CN113779863B - Ground surface temperature downscaling method based on data mining - Google Patents
Ground surface temperature downscaling method based on data mining Download PDFInfo
- Publication number
- CN113779863B CN113779863B CN202110878728.4A CN202110878728A CN113779863B CN 113779863 B CN113779863 B CN 113779863B CN 202110878728 A CN202110878728 A CN 202110878728A CN 113779863 B CN113779863 B CN 113779863B
- Authority
- CN
- China
- Prior art keywords
- data
- surface temperature
- window
- model
- resolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Geometry (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Radiation Pyrometers (AREA)
Abstract
The invention discloses a land surface temperature downscaling method based on data mining, which comprises the following steps of: selecting appropriate low-spatial-resolution earth surface temperature data and high-spatial-resolution visible light data as data sources according to the experimental area; obtaining a global model result and a local model result through the steps of data preprocessing, sample acquisition and the like; model-based Weight index (Weight) i ) Merging the results of the global model and the local model; calculating the temperature residual error in a scale-by-scale window based on a modified energy balance formula, and applying the adjustment result to each pixel in the window to complete the scale reduction process of the low-resolution earth surface temperature data; and calculating MAE, R and RMSE between the reduced-scale ground surface temperature and other high-resolution ground surface temperature products, and verifying the accuracy and applicability of the algorithm for reducing the scale of the ground surface temperature. The method is based on the full mining of the self-contained spectral information of the multi-source remote sensing data, and the improvement of the spatial resolution of the low-spatial-resolution earth surface temperature data is realized.
Description
Technical Field
The invention relates to the technical field of agricultural remote sensing, in particular to a method for reducing the scale (improving the spatial resolution) of a low-spatial-resolution earth surface temperature product.
Background
Surface Temperature (LST) is often defined as the surface temperature of the earth. As an important parameter for environmental research and resource management, the method is widely applied to the fields of drought monitoring, evapotranspiration, soil moisture estimation, forest fire detection and the like. In the 70 s of the 20 th century, the remote sensing technology is applied to surface temperature inversion, and the advantages of macroscopicity, rapidness, economy and the like make up for the defects of the traditional ground monitoring on spatial distribution. High-Resolution Radiometer (AVHHR) and medium-Resolution Imaging Spectrometer (MODIS) are two kinds of thermal infrared sensors widely used at present, and both can achieve at least one day of earth-covering observation under the condition that the spatial Resolution is 1 km. Fengyun-3(FY-3) is a second generation polar orbit meteorological satellite in China, and aims to realize all-weather, multi-spectral and three-dimensional observation of global atmospheric and geophysical elements. The VIRR is one of key effective loads carried on an FY-3A/B/C meteorological satellite, has 10 channels (table 2-1), has a wavelength range of 0.44-12.50 mu m, and provides visible light and infrared spectrum for global observation in atmosphere, sea and land. Wherein, two adjacent thermal infrared channels (IR1, 10.30-11.30 μm and IR2, 11.50-12.50 μm) provide a better opportunity for the inversion of LST by using the split window algorithm.
However, due to the limitation of the existing satellite-borne heat infrared sensor technology, the spatial resolution of the existing satellite-borne heat infrared image is generally low (sixty meters to tens of kilometers), the application of the earth surface temperature information with high spatial resolution in the research of some smaller-scale thermal environments is limited, and the analysis of the earth surface covering thermal characteristics on a small scale can help to carry out deeper understanding on the change of a small-scale weather system.
The current approaches for improving the spatial resolution of thermal infrared images can be roughly divided into two approaches: one is that the performance of an imaging system of the thermal infrared detector is improved, the spatial resolution of the thermal infrared image is essentially improved, in recent years, some airborne high-resolution thermal infrared sensors are successfully researched and developed, but due to the problems of high acquisition difficulty, high cost, long acquisition period, complex pretreatment process and the like, the actual application and popularization are difficult to obtain; the other method is based on an image processing method, utilizes the existing high spatial resolution data of visible light-near infrared, adopts a proper algorithm to carry out downscaling on the thermal infrared image with low spatial resolution, and enables the spatial resolution to reach the level of the visible light-near infrared data.
