second vector fishing net is generated with the preset area as the scope. The location selection tool is used to remove unnecessary fishing net units. After successful generation, the various driving element data of the 1km area are statistically added to the 1km second vector fishing net to form a fine fishing net. After the statistics are compiled, the 1km fine fishing net is exported as a CSV file (plain text file).

[0097] Optionally, the initial generalized additive model includes a generalized additive model for each land type, and the driving element data includes multiple driving elements;

[0098] The coarse and fine fishing nets are input into a pre-defined initial generalized additive model for downscaling to obtain the predicted area of ​​intermediate land types for the pre-defined region, including:

[0099] The coarse and fine fishing nets are input into the preset initial generalized additive model. The land use generalized additive model is used to make predictions based on the driving elements in the fine fishing net, and the original area prediction results of the corresponding land use type in the fine fishing net are obtained. The original area prediction results include the original land use area value and the original standard error.

[0100] Based on multiple original area prediction results, an initial land category area prediction result is generated for the preset area.

[0101] The initial land area prediction results are corrected based on coarse and fine fishing nets to obtain intermediate land area prediction results for the preset area.

[0102] In the embodiment provided in this application, a coarse and fine fishing net are input into a preset initial generalized additive model. The land use generalized additive model is used to predict the corresponding land use type based on the driving elements in the fine fishing net, obtaining the original area prediction result of the corresponding land use type in the fine fishing net. Based on multiple original area prediction results, an initial land use area prediction result for a preset area is formed. The initial land use area prediction result is corrected based on the coarse and fine fishing nets to achieve downscaling of the land use scenario data in the LUH2 dataset corresponding to the preset area in the coarse fishing net, obtaining an intermediate land use area prediction result for the preset area. This facilitates subsequent optimization of the initial generalized additive model based on the intermediate land use area prediction result, thereby improving the downscaling accuracy of the optimized target generalized additive model. This ensures that the target generalized additive model can meet the downscaling requirements, and thus enables the target land use area prediction result for the preset area output by the target generalized additive model to meet the downscaling requirements, improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset. The standard error refers to the standard deviation of the sampling distribution of a sample statistic (such as the mean, regression coefficient, etc.) under repeated sampling conditions. It measures the stability or reliability of the statistic.

[0103] In an exemplary embodiment provided in this application, a land use generalized additive model is used to predict the original area of ​​the corresponding land use type in the fine fishing net based on the driving elements in the fine fishing net. The specific steps are as follows: for each driving element in the fine fishing net, the driving element is input into the land use generalized additive model for prediction to obtain the corresponding original predicted area value; the average of multiple original predicted area values ​​is determined as the original land use area value of the corresponding land use type in the fine fishing net; the standard error is calculated based on the multiple original predicted area values; and the original land use area value and the standard error are used to form the original area prediction result of the corresponding land use type in the fine fishing net.

[0104] Using the constructed initial generalized additive model for different land use types, each driving factor in the fishing net unit is taken as input to predict the original area (including original land use area value and original standard error) of different land use types in a 1km fine fishing net unit. Based on multiple original area prediction results, an initial land use area prediction result for the preset area is formed. The initial land use area prediction result of the first prediction is exported as a separate CSV file (plain text file).

[0105] The initial land use area prediction results are corrected based on coarse and fine fishing nets to obtain intermediate land use area prediction results for the preset area. The specific steps are as follows:

[0106] First, after obtaining the initial land use area prediction results for different land use scenarios corresponding to the preset area in the LUH2 dataset with a 1km fine fishing net, the intersection function in QGIS 3.40.1 software can be used to attach the coarse fishing net ID to the fine fishing net. After successful processing, the resulting mixed fishing net will have two fields: the fine fishing net ID field ID_2 and the coarse fishing net ID field ID_1 to which each cell belongs.

