A method for estimating the spatio-temporal distribution of the lapse rate of mountain air temperature based on the FY4A AGRI land surface temperature

Abstract

The present invention discloses a method for estimating the spatio-temporal distribution of the lapse rate of mountain air temperature based on FY4A AGRI land surface temperature, including: obtaining the FY4A AGRI thermal infrared remote sensing land surface temperature and its influencing factor data and preprocessing them to obtain data at a resolution of 4 km; using the random forest regression algorithm to perform spatial dimension reconstruction on the missing land surface temperature data, and using the S-G filter to smooth the spatial dimension reconstruction result in the time dimension to obtain the spatio-temporal dimension reconstruction result of the land surface temperature; using the random forest regression algorithm to construct a land surface temperature downscaling model to reduce the land surface temperature from 4 km to 1 km resolution; separating the influence of altitude on the land surface temperature and calculating the standardized land surface temperature; using the sliding window method to establish a linear regression relationship between the standardized land surface temperature and altitude, calculating the regression coefficient, and further obtaining the lapse rate of air temperature. It provides necessary data and technical support for related research fields such as weather forecasting and climate model applications.

Description

Technical Field

[0001] The present invention relates to a method for estimating the spatio-temporal distribution of the lapse rate of mountain air temperature based on FY4A AGRI land surface temperature, belonging to the technical field of lapse rate of air temperature estimation. Background Art

[0002] The temperature lapse rate (TLR) refers to the rate at which the atmospheric temperature changes with altitude, also known as the vertical temperature gradient. In meteorology, TLR can be used to evaluate the stability of the atmosphere, plays an important role in the generation, development and change of atmospheric motion, plays a key role in the estimation of air temperature spatial distribution and the correction of air temperature forecasting, and is also one of the important indicators for studying climate change and validating climate models.

[0003] The temperature lapse rate is closely related to many factors such as the terrain undulation, surface type and coverage, and local solar incidence angle, and varies violently with time and space. Using a constant value cannot accurately express the spatio-temporal variation characteristics of the temperature lapse rate in complex regions such as mountainous areas, thus affecting related fields such as the accurate correction of mountain air temperature forecasting. Therefore, how to obtain a spatio-temporally continuous near-surface temperature lapse rate is of great significance.

[0004] Calculating the temperature lapse rate using meteorological station observation data is a most commonly used method. This method uses the temperature observation values and altitudes of multiple stations at different altitudes for regression fitting to estimate the temperature lapse rate. However, calculating the temperature lapse rate using station observation data is affected by many factors, including the distribution density of observation stations, the geographical area covered by the stations, the spatial representativeness of the stations, etc., and usually only one temperature lapse rate data can be calculated in a region. Therefore, this method is not suitable for obtaining the spatial distribution of the temperature lapse rate in mountainous areas with complex terrain and underlying surface conditions.

[0005] Remote sensing technology has the advantage of spatially continuous monitoring. Due to the high correlation between land surface temperature and air temperature, establishing a regression relationship between the instantaneous land surface temperature obtained by thermal infrared remote sensing inversion and the digital elevation model DEM can indirectly express the TLR characteristics. However, the commonly used thermal infrared remote sensing data is limited by factors such as satellite revisit cycle and weather conditions, and cannot provide land surface temperature data with high spatial resolution and strong time continuity, which affects the estimation of the temperature lapse rate with high spatio-temporal resolution.

[0006] Through comprehensive analysis of the research status at home and abroad, it can be found that the related research on using remote sensing technology for estimating the lapse rate of air temperature is still in its infancy. The lapse rate of air temperature estimated in most studies is the average value within a region and over a period of time. Individual studies only estimate the spatial distribution of TLR using clear-sky remote sensing images, and the research on estimating the continuous spatio-temporal distribution of TLR needs to be further carried out in depth. So far, there is no research on estimating the lapse rate of air temperature with high spatio-temporal resolution all day and all weather at home and abroad. Therefore, how to utilize the characteristics and advantages of remote sensing technology to achieve continuous spatio-temporal estimation of the lapse rate of air temperature is in urgent need to meet the meteorological forecasting and series of applications in mountainous areas with large topographic undulations. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for estimating the spatio-temporal distribution of the lapse rate of air temperature in mountainous areas based on the FY4A AGRI land surface temperature, which can obtain the lapse rate data with hourly and 1-km spatial resolution, and provide necessary data and technical support for related fields such as weather forecasting and climate model applications.