At present, the widely applied ground surface temperature downscaling method is a downscaling method based on ground surface parameter statistical regression, such as a classic Dis track (differentiation procedure for radiometric surface temperature) model, a TsHARP (algorithm for learning thermal image) algorithm, an improved algorithm of Dis track and TsHARP algorithms such as E-track, NL-track, etc., and a geographical weighting regression model such as GWR, GTWR, etc. which has been developed recently, but in general, the commonly used ground surface temperature downscaling algorithm has two common problems: (1) the method is characterized in that the input of high-order remote sensing products such as various waveband indexes (NDVI, NDBI, EVI, SAVI and the like), land utilization types, DEM and other information is relied on, and most algorithms only adopt fewer indexes, so that the accuracy of the algorithms in areas with complex earth surface coverage is lower; (2) some new algorithms which rise in recent years apply machine learning methods such as artificial neural networks and support vector machines, and adopt more indexes, compared with the traditional modeling method, the machine learning methods have higher requirements on sample data, sample collection based on the traditional downscaling mode usually collects samples by manually selecting interested regions or supervising classification, the methods are difficult to collect a large number of samples with complete attribute information, cannot identify space details, neglects the mutual influence of adjacent elements, and the defect restricts the precision of the downscaling algorithm results.
Disclosure of Invention
The invention aims to solve the technical problem of providing a ground surface temperature downscaling method based on data mining aiming at the defects of the prior art, in particular to a method for realizing low-spatial-resolution ground surface temperature downscaling based on mining of high-spatial-resolution short-wave band information. The spatial resolution of the low spatial resolution earth surface temperature data is improved based on the multi-source remote sensing data and the self-contained spectral information of the fully-mined data.
In order to solve the technical problem, the invention provides a method for realizing the size reduction of the earth surface temperature with low spatial resolution based on mining high spatial resolution short wave band information, which comprises the following steps:
step 1: selecting appropriate low-spatial-resolution earth surface temperature data and high-spatial-resolution visible light data as data sources according to the experimental area;
step 2: sampling a bilinear interpolation method to resample the low-spatial-resolution earth surface temperature data to a high-spatial-resolution visible light data pixel size, spatially aggregating the high-spatial-resolution data to a low-spatial-resolution pixel size (the process is embodied in moving window sampling), and inputting the resampled earth surface temperature data and the original visible light data (an input waveband can be selected as required) into an algorithm (the input data area is ensured to be consistent);
and step 3: defining a coefficient of variation C v As a threshold value for sufficiently mining band information, C v The lower the value, the more uniform the sample collected can be considered;
and 4, step 4: according to the scale ratio scale of data input by different sensors, a scale-scale moving window is created, the moving window moves from left to right and from top to bottom, and C of pixels in the window is counted v Value below the predetermined C v The pel of the value, considered as a pure pel, is collected as a sample.
And 5: splitting sample data into a sample value and a label value, constructing a model by adopting a random forest method, and applying the model to the whole image area to obtain a global model result;
step 6: defining a local model prediction window prediction _ window and a local model sampling window samples _ window, wherein the sampling window is required to be larger than the prediction window so as to weaken the boundary effect of a local prediction result;
and 7: step 7-1: based on samples _ window, performing operation of the step 4 to obtain local sample data; step 7-2: continuing to perform the operation of the step 5, constructing a local model, applying the model to a prediction _ window range, and moving a window; repeating the step 7-1 and the step 7-2 until the whole image area is predicted to obtain a local model result;
and 8: and calculating the difference between the results of the global model and the local model and the original earth surface temperature data pixel by pixel, and calculating the weight of the global model and the local model when the global model and the local model are combined at the pixel based on the difference so as to combine the results of the global model and the local model.