[0107] Secondly, in R language, the average area of ​​different land use types in a 1km fine-net fishing cell under a 0.25° coarse-net fishing cell is calculated. This average area is compared with the original LUH2 area value in the 0.25° coarse-net fishing cell. Then, the predicted 1km land use value is recalibrated based on the correction coefficient obtained after the comparison. The specific formula is as follows:

[0108]

[0109] in, This indicates the intermediate land use area values ​​included in the predicted intermediate land use area after multiplicative scaling. This indicates the original land use area values ​​included in the original area prediction results. This represents the area value of the original LUH2 in a 0.25° coarse fishing net. This represents the average area of ​​different land types within a 1km fine fishing net unit, compared to a 0.25° coarse fishing net unit.

[0110] Optionally, the initial generalized additive model includes the generalized additive model for each land type, and the intermediate land area prediction results include the intermediate land area values ​​and intermediate standard errors of the generalized additive model for each land type.

[0111] Based on the intermediate land category area prediction results, the initial generalized additive model is optimized to obtain the target generalized additive model, including:

[0112] Using a pre-defined constraint optimization function, constraint values ​​are calculated based on multiple intermediate land area values, multiple intermediate standard errors, and corresponding multiple real land area values.

[0113] The average difference is calculated based on the difference between the predicted value and the corresponding actual land area value for each intermediate land category.

[0114] The network parameters in the initial generalized additive model are adjusted based on the constraint values ​​and average differences to obtain the target generalized additive model.

[0115] In the embodiment provided in this application, a constraint optimization function is used to calculate constraint values ​​based on multiple intermediate land use area values, multiple intermediate standard errors, and corresponding multiple real land use area values. An average difference is calculated based on the predicted difference between each intermediate land use area value and its corresponding real land use area value. The network parameters in the initial generalized additive model are then adjusted based on the constraint values ​​and the average difference to ensure that the initial generalized additive model converges to meet the downscaling accuracy requirements. This improves the accuracy of the target land use area prediction results for the preset area output by the adjusted target generalized additive model that meets the downscaling accuracy requirements, thus meeting the downscaling requirements and improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset.

[0116] In an exemplary embodiment provided in this application, when adjusting the network parameters in the initial generalized additive model based on constraint values ​​and average differences, the network parameters of the generalized additive model for each land type are actually adjusted so that the difference in the predicted values ​​corresponding to the generalized additive model for each land type is less than a preset value, so that the generalized additive model for each land type converges, thereby enabling the initial generalized additive model to converge and obtain the target generalized additive model that can meet the downscaling requirements.

[0117] The logic for setting the constraint optimization function is as follows:

[0118] Since the downscaling prediction results for each land use type are based on a separate generalized additive model, and the area values ​​of different land use types within each fishing net cell must sum to 1, with each land use type's area value ranging from 0 to 1, using different generalized additive models for prediction may lead to the predicted area of ​​different land use types within the same cell violating the aforementioned two constraints. Therefore, a constraint optimization function needs to be set to ensure that the prediction results meet the actual conditions. The formula for the constraint optimization function is:

[0119]

[0120] in, Indicates the constraint value. Indicates the number of land use types. Indicates land use type, Indicates the first The actual land area value corresponding to the land type. Indicates the first The predicted area of ​​intermediate land categories corresponding to different land types includes the area values ​​of those intermediate land categories. Indicates the first The intermediate standard error is included in the intermediate land area prediction results output by the generalized additive model corresponding to the land type.

[0121] After multiple rounds of iteration, ensuring that the constraints are optimized The smallest, and meets the general conditions for land area prediction ( (Less than or equal to the preset value), to obtain the target generalized additive model, and output the constrained optimized prediction result of the target land area for the preset region.

[0122] Optionally, the network parameters in the initial generalized additive model are adjusted based on the constraint values ​​and average differences to obtain the target generalized additive model, including:

[0123] When the constraint value is greater than the preset value, and / or the average difference is greater than the preset difference, the network parameters in the initial generalized additive model are adjusted to obtain the intermediate generalized additive model.

