A machine learning-based spei coarse resolution data space downscaling method and device

Abstract

The present application relates to a kind of machine learning-based SPEI coarse resolution data space downscaling method and device, the method includes: step 1: obtaining the original coarse resolution SPEI grid data of long time series in research area;Step 2: obtain the downscaling factor data in research area;Step 3: the downscaling factor is resampled to and the spatial resolution of original SPEI;Step 4: construct the sample data of all pixels, including SPEI value and downscaling factor value, and sample data is divided into test set and training set;Step 5: based on training set, construct spatial downscaling model, construct downscaling model using machine learning algorithm Gaussian process regression, the accuracy of model is evaluated using test set;Step 6: the high-resolution downscaling factor grid data of step 2 is input into the model trained, generates long time series high spatial resolution's downscaling SPEI data product.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, specifically to a method and apparatus for spatial downscaling of coarse-resolution data based on standardized precipitation evapotranspiration index (SPEI) using machine learning. Background Technology

[0002] Global warming will lead to more frequent and severe droughts, severely impacting regional water resources and terrestrial ecosystems. China is one of the countries most frequently affected by drought. In recent decades, influenced by climate change, China has experienced frequent droughts, causing serious consequences for the ecology, agriculture, and socio-economic development. Therefore, monitoring and mitigating droughts is crucial.

[0003] Drought types typically include meteorological drought, agricultural drought, hydrological drought, and socioeconomic drought. Drought indices are the foundation for quantifying and describing drought and are effective tools for drought early warning, monitoring, and assessment. The Standardized Precipitation Evapotranspiration Index (SPEI) is one of the most widely used meteorological drought indices. While existing SPEI data products have sufficiently long time series (1901-2021), their spatial resolution is relatively coarse (0.5°). This is suitable for global studies, but its spatial resolution is too coarse for assessing the spatiotemporal changes of drought in my country. The spatial scale of drought can range from a few kilometers to an entire region; therefore, it is necessary to assess drought in my country at a high-resolution grid scale. The lack of long-term, high-spatial-resolution drought datasets makes it difficult to quantify long-term drought in my country and to grasp the complex spatiotemporal characteristics of drought.

[0004] Currently, high-resolution, spatially continuous SPEI datasets can be obtained through interpolation using meteorological station SPEI data. However, this method is limited by the uneven distribution of stations and the lack of meteorological station data in some areas, often resulting in significant spatial uncertainty in the interpolated results. Geographically weighted regression is a commonly used statistical downscaling method, but its applicability to nonstationary and nonlinear problems is limited.

[0005] Therefore, there is an urgent need to develop a technical solution to address the above-mentioned shortcomings, construct an effective SPEI spatial downscaling model, improve the spatial resolution of my country's long-term SPEI data, help to better understand the spatiotemporal patterns and occurrence of drought in my country, and improve the reliability of drought risk assessment in my country. Summary of the Invention

[0006] To achieve the above objectives, this invention provides a machine learning-based SPEI spatial downscaling method to improve the spatial resolution of SPEI data products, meet the application needs of fine-scale drought monitoring and assessment as well as water resource management, and solve the problem of low spatial resolution in existing SPEI data products.

[0007] More specifically, this invention employs a machine learning-based SPEI coarse-resolution data spatial downscaling method, comprising:

[0008] Step 1: Obtain raw coarse-resolution SPEI raster data of long-term series within the study area;

[0009] Step 2: Obtain downscaling factor data for the study area, including high-resolution raster data of climate, topography, and geographic location factors, where the climate factors correspond to the long-term series of SPEI.

[0010] Step 3: Resample the downscaling factor to the same spatial resolution as the original SPEI;

[0011] Step 4: Construct sample data for all pixels. Using pixels as statistical units, extract the SPEI value and downscaling factor value of all pixels. Use all pixel values ​​as samples and divide the sample data into test set and training set.

[0012] Step 5: Construct a spatial downscaling model based on the training set, using the SPEI value as the dependent variable and the downscaling factor as the independent variable. Utilize the Gaussian Process Regression (GPR) machine learning algorithm to build the downscaling model. Model parameters are determined using Bayesian optimization, and the model's accuracy is evaluated using a test set.

[0013] Step 6: Input the high-resolution downscaling factor raster data from Step 2 into the trained model to generate a long-term high spatial resolution downscaled SPEI data product.

[0014] According to an embodiment of the present invention, the method further includes step 7: cross-validating the original SPEI from step 1 and validating the downscaled SPEI data product by means of drought events in the study area's historical records.

