A method and system for reconstructing land surface temperature at a continuous hundred-meter scale in all-weather space and time

Through multi-scale data processing and machine learning algorithms, a step-by-step downscale model is constructed, which solves the problem of missing surface temperature data in cloud and rainy weather, and realizes high-precision all-weather and space-time continuous 100-meter scale surface temperature data acquisition, which improves data coverage rate and resolution.

CN118313239BActive Publication Date: 2024-11-05CHINA AERO GEOPHYSICAL SURVEY & REMOTE SENSING CENT FOR LAND & RESOURCES +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410302703.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-18
Publication Date
2024-11-05
Estimated Expiration
2044-03-18

AI Technical Summary

Technical Problem

The existing technology is difficult to obtain surface temperature data at a large range of 24/7 space-time continuous 100 meters, especially in cloudy and rainy weather conditions, and the existing filling methods require difficult to obtain additional data and poor accuracy.

Method used

The surface temperature reconstruction method of the all-weather space-time continuous 100-meter scale is adopted, and the relationship model between low-resolution and high-resolution data is established through multi-scale data acquisition and preprocessing, and the surface temperature observations with clear sky and cloudy conditions are combined to carry out the Cell-by-cell error correction is used to build downscale and correction models using machine learning algorithms, and achieve step-by-step downscale from 10km to 100m, and obtain high-precision surface temperature data.

Benefits of technology

High-precision all-weather and space-time continuous surface temperature data acquisition under clear sky and cloudy conditions is achieved, improving the coverage, reliability and spatial resolution of surface temperature products, and filling the lack of data under the cloud.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118313239B_ABST
    Figure CN118313239B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for reconstructing the surface temperature at a continuous hundred-meter scale in all-weather space-time. The method comprises the following steps: S1, data acquisition and preprocessing; S2, acquisition of the surface temperature at a kilometer scale in all-weather; S3, correction of the surface temperature at a kilometer scale in all-weather; S4, acquisition of the surface temperature at a hundred-meter scale in all-weather. The system comprises: a data preprocessing module, an all-weather kilometer-scale surface temperature acquisition module, an all-weather kilometer-scale surface temperature correction module and an all-weather hundred-meter-scale surface temperature acquisition module. The present invention introduces an all-weather surface temperature reanalysis product, and uses medium- and high-resolution remote sensing actual observation data to correct the surface temperature error pixel by pixel, which can accurately make up for the lack of data under the cloud. In addition, the high-spatial-resolution all-weather surface temperature is obtained through the strategy of two-level downscaling and clear-sky-cloud modeling, which effectively improves the coverage, reliability and spatial resolution of the surface temperature product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for reconstructing land surface temperature, and in particular to a method and system for reconstructing land surface temperature at a continuous hundred-meter scale in all-weather time and space. Background Art

[0002] Land surface temperature is a key geophysical parameter that plays a key role in studying the earth's energy balance, water cycle and climate change, and has an important impact on the surface energy balance and water cycle processes at the regional and global scales. Satellite remote sensing, especially satellites with thermal infrared channels, is the main way to obtain land surface temperature data. Accurate land surface temperature data not only helps to assess the surface energy balance and hydrological cycle, but also provides information about the long-term trend of the earth's surface temperature. It has been widely used in many research fields, including global climate change, surface flux estimation, soil moisture estimation, agricultural drought monitoring, urban thermal environment and geothermal resource detection. However, current satellite sensors still face challenges in spatial resolution and large-scale monitoring. No single sensor can provide high spatial resolution and large-scale land surface temperature data at the same time. In addition, thermal infrared channel data are easily affected by adverse weather conditions such as clouds and rain, resulting in the loss of land surface temperature data, which further limits the industry application of land surface temperature data on a fine scale and large scale. At present, large-scale land surface temperature can only be obtained through satellites with large swaths, and their resolution is usually low, usually at the kilometer scale. In order to improve the spatiotemporal resolution of remote sensing surface temperature data, commonly used methods include downscaling and spatiotemporal fusion.

[0003] The downscaling method is to combine high-resolution auxiliary remote sensing data (such as land cover, vegetation index, etc.) with low-resolution surface temperature data to improve the spatial resolution of surface temperature by establishing statistical relationships; while the spatiotemporal fusion method is to combine remote sensing data of different phases, such as using surface temperature information from multiple sensors and high-resolution surface feature data, and reconstruct high-resolution surface temperature images through specific algorithms. However, the downscaling and spatiotemporal fusion methods are only applicable to clear sky conditions and still cannot effectively solve the problem of missing surface temperature data caused by cloudy and rainy weather. The existing cloud surface temperature filling method requires additional data, such as accumulated solar radiation, microwave data, etc., but the difficulty in obtaining additional data, poor accuracy, and time-space mismatch further limit the effective implementation of the algorithm, and it is impossible to obtain large-scale, all-weather, spatiotemporal and continuous hundred-meter-scale surface temperature data. Summary of the invention

[0004] In order to solve the shortcomings of the above-mentioned technology, the present invention provides a method and system for reconstructing the surface temperature at a continuous hundred-meter scale in all-weather time and space.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for reconstructing the surface temperature at a continuous hundred-meter scale in all-weather time and space, comprising the following steps:

[0006] S1. Data acquisition and preprocessing: Acquire multi-scale data and perform preprocessing. After the processing is completed, perform spatiotemporal registration so that all data have the same projection coordinate system and coverage;

[0007] S2. Acquisition of all-weather kilometer-scale surface temperature: Establish a relationship model between the auxiliary dataset at a low resolution of 10 km and the surface temperature. L , substitute the auxiliary parameters of 1km spatial resolution into f L , calculate and obtain all-weather surface temperature data with medium resolution, i.e. 1km spatial resolution;

[0008] S3, Correction of all-weather kilometer-scale surface temperature: Correct the all-weather surface temperature data with a spatial resolution of 1 km obtained in S2, and combine the surface temperature observations under clear and cloudy conditions to form a high-precision all-weather surface temperature observation dataset LST with a spatial resolution of 1 km 1km ;

[0009] S4. Acquisition of all-weather hundred-meter-scale surface temperature: based on LST 1km , establish the relationship model between the auxiliary data set with medium resolution (i.e. 1 km spatial resolution) and the surface temperature M , f M Applied to the auxiliary parameter data with high resolution at the scale of 100 meters, the surface temperature at the scale of 100 meters is obtained all day long.