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] A method for estimating the spatio-temporal distribution of the lapse rate of air temperature in mountainous areas based on the FY4A AGRI land surface temperature includes the following steps:

[0010] Step 1: Obtain the FY4A AGRI thermal infrared remote sensing land surface temperature data of the mountainous area to be estimated, as well as the data of the influencing factors of the land surface temperature, and preprocess all the data to obtain the data with a spatial resolution of 4 km.

[0011] Step 2: Based on all the data with a spatial resolution of 4 km, use the random forest regression algorithm to reconstruct the missing data of the land surface temperature affected by cloud cover in the spatial dimension to obtain the spatial dimension reconstruction result of the land surface temperature, and use the S-G filter to smooth the spatial dimension reconstruction result in the time dimension to obtain the spatio-temporal dimension reconstruction result of the land surface temperature.

[0012] Step 3: On the basis of Step 2, use the random forest regression algorithm to construct a land surface temperature downscaling model to reduce the land surface temperature from a spatial resolution of 4 km to a spatial resolution of 1 km, and obtain the downscaled land surface temperature with a spatial resolution of 1 km.

[0013] Step 4: Based on the downscaled land surface temperature with a spatial resolution of 1 km, use the random forest regression algorithm to separate the influence of altitude on the land surface temperature and calculate the standardized land surface temperature; adopt the sliding window method to establish a linear regression relationship between the standardized land surface temperature and altitude, calculate the regression coefficient, and then obtain the lapse rate of air temperature.

[0014] Compared with the prior art by adopting the above technical solutions, the present invention has the following technical effects:

[0015] In view of the current situation that there is a lack of all-day and all-weather high spatio-temporal resolution air temperature lapse rate estimation methods at home and abroad, the present invention innovatively proposes a spatio-temporal distribution estimation method of mountain air temperature lapse rate based on FY4A AGRI land surface temperature by constructing a hierarchical progressive technical system of "reconstruction of missing land surface temperature data, downscaling of land surface temperature, and estimation of spatial distribution of air temperature lapse rate". Aiming at generating hourly air temperature lapse rate data with a spatial resolution of 1 km, the reanalysis data of surface cumulative net radiation flux is first added to the influencing factors to realize the reconstruction of missing FY4A AGRI hourly land surface temperature data. Then, based on the machine learning method, the 4-km resolution FY4A AGRI land surface temperature is downscaled to 1 km. Finally, the sliding window method is used to carry out the research on the estimation of the spatial distribution of the air temperature lapse rate. The present invention can effectively obtain the spatio-temporal distribution of the air temperature lapse rate by remote sensing means, and can provide technical support for related research such as air temperature forecast correction and climate model application. Description of the Drawings

[0016] Figure 1 is a flowchart of the spatio-temporal distribution estimation method of mountain air temperature lapse rate based on FY4A AGRI land surface temperature of the present invention;

[0017] Figure 2 are the FY4A land surface temperature reconstruction results (4-km resolution) at two moments on February 20th in Chongli District of the embodiment of the present invention, where (a) is 10:00 and (b) is 22:00;

[0018] Figure 3 are the land surface temperature downscaling results (1-km resolution) at two moments on February 20th in Chongli District of the embodiment of the present invention, where (a) is 10:00 and (b) is 22:00;

[0019] Figure 4 are the air temperature lapse rate estimation results at two moments on February 20th in Chongli District of the embodiment of the present invention, where (a) is 10:00 and (b) is 22:00. Detailed Embodiments

[0020] The following details the embodiments of the present invention, and the examples of the embodiments are shown in the drawings. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0021] The FY4A satellite is the second-generation geostationary meteorological satellite series in China. The multi-channel scanning radiometer imager (AGRI) it carries has 14 channels, can generate a full-disk observation image every 15 minutes, and has the characteristics of high temporal resolution and high precision. The surface temperature retrieved using the AGRI sensor has a high temporal resolution. After performing spatial downscaling on it, it has the basis for estimating the lapse rate of air temperature with high temporal and high spatial resolution.