And step 9: calculating a temperature residual error in a scale-by-scale window based on a modified energy balance formula, and applying a residual error result to each pixel in the window to complete the low-resolution ground surface temperature data downscaling process;
step 10: and calculating MAE (mean Absolute error), R and RMSE (mean Absolute error), between the ground surface temperature after the downscaling and the original low-resolution ground surface temperature value, and verifying the accuracy and the applicability of the algorithm for performing the downscaling on the ground surface temperature.
The original surface temperature data adopted in all the steps are MODIS/TERRA MOD11A2 LST daily product data, and the data format is HDF format; and converting the original HDF format file into a TIFF format file with geographic information by combining the quality control file (QC file) and the coordinate information.
The coefficient of variation (C) in said steps 3, 4 v ) The calculation formula is as follows:
in the formula, σ i Is the standard deviation, mu, of the pixel values within a moving window i Is the average value of the pixel values within the moving window.
In the step 8, based on the weights of the global and local models at different pixel positions, the weight ratio is calculated according to the following formula:
in the formula, Weight g Weight for the Global model, Weight l Weights are occupied by the local models; w is a g Is a global model weight index, w l The calculation formula of the local model weight index is as follows:
in the formula (d) g Deviation of global model results from the original ground temperature data at the pixels, d l Deviation of local model result from original ground temperature data at pixel position; if | d g |<0.1 or | d l |<When 0.1, it corresponds to w g Or w l The value is defined as 100.
Based on the modified energy balance equation in step 9, the pixel energy residual error calculation formula is as follows:
wherein n is the total number of pixels in the energy balance window in step 9, T i And merging the pixel values of the model result for the corresponding positions in the window. The energy residual field application process is shown by the following formula:
T apply ＝T combined +(T org -T res )
in the formula, T combined To merge model results, T org Is the original ground temperature data.
In step 10, the average absolute error MAE is calculated as follows:
wherein N is the total pixel number of the image of the ground temperature reduction scale result, T residual_i Is the ground temperature value T of a certain pixel after energy balance org_i Is the original low-resolution surface temperature value at the pixel.
The correlation coefficient R is the correlation between two variables and its correlation direction, and its calculation formula is as follows:
wherein N is the total pixel number of the image of the ground temperature reduction scale result, T residual_i Is the ground temperature value T of a certain pixel after energy balance org_i Is the original low-resolution surface temperature value at the pixel.
The root mean square error RMSE is the square root of the ratio of the square of the deviation of the observed value from the true value to the number of observations N. The calculation formula is as follows:
RMSE is root mean square error, N is total image element number of scale reduction result of ground temperature, T residual_i Is the ground temperature value T of a certain pixel after energy balance org_i Is the original low-resolution surface temperature value at the pixel.
Advantageous effects
(1) In order to efficiently process a large number of input variables that may affect the surface temperature distribution of different land use types, the present invention employs a data mining method for adaptively searching samples in data, creating and applying models to achieve a surface temperature downscaling. According to the mean coefficient of variation (C) v ) The method extracts earth surface temperature and spectral reflectivity samples from all areas of an application scene, ensures that the sample size is large enough and the distribution is uniform enough, and therefore, the model prediction result can be evaluated by a standard statistical method;
(2) because the space distribution and the change of the earth surface temperature are nonlinear, in order to make the downscaling process more consistent with the actual situation, a random forest method is selected for modeling, the difference between the global prediction model result and the local prediction model result in the space distribution and the earth surface environment is comprehensively analyzed in the model application process, and the weight coefficient (W) is used i ) The two model results are combined, so that the advantage that the global model result is more fit with the original ground temperature data on the overall distribution trend of the ground surface temperature is kept, and the accuracy of the local model on the details can be ensured.