[0010] Preferably, in S1, the multi-scale data includes:

[0011] Land surface temperature data at different scales, including the hourly Era5 land surface temperature reanalysis data (LST) with a spatial resolution of 0.1 degrees ERA5_0.1D , 1km spatial resolution daily MODIS land surface temperature product LST mod_1km And 90m spatial resolution Landsat surface temperature product LST lan_90m ;

[0012] Surface reflectance data of different scales, including 500m spatial resolution MODIS surface reflectance 16-day product and 30m spatial resolution Landsat surface reflectance product;

[0013] NASA DEM digital elevation data with a spatial resolution of 30m.

[0014] Preferably, the data preprocessing includes 100-meter scale data preprocessing:

[0015] The 30m spatial resolution Landsat surface reflectance product within a certain time range is selected, and the auxiliary Landsat surface reflectance synthetic data ρ that is seamless in time and space is obtained through data synthesis technology. lc_30m , the spatial resolution of 90 m is obtained by upscaling with spatial aggregation technology. lc_90m , and obtain ρ with a spatial resolution of 100 m by bilinear interpolation lc_100m ;

[0016] Based on the NASA DEM digital elevation data with a spatial resolution of 30 m, the corresponding elevation, slope and slope direction are obtained and combined to form an auxiliary terrain parameter dataset with a spatial resolution of 30 m. 30m ; The H was upscaled to a 90m spatial resolution by spatial aggregation technology. 90m , and obtain H with a spatial resolution of 100m by bilinear interpolation 100m ;

[0017] By LST lan_90m , divide the pixels into two different conditions, cloudy and clear sky, to establish masks, mark them as cloudy and clear sky pixels, respectively, and represent them with 0 and 1, which are used as weather condition parameters. Then, bilinear interpolation is used to obtain the weather condition parameter W with a spatial resolution of 100m 100m .

[0018] Preferably, data preprocessing includes kilometer-scale data preprocessing:

[0019] The 500m spatial resolution MODIS surface reflectance within a certain time range is selected, and the auxiliary MODIS surface reflectance synthetic data ρ that is seamless in time and space is obtained through data synthesis technology. mc_500m ;

[0020] The above data were projected and transformed, and the spatial aggregation technology was used to upscale the auxiliary MODIS surface reflectance synthetic data with a spatial resolution of 1 km. mc_1km ;

[0021] At the same time, through spatial aggregation technology, ρ lc_100m Upscaling to obtain auxiliary Landsat surface reflectance synthetic data ρ with a spatial resolution of 1 km lc_1km ;

[0022] Through spatial aggregation technology, H 100m Upscaling to obtain an auxiliary terrain parameter dataset H with a spatial resolution of 1 km 1km ;

[0023] By LST mod_1km , the pixels are divided into two different conditions, cloudy and clear sky, to establish masks, marking them as cloudy and clear sky pixels, respectively, represented by 0 and 1, which is used as the weather condition parameter W1km .

[0024] Preferably, data preprocessing includes 10km scale data preprocessing:

[0025] The hourly Era5 land surface temperature reanalysis data LST with a spatial resolution of 0.1 degrees ERA5_0.1D Preprocessing, including projection conversion, and bilinear interpolation of the data to obtain 10km resolution land surface temperature reanalysis data LST ERA5_10km ;

[0026] Through spatial aggregation technology, ρ mc_1km Upscaling to obtain auxiliary MODIS surface reflectance synthetic data ρ with a spatial resolution of 10 km mc_10km ;

[0027] Through spatial aggregation technology, H 1km Upscaling to obtain an auxiliary terrain parameter dataset H with a spatial resolution of 10 km 10km .

[0028] Preferably, in S2, the method for obtaining the all-weather kilometer-scale surface temperature is as follows:

[0029] Based on machine learning algorithms and LST ERA5_10km , H 10km , mc_10km The low-resolution model is obtained through training, i.e., the surface temperature downscaling model at the scale of 10 km;

[0030] After the model training is completed, the relationship model f between the auxiliary data set and the surface temperature at a low resolution of 10 km is obtained. L :

[0031] LST ERA5_10km =f L (ρ mc_10km ,H 10km )+R LST_L10km (1)

[0032] Among them, LST ERA5_10km is the surface temperature reanalysis data with a spatial resolution of 10 km, f L is the regression model of auxiliary parameters obtained from low-resolution training and surface temperature; ρ mc_10km It is the auxiliary MODIS surface reflectance synthetic data with a spatial resolution of 10km; H 10km It is an auxiliary terrain parameter dataset with a spatial resolution of 10km, including elevation, slope and aspect; R LST_L10km is through the model f L the residual between the predicted surface temperature and the true low-resolution surface temperature;

[0033] According to the scale invariance assumption of the downscaling method, the auxiliary parameter data with a resolution of 1 km is substituted into f L The all-weather medium-resolution surface temperature is obtained from the surface temperature at a spatial resolution of 1 km:

[0034] LST dsl_1km =f L (ρ mc_1km ,H 1km )+R LST_L1km (2)

[0035] Among them, LST dsl_1km is the all-weather surface temperature at a spatial resolution of 1 km obtained by downscaling; ρ mc_1km It is the auxiliary MODIS surface reflectance synthetic data with 1km spatial resolution; H 1km It is an auxiliary terrain parameter dataset with a spatial resolution of 1 km, including elevation, slope and aspect; R LST_L1km is through the model f L The residual between the predicted surface temperature and the true medium-resolution surface temperature is given by R LST_L10km Obtained through kriging interpolation.