[0124] The coarse and fine fishing nets are input into the intermediate generalized summation model for downscaling to obtain the predicted area of ​​transitional land types for the preset area.

[0125] When the constraint value corresponding to the predicted area of ​​transitional land is less than or equal to the preset value, and the corresponding average difference is less than or equal to the preset difference, the corresponding intermediate generalized additive model is determined as the target generalized additive model.

[0126] In the embodiment provided in this application, during the training and adjustment of the initial generalized additive model, convergence values ​​(preset values ​​and preset differences) are set for constraint values ​​and average differences. When both constraint values ​​and average differences meet their corresponding convergence values, the intermediate generalized additive model obtained from this training is determined as the target generalized additive model. This ensures that the target generalized additive model converges to meet the downscaling accuracy requirements, thereby improving the accuracy of the target land use area prediction results for the preset region output by the adjusted target generalized additive model that meets the downscaling accuracy requirements. This allows the model to meet the downscaling requirements and improves the resolution of the land use scenario data corresponding to the preset region in the LUH2 dataset. The preset difference can be 0.001.

[0127] In an exemplary embodiment provided in this application, importing the intermediate land area prediction results into QGIS software reveals that the downscaling results still have certain problems. Therefore, when adjusting the network parameters in the initial generalized additive model based on constraint values ​​and average differences, multiple rounds of iterative adjustment are required to obtain the target generalized additive model. The specific process of multiple rounds of iterative adjustment is as follows:

[0128] The intermediate land area prediction results are imported into the previously trained initial generalized additive model as the response variable. This step may have the problem of excessive data volume. Therefore, sampling can be carried out according to hardware conditions. Taking 20% ​​equally spaced random sampling as an example, all data pixels are divided into 10 equally spaced intervals. 3,000 to 5,000 pixels are extracted from each interval, and a total of 50,000 to 100,000 constraint optimization results are extracted as the response variable and imported.

[0129] When the constraint value is greater than the preset value, and / or the average difference is greater than the preset difference, the network parameters in the initial generalized additive model are adjusted to retrain the initial generalized additive model and obtain the intermediate generalized additive model.

[0130] The intermediate land use area prediction results are imported into the intermediate generalized additive model for processing to obtain the transitional land use area prediction results. These results are then compared with the intermediate land use area prediction results to determine the difference in predicted values ​​for different land use types in different pixels. This is achieved by subtracting the two values. If the average difference exceeds 0.001 (preset difference) and / or the constraint value for this training is greater than the preset value, the transitional land use area prediction results for this round are used as the response variable and imported into the intermediate generalized additive model for retraining. This process is repeated until the average difference is less than or equal to 0.001 (preset difference) and the constraint value is less than or equal to the preset value. At this point, the model has converged, and the target generalized additive model is obtained.

[0131] The predicted area of ​​the target land type, output by the target generalized additive model, is the final downscaling result for the preset area.

[0132] The final downscaling results are saved in CSV format and imported into QGIS software. The results are then connected to the fishing net based on ID_1 and visualized. Finally, the vector-to-raster tool is used to convert the downscaling results into a visualized 1km raster file, thus obtaining LUH2 downscaling results for 5 land cover types at 1km.

[0133] Please see Figure 2 , Figure 2 As an exemplary embodiment of this application, the application provides a flowchart illustrating the land use data downscaling method based on a generalized additive model, as shown below. Figure 2 As shown, the application process steps are as follows:

[0134] Input LUH2 land use data (LUH2 dataset) with a resolution of 0.25°, and driving element data with an independent variable layer and a resolution of 1km, to form a coarse fishing net and a fine fishing net for the preset area;

[0135] The coarse and fine fishing nets are input into a predefined initial generalized additive model. This model defines the explanatory and predictor variables, as well as the explained variable. The coarse and fine fishing nets are then downscaled within the initial model. After the first iteration, the following results can be obtained: Figure 3 The results shown are the first predictions (intermediate land area predictions) with a resolution of 1 km.