(3) In order to reduce the deviation of the model in practical application, the invention provides an energy balance method, the difference between the application result of the downscaling model and the original low-spatial resolution earth surface temperature data is counted by an energy balance formula modified based on the Stefan-Boltzmann law, and the difference is applied to the result image to improve the downscaling precision. The method does not depend on the relation between any other predefined earth surface temperature and other environment variables, only the characteristic information is mined from the scene according to the method, the method is allowed to be applied at any time in different areas and under different conditions, and the possibility is provided for the application of subsequent earth surface temperature data in small-scale areas such as urban heat island monitoring, soil moisture estimation, forest fire risk monitoring and the like.
Drawings
FIG. 1 is a flow chart of algorithm implementation;
FIG. 2 shows the algorithm input high spatial resolution visible light and low spatial resolution earth surface temperature data (a is visible light data, b is earth surface temperature data);
FIGS. 3-5 are graphs showing downscaling results (a is the result before downscaling, b is the result after downscaling, and c is the region-corresponding visible wavelength band data);
FIG. 6 is a diagram of a uniformly selected 25 point accuracy validation;
Detailed Description
The present invention will be described in detail with reference to specific examples.
Example 1:
the invention provides a method for realizing ground surface temperature downscaling with low spatial resolution based on mining of high spatial resolution short wave band information, which comprises the following steps of:
step 1: selecting appropriate low-spatial-resolution earth surface temperature data and high-spatial-resolution visible light data as data sources according to the experimental area;
step 2: sampling a bilinear interpolation method to resample the low-spatial-resolution earth surface temperature data to a high-spatial-resolution visible light data pixel size, spatially aggregating the high-spatial-resolution data to a low-spatial-resolution pixel size (the process is embodied in moving window sampling), and inputting the resampled earth surface temperature data and the original visible light data (an input waveband can be selected as required) into an algorithm (the input data area is ensured to be consistent);
and 3, step 3: defining a coefficient of variation C v As a threshold value for sufficiently mining band information, C v The lower the value, the more uniform the sample collected, C v The calculation formula of (a) is as follows:
in the formula, σ i Is the standard deviation, mu, of the pixel values within a moving window i Is the average value of the pixel values within the moving window.
And 4, step 4: according to the scale ratio scale of data input by different sensors, a scale-scale moving window is created, the moving window moves from left to right and from top to bottom, and C of pixels in the window is counted v Value below the predetermined C v The pel of the value, considered as a pure pel, is collected as a sample.
And 5: splitting sample data into a sample value and a label value, constructing a model by adopting a random forest method, and applying the model to the whole image area to obtain a global (global) model result;
step 6: defining a local model prediction window prediction _ window and a local model sampling window samples _ window, wherein the sampling window is required to be larger than the prediction window so as to weaken the boundary effect of a local prediction result;
and 7: based on samples _ window, performing the operation of step 4 to obtain local sample data, continuing the operation of step 5, constructing a local model, applying the model to a predict _ window range, moving a window, and repeating the steps 4 and 5 until the whole image area is predicted to obtain a local model result;
and 8: calculating the difference between the results of the global model and the local model and the original earth surface temperature data pixel by pixel, calculating the weight of the global model and the local model when the global model and the local model are combined at the pixel based on the difference, combining the results of the global model and the local model, and calculating the weight ratio according to the following formula:
in the formula, Weight g Weight for the Global model, Weight l Weights are occupied by the local models; w is a g Is a global model weight index, w l The calculation formula of the local model weight index is as follows:
in the formula (d) g Deviation of global model results from the original ground temperature data at the pixels, d l Deviation of local model result from original ground temperature data at pixel position; if | d g |<0.1 or | d l |<When 0.1, it corresponds to w g Or w l The value is defined as 100.
And step 9: based on a modified energy balance formula, calculating a temperature residual error in a scale-by-scale window, applying a residual error result to each pixel in the window, and completing a low-resolution earth surface temperature data downscaling process, wherein the pixel energy residual error calculation formula is as follows:
in the formula, n is the total number of pixels in the energy balance window, T i And combining the pixel values of the model results for the corresponding positions in the window. The energy residual field application process is shown by the following formula:
T apply ＝T combined +(T org -T res )
in the formula, T combined To merge model results, T org Is the original ground temperature data.