[0022] Aiming at the problem of estimating the spatio-temporal distribution of the all-weather and all-day lapse rate of air temperature TLR, the present invention proposes a method for estimating the spatio-temporal distribution of the lapse rate of air temperature in mountainous areas based on the FY4A AGRI surface temperature. Based on the reconstruction of the missing FY4A AGRI surface temperature data, this method downscales it to 1 km based on a machine learning method, and then uses the sliding window method to carry out research on the estimation of the lapse rate of air temperature, obtaining the lapse rate of air temperature data with an hourly and 1-km spatial resolution. As Figure 1 shown, the specific scheme is as follows:

[0023] 1) Obtain the FY4A AGRI thermal infrared remote sensing surface temperature data and the relevant data of the surface temperature influencing factors, and perform necessary data preprocessing.

[0024] First, for the obtained FY4A AGRI surface temperature data and the surface temperature influencing factor data, including the fifth-generation global climate atmospheric reanalysis (ERA5) data of the European Centre for Medium-Range Weather Forecasts (ECMWF), the remote sensing data of the Moderate Resolution Imaging Spectroradiometer (MODIS), and the digital elevation model (DEM) data of the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), etc., perform format conversion to ensure the unity of the data format; secondly, perform geometric correction on the FY4A AGRI data, and use the vector file of the study area to crop the surface temperature data and its influencing factor data, and then perform spatial registration on the cropped data to ensure the consistency of the spatial range;

[0025] For the cropped data, resample it to 4 km using the aggregation averaging method to ensure the consistency of the spatial resolution of the data.

[0026] 2) Use the FY4A AGRI surface temperature remote sensing data. First, perform surface temperature spatial dimension reconstruction based on the random forest algorithm. Secondly, for the preliminary reconstruction result of the surface temperature spatial dimension, use the S-G filter to perform pixel-by-pixel temporal dimension smoothing to achieve the spatio-temporal dimension reconstruction of the surface temperature.

[0027] The core idea of land surface temperature (LST) reconstruction is to explore the relationship between LST and predictive variables. Since the relationship between LST and its influencing factors is complex, it is difficult to construct a suitable numerical model to reflect this relationship. Therefore, the random forest regression model in machine learning is used to describe the non-linear relationship between LST and predictive variables, and an LST spatial dimension reconstruction model is constructed. At a spatial resolution of 4 km, the land surface temperature of FY4A clear-sky pixels is used as the dependent variable, and the surface cumulative net radiation flux, altitude, slope, NDVI, and NDWI at the corresponding pixel positions are used as independent variables. The random forest regression algorithm is used to train the reconstruction model to establish a non-linear regression relationship between the land surface temperature and the independent variables. The regression model can be expressed by formula (1):

[0028] (1)

[0029] In the formula, represents the land surface temperature; represents the surface cumulative net radiation flux; represents the normalized difference vegetation index; represents the normalized difference water index; represents the surface elevation; represents the surface slope; represents the random forest regression model.

[0030] After the LST spatial dimension reconstruction model is trained, the surface cumulative net radiation flux, altitude, slope, NDVI, and NDWI values corresponding to the land surface temperature missing pixels under cloud cover are used as independent variables and input into the model to calculate the corresponding land surface temperature under the cloud to fill in the missing values.

[0031] LST reconstruction based on the random forest algorithm is a spatial dimension LST reconstruction method. Although it can fill in the missing values, this method only considers the spatial distribution characteristics of LST and does not consider the temporal variation characteristics of LST, making it difficult to ensure the continuity and variation characteristics of the reconstructed LST in the time dimension. To ensure that the reconstructed result is the true land surface temperature under the cloud, it is necessary to smooth the preliminary LST reconstruction result in the time dimension to process the outliers in the spatial dimension reconstruction. The Savitzky-Golay (S-G) filter is a filtering method based on local polynomial least squares fitting in the time domain. Statistically analyzing the land surface temperature reconstruction result in the time dimension and using S-G filtering for smoothing can filter out noise while keeping the shape of the land surface temperature time series unchanged. The S-G filtering process can be expressed by formula (2):

[0032] (2)

[0033] In the formula, is the reconstructed dataset of smoothed land surface temperature; is the reconstructed land surface temperature data; is the filtering coefficient; is the number of data in the sliding window ( ), is the window width.

