Complex terrain air temperature refined downscaling method based on TopoMW model

Abstract

The invention relates to a complex terrain air temperature refined downscaling method based on a TopoMW model, and belongs to the technical field of meteorological element downscaling, and the method comprises the steps: obtaining multi-source data of a target region, carrying out the preprocessing of the multi-source data, and fusing the multi-source data into a feature set; constructing a complex terrain multi-weighting matrix based on the preprocessed multi-source data; a TopoMW model improved based on UNet is established, a weighted MSE loss function is adopted to train the TopoMW model on the basis of a feature set in combination with a complex terrain multiple weighting matrix, an optimal downscaling model is obtained, and the contradiction between model statistic multiplexing and performance retention is solved; inputting a feature set to be predicted into the optimal downscaling model to obtain an initial prediction result; the initial prediction result is corrected based on the complex terrain multiple weighting matrix to obtain a final result, extreme deviation is eliminated, and accurate generation of the 30-meter high-resolution air temperature under the complex terrain is achieved.

Description

Technical Field

[0001] This invention relates to the field of meteorological element downscaling technology, specifically to a refined downscaling method for temperature in complex terrain based on the TopoMW model. Background Technology

[0002] Meteorological elements (especially temperature and precipitation) are key parameters in fields such as surface energy balance, ecological environment monitoring, and agricultural production management, and their spatial resolution directly affects the application results. Existing meteorological data (such as ERA5) are mostly low-resolution products at the kilometer level, which cannot capture the microclimate differences in small-scale areas (such as hillsides, woodlands, and urban blocks) under complex terrain, thus limiting their application in refined scenarios. Meteorological element downscaling technology, by fusing high-resolution auxiliary data, upscales low-resolution meteorological data to a finer scale, becoming a core means to solve this problem.

[0003] Existing meteorological element downscaling methods are mainly divided into two categories: dynamic downscaling and statistical downscaling. Dynamic downscaling is based on physical models and has a clear theoretical foundation, but it is computationally expensive and has stringent hardware requirements. Statistical downscaling (such as the TsHARP algorithm) relies on the linear relationship between meteorological elements and factors such as vegetation indices and topography. It is computationally simple, but it has the following limitations: 1. Linear relationship limitation: Traditional statistical methods are difficult to characterize the nonlinear relationship between meteorological elements and auxiliary factors under complex terrain, especially in areas with strong vegetation cover heterogeneity and large topographic relief, resulting in low downscaling accuracy; 2. Statistical mismatch problem: If the normalized statistics (mean, standard deviation) are inconsistent during model training and validation, it will lead to inconsistencies in the prediction results. 1. Extreme bias, failing to reflect the true distribution of meteorological elements; 2. Low effective data utilization: Traditional methods rely heavily on sparse data from meteorological stations for verification, neglecting the large-scale verification value of high-resolution raster data, resulting in incomplete accuracy evaluation; 3. Poor adaptability to complex terrain: Traditional methods do not consider the impact of terrain complexity differences on prediction accuracy, and are insufficient in characterizing the spatial distribution of meteorological elements in extremely complex and highly complex areas, leading to significant deviations between prediction results and reality; Therefore, there is an urgent need for a refined downscaling method for meteorological elements that can overcome the limitations of linear relationships, solve the problem of statistical mismatch, enhance adaptability to complex terrain, and achieve large-scale accurate verification. Summary of the Invention

[0004] The purpose of this invention is to provide a refined downscaling method for temperature in complex terrain based on the TopoMW model, in order to solve the problems presented in the background art.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: a refined downscaling method for temperature in complex terrain based on the TopoMW model, comprising the following steps: Acquire multi-source data for the target region, and then fuse the multi-source data into a feature set after preprocessing. Construct a complex terrain weighted matrix based on preprocessed multi-source data; An improved TopoMW model based on UNet was established, and the TopoMW model was trained using a weighted MSE loss function based on the feature set and multiple weighting matrices for complex terrain to obtain the optimal downscaling model. The feature set to be predicted is input into the optimal downscaling model to obtain the initial prediction results; The final result is obtained by correcting the initial prediction results using multiple weighted matrices based on complex terrain.