Step 10: re-aggregating the downscaled earth surface temperature data to the original earth surface temperature data scale, calculating MAE (mean Absolute error), R and RMSE (maximum mean Absolute error) between the downscaled earth surface temperature and the original low-resolution earth surface temperature value, and verifying the accuracy and the applicability of the algorithm for performing the downscaling of the earth surface temperature, wherein the MAE calculation formula is as follows:
wherein N is the total pixel number of the image of the ground temperature reduction scale result, T residual_i Is the ground temperature value T of a certain pixel after energy balance org_i Is the original low-resolution surface temperature value at the pixel.
The correlation coefficient R is the correlation between two variables and its correlation direction, and its calculation formula is as follows:
wherein N is the total pixel number of the image obtained by reducing the scale of the earth temperature, T residual_i As energyTemperature value of certain pixel after balance, T org_i Is the original low-resolution surface temperature value at the pixel.
The root mean square error RMSE is the square root of the ratio of the square of the deviation of the observed value from the true value to the number of observations N. The calculation formula is as follows:
RMSE is root mean square error, N is total image element number of scale reduction result of ground temperature, T residual_i Is the ground temperature value T of a certain pixel after energy balance org_i Is the original low-resolution surface temperature value at the pixel.
The original surface temperature data adopted in all the steps are MODIS/TERRA MOD11A2 LST daily product data, and the data format is HDF format; and converting the original HDF format file into a TIFF format file with geographic information by combining the quality control file (QC file) and the coordinate information.
The research area is located in a Chongqing city main city area, is located at 106 degrees E-107 degrees E and 29 degrees N-30.5 degrees N and comprises a north orange area, an Yubei orange area, a Jiangbei orange area, a sandlawn dam area, an Yuzhong orange area, a southern river area, a Jiulong river area, a great crossing area and a south-bara area. MODIS LAT product data (MOD11A2) in 7, 23 and 2017 and Sentinel-2A visible light remote sensing data are obtained as downscaling data sources, and according to the downscaling process:
1. according to the terrain and landform conditions of the Chongqing city in the example, Sentinel-2A high-spatial-resolution visible light data and MODIS LST products are selected as a ground surface temperature downscaling data source;
2. resampling a MODIS LST (1KM resolution) product to 20m scale by adopting a bilinear interpolation method, and inputting the product and Sentinel-2A data (5 th, 6 th, 7 th, 8a th, 11 th and 12 th wave bands) into an algorithm;
3. first, the coefficient of variation C is set v A value of 0.2;
4. the sampling window size was set to 50 x 50 (1000/20), window C according to the input MODIS (1KM) and Sentinel-2A (20M) v Adding the window center pixel with the value lower than 0.2 into the sample set;
5. secondly, splitting the sample obtained in the previous step into a sample part and a label part, performing random forest modeling, and applying the model to the whole Sentinel-2A image to obtain a global model result;
6. a local model prediction window prediction _ window (50-100 times the global sampling window size, which in this example is set to 50 times, i.e. 2500 x 2500) and a local model sampling window samples _ window (1.25 times the prediction window, which in this example is set to 3125 x 3125) are defined.
7. Based on samples _ window, performing the operation of step 4 to obtain local sample data, continuing the operation of step 5, applying the model to a prediction _ window range, moving the window, and repeating the steps 4 and 5 until the whole image area is predicted to obtain a local model result;
8. calculating the difference between the results of the global model and the local model and the original earth surface temperature data pixel by pixel, calculating the weight of the global model and the local model when the global model and the local model are combined at the pixel based on the difference, and combining the results of the global model and the local model to obtain a combined model result;
9. calculating temperature residual errors in a 50 x 50 window based on a modified energy balance formula, applying a residual error result to each combined model result pixel in the window, and completing a low-resolution ground surface temperature data downscaling process to obtain an energy balance result;
10. and (3) re-aggregating the ground surface temperature data after the dimension reduction to the dimension of the original ground surface temperature data, calculating MAE (mean Absolute error), R and RMSE (maximum mean Absolute error) between the ground surface temperature after the dimension reduction and the original low-resolution ground surface temperature value, and verifying the accuracy and the applicability of the algorithm for the ground surface temperature dimension reduction.