[0034] 3) Using the reconstructed FY4A AGRI seamless land surface temperature data, a relationship model between land surface temperature and its influencing factors is constructed based on the random forest algorithm to downscale the land surface temperature data.

[0035] When establishing the land surface temperature (LST) downscaling model, the complex relationships among land surface temperature and various influencing factors such as solar radiation (SR), normalized difference vegetation index (NDVI), and albedo need to be considered. The main steps for establishing the downscaling model are as follows:

[0036] Use the data with a resolution of 4 km as the training set, define the loss function (such as root mean square error RMSE) and select an appropriate optimization algorithm to train the land surface temperature - influencing factor relationship model. The model form is as follows:

[0037] (3)

[0038] where, is the land surface temperature at a resolution of 4 km, are the land surface and terrain influencing factors at a resolution of 4 km.

[0039] Input the influencing factor data with a resolution of 1 km into the trained model, and perform residual correction to downscale the land surface temperature at a resolution of 1 km.

[0040] The land surface temperature downscaling formula can be expressed as:

[0041] (4)

[0042] where, is the downscaled land surface temperature at a resolution of 1 km, are the influencing factors at a resolution of 1 km, is the mapping function obtained from the corresponding data with a resolution of 4 km; is the fitting residual of formula (3), resampled to a spatial resolution of 1 km by bilinear interpolation method.

[0043] 4) Using the downscaled FY4A AGRI seamless land surface temperature data, first use the random forest model to separate the influence of the altitude factor on the land surface temperature, obtain the land surface temperature excluding the influence of the altitude-based land surface characteristics, and then calculate the normalized land surface temperature NLST; secondly, adopt the sliding window method, and use windows of different sizes (such as etc.) to establish a linear regression relationship between the normalized land surface temperature and the altitude, calculate the regression coefficient, and thus obtain the lapse rate result; finally, select the window with the best effect to obtain the lapse rate with the best accuracy.

[0044] First, at a spatial resolution of 1 km, establish a relationship model between AGRI LST and land surface characteristic parameters:

[0045] (5)

[0046] Among them, is the land surface temperature at a resolution of 1 km after downscaling, is a non-linear relationship model, (surface slope), (aspect), (topographic position index), (normalized difference vegetation index), (local solar incidence angle) and (altitude) are land surface characteristics affecting LST at a spatial resolution of 1 km.

[0047] Secondly, separate the influence of altitude on the land surface temperature to obtain the normalized land surface temperature NLST:

[0048] (6)

[0049] (7)

[0050] Among them, is the LST model value based on characteristics other than altitude calculated by the random forest regression model, is the LST after separating the influence of altitude, is a non-linear fitting model.

[0051] Finally, use the sliding window method to establish a linear regression model between LST and elevation within sliding windows of different sizes (such as ). The slope of the regression model is used as the lapse rate of the central pixel within the window and compared with the TLR calculated from the measured data to determine the optimal sliding window size, and then calculate the spatial distribution of the lapse rate TLR in the study area. The formula is expressed as:

[0052] (8)

[0053] In Equation (8), is the normalized LST value of each 1-kilometer resolution pixel, and are regression coefficients, is the estimated TLR of the central pixel of the moving window.

[0054] Calculate the TLR using the site-observed air temperature data to verify the values calculated by different sliding windows, and select the sliding window with the highest accuracy, and slide to calculate the TLR value of each pixel position in the entire area. Embodiment

[0055] Based on the reconstruction of the missing FY4A AGRI land surface temperature data, this invention downscales it to 1 km based on machine learning methods, and then uses the sliding window method to carry out research on the estimation of the lapse rate of air temperature. Finally, hourly, 1-km spatial resolution lapse rate of air temperature data is obtained. This invention has been applied in Chongli District, Hebei Province, China, and the results show that the method proposed by this invention can obtain high-precision spatio-temporal distribution of the lapse rate of air temperature.