[0015] According to an embodiment of the present invention, in step 2, the climatic factors include monthly average temperature, monthly maximum temperature, monthly minimum temperature, and monthly precipitation; the topographic factors include elevation, slope, and aspect; and the geographic location factors include longitude and latitude. For example, monthly precipitation, monthly average temperature, maximum temperature, and minimum temperature data can be obtained from the National Tibetan Plateau Scientific Data Center (https: / / data.tpdc.ac.cn / ), with a spatial resolution of 0.0083333° (approximately 1 km) and a time period from January 1901 to December 2021. Those skilled in the art can also select other high-resolution meteorological data products according to specific circumstances; the topographic factors can be calculated using a digital elevation model (DEM), and the DEM data can be SRTM DEM Version 4. Those skilled in the art can also select other DEM data products according to specific circumstances.

[0016] According to an embodiment of the present invention, in step 1, the coarse resolution is 0.5° × 0.5°. For example, the original SPEI database was constructed by Sergio M. Vicente-Serrano, the proposer of SPEI, with a spatial resolution of 0.5°, and the spatial resolution of the downscaling factor is 0.0083°.

[0017] According to an embodiment of the present invention, in step 5, the hyperparameters of the Gaussian process regression (GPR) model are obtained using Bayesian optimization, and the downscaling model is constructed based on monthly data.

[0018] The Gaussian Probabilistic Regression (GPR) method is a nonparametric probabilistic model based on kernel functions, with its core theory being Bayesian methods. GPR performs well on small datasets and is generally not affected by overfitting. A GPR with Gaussian noise can be represented as follows:

[0019] y = f(x) + ε

[0020] In the formula, ε is the noise that follows a Gaussian distribution (ε ~ ​​N(0, σ). 2 Let x represent the input and y represent the output. f(x) is the regression function, which can be expressed as:

[0021] f(x)~GP(m(x),k(x,x′))

[0022] In the formula, m(x) is the mean function and k(x, x′) is the covariance function.

[0023] According to an embodiment of the present invention, step 5 further includes evaluating the accuracy of the model by calculating the coefficient of determination (R²) using a test set. 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the model.

[0024] Accuracy evaluation index R 2 The formulas for calculating RMSE and MAE are as follows:

[0025]

[0026]

[0027]

[0028] In the formula: O i E represents the actual value of the i-th data point. i Let be the simulated value of the i-th site; n represents the number of samples. This is the average of the actual values.

[0029] According to an embodiment of the present invention, step 7 includes calculating the correlation between the original SPEI and the downscaled SPEI to ensure that the accuracy of the downscaled data is not reduced, and using historical drought events to evaluate the effectiveness of the downscaled SPEI in identifying drought.

[0030] According to an embodiment of the present invention, in step 4, the samples are divided into a 70% training set for training the model and a 30% test set for evaluating the model accuracy.

[0031] It should be understood that SPEI data can include multiple time scales, such as January SPEI, March SPEI, June SPEI, etc.

[0032] The present invention also provides an apparatus for a machine learning-based SPEI coarse-resolution data spatial downscaling method, comprising:

[0033] The coarse-resolution data acquisition module is used to acquire raw coarse-resolution SPEI raster data of long-term series within the study area;

[0034] The downscaling factor data acquisition module is used to acquire downscaling factor data within the study area;

[0035] The downscaling factor resampling module is used to resample the downscaling factor to the same spatial resolution as the original SPEI;

[0036] The sample data construction module is used to construct sample data for all pixels and divide the sample data into a test set and a training set.

[0037] The spatial downscaling model building module is used to construct spatial downscaling machine learning models based on the training set and evaluate the model's accuracy based on the test set; and

[0038] The downscaling SPEI data product generation module is used to input high-resolution downscaling factor raster data into a trained model to generate high spatial resolution downscaling SPEI data products for long-term series.

[0039] According to an embodiment of the present invention, the apparatus further includes a verification module for cross-validating the original SPEI from step 1 and verifying the downscaled SPEI data product using drought events recorded in the study area.

[0040] The present invention also provides an electronic device, including a memory and one or more processors;

[0041] The memory is used to store one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method of the present invention.

[0042] This invention proposes a coarse-resolution SPEI spatial downscaling method based on the machine learning GPR algorithm. The factors of the constructed downscaling model are easily obtained, and it does not require on-site measured data, resulting in high model accuracy. This method can generate high spatial resolution SPEI data, which can provide data support for small-scale, high-precision drought monitoring and assessment. Attached Figure Description

[0043] Figure 1 The flowchart below illustrates a machine learning-based SPEI coarse-resolution data spatial downscaling method according to an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram showing the accuracy results of the downscaling model on SPEI-6 in the SPEI coarse-resolution data spatial downscaling method based on machine learning according to an embodiment of the present invention.