The operation results are as follows, wherein fig. 2 is the input data of the algorithm downscaling, a in the figure is Sentinel-2A visible light data, b is the MODIS LST 1KM product, as the data to be downscaled, the pixel of the low spatial resolution surface temperature in the research area is downscaled according to the steps 3-9, fig. 1 is a flowchart of the downscaling algorithm, the result of the downscaling of the surface temperature based on the algorithm is shown in fig. 3-5, a in fig. 3 is the surface temperature of the town area before downscaling, b is the surface temperature of the town area after downscaling, c is the visible light wave section data in the area, a in fig. 4 is the surface temperature of the river airport before downscaling, b is the surface temperature of the river north airport after downscaling, c is the visible light wave section data in the river north airport, a in fig. 5 is the surface temperature of the suburban area before downscaling, b is the suburban area surface temperature after downscaling, c is suburban visible light band data. Verifying the land surface temperature data before and after the downscaling according to the step 10, wherein the result is shown in fig. 6 and table 1;
TABLE 1
As can be seen from FIG. 6, the coefficient of determination R between the surface temperature before and after the downscaling 2 0.8547 is reached, which shows that the ground surface temperature data greatly improves the spatial resolution after the scale reduction (1KM->20M), the original surface temperature information is better preserved, and the lower mean absolute error MAE (around 0.527233 ℃) and the higher correlation coefficient R (0.922854) in table 1 also confirm this result.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (2)
1. A land surface temperature downscaling method based on data mining is characterized in that low spatial resolution land surface temperature downscaling is achieved based on mining high spatial resolution short wave band information, and the method comprises the following steps:
step 1: selecting appropriate low-spatial-resolution earth surface temperature data and high-spatial-resolution visible light data as data sources according to the experimental area;
and 2, step: resampling low-spatial-resolution earth surface temperature data to the pixel size of high-spatial-resolution visible light data by adopting a bilinear interpolation method, spatially aggregating the high-spatial-resolution data to the pixel size of the low-spatial-resolution, and inputting the resampled earth surface temperature data and original visible light data;
and step 3: defining a coefficient of variation C v As a threshold value for sufficiently mining band information, C v The lower the value, the more uniform the collected sample can be considered;
and 4, step 4: according to the scale ratio scale of data input by different sensors, a scale-scale moving window is created, the moving window moves from left to right and from top to bottom, and C of pixels in the window is counted v Value below the predetermined C v The pixel of the value is regarded as a pure pixel and is collected as a sample;
coefficient of variation (C) v ) The calculation formula is as follows:
in the formula, σ i Is the standard deviation, mu, of the pixel values within a moving window i The average value of the pixel values in the moving window is shown, and n is the total number of the pixels in the moving window;
and 5: splitting sample data into a sample value and a label value, constructing a model by adopting a random forest method, and applying the model to the whole image area to obtain a global model result;
step 6: defining a local model prediction window prediction _ window and a local model sampling window samples _ window, wherein the sampling window is required to be larger than the prediction window so as to weaken the boundary effect of a local prediction result;
and 7: step 7-1: based on samples _ window, performing operation of the step 4 to obtain local sample data; step 7-2: continuing to perform the operation of the step 5, constructing a local model, applying the model to a prediction _ window range, and moving a window; repeating the step 7-1 and the step 7-2 until the whole image area is predicted to obtain a local model result;
and 8: calculating the difference between the results of the global model and the local model and the original earth surface temperature data pixel by pixel, calculating the weight of the global model and the local model when the global model and the local model are combined at the pixel based on the difference, and comparing the global model and the local modelThe results of the local models are merged to obtain T combined ；
In the step 8, the results of the global model and the local model are combined, and based on the weights of the global model and the local model at different pixel positions, the weight ratio is calculated according to the following formula:
in the formula, Weight g Weight for the Global model, Weight l Weights are occupied by the local models; w is a g Is a global model weight index, w l The calculation formula of the local model weight index is as follows:
in the formula (d) g Deviation of global model results from the original ground temperature data at the pixels, d l Deviation of local model result from original ground temperature data at pixel position; if | d g < 0.1 or | d l If | is less than 0.1, it corresponds to w g Or w l The value is defined as 100;
and step 9: calculating a temperature residual error in a scale-by-scale window based on an energy balance formula, and applying a residual error result to each pixel in the window to complete a low-resolution ground surface temperature data downscaling process;
the energy balance formula is as follows:
in the formula, n is total number of pixels in scale window, T i Merging the pixel values of the model result for the corresponding positions in the window; the energy residual field application process is shown in the following formula:
T apply ＝T combined +(T org -T res )
in the formula, T combined To merge model results, T org Is the original ground temperature data.