[0056] 1) Selection of the test area: Select Chongli District, Zhangjiakou City, Hebei Province as the test area, with a total area of about 2,334 square kilometers, between east longitude and , north latitude and , belonging to the semi-arid monsoon climate zone. Chongli District is located in the northwest of Hebei Province, at the transitional zone between the Inner Mongolia Plateau and the North China Plain, leaning on the Inner Mongolia grassland in the north and facing the central urban area of Zhangjiakou in the south. 80% of the territory of Chongli District is mountainous, and the forest coverage rate reaches 52.38%. The territory is the mountainous area in the northwest of Hebei, mostly in the northeast-southwest and east-west directions. The altitude range of Chongli District is 813 - 2,174 m, the maximum altitude difference is 1,361 m, and the average altitude is 1,485 m. Its geomorphic feature is "mountains are connected, continuous, and there are countless gullies within gullies". Due to the large terrain undulation in this area, it is an ideal area for verifying the lapse rate of air temperature estimation methods.

[0057] 2) Considering the principles of land surface temperature reconstruction, the relationship model between land surface temperature and downscaling influencing factors, and the influencing factors, temporal and spatial variation characteristics of the lapse rate of air temperature comprehensively, land surface temperature data, as well as solar radiation (SR), normalized difference vegetation index (NDVI), albedo, terrain, etc. are selected as independent variables affecting land surface temperature. The FY4A AGRI land surface temperature data is downloaded from the Fengyun satellite remote sensing data service network. The MODIS land surface temperature data and land surface reflectance data in the same time period are downloaded from the Google Earth Engine official website. The ASTER DEM version 3 data with a resolution of 30 meters is downloaded from the ASTER Global Digital Elevation Model website. The hourly air temperature observation data of 74 automatic weather stations in Chongli District are collected. With the help of ENVI 5.3 software and ArcMap 10.8 software, operations such as image resampling, image cropping, and spatial registration are carried out on the collected data to obtain various remote sensing images required for estimating the lapse rate of air temperature.

[0058] 3) Reconstruction of missing data of FY4A AGRI land surface temperature: At a spatial resolution of 4 km, taking the land surface temperature of clear-sky pixels in the study area as the dependent variable, and taking the cumulative net radiation flux, altitude, slope, NDVI, and NDWI values of the corresponding preprocessed pixels as independent variables, setting the number of decision trees in the hyperparameters of the random forest algorithm to 500, the minimum number of leaf trees to 10, and the training set division ratio to 80%, training to obtain the random forest land surface temperature reconstruction model (Equation (1)), and then inputting the various radiation factors, spectral factors, and terrain factor parameters corresponding to cloud-covered pixels to obtain the land surface temperature values of the corresponding cloud-covered pixels. Since the reconstruction result of the random forest algorithm only considers the influence of the spatial dimension and there may be outliers, the Savitzky-Golay (S-G) filter is used to process the outliers of the reconstruction result of cloud-covered pixels in the time dimension. First, the reconstruction results of each pixel are statistically analyzed according to the time series, and then each pixel is smoothed in the time dimension using Equation (2) to process the outliers of the land surface temperature in the reconstruction result. Figure 2 Figures (a) and (b) are the FY4A land surface temperature reconstruction results at 10:00 and 22:00 respectively.

[0059] 4) Downscaling of FY4A AGRI land surface temperature: At a spatial resolution of 4 km, taking the reconstructed land surface temperature in the study area as the dependent variable, and taking various influencing factors such as land surface temperature and solar radiation (SR), normalized difference vegetation index (NDVI), albedo, etc. as independent variables, setting the number of decision trees in the hyperparameters of the random forest algorithm to 500, the minimum number of leaf trees to 10, and the training set division ratio to 80%, training to obtain the random forest land surface temperature - influencing factor relationship model (Equation (3)), and then inputting the influencing factor data with a resolution of 1 km into the trained model and performing residual correction (Equation (4)) to downscale the land surface temperature at a resolution of 1 km. Figure 3(a) and (b) are the surface temperature downscaling results at 10:00 and 22:00, respectively.