[0036] Preferably, in S3, the correction process of the all-weather kilometer-scale surface temperature is specifically as follows:

[0037] The machine learning algorithm is used to construct the correction model. After the model training is completed, the medium-resolution correction model is obtained, that is, the correction regression model G of the all-weather surface temperature at a spatial resolution of 1 km. M :

[0038] LST mod_1km =G M (LST dsl_1km ,ρ mc_1km ,H 1km ) (3)

[0039] Among them, LST mod_km1 LST is the land surface temperature observation value with a spatial resolution of 1 km under clear sky conditions; dsl_1km is the all-weather surface temperature with a spatial resolution of 1 km obtained by downscaling; G M The modified regression model for all-weather surface temperature with a spatial resolution of 1 km was trained;

[0040] The medium-resolution all-weather surface temperature data obtained by downscaling under cloudy conditions are corrected by formula (3) to obtain the corrected medium-resolution all-weather surface temperature observation value in the cloudy area;

[0041] Combining the surface temperature observations under clear sky conditions and cloud conditions, a high-precision all-weather surface temperature observation dataset LST with a spatial resolution of 1km is formed 1km .

[0042] Preferably, in S4, the process of obtaining the all-weather 100-meter scale surface temperature is as follows:

[0043] Based on LST 1km , using the 1km spatial resolution W 1km , lc_1km , H 1km , jointly construct a medium-resolution downscaling model, that is, a surface temperature downscaling model at a spatial resolution of 1 km;

[0044] After the model training is completed, the relationship model f between the auxiliary parameter dataset and the surface temperature at a medium resolution of 1 km is obtained. M :

[0045] LST 1km =f M (ρ lc_1km ,H 1km ,W 1km )+R LST_M1km (4)

[0046] Among them, LST 1km It is a high-precision all-weather surface temperature observation dataset with a spatial resolution of 1 km; M is the relationship model between the auxiliary parameter set and the surface temperature at a spatial resolution of 1 km; ρ lc_1km It is the auxiliary Landsat surface reflectance synthetic data with 1km spatial resolution; W 1km is the weather condition parameter with a spatial resolution of 1 km, with values ​​of 0 and 1, representing two different weather conditions: cloudy and clear sky, respectively; R LST_M1km is the residual between the surface temperature predicted by the model and the true low-resolution surface temperature;

[0047] According to the scale invariance assumption of the downscaling method, f M Applied to the auxiliary parameter data with high resolution at the scale of 100 meters, the all-weather high-resolution surface temperature, that is, the surface temperature at the scale of 100 meters, is obtained:

[0048] LST dsl_100m =f M (ρ lc_100m ,H 100m ,W 100m )+R LST_M100m (5)

[0049] Among them, LST dsl_100mis the all-weather surface temperature with a spatial resolution of 100 m obtained by downscaling; ρ lc_100m It is the auxiliary Landsat surface reflectance synthetic data with a spatial resolution of 100m; H 100m It is auxiliary terrain parameter data with a spatial resolution of 100m; W 100m is the weather condition parameter corresponding to the 100m spatial resolution pixel, with values ​​of 0 and 1, indicating two different weather conditions: cloudy and clear sky, respectively; R LST_M100m By model f M The residual between the predicted surface temperature and the true high-resolution surface temperature is given by R LST_M1km Obtained through kriging interpolation.

[0050] A system for reconstructing the surface temperature at a hundred-meter scale in all-weather time and space, the reconstruction system comprising the following modules:

[0051] Data preprocessing module: used to perform projection transformation, spatial aggregation, bilinear interpolation and spatiotemporal registration processing on data;

[0052] All-weather kilometer-scale surface temperature acquisition module: used to train a low-resolution, 10km-scale surface temperature regression model; then obtain the all-weather surface temperature with a spatial resolution of 1km through a 10km to 1km downscaling model;

[0053] All-weather kilometer-scale surface temperature correction module: used to correct the all-weather surface temperature with a spatial resolution of 1km, and obtain a high-precision estimate of the all-weather surface temperature with a spatial resolution of 1km;

[0054] All-weather 100-meter scale surface temperature acquisition module: used to construct a surface temperature downscaling model at a spatial resolution of 1km, and then obtain the all-weather surface temperature at a spatial resolution of 100m through the 1km to 100m downscaling model.

[0055] The present invention discloses a method and system for reconstructing the land surface temperature at a continuous hundred-meter scale in all-weather space and time, which solves the problems existing in the existing methods in obtaining large-scale, all-weather, space-time, continuous hundred-meter-scale land surface temperature. It introduces the all-weather land surface temperature reanalysis product, and uses the actual observation data of medium and high resolution remote sensing to correct the error of the land surface temperature pixel by pixel, which can accurately fill the missing data under the cloud. In addition, the high spatial resolution all-weather land surface temperature is obtained through the strategy of two-level downscaling and clear sky-cloud sky modeling, which effectively improves the coverage, reliability and spatial resolution of the land surface temperature product. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the overall technical process of the present invention.

[0057] Figure 2It is a schematic diagram of the reconstruction system module of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0059] The present invention proposes a method for reconstructing land temperature at a scale of 100 meters that is continuous in all-weather space and time over a large range. The method uses all-weather land temperature reanalysis data to effectively fill the missing land temperature information under cloud cover, and generates high-precision land temperature products at a scale of 100 meters that is continuous in all-weather space and time over a large range through error correction and two-level downscaling.