[0045] Figure 3 This is a spatial comparison diagram of the original SPEI and the downscaled SPEI in the SPEI coarse-resolution data spatial downscaling method based on machine learning according to an embodiment of the present invention.

[0046] Figure 4 The correlation coefficient between the original SPEI and the downscaled SPEI in the monthly data spatial downscaling method based on machine learning according to the embodiment of the present invention.

[0047] Figure 5 To illustrate how the machine learning-based spatial downscaling method for SPEI coarse-resolution data according to an embodiment of the present invention is applied to the spatial distribution map of SPEI-6 in a certain study area in 2006;

[0048] Figure 6A schematic diagram of a machine learning-based SPEI coarse-resolution data spatial downscaling device according to an embodiment of the present invention; and

[0049] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Figure 1 This is a flowchart illustrating a machine learning-based SPEI coarse-resolution data spatial downscaling method according to an embodiment of this application. The method of this invention is described in detail below, using a city in China as the research area. As shown in the figure, the method of an embodiment may include:

[0052] (1) A city in China was selected as the study area. The original coarse resolution 0.5° long-term SPEI spatial dataset (SPEIbase v2.7) constructed by Sergio M.Vicente-Serrano was obtained. Taking the SPEI at the 6-month scale as an example, the time span is from 1901 to 2020.

[0053] (2) Obtain downscaling factor data and obtain monthly precipitation and temperature data for my country from 1901 to 2020. The data is obtained from the National Tibetan Plateau Scientific Data Center with a spatial resolution of 0.0083333° (about 1km). Specifically, it includes monthly average temperature, monthly maximum temperature, monthly minimum temperature, and monthly precipitation. Obtain DEM data of STRM version 4 and calculate elevation, slope and aspect, and generate longitude and latitude spatial data.

[0054] (3) Unify the spatial resolution of the downscaling factor raster data to 0.5°, while maintaining the same grid size as the coarse resolution SPEI.

[0055] (4) Summarize the SPEI value and downscaling factor value of each pixel, with the original SPEI value as the dependent variable and the downscaling factors temperature, precipitation, and topography as independent variables, to construct sample data for all pixels. The sample data is then divided into a 70% training set for training the model and a 30% test set for evaluating the model accuracy.

[0056] (5) Based on the machine learning GPR algorithm, a downscaling model is trained monthly using the training set. During the GPR model construction process, the model hyperparameters are optimized using Bayesian methods. The optimized basic function is constant, and the kernel function is a rational quadratic kernel function. The data inside the model is standardized. Based on the test set, R... 2 The RMSE and MAE indicators are used to evaluate the accuracy of the model, and the model has high prediction accuracy. Figure 2 This demonstrates the accuracy of SPEI-6, showing the R-value for each month. 2 The medians are all above 0.95, and both RMSE and MAE are less than 0.2.

[0057] (6) Use the constructed downscaling model to generate long-term series SPEI spatial data products for my country. Figure 3 The comparison between the original coarse SPEI and the downscaled SPEI shows that the downscaled SPEI is smoother and has more detail.

[0058] (7) Then, cross-validation was performed using the downscaled SPEI and the original SPEI. The results showed that the two had a high correlation, with the correlation greater than 0.9 for each month. The results are as follows: Figure 4 As shown;

[0059] Furthermore, historical drought events were used to verify the reliability of the downscaling data. Taking the summer drought event in the study area in 2006 as an example, it was found that the model could accurately identify historical drought events in small areas. The SPEI-6 was low starting in June, with the drought being most severe in August, thus verifying the reliability of the data produced by the downscaling model. The results are as follows: Figure 5 As shown.

[0060] Figure 6 This is a schematic diagram of a machine learning-based SPEI coarse-resolution data spatial downscaling device according to an embodiment of the present invention. Figure 6As shown, the apparatus includes: a coarse-resolution data acquisition module 610 for acquiring original coarse-resolution SPEI raster data of a long-term series within the study area; a downscaling factor data acquisition module 620 for acquiring downscaling factor data of the study area; a downscaling factor resampling module 630 for resampling the downscaling factor to the same spatial resolution as the original SPEI; a sample data construction module 640 for constructing sample data for all pixels and dividing the sample data into a test set and a training set; a spatial downscaling model construction module 650 for constructing a spatial downscaling machine learning model based on the training set and evaluating the accuracy of the model based on the test set; a downscaling SPEI data product formation module 660 for inputting high-resolution downscaling factor raster data into the trained model to generate a long-term series high spatial resolution downscaling SPEI data product; and a validation module 670 for cross-validating the original SPEI from step 1 and validating the downscaling SPEI data product using drought events recorded in the study area.