[0136] The initial generalized additive model is optimized based on the intermediate land category area prediction results to obtain the target generalized additive model. The optimization and adjustment process is as follows:

[0137] During model training, the network parameters in the generalized additive model for each land type are adjusted. If a generalized additive model for a land type has converged in a certain iteration, that is, the difference in the predicted values ​​corresponding to the generalized additive models for land types is less than or equal to 0.001, it indicates that the generalized additive model for that land type has converged. Then, the network parameters of the generalized additive model for that land type after convergence are fixed (in the next iteration, only the network parameters in the generalized additive models for land types that have not been fixed are adjusted), and the predicted area of ​​the transitional land type output by each round of the model is imported into the intermediate generalized additive model after adjusting the network parameters in the next round, until the generalized additive models for all land types in the intermediate generalized additive model have converged, and the final target generalized additive model is output.

[0138] Then, based on the objective generalized additive model, the output is as follows: Figure 4 The image shows the final iterative results (target land area prediction results) for the preset area.

[0139] In summary, the land use data downscaling method based on the generalized additive model proposed in this application can improve downscaling accuracy, solve spatial heterogeneity problems, improve computational efficiency, and enhance the universality of the model.

[0140] Improving downscaling accuracy is achieved by introducing spatial statistical models such as the Generalized Additive Model (GAM) and combining multiple driving factors (such as climate, soil, and topography) for downscaling. This allows for a more accurate simulation of the spatial distribution of land use types, particularly effective in handling regions with significant spatial heterogeneity. This method provides accurate land use change predictions while maintaining high computational efficiency, downscaling global-scale LUH2 data (0.25° resolution) to higher resolutions (such as 1km or finer resolution). This downscaling process not only ensures accuracy but also effectively improves computational efficiency, making it suitable for land use studies in various regions.

[0141] Addressing the issue of spatial heterogeneity involves introducing multiple geographical and environmental drivers (such as climate, soil, and topography) to capture the spatial heterogeneity of land use change. This approach considers the spatial heterogeneity of land use types and ecological environment characteristics, uses multi-source drivers as input variables, and employs regression analysis to capture the dynamic characteristics of land use change. This effectively reflects the spatial distribution and changing patterns of different land use types. Compared to traditional single-variable methods, the method presented in this application more realistically reflects the spatial dynamics and ecosystem impacts of different land use types.

[0142] The improved computational efficiency is reflected in the use of R language and Geographic Information System (GIS) for data processing, the application of efficient spatial analysis methods and optimized algorithms (such as stepwise regression and VIF collinearity detection), and the use of stepwise regression, VIF collinearity detection, and AIC value optimization to ensure that the constructed model can reflect complex land use change patterns. This maintains high generalization ability and accuracy in a variable environment, guaranteeing both high accuracy and reduced computation time during downscaling, making it suitable for large-scale data processing and complex regional analysis. Furthermore, by combining R language with GIS technology, using GIS software (such as QGIS) for data preprocessing, region clipping, and grid generation, and leveraging R language's powerful data processing capabilities, the data processing workflow is simplified, significantly improving the efficiency and accuracy of the downscaling process.

[0143] The enhanced model's universality is reflected in its ability to be flexibly adjusted according to the characteristics of different study areas (such as different land use types and ecological environment backgrounds). Even with insufficient or incomplete data, it can provide relatively accurate downscaling results by selecting appropriate driving factors and optimizing model parameters. It provides reliable downscaling results in both data-rich and data-scarce regions, demonstrating strong universality.

[0144] Please see Figure 5 , Figure 5 An exemplary embodiment of this application illustrates a land use data downscaling system based on a generalized additive model, such as... Figure 5 As shown, this application provides a land use data downscaling system 500 based on a generalized additive model, comprising:

[0145] Module 501 is used to acquire the LUH2 dataset and the downscaling requirements. The LUH2 dataset includes data of various land cover types.

[0146] Fishing net construction module 502 is used to construct fishing nets in a preset area based on the LUH2 dataset, downscaling requirements, and preset driving element data, to obtain coarse and fine fishing nets.