[0060] 5) Estimation of the temporal and spatial distribution of the temperature lapse rate: First, the downscaled surface temperature at 1 km was used as the dependent variable, and Slope, Aspect, TPI (Topographic Position Index), NDVI (Normalized Difference Vegetation Index), (local solar incidence angle) and DEM (digital elevation model) as independent variables, the number of decision trees in the hyperparameters of the random forest algorithm was set to 500, the minimum leaf tree was set to 10, and the training set division ratio was set to 80%. The relationship model between AGRI LST and surface characteristics was established (Equation (5)). Secondly, the influence of influencing parameters other than DEM on surface temperature was calculated (Equation (6)) to obtain the standardized LST (Equation (7)). Finally, the sliding window method was used to calculate the influence of influencing parameters other than DEM on surface temperature within sliding windows of different sizes (such as ) A linear regression model between LST and elevation was established (Equation (8)). The slope of the regression model was used as the temperature lapse rate of the central pixel in the window. By comparing it with the TLR calculated from the measured data, the optimal sliding window size was determined, and then the spatial distribution of the temperature lapse rate TLR in the study area was calculated.

[0061] 6) Experimental results analysis: The estimated temperature lapse rate of Chongli District at all hours on February 20 was compared with the temperature lapse rate calculated from the measured temperature of representative ground meteorological stations, and the determination coefficient R was calculated. 2 , mean absolute error MAE and root mean square error RMSE. After experimental comparison, the estimated temperature lapse rate results are more accurate when the window size is 13×13. Taking 10:00 and 22:00 on February 20 as an example, Figure 4 The results of the estimated temperature lapse rate at two moments with a window size of 13×13 are given. The temperature lapse rate calculated using the station observed temperature is used as a reference. The root mean square error of the temperature lapse rate TLR at 10:00 and 22:00 decreased by 15.3% and 34.2% respectively. Figure 4 (a) At 10 a.m., the temperature lapse rate in Chongli District is generally high in the east and low in the west. The central and western parts are valley areas. At this time, the solar radiation is strong, and the radiation heating in the valley area is more obvious, resulting in a faster temperature rise. At this time, the ground is heated by solar radiation, and the heat begins to transfer to the atmosphere. However, since heat transfer takes time, the air temperature near the ground rises rapidly, and as the altitude increases, the air temperature drops, forming a lower temperature lapse rate. According to the results of the present invention, the temperature lapse rates in the central and western parts of Chongli District at this moment are distributed in arrive The eastern region is distributed between arrive between. According to Figure 4 in (b) of Figure 4 , at 22:00, the ground loses the heat of solar radiation and begins to release long-wave radiation to the atmosphere. The ground gradually cools through radiation, and the air temperature drops. Due to the weak radiation cooling effect in the valley area at night, the air temperature near the ground drops slowly, while with the increase in altitude, the air temperature drops faster. Therefore, the lapse rate of air temperature is relatively high. According to the results of the present invention, the lapse rate of air temperature in the western part of Chongli District at 22:00 is distributed between and between, and the eastern region is distributed between and between. It can be seen from this that the lapse rate of air temperature in most areas has a large gap with the fixed value, and even temperature inversion occurs in some local areas, so the fixed value of cannot be used to replace the lapse rate of air temperature in mountainous areas.

[0062] 7) Experimental conclusion: The results of the present invention fully illustrate that the method for estimating the spatio-temporal distribution of the lapse rate of air temperature in mountainous areas based on FY4A AGRI land surface temperature reconstruction, land surface temperature downscaling, and sliding window can effectively obtain the lapse rate of air temperature data with hourly and 1 km resolution with high accuracy, and can provide technical support for related research such as weather forecasting and climate model applications.

[0063] Based on the same inventive concept, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the foregoing method for estimating the spatio-temporal distribution of the lapse rate of air temperature in mountainous areas based on FY4A AGRI land surface temperature are implemented.

[0064] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the foregoing method for estimating the spatio-temporal distribution of the lapse rate of air temperature in mountainous areas based on FY4A AGRI land surface temperature are implemented.

[0065] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0067] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0069] The above embodiments are only used to illustrate the technical idea of the present invention, and the protection scope of the present invention cannot be limited thereby. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the present invention.