[0006] Preferably, the multi-source data includes low-resolution ERA5 meteorological data, station observation data, and digital elevation model (DEM) data; wherein, the ERA5 meteorological data includes at least precipitation and temperature factors, and the station observation data includes the station's latitude and longitude, precipitation, and temperature measurements for the corresponding time period.

[0007] Preferably, the preprocessing includes: ERA5 meteorological data unit conversion (temperature from Kelvin K to Celsius °C, precipitation from meters m to millimeters mm), and bilinear interpolation was used to resample the data to 30-meter resolution; Outliers in the station observation data were removed based on the 3σ criterion, missing data were filled by linear interpolation, and the station coordinates were spatially matched with a 30-meter grid. Topographic feature factors are extracted from DEM data, including at least elevation, slope and aspect, and all topographic feature factors are resampled to a target 30-meter resolution grid.

[0008] Preferably, the feature set consists of 5 channels, including ERA5 precipitation data, ERA5 temperature data, DEM elevation, DEM slope, and DEM aspect.

[0009] Preferably, the complex terrain multiple weighting matrix design employs a three-layer weighting strategy, including: The first layer of terrain complexity is weighted: the terrain is divided into four levels according to the slope and the macro-elevation undulation: extremely complex area, highly complex area, moderately complex area and flat area. Different levels are assigned different basic weights, with the extremely complex area having the highest weight and the flat area having the lowest weight. The second layer of multi-scale terrain heterogeneity weighting: The standard deviation of terrain slope is calculated using 3×3 small windows and 9×9 medium windows respectively to quantify micro and macro terrain heterogeneity. The heterogeneity index is standardized and used as the enhancement weight. The third layer IDW interpolation diffusion weighting: the basic weights and enhanced weights corresponding to the site are diffused to the entire target grid through three interpolations to form the final complex terrain multi-weighted matrix.

[0010] Preferably, the classification criteria for the hierarchical weighted terrain complexity are as follows: slope ≥ 30° and elevation fluctuation ≥ 500m are classified as extremely complex areas, with a basic weight of 0.9-1.0; slope 15°-30° and elevation fluctuation 200-500m are classified as highly complex areas, with a basic weight of 0.7-0.8; slope 5°-15° and elevation fluctuation 50-200m are classified as moderately complex areas, with a basic weight of 0.4-0.6; and slope < 5° and elevation fluctuation < 50m are classified as gently sloping areas, with a basic weight of 0.1-0.3. The standardized range of the multi-scale topographic heterogeneity weighting is [0, 0.3], and the total weight range after superimposing with the basic weight is [0.1, 1.3].

[0011] Preferably, the weighted MSE loss function is: ; in, The weight of the i-th pixel in the multiple weighting matrix for complex terrain is... The label values ​​are the spatially interpolated values ​​of the site observation data; These are the predicted values ​​for the corresponding elements in the TopoMW model.

[0012] Preferably, the correction includes: A preset threshold is used to filter and retain valid pixels in complex areas of the initial prediction results; Based on the filtered valid pixels, and using the weights of the multiple weighting matrix for complex terrain as sample weights, linear regression relationships are established between the original model output values ​​and the spatial interpolation labels of station observation data for two types of elements: temperature and precipitation. The regression coefficients for each are calculated using the following formulas: ;in: This is the calibrated value. These are the original output values ​​of the model; Differential calibration of the target region is performed based on regression coefficients.

[0013] Preferably, the effective pixel selection thresholds include: true temperature values ​​∈ [1.0℃, 40℃], true precipitation values ​​∈ [0mm, 500mm], and a total weight ≥ 0.6 in the multiple weighted matrix for complex terrain.

[0014] Beneficial Effects: This invention fuses multi-source data into a feature set after preprocessing and constructs a multi-weighted matrix for complex terrain, ensuring consistency with the multi-weighted matrix data source. A weighted MSE loss function is designed, and the parameter optimization process is embedded with the weight information of the weighted matrix, allowing the multi-weighted strategy to directly guide the model's learning direction. Furthermore, this function assigns higher training weights to high-weight regions (extremely complex and highly complex regions), forcing the model to focus on learning the distribution patterns of meteorological elements in complex terrain areas. This solves the problem of insufficient accuracy in complex terrain areas caused by the traditional UNet's "equal weighting." The multi-weighted matrix for complex terrain is used for correction during output, and differential calibration improves the accuracy in complex terrain areas. Ultimately, the innovative value of the multi-weighted mechanism is realized in the downscaling results; achieving accurate generation of 30-meter high-resolution air temperature under complex terrain conditions.