[0147] The downscaling module 503 is used to input coarse and fine fishing nets into a preset initial generalized additive model for downscaling processing, and to obtain the predicted area of ​​intermediate land types for the preset area.

[0148] The model optimization module 504 is used to optimize the initial generalized additive model based on the intermediate land area prediction results to obtain the target generalized additive model;

[0149] The result output module 505 is used to output the target land area prediction results for the preset area based on the target generalized additive model.

[0150] The land use data downscaling system 500 based on the generalized additive model provided in this application first constructs a coarse and fine net for a preset area using a net construction module 502 based on the LUH2 dataset acquired by the acquisition module 501, downscaling requirements, and preset driving element data. The downscaling module 503 inputs the coarse and fine nets into a preset initial generalized additive model for downscaling. The model optimization module 504 then optimizes the initial generalized additive model using the intermediate land use area prediction results obtained from the downscaling process, thereby improving the downscaling accuracy of the optimized target generalized additive model. This ensures that the target land use area prediction results for the preset area output by the target generalized additive model in the result output module 505 meet the downscaling requirements, thus improving the resolution of the land use scenario data corresponding to the preset area in the LUH2 dataset.

[0151] Optionally, the fishing net construction module 502 is specifically used for:

[0152] Based on the LUH2 dataset and preset driving element data, a fishing net is constructed in a preset area to obtain a coarse fishing net;

[0153] Based on the downscaling requirements and driving factor data, a fishing net is constructed in the preset area to obtain a fine fishing net.

[0154] Optionally, the fishing net construction module 502 is specifically used for:

[0155] The LUH2 dataset is cropped based on a preset region to obtain a low-resolution dataset for the preset region;

[0156] A first vector fishing net matching the size of the low-resolution dataset is constructed;

[0157] The low-resolution dataset and the preset driving element data are added to the first vector fishing net to form a coarse fishing net.

[0158] Optionally, the downscaling requirement is the requirement to downscale the LUH2 dataset to the target resolution;

[0159] Fishing net construction module 502 is specifically used for:

[0160] A second vector fishing net is constructed corresponding to the preset area, and the resolution of the second vector fishing net is the same as the target resolution;

[0161] The driving element data is added to the second vector fishing net to form a fine fishing net.

[0162] Optionally, the initial generalized additive model includes a generalized additive model for each land type, and the driving element data includes multiple driving elements;

[0163] Downscaling module 503 is specifically used for:

[0164] The coarse and fine fishing nets are input into the preset initial generalized additive model. The land use generalized additive model is used to make predictions based on the driving elements in the fine fishing net, and the original area prediction results of the corresponding land use type in the fine fishing net are obtained. The original area prediction results include the original land use area value and the original standard error.

[0165] Based on multiple original area prediction results, an initial land category area prediction result is generated for the preset area.

[0166] The initial land area prediction results are corrected based on coarse and fine fishing nets to obtain intermediate land area prediction results for the preset area.

[0167] Optionally, the initial generalized additive model includes the generalized additive model for each land type, and the intermediate land area prediction results include the intermediate land area values ​​and intermediate standard errors of the generalized additive model for each land type.

[0168] Model optimization module 504 is specifically used for:

[0169] Using a pre-defined constraint optimization function, constraint values ​​are calculated based on multiple intermediate land area values, multiple intermediate standard errors, and corresponding multiple real land area values.

[0170] The average difference is calculated based on the difference between the predicted value and the corresponding actual land area value for each intermediate land category.

[0171] The network parameters in the initial generalized additive model are adjusted based on the constraint values ​​and average differences to obtain the target generalized additive model.

[0172] Optionally, the model optimization module 504 is specifically used for:

[0173] When the constraint value is greater than the preset value, and / or the average difference is greater than the preset difference, the network parameters in the initial generalized additive model are adjusted to obtain the intermediate generalized additive model.