[0060] It mainly solves the following technical problems: In view of the small coverage of existing surface temperature product data and the spatiotemporal discontinuity caused by cloudy and rainy weather, all-weather surface temperature reanalysis data are introduced, and an error correction model is constructed using clear sky surface temperature data based on actual remote sensing observations to ensure that high-precision surface temperature data can be obtained under both clear and cloudy conditions; in view of the shortcomings of low spatial resolution of large-scale surface temperature product data, a two-level downscaling strategy is established taking into account the different situations under clear and cloudy conditions, gradually reducing from ten kilometers to kilometers and then from kilometers to hundreds of meters, thereby achieving the acquisition of large-scale and spatiotemporally continuous surface temperature at the hundred-meter scale.

[0061] The overall technical process is as follows Figure 1 As shown, the following steps are included:

[0062] S1. Data acquisition and preprocessing: Acquire multi-scale data and perform preprocessing. After the processing is completed, perform spatiotemporal registration so that all data have the same projection coordinate system and coverage;

[0063] S2. Acquisition of all-weather kilometer-scale surface temperature: Establish a relationship model between the auxiliary dataset at a low resolution of 10 km and the surface temperature. L , substitute the auxiliary parameters of 1km spatial resolution into f L , calculate and obtain all-weather surface temperature data with medium resolution, i.e. 1km spatial resolution;

[0064] S3, Correction of all-weather kilometer-scale surface temperature: Correct the all-weather surface temperature data with a spatial resolution of 1 km obtained in S2, and combine the surface temperature observations under clear and cloudy conditions to form a high-precision all-weather surface temperature observation dataset LST with a spatial resolution of 1 km 1km ;

[0065] S4. Acquisition of all-weather hundred-meter-scale surface temperature: based on LST 1km , establish the relationship model between the auxiliary data set with medium resolution (i.e. 1 km spatial resolution) and the surface temperature M , fM Applied to the auxiliary parameter data with high resolution at the scale of 100 meters, the surface temperature at the scale of 100 meters is obtained all day long.

[0066] Furthermore, the surface temperature reconstruction method of the present invention is described in detail:

[0067] 1) Acquisition and preprocessing of multi-scale data

[0068] The multi-scale data obtained specifically include: obtaining surface temperature data at different scales, including 0.1 degree (about 11.1km) spatial resolution hourly Era5 surface temperature reanalysis data LST ERA5_0.1D , 1km spatial resolution daily MODIS land surface temperature product LST mod_1km And 90m spatial resolution Landsat surface temperature product LST lan_90m ;

[0069] Obtain surface reflectance data of different scales, including 500m spatial resolution MODIS surface reflectance 16-day product and 30m spatial resolution Landsat surface reflectance product;

[0070] Obtain NASADEM digital elevation data with a spatial resolution of 30m.

[0071] The preprocessing process is:

[0072] (a) 100-meter scale data preprocessing: To eliminate the problem of missing reflectance data caused by cloudy and rainy weather, firstly, we select 30-meter spatial resolution Landsat surface reflectance products within a certain time range (e.g., one month before and after the target date), and obtain auxiliary Landsat surface reflectance synthetic data ρ with 30-meter spatial resolution through data synthesis technology. lc_30m If some reflectance data are still missing, the time range of the data can be appropriately expanded, such as selecting 2 months before and after the target date, until time-space seamless reflectance synthetic data can be obtained.

[0073] Subsequently, the auxiliary Landsat surface reflectance data with a spatial resolution of 90 m were obtained by upscaling using spatial aggregation technology. lc_90m , and the auxiliary Landsat surface reflectance data ρ with a spatial resolution of 100 m is obtained by bilinear interpolation lc_100m .

[0074] Finally, based on the NASA DEM digital elevation data with a spatial resolution of 30 m, the corresponding elevation, slope and slope aspect are obtained and combined to form an auxiliary terrain parameter dataset H with a spatial resolution of 30 m. 30m The auxiliary terrain parameter dataset H with a spatial resolution of 90m was obtained by upscaling the spatial aggregation technology.90m , and obtain the auxiliary terrain parameter dataset H with a spatial resolution of 100m by bilinear interpolation 100m .

[0075] In addition, the surface temperature data is actually observed through high-resolution surface temperature, that is, the 90m spatial resolution Landsat surface temperature product LST lan_90m , divide the pixels into two different conditions, cloudy and clear sky, to establish masks, mark them as cloudy and clear sky pixels, respectively, and represent them with 0 and 1, which are used as weather condition parameters. Then, bilinear interpolation is used to obtain the weather condition parameter W with a spatial resolution of 100m 100m .

[0076] (b) 1 km scale data preprocessing: To eliminate the problem of missing reflectance data caused by cloudy and rainy weather, firstly, the 500 m spatial resolution MODIS surface reflectance within a certain time range (e.g., one month before and after the target date) is selected, and the auxiliary MODIS surface reflectance synthetic data ρ with a spatial resolution of 500 m is obtained through data synthesis technology. mc_500m If some reflectance data are still missing, the time range of the data can be appropriately expanded, such as selecting 2 months before and after the target date, until the reflectance synthetic data with seamless time and space can be obtained. Then, these data are processed by projection conversion and upscaling through spatial aggregation technology to obtain auxiliary MODIS surface reflectance synthetic data with a spatial resolution of 1 km. mc_1km .

[0077] At the same time, the auxiliary Landsat surface reflectance data with a spatial resolution of 100m was aggregated using spatial aggregation technology. lc_100m Upscaling to obtain auxiliary Landsat surface reflectance synthetic data ρ with a spatial resolution of 1 km lc_1km Through spatial aggregation technology, the auxiliary terrain parameter dataset H 100m Upscaling to obtain an auxiliary terrain parameter dataset H with a spatial resolution of 1 km 1km .