Claims

1. A method for estimating the spatiotemporal distribution of the temperature lapse rate in mountainous areas based on the FY4A AGRI surface temperature, characterized in that: The steps include: Step 1: Obtain the FY4A AGRI thermal infrared remote sensing surface temperature data of the mountain area to be estimated, as well as the surface temperature influencing factor data, and preprocess all the data to obtain data at a spatial resolution of 4 km; Step 2: Based on all data at a spatial resolution of 4 km, the random forest regression algorithm is used to reconstruct the missing data of surface temperature under the influence of cloud cover in spatial dimension, and the spatial dimension reconstruction result of the surface temperature is obtained. The SG filter is used to smooth the spatial dimension reconstruction result in time dimension, and the spatiotemporal dimension reconstruction result of the surface temperature is obtained. The specific process is as follows: Step 2.1, based on the surface temperature data and its influencing factor data at a spatial resolution of 4 km obtained by preprocessing in step 1, the surface temperature of the clear sky pixel is used as the dependent variable, and the surface cumulative net radiation flux, altitude, slope, normalized difference vegetation index and normalized difference water index corresponding to the clear sky pixel position are used as independent variables. A nonlinear regression model between the surface temperature and the independent variables is established and trained to obtain a trained surface temperature spatial dimension reconstruction model; Among them, the nonlinear regression model between surface temperature and independent variables is expressed as follows: LST=f(SNR,NDVI,NDWI,Z,Slope) Wherein, LST represents the land surface temperature, SNR represents the accumulated net radiation flux of the land surface, NDVI represents the normalized difference vegetation index, NDWI represents the normalized difference water index, Z represents the surface elevation, Slope represents the surface slope, and f represents the random forest regression model; Step 2.2, the surface cumulative net radiation flux, altitude, slope, normalized difference vegetation index and normalized difference water index corresponding to the missing pixels of surface temperature under the influence of cloud cover are input as independent variables into the trained surface temperature spatial dimension reconstruction model to obtain the surface temperature spatial dimension reconstruction result; Step 2.3: For the spatial dimension reconstruction result of the surface temperature, statistics are performed in the time dimension, and the SG filter is used for smoothing to obtain the spatial and temporal dimension reconstruction result of the surface temperature; the smoothing process of the SG filter is as follows: In the formula, The smoothed land surface temperature reconstruction data, LST i+j is the spatial reconstruction data of surface temperature, C i is the filter coefficient, N is the number of data in the sliding window, N=2m+1, m is the window width; Step 3: Based on step 2, a random forest regression algorithm is used to construct a surface temperature downscaling model to reduce the surface temperature from 4 km spatial resolution to 1 km spatial resolution, and obtain the downscaled 1 km spatial resolution surface temperature; Step 4: Based on the downscaled 1km spatial resolution surface temperature, the random forest regression algorithm is used to separate the effect of altitude on the surface temperature and calculate the standardized surface temperature. The sliding window method is used to establish a linear regression relationship between the standardized surface temperature and altitude, and the regression coefficient is calculated to obtain the temperature lapse rate. The specific process is as follows: Step 4.1, establish the relationship model between the downscaled 1km spatial resolution surface temperature and surface characteristic parameters: LST 1km =f(Slope 1km ,Aspect 1km ,TPI 1km ,NDVI 1km ,θ 1km ,Z 1lm ) Where, LST 1km represents the downscaled 1km spatial resolution surface temperature, Slope 1km Represents the surface slope at 1km spatial resolution, Aspect 1km Represents 1km spatial resolution direction, TPI 1km Represents 1km spatial resolution terrain position index, NDVI 1km Represents the normalized difference vegetation index at 1 km spatial resolution, θ 1km Represents the local solar incidence angle at 1 km spatial resolution, Z 1km Represents the altitude at 1km spatial resolution; Step 4.2, separate the effect of altitude on surface temperature, establish the relationship model between surface characteristic parameters other than altitude and surface temperature with 1 km spatial resolution, and obtain the standardized surface temperature: LST m =g(Slope 1km ,Aspect 1km ,TPI 1km ,NDVI 1km ,θ 1km ) NLST=LST 1km -LST m Where, LST m is the land surface temperature model value excluding the altitude parameter, NLST represents the standardized land surface temperature after separating the altitude effect, and g represents the nonlinear fitting model; Step 4.3, using the sliding window method, establish a linear regression model between the standardized surface temperature and altitude in sliding windows of different sizes: NLST=a×Z 1km +b In the formula, a and b are regression coefficients, and the value of a is the estimated value of the temperature lapse rate of the central pixel of the sliding window; Calculate the a value corresponding to sliding windows of different sizes, calculate the error between the estimated value of the temperature lapse rate and the actual observed value under sliding windows of different sizes, take the sliding window size corresponding to the minimum error, and perform sliding calculation to obtain the temperature lapse rate at each pixel position in the entire mountainous area.

2. The method for estimating the spatiotemporal distribution of the temperature lapse rate in mountainous areas based on the FY4A AGRI surface temperature according to claim 1 is characterized in that: The specific process of step 1 is as follows: Step 1.1, convert the obtained FY4A AGRI thermal infrared remote sensing surface temperature data of the mountainous area to be estimated, as well as the surface temperature influencing factor data, so that all data formats are unified; the surface temperature influencing factor data include the fifth generation atmospheric reanalysis data of the European Centre for Medium-Range Weather Forecasts, remote sensing data of the Moderate Resolution Imaging Spectrometer, and digital elevation model data of the Advanced Spaceborne Thermal Emission and Reflection Radiometer; Step 1.2, based on step 1.1, geometrically correct the land surface temperature data, clip the land surface temperature data and its influencing factor data, and then spatially register the clipped land surface temperature data and land surface temperature influencing factor data; Step 1.3: Based on step 1.2, the spatially registered surface temperature data and its influencing factor data are resampled to 4 km using the aggregation averaging method to obtain data at a spatial resolution of 4 km.

3. The method for estimating the spatiotemporal distribution of the temperature lapse rate in mountainous areas based on the FY4A AGRI surface temperature according to claim 1 is characterized in that: The specific process of step 3 is as follows: Step 3.1: Use the average aggregation method to resample the surface temperature influencing parameters at 1 km spatial resolution to 4 km resolution. The surface temperature influencing parameters include normalized difference vegetation index, surface albedo, solar radiation, altitude and latitude. Based on the relationship between the surface temperature and the influencing parameters at 4 km spatial resolution, use the random forest regression algorithm to construct and train the surface temperature downscaling model to obtain the trained surface temperature downscaling model. The surface temperature downscaling model is expressed as follows: LST 4km =f(NDVI 4km ,Albedo 4km ,SR 4km ,Z 4km ,Lat 4km ) Where, LST 4km Represents 4km spatial resolution surface temperature, NDVI 4km Represents the normalized difference vegetation index at 4km spatial resolution, Albedo 4km Represents the surface albedo at 4 km spatial resolution, SR 4km Represents 4km resolution solar radiation, Z 4km Represents 4km resolution altitude, Lat 4km represents the 4km spatial resolution latitude, and f represents the random forest regression model; Step 3.2: Calculate the fitting residual ε using the predicted value and actual value of the surface temperature model with a spatial resolution of 4 km 4km : ε 4km =LST′ 4km -LST 4km Where LST′ 4km Represents the actual value of AGRI land surface temperature at 4 km spatial resolution after reconstruction, LST 4km Represents the surface temperature with a spatial resolution of 4 km predicted by the surface temperature downscaling model; Step 3.3: Input the normalized difference vegetation index, surface albedo, solar radiation, altitude, and latitude at a spatial resolution of 1 km into the trained surface temperature downscaling model, and use the bilinear interpolation method to convert ε 4km Resample to 1 km spatial resolution and perform fitting residual correction to obtain the surface temperature at 1 km spatial resolution; the surface temperature at 1 km spatial resolution is expressed as follows: LST 1km =f(NDVI 1km ,Albedo 1km ,SR 1km ,WITH 1km ,Lat 1km )+ε 1km Where, LST 1km Represents the downscaled 1km spatial resolution surface temperature, NDVI 1km Represents the normalized difference vegetation index at 1km spatial resolution, Albedo 1km Represents the surface albedo with a spatial resolution of 1 km, SR 1km Represents solar radiation at 1km spatial resolution, Z 1km Represents the altitude at 1km spatial resolution, Lat 1km represents 1km spatial resolution latitude, ε 1km is the fitting residual of 1 km spatial resolution after resampling.

4. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the steps of the method for estimating the spatiotemporal distribution of the air temperature lapse rate in mountainous areas based on the FY4A AGRI surface temperature are implemented as described in any one of claims 1 to 3.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for estimating the spatiotemporal distribution of the temperature lapse rate in mountainous areas based on the FY4A AGRI land surface temperature are implemented as described in any one of claims 1 to 3.

Images