[0174] The coarse and fine fishing nets are input into the intermediate generalized summation model for downscaling to obtain the predicted area of ​​transitional land types for the preset area.

[0175] When the constraint value corresponding to the predicted area of ​​transitional land is less than or equal to the preset value, and the corresponding average difference is less than or equal to the preset difference, the corresponding intermediate generalized additive model is determined as the target generalized additive model.

[0176] It should be noted that the land use data downscaling system based on the generalized additive model provided in the above embodiments and the land use data downscaling method based on the generalized additive model provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the land use data downscaling system based on the generalized additive model provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0177] A computing device according to an embodiment of this application includes a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements some or all of the steps of the land use data downscaling method based on the generalized additive model described above.

[0178] The computing device can be a computer, and the corresponding program is computer software. The parameters and steps of the computing device described above can be referred to the parameters and steps of the embodiment of the land use data downscaling method based on the generalized additive model in the above text, and will not be repeated here.

[0179] This application provides a computer-readable storage medium storing instructions that, when executed, perform the steps of the land use data downscaling method based on a generalized additive model.

[0180] The computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0181] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of this disclosure. The aforementioned computer-readable storage medium can be a non-transitory computer-readable storage medium, including: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, and other media capable of storing program code; it can also be a transient computer-readable storage medium.

[0182] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0183] Those skilled in the art will recognize that this application can be implemented as a system, method, or computer program product. Therefore, this disclosure can be implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "module" or "system." Furthermore, in some embodiments, this application can also be implemented as a computer program product contained in one or more computer-readable media, which contains computer-readable program code. Computer-readable storage media can be, for example, but not limited to—electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof.

[0184] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0185] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A land use data downscaling method based on generalized additive model, characterized in that, The method comprises the following steps: obtaining an LUH2 data set and a downscaling requirement, the LUH2 data set comprising data of a plurality of land class types; performing mesh construction on a preset region based on the LUH2 data set, the downscaling requirement, and preset driving factor data to obtain a coarse mesh and a fine mesh; inputting the coarse mesh and the fine mesh into a preset initial generalized additive model for downscaling processing to obtain an intermediate land class area prediction result for the preset region; wherein the initial generalized additive model comprises a land class generalized additive model corresponding to each land class type, and the intermediate land class area prediction result comprises an intermediate land class area value of each land class generalized additive model and an intermediate standard error; calculating a constraint value based on a plurality of intermediate land class area values, a plurality of intermediate standard errors, and a plurality of corresponding true land class area values by using a preset constraint optimization function; the formula of the constraint optimization function is: wherein, represents a constraint value, represents a number of land class types, represents a land class type, represents a true land class area value corresponding to the th land class type, represents an intermediate land class area value included in the intermediate land class area prediction result corresponding to the th land class type, represents an intermediate standard error included in the intermediate land class area prediction result output by the land class generalized additive model corresponding to the th land class type. calculating an average difference based on a prediction value difference between each intermediate land class area value and a corresponding true land class area value; when the constraint value is greater than a preset value and / or the average difference is greater than a preset difference, adjusting network parameters in the initial generalized additive model to obtain an intermediate generalized additive model; inputting the coarse mesh and the fine mesh into the intermediate generalized additive model for downscaling processing to obtain a transition land class area prediction result for the preset region; when a constraint value corresponding to the transition land class area prediction result is less than or equal to the preset value and a corresponding average difference is less than or equal to the preset difference, determining the corresponding intermediate generalized additive model as a target generalized additive model; outputting a target land class area prediction result for the preset region based on the target generalized additive model.

2. The method of claim 1, wherein, The method comprises the following steps: performing mesh construction on a preset region based on the LUH2 data set and preset driving factor data to obtain a coarse mesh; performing mesh construction on the preset region based on the downscaling requirement and the driving factor data to obtain a fine mesh.