[0061] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 7 As shown, the electronic device includes a processor 710, a memory 720, an input device 730, and an output device 740; the number of processors 710 in the electronic device can be one or more. Figure 7 Taking a processor 710 as an example; the processor 710, memory 720, input device 730, and output device 740 in the electronic device can be connected via a bus or other means. Figure 7 Taking the example of a connection between China and Israel via a bus.

[0062] The memory 720, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the machine learning-based SPEI coarse-resolution data spatial downscaling method in this embodiment of the invention (e.g., coarse-resolution data acquisition module 610; downscaling factor data acquisition module 620; downscaling factor resampling module 630; sample data construction module 640; spatial downscaling model construction module 650; downscaling SPEI data product formation module 660; and verification module 670). The processor 710 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 720, thereby implementing the aforementioned machine learning-based SPEI coarse-resolution data spatial downscaling method.

[0063] The memory 720 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 720 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory 720 may further include memory remotely located relative to the processor 710, which can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The input device 730 can be used to receive input digital or character information and generate key signal inputs related to user settings and function control of the electronic device. The output device 740 may include a display device such as a screen.

[0064] Based on the existing coarse-resolution long-term SPEI dataset, this invention utilizes the advantages of the Gaussian process regression machine learning algorithm to construct a downscaling model that considers climate, topography, and geographical location factors, thereby obtaining high spatial resolution (0.0083°) SPEI spatial data products of my country for long-term series, providing a reference for drought assessment at a more refined scale.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based SPEI coarse-resolution data spatial downscaling method, characterized in that, include: Step 1: Obtain raw coarse-resolution SPEI raster data of long-term series within the study area; Step 2: Obtain downscaling factor data for the study area, including high-resolution raster data of climate, topography, and geographic location factors, where the climate factors correspond to the long-term series of SPEI. Step 3: Resample the downscaling factor to the same spatial resolution as the original SPEI; Step 4: Construct sample data for all pixels. Using pixels as statistical units, extract the SPEI value and downscaling factor value of all pixels. Use all pixel values ​​as sample data and divide the sample data into test set and training set. Step 5: Construct a spatial downscaling model based on the training set, using the SPEI value as the dependent variable and the downscaling factor as the independent variable. Use the Gaussian process regression machine learning algorithm to construct the downscaling model. The model parameters are determined through Bayesian optimization, and the accuracy of the model is evaluated using the test set. as well as Step 6: Input the high-resolution downscaling factor raster data from Step 2 into the trained model to generate a long-term high spatial resolution downscaled SPEI data product.

2. The method according to claim 1, characterized in that, It also includes step 7: cross-validation using the original SPEI from step 1 and validation of the downscaled SPEI data product using drought events from the study area’s historical records.

3. The method according to claim 1, characterized in that, In step 2, the climate factors include monthly average temperature, monthly maximum temperature, monthly minimum temperature, and monthly precipitation; the topographic factors include elevation, slope, and aspect; and the geographical location factors include longitude and latitude.

4. The method according to claim 1, characterized in that, In step 1, the coarse resolution is 0.5°×0.5°, and the spatial resolution of the downscaling factor is 0.0083°.

5. The method according to claim 1, characterized in that, In step 5, the hyperparameters of the Gaussian process regression model are obtained using Bayesian optimization, and the downscaling model is constructed based on monthly data.

6. The method according to claim 1, characterized in that, Step 5 also includes evaluating the accuracy of the model by calculating the coefficient of determination (R²) using the test set. 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the model.

7. The method according to claim 2, characterized in that, Step 7 includes calculating the correlation between the original SPEI and the downscaled SPEI to ensure that the accuracy of the downscaled data is not reduced, and using historical drought time to evaluate the effectiveness of the downscaled SPEI in identifying drought.

8. The method according to claim 1, characterized in that, In step 4, the samples are divided into a 70% training set for training the model and a 30% test set for evaluating the model's accuracy.

9. A machine learning-based SPEI coarse-resolution data spatial downscaling method apparatus, characterized in that, include: The coarse-resolution data acquisition module is used to acquire raw coarse-resolution SPEI raster data of long-term series within the study area; The downscaling factor data acquisition module is used to acquire downscaling factor data within the study area; The downscaling factor resampling module is used to resample the downscaling factor to the same spatial resolution as the original SPEI; The sample data construction module is used to construct sample data for all pixels and divide the sample data into a test set and a training set. The spatial downscaling model building module is used to build spatial downscaling machine learning models based on the training set and evaluate the accuracy of the models based on the test set. as well as The downscaling SPEI data product generation module is used to input high-resolution downscaling factor raster data into a trained model to generate high spatial resolution downscaling SPEI data products for long-term series.

10. An electronic device, characterized in that, include: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-8.

Images