[0078] The actual surface temperature observations are obtained through the medium-resolution surface temperature, i.e. the daily MODIS surface temperature product LST with a spatial resolution of 1 km. mod_1km , the pixels are divided into two different conditions, cloudy and clear sky, to establish masks, marking them as cloudy and clear sky pixels, respectively, represented by 0 and 1, which is used as the weather condition parameter W 1km .

[0079] (c) 10 km scale data preprocessing: 0.1 degree (about 11.1 km) spatial resolution hourly Era5 surface temperature reanalysis data LST ERA5_0.1DPreprocessing, including projection conversion, and bilinear interpolation of the data to obtain 10km resolution land surface temperature reanalysis data LST ERA5_10km Through spatial aggregation technology, the auxiliary MODIS surface reflectance synthesis data with a spatial resolution of 1 km is synthesized mc_1km Upscaling to obtain auxiliary MODIS surface reflectance synthetic data ρ with a spatial resolution of 10 km mc_10km Through spatial aggregation technology, the auxiliary terrain parameter dataset H 1km Upscaling to obtain an auxiliary terrain parameter dataset H with a spatial resolution of 10 km 10km .

[0080] After the above multi-scale data are acquired and preprocessed, spatiotemporal registration is required to ensure that all data have the same projection coordinate system and coverage for subsequent operations.

[0081] 2) Obtaining all-weather kilometer-scale surface temperature

[0082] Based on machine learning algorithms, the prepared 10km spatial resolution land surface temperature reanalysis data LST ERA5_10km , auxiliary terrain parameter dataset H 10km , auxiliary MODIS surface reflectance synthetic data ρ mc_10km The training results in a low-resolution model, namely a 10km-scale surface temperature downscaling model.

[0083] The present invention has no specific restrictions on the selection of downscaling models. Different regression models have different accuracy and computational costs. Different methods can be selected according to actual conditions to establish the relationship between auxiliary parameters and surface temperature. For example, when there is a requirement for computational speed, a simple linear regression method can be used; when there is a requirement for model complexity and accuracy, a machine learning method such as a neural network and a random forest can be used.

[0084] After the surface temperature downscaling model training is completed, the relationship model between the auxiliary dataset and the surface temperature at a low resolution of 10 km can be obtained. L :

[0085] LST ERA5_10km =f L (ρ mc_10km ,H 10km )+R LST_L10km (1)

[0086] Among them, LST ERA5_10km is the surface temperature reanalysis data with a spatial resolution of 10 km, f L is the regression model of auxiliary parameters obtained from low-resolution training and surface temperature; ρ mc_10kmIt is the auxiliary MODIS surface reflectance synthetic data with a spatial resolution of 10km; H 10km It is an auxiliary terrain parameter dataset with a spatial resolution of 10km, including elevation, slope and aspect; R LST_L10km is through the model f L The residual between the predicted land surface temperature and the true low-resolution land surface temperature.

[0087] According to the scale invariance assumption of the downscaling method, the auxiliary parameter data with a resolution of 1 km can be substituted into f L The all-weather medium-resolution surface temperature is obtained from the surface temperature at a spatial resolution of 1 km:

[0088] LST dsl_1km =f L (ρ mc_1km ,H 1km )+R LST_L1km (2)

[0089] Among them, LST dsl_1km is the all-weather surface temperature at a spatial resolution of 1 km obtained by downscaling; ρ mc_1km It is the auxiliary MODIS surface reflectance synthetic data with 1km spatial resolution; H 1km It is an auxiliary terrain parameter dataset with a spatial resolution of 1 km, including elevation, slope and aspect; R LST_L1km is through the model f L The residual between the predicted surface temperature and the true medium-resolution surface temperature is given by R LST_L10km Obtained through kriging interpolation.

[0090] 3) All-weather kilometer-scale temperature correction

[0091] Using machine learning algorithms and existing 1km spatial resolution non-all-weather land surface temperature observations (LST) under clear sky conditions mod_k1m The all-weather kilometer-scale land surface temperature LST obtained by downscaling in step 2) dsl_1km The estimated value is corrected to improve the accuracy of the estimated value of the all-weather kilometer-scale surface temperature. This step uses a machine learning algorithm to construct a correction model, where the input of the machine learning is the weather condition parameter W 1km All-weather kilometer-scale land surface temperature LST under clear sky conditions dsl_1km Estimated value, output is the corresponding surface temperature observation value LST under clear sky conditions mod_1km This step also has no specific restrictions on the selection of the modified regression model based on machine learning.

[0092] After the model training is completed, the medium-resolution correction model can be obtained, that is, the correction regression model G of the all-weather surface temperature at a spatial resolution of 1 kmM :

[0093] LST mod_1km =G M (LST dsl_1km ,ρ mc_1km ,H 1km ) (3)

[0094] Among them, LST mod_1km It is the non-all-weather surface temperature observation value with a spatial resolution of 1 km; LST dsl_1km is the all-weather surface temperature with a spatial resolution of 1 km obtained by downscaling; G M is the modified regression model of all-weather surface temperature with a spatial resolution of 1 km trained; ρ mc_1km It is the auxiliary MODIS surface reflectance synthetic data with 1km spatial resolution; H 1km It is an auxiliary terrain parameter dataset with a spatial resolution of 1km, including elevation, slope and aspect.

[0095] After obtaining the corrected regression model of all-weather surface temperature at 1km spatial resolution, the medium-resolution all-weather surface temperature data obtained by downscaling under cloudy conditions can be corrected by formula (3) to obtain the corrected medium-resolution all-weather surface temperature observation value in the cloudy area. Combined with the existing 1km spatial resolution non-all-weather surface temperature observation value LST under clear sky conditions mod_1km The combination of the two forms a high-precision all-weather surface temperature observation dataset LST with a spatial resolution of 1km 1km .

[0096] 4) Obtaining all-weather surface temperature at the hundred-meter scale

[0097] Based on the modified regression model in step 3), we can obtain the modified 1km spatial resolution high-precision all-weather land surface temperature observation dataset LST 1km , using the weather condition parameter W with a spatial resolution of 1 km 1km , auxiliary Landsat surface reflectance synthetic data ρ lc_1km , auxiliary terrain parameter dataset H 1km , jointly construct a medium-resolution downscaling model, that is, a surface temperature downscaling model with a spatial resolution of 1 km.

[0098] This step does not have any specific restrictions on the choice of downscaling model. Different regression models have different accuracy and computational costs. Different methods can be selected to establish the relationship between parameters and surface temperature according to actual conditions. For example, if there are requirements for computational speed, a simple linear regression method can be used; if there are requirements for model complexity and accuracy, machine learning methods such as neural networks and random forests can be used.

[0099] After the surface temperature downscaling model training is completed, the medium resolution, i.e., 1 km spatial resolution, relationship model between the auxiliary parameter dataset and the surface temperature can be obtained. M :

[0100] LST 1km =f M (ρ lc_1km ,H 1km ,W 1km )+R LST_M1km (4)

[0101] Among them, LST 1km It is a high-precision all-weather surface temperature observation dataset with a spatial resolution of 1 km; M is the relationship model between the auxiliary parameter set with 1 km spatial resolution and the surface temperature; ρ lc_1km It is the auxiliary Landsat surface reflectance synthetic data with 1km spatial resolution; H 1km It is an auxiliary terrain parameter dataset with a spatial resolution of 1km, including elevation, slope and aspect; W 1km is the weather condition parameter with a spatial resolution of 1 km, with values ​​of 0 and 1, representing two different weather conditions: cloudy and clear sky, respectively; R LST_M1km is the residual between the surface temperature predicted by the model and the true low-resolution surface temperature.

[0102] According to the scale invariance assumption of the downscaling method, f M Applied to the auxiliary parameter data with high resolution at the scale of 100 meters, the all-weather high-resolution surface temperature, that is, the surface temperature at the scale of 100 meters, is obtained:

[0103] LST dsl_100m =f M (ρ lc_100m ,H 100m ,W 100m )+R LST_M100m (5)

[0104] Among them, LST dsl_100m is the all-weather surface temperature with a spatial resolution of 100 m obtained by downscaling; ρ lc_100m It is the auxiliary Landsat surface reflectance synthetic data with a spatial resolution of 100m; H 100m It is auxiliary terrain parameter data with a spatial resolution of 100m; W 100m is the weather condition parameter corresponding to the 100m spatial resolution pixel, with values ​​of 0 and 1, indicating two different weather conditions: cloudy and clear sky, respectively; R LST_M100m By model f M The residual between the predicted surface temperature and the true high-resolution surface temperature is given by RLST_M1km Obtained through kriging interpolation.

[0105] Based on the above-mentioned all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method, the present invention also provides an all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction system, as shown in the system module connection diagram of the figure, which includes four modules: a data preprocessing module, an all-weather kilometer scale surface temperature acquisition module, an all-weather kilometer scale surface temperature correction module and an all-weather hundred-meter scale surface temperature acquisition module.

[0106] Data preprocessing module: used to perform projection conversion, spatial aggregation, bilinear interpolation and spatiotemporal registration on the data, so that all data have the same projection coordinate system and coverage, and the spatial resolution of the low-resolution image is an integer multiple of the medium-resolution image, and the spatial resolution of the medium-resolution image is an integer multiple of the high-resolution image;

[0107] All-weather kilometer-scale surface temperature acquisition module: Use the prepared all-weather 10km spatial resolution surface temperature reanalysis data, terrain parameter auxiliary data set and auxiliary reflectivity data to train a low-resolution, i.e. 10km scale surface temperature regression model; use the 1km spatial resolution terrain parameter auxiliary data set and auxiliary MODIS reflectivity synthesis data, etc., through the 10km to 1km downscaling model, to obtain the 1km spatial resolution all-weather surface temperature.

[0108] All-weather kilometer-scale surface temperature correction module: Use the existing surface temperature observations under clear sky conditions to correct the 1km spatial resolution all-weather surface temperature obtained by the all-weather kilometer-scale surface temperature acquisition module to obtain a high-precision estimate of the all-weather 1km spatial resolution surface temperature.

[0109] All-weather surface temperature acquisition module at the hundred-meter scale: The all-weather kilometer-scale surface temperature correction module is used to obtain the corrected 1km resolution all-weather surface temperature observation data. A surface temperature downscaling model with a spatial resolution of 1km is constructed according to two different conditions: cloudy and clear sky. Then, the terrain parameter auxiliary data set with a spatial resolution of 100m, the auxiliary Landsat surface reflectivity synthetic data, etc. are used to obtain the all-weather surface temperature with a spatial resolution of 100m through the 1km to 100m downscaling model.

[0110] In summary, for the all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method and system proposed in the present invention, in view of the application limitations of current downscaling methods and spatiotemporal fusion methods under cloudy conditions, and the difficulty in obtaining additional data required by existing under-cloud surface temperature filling methods, the accuracy is greatly affected, and data mismatch limits the implementation of the algorithm. The reanalysis data is combined with the actual remote sensing observation data, and the clear sky data is used to correct the surface temperature reanalysis data. By implementing a two-level downscaling strategy, a downscaling model is constructed from low resolution to medium resolution and from medium resolution to high resolution, thereby ensuring the acquisition of high-precision all-weather spatiotemporal continuous surface temperature data under clear sky and cloudy conditions.

[0111] The above implementation modes are not limitations of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also belong to the protection scope of the present invention.

Claims

1. A method for reconstructing land surface temperature at a continuous hundred-meter scale in all-weather space and time, characterized by: The following steps are involved: S1. Data acquisition and preprocessing: Acquire multi-scale data and perform preprocessing. After the processing is completed, perform spatiotemporal registration so that all data have the same projection coordinate system and coverage; S2. Acquisition of all-weather kilometer-scale surface temperature: Establish a relationship model between the auxiliary dataset at a low resolution of 10 km and the surface temperature. L , substitute the auxiliary parameters of 1km spatial resolution into f L , calculate and obtain all-weather surface temperature data with medium resolution, i.e. 1km spatial resolution; S3, Correction of all-weather kilometer-scale surface temperature: Correct the all-weather surface temperature data with a spatial resolution of 1 km obtained in S2, and combine the surface temperature observations under clear and cloudy conditions to form a high-precision all-weather surface temperature observation dataset LST with a spatial resolution of 1 km 1km ; S4. Acquisition of all-weather hundred-meter-scale surface temperature: based on LST 1km , establish the relationship model between the auxiliary data set with medium resolution (i.e. 1 km spatial resolution) and the surface temperature M , f M It is applied to the auxiliary parameter data with high resolution at the hundred-meter scale to obtain the surface temperature at the hundred-meter scale all day long; In S2, the methods for obtaining all-weather kilometer-scale surface temperature are as follows: Based on machine learning algorithms and LST ERA5_10km , H 10km , mc_10km The low-resolution model is obtained through training, i.e., the surface temperature downscaling model at the scale of 10 km; After the model training is completed, the relationship model f between the auxiliary data set and the surface temperature at a low resolution of 10 km is obtained. L : LST ERA5_10km =f L (ρ mc_10km ,H 10km )+R LST_L10km (1) Among them, LST ERA5_10km is the surface temperature reanalysis data with a spatial resolution of 10 km, f L is the regression model of auxiliary parameters obtained from low-resolution training and surface temperature; ρ mc_10km It is the auxiliary MODIS surface reflectance synthetic data with a spatial resolution of 10km; H 10km It is an auxiliary terrain parameter dataset with a spatial resolution of 10km, including elevation, slope and aspect; R LST_L10km is through the model f L the residual between the predicted surface temperature and the true low-resolution surface temperature; According to the scale invariance assumption of the downscaling method, the auxiliary parameter data with a resolution of 1 km is substituted into f L The all-weather medium-resolution surface temperature is obtained from the surface temperature at a spatial resolution of 1 km: LST dsl_1km =f L (ρ mc_1km ,H 1km )+R LST_L1km (2) Among them, LST dsl_1km is the all-weather kilometer-scale surface temperature obtained by downscaling with a spatial resolution of 1 km; ρ mc_1km It is the auxiliary MODIS surface reflectance synthetic data with 1km spatial resolution; H 1km It is an auxiliary terrain parameter dataset with a spatial resolution of 1 km, including elevation, slope and aspect; R LST_L1km is through the model f L The residual between the predicted surface temperature and the true medium-resolution surface temperature is given by R LST_L10km Obtained by Kriging interpolation; In S4, the process of obtaining the all-weather 100-meter scale surface temperature is as follows: Based on LST 1km , using the 1km spatial resolution W 1km , lc_1km , H 1km , jointly construct a medium-resolution downscaling model, that is, a surface temperature downscaling model at a spatial resolution of 1 km; After the model training is completed, the relationship model f between the auxiliary parameter dataset and the surface temperature at a medium resolution of 1 km is obtained. M : LST 1km =f M (ρ lc_1km ,H 1km ,W 1km )+R LST_M1km (4) Among them, LST 1km It is a high-precision all-weather surface temperature observation dataset with a spatial resolution of 1 km; M is the relationship model between the auxiliary parameter set and the surface temperature at a spatial resolution of 1 km; ρ lc_1km It is the auxiliary Landsat surface reflectance synthetic data with 1km spatial resolution; W 1km is the weather condition parameter with a spatial resolution of 1 km, with values ​​of 0 and 1, representing two different weather conditions: cloudy and clear sky, respectively; R LST_M1km is the residual between the surface temperature predicted by the model and the true low-resolution surface temperature; According to the scale invariance assumption of the downscaling method, f M Applied to the auxiliary parameter data with high resolution at the scale of 100 meters, the all-weather high-resolution surface temperature, that is, the surface temperature at the scale of 100 meters, is obtained: LST dsl_100m =f M (ρ lc_100m ,H 100m ,W 100m )+R LST_M100m (5) Among them, LST dsl_100m is the all-weather surface temperature with a spatial resolution of 100 m obtained by downscaling; ρ lc_100m It is the auxiliary Landsat surface reflectance synthetic data with a spatial resolution of 100m; H 100m It is auxiliary terrain parameter data with a spatial resolution of 100m; W 100m is the weather condition parameter corresponding to the 100m spatial resolution pixel, with values ​​of 0 and 1, indicating two different weather conditions: cloudy and clear sky, respectively; R LST_M100m By model f M The residual between the predicted surface temperature and the true high-resolution surface temperature is given by R LST_M1km Obtained through kriging interpolation.

2. The all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method according to claim 1 is characterized by: In S1, multi-scale data includes: Land surface temperature data at different scales, including the hourly Era5 land surface temperature reanalysis data (LST) with a spatial resolution of 0.1 degrees ERA5_0.1D , 1km spatial resolution daily MODIS land surface temperature product LST mod_1km And 90m spatial resolution Landsat surface temperature product LST lan_90m ; Surface reflectance data of different scales, including 500m spatial resolution MODIS surface reflectance 16-day product and 30m spatial resolution Landsat surface reflectance product; NASA DEM digital elevation data with a spatial resolution of 30m.

3. The all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method according to claim 2 is characterized by: Data preprocessing includes 100-meter scale data preprocessing: The 30m spatial resolution Landsat surface reflectance product within a certain time range is selected, and the auxiliary Landsat surface reflectance synthetic data ρ with seamless time and space is obtained through data synthesis technology. lc_30m , the spatial resolution of 90 m is obtained by upscaling with spatial aggregation technology. lc_90m , and obtain ρ with a spatial resolution of 100 m by bilinear interpolation lc_100m ; Based on the NASA DEM digital elevation data with a spatial resolution of 30 m, the corresponding elevation, slope and slope direction are obtained and combined to form an auxiliary terrain parameter dataset with a spatial resolution of 30 m. 30m ; The H was upscaled to a 90m spatial resolution by spatial aggregation technology. 90m , and obtain H with a spatial resolution of 100m by bilinear interpolation 100m ; By LST lan_90m , divide the pixels into two different conditions, cloudy and clear sky, to establish masks, mark them as cloudy and clear sky pixels, respectively, and represent them with 0 and 1, which are used as weather condition parameters. Then, bilinear interpolation is used to obtain the weather condition parameter W with a spatial resolution of 100m 100m .

4. The all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method according to claim 3 is characterized by: Data preprocessing includes kilometer-scale data preprocessing: The 500m spatial resolution MODIS surface reflectance within a certain time range is selected, and the auxiliary MODIS surface reflectance synthetic data ρ that is seamless in time and space is obtained through data synthesis technology. mc_500m ; The above data were projected and transformed, and the spatial aggregation technology was used to upscale the auxiliary MODIS surface reflectance synthetic data with a spatial resolution of 1 km. mc_1km ; At the same time, through spatial aggregation technology, ρ lc_100m Upscaling to obtain auxiliary Landsat surface reflectance synthetic data ρ with a spatial resolution of 1 km lc_1km ; Through spatial aggregation technology, H 100m Upscaling to obtain an auxiliary terrain parameter dataset H with a spatial resolution of 1 km 1km ; By LST mod_1km , the pixels are divided into two different conditions, cloudy and clear sky, to establish masks, marking them as cloudy and clear sky pixels, respectively, represented by 0 and 1, which is used as the weather condition parameter W 1km .

5. The all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method according to claim 4 is characterized by: Data preprocessing includes 10km scale data preprocessing: The hourly Era5 land surface temperature reanalysis data LST with a spatial resolution of 0.1 degrees ERA5_0.1D Preprocessing, including projection conversion, and bilinear interpolation of the data to obtain 10km resolution land surface temperature reanalysis data LST ERA5_10km ; Through spatial aggregation technology, ρ mc_1km Upscaling to obtain the auxiliary MODIS surface reflectance synthetic data ρ with a spatial resolution of 10 km mc_10km ; Through spatial aggregation technology, H 1km Upscaling to obtain an auxiliary terrain parameter dataset H with a spatial resolution of 10 km 10km .

6. The all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method according to claim 5 is characterized by: In S3, the correction process of all-weather kilometer-scale surface temperature is as follows: The machine learning algorithm is used to construct the correction model. After the model training is completed, the medium-resolution correction model is obtained, that is, the correction regression model G of the all-weather surface temperature at a spatial resolution of 1 km. M : LST mod_1km =G M (LST dsl_1km ,ρ mc_1km ,H 1km ) (3) Among them, LST mod_1km LST is the land surface temperature observation value with a spatial resolution of 1 km under clear sky conditions; dsl_1km is the all-weather surface temperature with a spatial resolution of 1 km obtained by downscaling; G M The modified regression model for all-weather surface temperature with a spatial resolution of 1 km was trained; The medium-resolution all-weather surface temperature data obtained by downscaling under cloudy conditions are corrected by formula (3) to obtain the corrected medium-resolution all-weather surface temperature observation value in the cloudy area; Combining the surface temperature observations under clear sky conditions and cloud conditions, a high-precision all-weather surface temperature observation dataset LST with a spatial resolution of 1km is formed 1km .

7. A system for the all-weather spatiotemporal continuous hundred-meter scale surface temperature reconstruction method according to any one of claims 1 to 6, characterized in that: The reconstruction system includes the following modules: Data preprocessing module: used to perform projection transformation, spatial aggregation, bilinear interpolation and spatiotemporal registration processing on data; All-weather kilometer-scale surface temperature acquisition module: used to train a low-resolution, 10km-scale surface temperature regression model; then obtain the all-weather surface temperature with a spatial resolution of 1km through a 10km to 1km downscaling model; All-weather kilometer-scale surface temperature correction module: used to correct the all-weather surface temperature with a spatial resolution of 1km, and obtain a high-precision estimate of the all-weather surface temperature with a spatial resolution of 1km; All-weather 100-meter scale surface temperature acquisition module: used to construct a surface temperature downscaling model at a spatial resolution of 1km, and then obtain the all-weather surface temperature at a spatial resolution of 100m through the 1km to 100m downscaling model.

Citation Information

Patent Citations

  • Downscaling and fusion combined surface temperature temporal-spatial resolution enhancement method

    CN115630308A

  • Method for generating near-real-time hourly all-weather surface temperature

    CN115855272A

  • High-resolution thermal infrared surface temperature all-weather reconstruction method

    CN116011342A

  • Meteorological big data fusion method based on deep learning

    US20230351164A1

  • Accurate large-area inversion method for near-surface air temperature

    WO2018145229A1