2. The method for downscaling the earth surface temperature according to claim 1, wherein the adopted original earth surface temperature data is MODIS/TERRA MOD11A2 LST daily product data, and the data format is HDF format; and converting the original HDF5 format file into a TIFF format file with geographic information by combining a quality control file (QC file) and image coordinate information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110878728.4A CN113779863B (en) | 2021-08-02 | 2021-08-02 | Ground surface temperature downscaling method based on data mining |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110878728.4A CN113779863B (en) | 2021-08-02 | 2021-08-02 | Ground surface temperature downscaling method based on data mining |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113779863A CN113779863A (en) | 2021-12-10 |
| CN113779863B true CN113779863B (en) | 2022-08-09 |
Family
ID=78836384
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110878728.4A Active CN113779863B (en) | 2021-08-02 | 2021-08-02 | Ground surface temperature downscaling method based on data mining |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113779863B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105427309A (en) * | 2015-11-23 | 2016-03-23 | 中国地质大学（北京） | Multiscale hierarchical processing method for extracting object-oriented high-spatial resolution remote sensing information |
| CN107423537A (en) * | 2017-01-25 | 2017-12-01 | 河海大学 | A kind of method of the surface temperature NO emissions reduction based on adaptive threshold |
| CN107748736A (en) * | 2017-10-13 | 2018-03-02 | 河海大学 | A kind of multiple-factor Remote Sensing temperature space NO emissions reduction method based on random forest |
| CN108896185A (en) * | 2018-05-14 | 2018-11-27 | 河海大学 | Remote Sensing temperature space NO emissions reduction method based on normalization desert index |
| CN109635309A (en) * | 2018-10-17 | 2019-04-16 | 广州地理研究所 | A kind of surface temperature space NO emissions reduction method |
| CN109741261A (en) * | 2019-01-03 | 2019-05-10 | 北京师范大学 | A kind of Remote Sensing temperature NO emissions reduction algorithm based on object-oriented window |
| CN109885959A (en) * | 2019-03-05 | 2019-06-14 | 中国科学院地理科学与资源研究所 | A kind of surface temperature robust NO emissions reduction method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101404430B1 (en) * | 2013-06-11 | 2014-06-10 | 서울시립대학교 산학협력단 | Method for estimation of surface temperature lapse rate Using thermal infrared images |
-
2021
- 2021-08-02 CN CN202110878728.4A patent/CN113779863B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105427309A (en) * | 2015-11-23 | 2016-03-23 | 中国地质大学（北京） | Multiscale hierarchical processing method for extracting object-oriented high-spatial resolution remote sensing information |
| CN107423537A (en) * | 2017-01-25 | 2017-12-01 | 河海大学 | A kind of method of the surface temperature NO emissions reduction based on adaptive threshold |
| CN107748736A (en) * | 2017-10-13 | 2018-03-02 | 河海大学 | A kind of multiple-factor Remote Sensing temperature space NO emissions reduction method based on random forest |
| CN108896185A (en) * | 2018-05-14 | 2018-11-27 | 河海大学 | Remote Sensing temperature space NO emissions reduction method based on normalization desert index |
| CN109635309A (en) * | 2018-10-17 | 2019-04-16 | 广州地理研究所 | A kind of surface temperature space NO emissions reduction method |
| CN109741261A (en) * | 2019-01-03 | 2019-05-10 | 北京师范大学 | A kind of Remote Sensing temperature NO emissions reduction algorithm based on object-oriented window |