[0015] In this invention, the correction process involves selecting effective pixels in complex areas, constructing a weighted linear mapping relationship, and performing differentiated full-area calibration, forming a dual-weighted optimization closed loop of training weighting and prediction calibration. This eliminates extreme prediction bias caused by statistical mismatch and improves the accuracy consistency in complex terrain areas. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method for fine-grained downscaling of air temperature in complex terrain based on the TopoMW model of the present invention; Figure 2 The image shown is before the 30m result image of the surface temperature downscaling in this embodiment of the invention. Figure 3 This is a 30m result diagram of the surface temperature downscaling in an embodiment of the present invention. Figure 4 This is the image before processing the 30m result map of precipitation downscaling in the implementation of this invention; Figure 5 This is a graph showing the results of downscaling precipitation data with a resolution of 30m in the implementation of this invention.

[0017] Figure 6 This is a DEM topographic image of the study area in this invention. Detailed Implementation

[0018] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0019] Example, reference Figure 1 A refined downscaling method for temperature in complex terrain based on the TopoMW model includes the following steps: Acquire multi-source data for the target area, including low-resolution ERA5 meteorological data, station observation data, and digital elevation model (DEM) data. The ERA5 meteorological data must include at least precipitation and temperature factors, while the station observation data must include the station's latitude and longitude, precipitation, and measured temperature values ​​for the corresponding time period. Preprocessing of multi-source data includes: ERA5 meteorological data unit conversion (temperature from Kelvin K to Celsius °C, precipitation from meters m to millimeters mm), and bilinear interpolation was used to resample the data to 30-meter resolution; Outliers (such as abnormal records with temperatures >40℃ or <-10℃) in the station observation data were removed based on the 3σ criterion. Missing data were filled in using linear interpolation, and the station coordinates were spatially matched with a 30-meter grid. Topographic feature factors are extracted from DEM data. The topographic feature factors include at least elevation, slope and aspect. All topographic feature factors are then resampled to a target 30-meter resolution grid. The preprocessed ERA5 data and terrain feature factors were unified into the WGS84 coordinate system to ensure accurate spatial location matching, and finally a 5-channel input feature set of "ERA5 precipitation + ERA5 temperature + DEM elevation + slope + aspect" was constructed. Feature engineering construction (construction of multiple weighted matrices for complex terrain): Based on the preprocessed multi-source data, feature engineering and the construction of the core complex terrain weighted matrix were completed: Feature standardization: Normalize the 5-channel input feature set (standardization formula: Eliminate the impact of differences in the scale of different factors on model training; A three-layer weighting strategy was designed, combining station data and terrain feature factors to construct a 30-meter resolution weighted matrix for the entire region. The specific process is as follows: The first layer of terrain complexity is weighted: the terrain is divided into four levels according to the slope and the macro-elevation undulation: extremely complex area, highly complex area, moderately complex area and flat area. Different levels are assigned different basic weights, with the extremely complex area having the highest weight and the flat area having the lowest weight. In one specific embodiment, the classification criteria for the hierarchical weighted classification of terrain complexity are as follows: slope ≥ 30° and elevation fluctuation ≥ 500m are classified as extremely complex areas, with a basic weight of 0.9-1.0; slope 15°-30° and elevation fluctuation 200-500m are classified as highly complex areas, with a basic weight of 0.7-0.8; slope 5°-15° and elevation fluctuation 50-200m are classified as moderately complex areas, with a basic weight of 0.4-0.6; and slope < 5° and elevation fluctuation < 50m are classified as gentle areas, with a basic weight of 0.1-0.3. The second layer of multi-scale terrain heterogeneity weighting: The standard deviation of terrain slope is calculated using 3×3 small windows and 9×9 medium windows respectively to quantify micro and macro terrain heterogeneity. The heterogeneity index is standardized and used as the enhancement weight. In one specific embodiment, the heterogeneity index is standardized to the range of [0, 0.3] and used as the enhanced weight. After being superimposed with the basic weight, the total weight range is [0.1, 1.3] (temporary weight = basic weight + enhanced weight). The third layer IDW interpolation diffusion weighting: the basic weights and enhanced weights corresponding to the site are diffused to the entire target grid through three interpolations to form the final complex terrain multi-weighted matrix.

[0020] The TopoMW model, based on an improved version of UNet, is established. The TopoMW model architecture design includes: The encoder module: Based on the original input layer of UNet, a 5-channel multi-source feature input interface is designed to accurately match the input feature set of ERA5 precipitation + ERA5 temperature + DEM elevation + slope + aspect, realizing the deep binding between model input and multi-layer weighting mechanism; Based on the original 3-layer convolution-pooling module of UNet, a dual-scale feature extraction branch of 3×3 small window + 9×9 medium window is added to accurately capture the spatial features corresponding to micro and macro terrain heterogeneity. The heterogeneity difference reflected by the standard deviation of terrain slope is quantified by convolution operation of different window sizes, providing feature support for multi-scale terrain heterogeneity weighting; Bottleneck layer: Set up a 256-channel convolutional unit layer to integrate the deep features output by the encoder and improve the feature representation capability; Decoder module: Contains 3 layers of transposed convolution-feature concatenation units. The transposed convolution kernel size is 3×3. It fuses shallow features (including non-linear correlation features of terrain-meteorological elements) of the corresponding layer of the encoder through skip connections to achieve high-resolution feature reconstruction. In the TopoMW model, the model is trained using a weighted MSE loss function based on a feature set combined with a complex terrain weighting matrix, to obtain the optimal downscaling model. This function forces the model to focus on learning the distribution patterns of meteorological elements in complex terrain areas by assigning higher training weights to high-weight areas (extremely complex areas and highly complex areas), thus solving the problem of insufficient accuracy in complex terrain areas caused by the traditional UNet's "equal weighting".

[0021] Training parameter settings: The multi-source data was divided into a training set (80%) and a validation set (20%), with the 30-meter grid data after spatial interpolation of the station observation data used as the label; The weighted MSE loss function is used, and the weighted MSE loss function is as follows: ; in, The weight of the i-th pixel in the multiple weighting matrix for complex terrain is... The label values ​​are the spatially interpolated values ​​of the site observation data; These are the predicted values ​​for the corresponding elements in the TopoMW model; The Adam optimizer was used, with an initial learning rate of 1e-4, a batch size of 8, and 50 training epochs. When the validation set loss did not decrease for 5 consecutive epochs, the learning rate was decayed (decay coefficient 0.5) to prevent overfitting, and the optimal downscaling model was finally obtained. Output layer: A 1×1 convolutional kernel is used to map the decoder output features into 2-channel results (temperature and precipitation), and finally outputs a 30-meter resolution prediction result; the feature set to be predicted is input into the optimal downscaling model to obtain the initial prediction result; The final result is obtained by correcting the initial prediction results using a weighted matrix based on complex terrain; the corrections include: Effective pixel selection and weighted adaptation: Effective pixels with "true temperature values ​​∈ [1.0℃, 40℃] and true precipitation values ​​∈ [0mm, 500mm]" were collected from the validation set, and a total of 11.547 million effective pixels were selected; pixels in terrain-complex areas (extremely complex areas and highly complex areas) with a weight of ≥0.6 in the weighted matrix were preferentially retained, and the proportion of complex terrain areas in the calibration samples was not less than 60%, ensuring that the calibration process focused on covering the areas sensitive to model prediction errors; Construction of weighted linear mapping relationship: Based on the filtered effective pixels, and using the weights of the multiple weighting matrix of complex terrain as the sample weights, linear regression calibration models for two elements, temperature and precipitation, are established respectively. The calculation formula is as follows: ,in This is the calibrated value. The original output values ​​of the model are denoted as , and 'a' (slope) and 'b' (intercept) are the regression coefficients. Application of weighted calibration across the entire region: The regression coefficients mentioned above are applied to the preliminary prediction results of temperature and precipitation in the entire target region. The calibration effect is enhanced in the high-weighted regions (complex terrain areas) of the weighted matrix, while the original prediction trend is appropriately preserved in the low-weighted regions (flat areas), thus eliminating the problem of extreme prediction values ​​caused by statistical bias.

[0022] This application also provides a specific embodiment that comprehensively analyzes the effectiveness and superiority of the method provided above; the following settings are made: The target area is a region (including complex terrain such as mountains, plains, and towns). We acquire multi-source basic data within the study area and complete systematic preprocessing to provide high-quality input for subsequent model training and downscaling prediction.

[0023] The data types and their sources are as follows: 1. ERA5 meteorological reanalysis data: Acquired low-resolution (11km) data for the entire year of 2023, including two core meteorological factors: temperature and precipitation, such as... Figure 2 (Raw surface temperature image from ERA5 at 11km resolution) Figure 4 (ERA5 raw 11km resolution precipitation image). 2. Station observation data: Concurrent measured data from 6 meteorological stations in the study area were collected, including station latitude and longitude, daily temperature (°C), and precipitation (mm). 3. DEM Topographic Data: Obtain 30-meter resolution digital elevation model data for extracting topographic feature factors.

[0024] After data preprocessing, a 5-channel input feature set is finally constructed, consisting of "ERA5 precipitation + ERA5 temperature + DEM elevation + slope + aspect". Figure 6 (The DEM topographic image of the study area in this invention is shown); The temporary weights corresponding to each station are diffused to the entire study area using cubic inverse distance weighting (IDW) interpolation to form the final complex terrain multi-weighted matrix; The data from January to October 2023 was used as the training set (80%), and the data from November to December was used as the validation set (20%). The TopoMW model constructed in this invention was trained using 30-meter grid data after spatial interpolation of the station observation data as labels to obtain the optimal downscaling model. Preliminary downscaling prediction: The preprocessed 5-channel features from November to December 2023 were input into the optimal downscaling model, and the preliminary prediction results of temperature and precipitation in the study area at a resolution of 30 meters were output. The preliminary prediction results were corrected using the method of this invention. The calculated calibration coefficients for temperature were (a=0.98) and (b=1.25), and the calibration coefficients for precipitation were (a=1.02) and (b=0.83). After weighted calibration, the final results of the spatial distribution of air temperature and precipitation in the study area, with a high resolution and high precision of 30 meters, are output, as follows: Figure 3 and Figure 5 (A schematic diagram of meteorological element results at 30-meter resolution after downscaling) is shown; Using validation set data and on-site measured data, a comprehensive accuracy evaluation of the final downscaling results is conducted to verify the applicability of the method. Evaluation data: 11.547 million effective pixels of validation data from November-December 2023 were used, combined with measured data from 6 meteorological stations for double validation; Evaluation indicators: Three core indicators were used: coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE), calculated as follows: 1. Coefficient of determination (R²): ,in The mean of the true values; 2. Root Mean Square Error (RMSE): , where n is the number of valid pixels; 3. Mean Absolute Error (MAE): .

[0025] Evaluation results: Downscaling results: R²=0.7992, RMSE=3.19℃, MAE=2.53℃; Downscaling results for precipitation: R²=0.68, RMSE=18.2mm, MAE=12.5mm; The results show that the present invention can accurately capture the spatial heterogeneity of meteorological elements under complex terrain, especially in terms of temperature downscaling, with a significantly higher accuracy than traditional statistical downscaling methods (the traditional TsHARP algorithm has a temperature RMSE of about 4.8℃), which fully meets the application needs of scenarios such as precision agricultural management and urban heat island analysis.

[0026] The embodiments of the present invention have been described in detail above with reference to the examples. However, the present invention is not limited to the above embodiments. For those skilled in the art, after learning the contents described in the present invention, several equivalent changes and substitutions can be made without departing from the principle of the present invention. These equivalent changes and substitutions should also be considered to fall within the protection scope of the present invention.

Claims

1. A refined downscaling method for temperature in complex terrain based on the TopoMW model, characterized by: Includes the following steps: Acquire multi-source data for the target region, and then fuse the multi-source data into a feature set after preprocessing. Construct a complex terrain weighted matrix based on preprocessed multi-source data; An improved TopoMW model based on UNet was established, and the TopoMW model was trained using a weighted MSE loss function based on the feature set and multiple weighting matrices for complex terrain to obtain the optimal downscaling model. The feature set to be predicted is input into the optimal downscaling model to obtain the initial prediction results; The final result is obtained by correcting the initial prediction results using multiple weighted matrices based on complex terrain.

2. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 1, characterized in that: The multi-source data includes low-resolution ERA5 meteorological data, station observation data, and DEM data; among which, the ERA5 meteorological data includes at least precipitation and temperature factors, and the station observation data includes the station's latitude and longitude, precipitation, and temperature measurements for the corresponding time period.

3. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 2, characterized in that: The preprocessing includes: ERA5 meteorological data unit conversion, and bilinear interpolation method was used to resample the data to 30-meter resolution; Outliers in the station observation data were removed based on the 3σ criterion, missing data were filled by linear interpolation, and the station coordinates were spatially matched with a 30-meter grid. Topographic feature factors are extracted from DEM data, including at least elevation, slope and aspect, and all topographic feature factors are resampled to a target 30-meter resolution grid.

4. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 3, characterized in that: The feature set consists of 5 channels, including ERA5 precipitation data, ERA5 temperature data, DEM elevation, DEM slope, and DEM aspect.

5. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 4, characterized in that: The complex terrain multiple weighted matrix design employs a three-layer weighting strategy, including: The first layer of terrain complexity is weighted: the terrain is divided into four levels according to the slope and the macro-elevation undulation: extremely complex area, highly complex area, moderately complex area and flat area. Different levels are assigned different basic weights, with the extremely complex area having the highest weight and the flat area having the lowest weight. The second layer of multi-scale terrain heterogeneity weighting: The standard deviation of terrain slope is calculated using 3×3 small windows and 9×9 medium windows respectively to quantify micro and macro terrain heterogeneity. The heterogeneity index is then standardized and used as the enhancement weight. The third layer IDW interpolation diffusion weighting: the basic weights and enhanced weights corresponding to the site are diffused to the entire target grid through three interpolations to form the final complex terrain multi-weighted matrix.

6. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 5, characterized in that: The classification criteria for the weighted stratification of terrain complexity are as follows: Slope ≥ 30° and elevation fluctuation ≥ 500m are classified as extremely complex areas, assigned a base weight of 0.9-1.0; slope 15°-30° and elevation fluctuation 200-500m are classified as highly complex areas, assigned a base weight of 0.7-0.8; slope 5°-15° and elevation fluctuation 50-200m are classified as moderately complex areas, assigned a base weight of 0.4-0.6; slope < 5° and elevation fluctuation < 50m are classified as gently sloping areas, assigned a base weight of 0.1-0.

3. The standardized range of the multi-scale topographic heterogeneity weighting is [0, 0.3], and the total weight range after superimposing with the basic weight is [0.1, 1.3].

7. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 1, characterized in that: The weighted MSE loss function is: ； in, The weight of the i-th pixel in the multiple weighting matrix for complex terrain is... The label values ​​are the spatially interpolated values ​​of the site observation data; These are the predicted values ​​for the corresponding elements in the TopoMW model.

8. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 5, characterized in that: The corrections include: A preset threshold is used to filter and retain valid pixels in complex areas of the initial prediction results; Based on the filtered valid pixels, and using the weights of the multiple weighting matrix for complex terrain as sample weights, linear regression relationships are established between the original model output values ​​and the spatial interpolation labels of station observation data for two types of elements: temperature and precipitation. The regression coefficients for each are calculated using the following formulas: ;in: This is the calibrated value. These are the original output values ​​of the model, where a is the slope and b is the intercept. Differential calibration of the target region is performed based on regression coefficients.

9. The method for refined downscaling of temperature in complex terrain based on the TopoMW model according to claim 8, characterized in that: The effective pixel selection thresholds include: true temperature values ​​∈ [1.0℃, 40℃], true precipitation values ​​∈ [0mm, 500mm], and a total weight ≥ 0.6 in the multiple weighted matrix for complex terrain.

Images