3. The method of claim 2, wherein, The method comprises the following steps: performing mesh construction on a preset region based on the LUH2 data set and preset driving factor data to obtain a coarse mesh; performing mesh construction on the preset region based on the downscaling requirement and the driving factor data to obtain a fine mesh. The method comprises the following steps:

4. The method of claim 2, wherein, performing mesh construction on a preset region based on the LUH2 data set and preset driving factor data to obtain a coarse mesh; performing mesh construction on the preset region based on the downscaling requirement and the driving factor data to obtain a fine mesh. The downscaling requirement is a requirement for downscaling the LUH2 data set to a target resolution; The method comprises the following steps: performing mesh construction on a preset region based on the LUH2 data set and preset driving factor data to obtain a coarse mesh; performing mesh construction on the preset region based on the downscaling requirement and the driving factor data to obtain a fine mesh.

5. The method according to any one of claims 1 to 4, characterized in that, The initial generalized additive model comprises a land class generalized additive model corresponding to each land class type, and the driving factor data comprises a plurality of driving factors; The coarse fishing net and the fine fishing net are input into a preset initial generalized additive model for downscaling processing to obtain an intermediate land class area prediction result for the preset area, including: The coarse fishing net and the fine fishing net are input into a preset initial generalized additive model, and the land class generalized additive model is used to predict based on the driving factors in the fine fishing net to obtain an original area prediction result of the corresponding land class type in the fine fishing net, the original area prediction result comprising an original land class area value and an original standard error; Based on a plurality of the original area prediction results, an initial land class area prediction result for the preset area is formed; Based on the coarse fishing net and the fine fishing net, the initial land class area prediction result is corrected to obtain an intermediate land class area prediction result for the preset area. 6.A land use data downscaling system based on generalized additive model, characterized in that, Including: An acquisition module is configured to acquire an LUH2 dataset and a downscaling requirement, the LUH2 dataset comprising data of a plurality of land class types; A fishing net construction module is configured to construct a fishing net for a preset area based on the LUH2 dataset, the downscaling requirement, and preset driving factor data to obtain a coarse fishing net and a fine fishing net; A downscaling module is configured to input the coarse fishing net and the fine fishing net into a preset initial generalized additive model for downscaling processing to obtain an intermediate land class area prediction result for the preset area; wherein the initial generalized additive model comprises a land class generalized additive model corresponding to each land class type, and the intermediate land class area prediction result comprises an intermediate land class area value and an intermediate standard error of each land class generalized additive model; A model optimization module is configured to calculate a constraint value by using a preset constraint optimization function based on a plurality of the intermediate land class area values, a plurality of the intermediate standard errors, and a plurality of corresponding true land class area values; The formula of the constraint optimization function is: wherein, represents a constraint value, represents a number of land class types, represents a land class type, represents a true land class area value corresponding to the th land class type, represents an intermediate land class area value included in the intermediate land class area prediction result corresponding to the th land class type, represents an intermediate standard error included in the intermediate land class area prediction result output by the land class generalized additive model corresponding to the th land class type. An average difference is calculated based on a prediction value difference between each intermediate land class area value and a corresponding true land class area value; When the constraint value is greater than a preset value, and / or the average difference is greater than a preset difference, a network parameter in the initial generalized additive model is adjusted to obtain an intermediate generalized additive model; The coarse fishing net and the fine fishing net are input into the intermediate generalized additive model for downscaling processing to obtain a transition land class area prediction result for the preset area; When a constraint value corresponding to the transition land class area prediction result is less than or equal to the preset value, and a corresponding average difference is less than or equal to the preset difference, the corresponding intermediate generalized additive model is determined as a target generalized additive model; A result output module is configured to output a target land class area prediction result for the preset area based on the target generalized additive model.

7. A computing device comprising a memory, a processor, and a program stored on the memory and running on the processor, wherein, The processor executes the program to implement the steps of the land use data downscaling method based on the generalized additive model according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions, and when the instructions run on the terminal device, the terminal device executes the steps of the land use data downscaling method based on the generalized additive model according to any one of claims 1 to 5.

Images