| CN109885959A (en) * | 2019-03-05 | 2019-06-14 | 中国科学院地理科学与资源研究所 | A kind of surface temperature robust NO emissions reduction method |
Non-Patent Citations (1)
| Title |
|---|
| A New Approach for Surface Urban Heat Island Monitoring Based on Machine Learning Algorithm and Spatiotemporal Fusion Model;YUAN YAO 等;《IEEE Access 》;20200907;第8卷;第164268-164281页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113779863A (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Spatially continuous and high-resolution land surface temperature product generation: A review of reconstruction and spatiotemporal fusion techniques | |
| Wang et al. | Fusion of Landsat 8 OLI and Sentinel-2 MSI data | |
| CN111191673B (en) | Ground surface temperature downscaling method and system | |
| Yoo et al. | Spatial downscaling of MODIS land surface temperature: Recent research trends, challenges, and future directions | |
| Imani | Particulate matter (PM2. 5 and PM10) generation map using MODIS Level-1 satellite images and deep neural network | |
| Huang et al. | A novel fusion method for generating surface soil moisture data with high accuracy, high spatial resolution, and high spatio‐temporal Continuity | |
| Hou et al. | On the value of available MODIS and Landsat8 OLI image pairs for MODIS fractional snow cover mapping based on an artificial neural network | |
| Kartal et al. | Prediction of MODIS land surface temperature using new hybrid models based on spatial interpolation techniques and deep learning models | |
| Liu et al. | Hyperspectral infrared sounder cloud detection using deep neural network model | |
| CN114241331B (en) | Remote sensing modeling method for ground biomass of reed in wetland by taking UAV as ground and Septinel-2 medium | |
| Verma et al. | A machine learning approach and methodology for solar radiation assessment using multispectral satellite images | |
| CN113408111B (en) | Atmospheric precipitation inversion method and system, electronic equipment and storage medium | |
| Tang et al. | Filling then spatio-temporal fusion for all-sky MODIS land surface temperature generation | |
| Premier et al. | A novel approach based on a hierarchical multiresolution analysis of optical time series to reconstruct the daily high-resolution snow cover area | |
| CN112285808B (en) | Method for reducing scale of APHRODITE precipitation data | |
| Xiong et al. | Fusing Landsat-7, Landsat-8 and Sentinel-2 surface reflectance to generate dense time series images with 10m spatial resolution | |
| Wang et al. | A taylor expansion algorithm for spatial downscaling of MODIS land surface temperature | |
| Sui et al. | Processing of multitemporal data and change detection | |
| CN113779863B (en) | Ground surface temperature downscaling method based on data mining | |
| Colaninno et al. | Towards an operational model for estimating day and night instantaneous near-surface air temperature for urban heat island studies: outline and assessment | |
| CN116522090A (en) | Soil moisture downscaling method considering vegetation memory and fine-scale vegetation index | |
| Liu et al. | Bi-LSTM model for time series leaf area index estimation using multiple satellite products | |
| Chen et al. | Urban land surface temperature retrieval from high spatial resolution thermal infrared image using a modified split-window algorithm | |
| CN112052720B (en) | High-space-time normalization vegetation index NDVI fusion model based on histogram clustering | |
| Zhou et al. | A station-data-based model residual machine learning method for fine-grained meteorological grid prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |