A method for estimating near-surface temperature under clear sky conditions with high spatial resolution

Abstract

The present invention relates to a method for estimating near-surface temperature with high spatial resolution under clear sky conditions. The method includes collecting the measured temperature of meteorological stations, the low spatial resolution surface temperature of satellite remote sensing, the temperature predicted by numerical models, the elevation of high spatial resolution, the vegetation index data and the latitude, longitude and time information. A low spatial resolution temperature model is constructed using the surface temperature with low spatial resolution, the temperature predicted by numerical models, the elevation, the vegetation index, the latitude, the longitude and the time information; it is assumed that the difference in near-surface temperature between the high spatial resolution sub-pixel and the low spatial resolution pixel where it is located is mainly caused by the difference in elevation and vegetation index between the two, thereby constructing a pixel temperature difference model; on the basis of the low spatial resolution temperature model and the pixel temperature difference model, a high spatial resolution temperature estimation model under clear sky conditions is derived. The present invention can obtain the spatial distribution of near-surface temperature under clear sky conditions with high precision that is more refined than satellite observations.

Description

Technical Field

[0001] The invention belongs to the technical field of atmospheric remote sensing, and in particular relates to a method for estimating near-ground temperature with high spatial resolution under clear sky conditions. Background Art

[0002] Surface air temperature (T air ) refers to the atmospheric temperature measured by a louvered box 1.5-2.0 meters above the ground. air It is a key indicator in many research fields such as agricultural ecological environment, urban heat island effect and climate change. air It is a very important input parameter in hydrological, meteorological, environmental, and climate models. air Affected by solar radiation, longitude and latitude, surface type, soil moisture, cloud cover, altitude and other factors, it presents obvious temporal and spatial heterogeneity. air It is very important for better understanding land surface processes and studying global changes.

[0003] T air It mainly relies on routine observations from meteorological stations, which can provide high-frequency and high-precision T air , but only discrete T air . Station observation T air It only represents the temperature of a limited area around the weather station. air There is significant spatial heterogeneity, and the stations cannot provide accurate temperature gradient changes over a large range. Spatial interpolation can realize the conversion of point observations to surface scales. Although interpolation methods are constantly developing and improving, limited station observations cannot obtain high-precision spatial distribution of temperature, especially in mountainous areas with complex terrain. The Global Forecast System (GFS) data provided by the National Centers for Environmental Prediction (NCEP) of the United States and the EAR5-Land reanalysis dataset provided by the European Center for Medium-Range Weather Forecasts (ECMWF) have been generated and are available to users. They can provide regional and global gridded T air Data. With station T air In comparison, these T air Products can provide seamless data. Typically, these T air The products perform well at macroscales (e.g. continental or global scales), but their spatial resolution is low (from 0.0625° to 0.25°). air This does not apply to the local scale T air applications, such as urban or mountainous areas.

[0004] With the development of meteorological satellite remote sensing technology, remote sensing data has been widely used inair Estimate. The land surface temperature (LST) sensed by meteorological satellites is related to T air has a strong correlation, so T air Remote sensing estimation of T is mainly based on LST. Meteorological satellites are divided into two types: polar and geostationary orbit satellites. Polar orbit satellites can provide global LST with a spatial resolution of about 1km, but their temporal resolution is poor, and the same observation point can only provide 1-2 observations a day. In contrast, geostationary meteorological satellites have a high temporal resolution (about 5-10 minutes), but their LST spatial resolution is only 2-4km. This makes it currently impossible to estimate T with high temporal and high spatial resolution based on meteorological satellite data. air .

[0005] With the development of relevant scientific research and practical application fields, such as urban heat island research, climate model refinement, and the application of temperature in local areas, higher requirements are placed on the spatial resolution of temperature. How to obtain clear sky T with high temporal and spatial resolution? air Data has become the focus of attention. To improve the spatial resolution of air temperature obtained based on geostationary meteorological satellite data, adding some high spatial resolution ground information that affects the temperature (such as NDVI, DEM) is a possible method. At present, there is no research on high spatial resolution clear-air temperature estimation using geostationary meteorological satellites and other ground-based auxiliary data. Summary of the invention

[0006] The purpose of the present invention is to solve the defects of the above-mentioned prior art and provide a method for estimating near-surface temperature with high spatial resolution under clear sky conditions. Specifically, it is assumed that the difference in near-surface temperature between a high spatial resolution sub-pixel and the low spatial resolution pixel in which it is located is mainly caused by the difference in elevation and vegetation index between the two. On this basis, a high spatial resolution temperature estimation model under clear sky conditions is derived. Low spatial resolution surface temperature, numerical model predicted temperature, elevation, vegetation index, latitude, longitude and time information, as well as high spatial resolution elevation and vegetation index are simultaneously used as inputs to the estimation model, and high-resolution temperature estimation is achieved based on a multi-layer forward neural network.

[0007] The present invention mainly solves the problem of estimating near-ground air temperature with high temporal and high spatial resolution using surface temperature data from geostationary meteorological satellites under clear sky conditions.

[0008] A method for estimating near-surface air temperature with high spatial resolution under clear sky conditions comprises the following steps:

[0009] Step 1. Collect the measured temperature at meteorological stations, low spatial resolution surface temperature from satellite remote sensing, temperature predicted by numerical models, high spatial resolution elevation and vegetation index, and other auxiliary data.

[0010] The other auxiliary data include: latitude, longitude, Julian day and hourly time information of the satellite observation point.

[0011] Step 2. Temporally and spatially match the data collected in step 1 according to the location and observation time of the meteorological station to obtain the measured temperature of the meteorological station, satellite surface temperature, numerical model predicted temperature, elevation and vegetation index at two spatial resolutions and other auxiliary data that match the time and space.

[0012] The temporal and spatial matching includes: selecting the satellite observation data closest to the observation time of the meteorological station according to the observation time. Further, using the nearest neighbor method to extract the surface temperature and longitude and latitude information of the nearest satellite observation pixel. The high spatial resolution elevation and vegetation index are averaged for the satellite observation pixel, and the low spatial resolution elevation and vegetation index corresponding to the nearest satellite pixel are extracted.

[0013] Step 3. Use low spatial resolution surface temperature, numerical model predicted temperature, elevation, vegetation index, latitude, longitude and time information to build a low spatial resolution temperature model based on the neural network model.

[0014] Step 4. Assuming that the difference in near-surface temperature between a high-spatial-resolution sub-pixel and its low-spatial-resolution pixel is mainly caused by the difference in elevation and vegetation index between the two, a temperature difference model between the high-spatial-resolution pixel and its low-spatial-resolution pixel is constructed.

[0015] Step 5. Based on the low spatial resolution temperature model and the temperature difference model between high spatial resolution and low spatial resolution pixels, a high spatial resolution temperature estimation model under clear sky conditions is derived.

[0016] Step 6. Based on the derived high spatial resolution temperature model and the spatiotemporally matched meteorological stations, satellite surface temperature, numerical model predicted temperature, elevation at two spatial resolutions, vegetation index and other auxiliary historical data, a high spatial resolution near-surface temperature estimation model under clear sky conditions is constructed based on a multi-layer forward neural network.

[0017] Furthermore, in the high spatial resolution near-surface air temperature estimation method under clear sky conditions as described above, the high spatial resolution elevation and vegetation index of step 1 are derived from the 30 m elevation of SRTM and the 250 m vegetation index of MODIS, respectively.

[0018] Furthermore, in the high spatial resolution near-surface temperature estimation method under clear sky conditions as described above, the two spatial resolution elevations and vegetation in step 2 include: the low spatial resolution elevation and vegetation index corresponding to the nearest satellite pixel; and the high spatial resolution elevation and vegetation index closest to the meteorological station.

[0019] Considering that the near-surface temperature within the pixel range of high spatial resolution (such as 250m) usually does not change much, it is reasonable to use the temperature of the meteorological station to represent the temperature of the 250m grid point when constructing the high spatial resolution temperature estimation model.

[0020] Further, in the above-mentioned high spatial resolution near-surface temperature estimation method under clear sky conditions, the low spatial resolution temperature model in step 3 includes the following contents:

[0021] T air,低 =f1(LST,DEM,NDVI,T GFS,a ,LAT,LON,JD,hour) (1)

[0022] Where T air,低 Low spatial resolution temperature, LST, DEM, NDVI, T GFS,a , LAT, LON, JD, and hour are low-resolution surface temperature, elevation, vegetation index, numerical model predicted temperature, latitude, longitude, Julian day, and hour, respectively. f1() is a nonlinear temperature estimation function implemented using a neural network.

[0023] Furthermore, in the high spatial resolution near-surface temperature estimation method under clear sky conditions as described above, the temperature difference model between the high spatial resolution pixel and the low spatial resolution pixel in step 4 includes the following contents:

[0024] ΔT air =T air,低 -T air,高 =f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (2)

[0025] Where, ΔT air is the temperature difference between the high spatial resolution pixel and the low spatial resolution pixel, T air,低 and T air,高 Low and high resolution temperatures, DEM 低 and DEM 高 They are low and high resolution elevation, NDVI 低 and NDVI 高 They are low and high resolution vegetation indices respectively. f2() is a nonlinear estimation function implemented using a neural network.

[0026] Further, in the method for estimating near-surface temperature with high spatial resolution under clear sky conditions as described above, step 5 deriving a model for estimating temperature with high spatial resolution under clear sky conditions includes the following contents:

[0027] The high spatial resolution temperature estimation model includes: According to formula (2),

[0028] T air,高 =T air,低 -f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (3)

[0029] Substituting formula (1) into formula (3), the high spatial resolution temperature can be expressed as:

[0030] T air,高 =f1(LST,DEM,NDVI,T GFS,a ,LAT,LON,JD,hour)-f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (4)

[0032] Considering that both f1() and f2() are nonlinear estimation functions and the errors of the two models will be superimposed, the high spatial resolution temperature estimation model can be expressed as:

[0033] T air,高 =f(LST,DEM 低 DEM 高 ,NDVI 低 ,NDVI 高 ,T GFS,a ,LAT,LON,JD,hour) (5)

[0034] Where f() is a nonlinear estimation function implemented using a neural network.

[0035] Beneficial effects of the present invention:

[0036] At present, the high spatial resolution of the surface temperature products of existing geostationary meteorological satellites is mostly 2-4km, so that the high spatial resolution of the near-surface temperature estimated based on the surface temperature products of geostationary meteorological satellites is also limited to 2-4km, which cannot meet the demand for refined near-surface temperature. The high spatial resolution estimation model based on high spatial resolution elevation and vegetation index data proposed in the present invention can increase the estimated near-surface temperature spatial resolution from 2-4km to 250m by introducing high and low spatial resolution elevation and vegetation index into the model.

[0037] The present invention directly realizes the estimation of near-surface temperature with high spatial resolution through a model, rather than first realizing the near-surface temperature estimation and then further improving the spatial resolution through a downscaling model, which effectively reduces the transmission of model errors.

[0038] Compared with the traditional algorithm, the present invention can realize the refined estimation of near-surface temperature based on the geostationary meteorological satellite. In addition, the present invention is also applicable to the near-surface temperature with high spatial resolution of polar orbit meteorological satellite. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a method for estimating near-surface temperature with high spatial resolution under clear sky conditions provided by an embodiment of the present invention;

[0040] Figure 2 A two-dimensional histogram of the temperature estimated by the embodiment of the present invention and the station temperature;

[0041] Figure 3 Graph 1 is a spatial distribution diagram of the root mean square error of the station according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] like Figure 1 As shown, the present invention mainly solves the problem of high spatial resolution near-surface temperature estimation under clear sky conditions based on meteorological satellite data. The embodiment of the present invention is based on the surface temperature data of the Fengyun-4 satellite imager (AGRI) to carry out high spatial resolution temperature estimation in Hunan. The high spatial resolution near-surface temperature estimation under clear sky conditions provided by the embodiment of the present invention includes the following steps:

[0044] Step 1. Collect the measured temperature at meteorological stations in Hunan, the low spatial resolution surface temperature from the Fengyun-4 satellite imager (AGRI), the temperature predicted by the numerical model, the high spatial resolution elevation and vegetation index, and other auxiliary data.

[0045] The other auxiliary data include: latitude, longitude, Julian day and hourly time information of the Fengyun-4 satellite observation point.

[0046] Step 2. Temporally and spatially match the data collected in step 1 according to the location and observation time of the meteorological station to obtain the measured temperature of the meteorological station, the FY-4 AGRI surface temperature, the temperature predicted by the numerical model, the elevation and vegetation index at two spatial resolutions, and other auxiliary data.

[0047] The temporal and spatial matching includes: selecting the satellite observation data closest to the observation time of the meteorological station according to the observation time. Further, using the nearest neighbor method to extract the surface temperature and longitude and latitude information of the Fengyun-4 satellite observation pixel closest to it. The high spatial resolution elevation and vegetation index are averaged for the satellite observation pixels, and the low spatial resolution elevation and vegetation index corresponding to the nearest Fengyun-4 satellite pixel are extracted.

[0048] Step 3. Using the AGRI low spatial resolution surface temperature, numerical model predicted temperature, elevation, vegetation index, latitude, longitude and time information, a low spatial resolution temperature model was constructed based on the neural network model.

[0049] Step 4. Assuming that the difference in near-surface temperature between the high spatial resolution sub-pixel and the AGRI low spatial resolution pixel is mainly caused by the difference in elevation and vegetation index between the two, a temperature difference model between the high spatial resolution and the low spatial resolution pixel is constructed.

[0050] Step 5. Based on the low spatial resolution temperature model and the temperature difference model between high spatial resolution and low spatial resolution pixels, a high spatial resolution temperature estimation model under clear sky conditions is derived.

[0051] Step 6. Based on the derived high spatial resolution temperature model and the spatiotemporally matched meteorological stations, the FY-4 satellite AGRI surface temperature, the numerical model predicted temperature, the elevation and vegetation index at two spatial resolutions and other auxiliary historical data, a high spatial resolution near-surface temperature estimation model under clear sky conditions is constructed based on a multi-layer forward neural network.

[0052] Furthermore, in the high spatial resolution near-surface air temperature estimation method under clear sky conditions as described above, the high spatial resolution elevation and vegetation index of step 1 are derived from the 30 m elevation of SRTM and the 250 m vegetation index of MODIS, respectively.

[0053] Furthermore, in the high spatial resolution near-surface temperature estimation method under clear sky conditions as described above, the two spatial resolution elevations and vegetation in step 2 include: the low spatial resolution elevation and vegetation index corresponding to the nearest Fengyun-4 satellite pixel; and the high spatial resolution elevation and vegetation index nearest to the meteorological station.

[0054] Considering that the near-surface temperature within the pixel range of high spatial resolution (such as 250m) usually does not change much, it is reasonable to use the temperature of the meteorological station to represent the temperature of the 250m grid point when constructing the high spatial resolution temperature estimation model.

[0055] Further, in the above-mentioned high spatial resolution near-surface temperature estimation method under clear sky conditions, the low spatial resolution temperature model in step 3 includes the following contents:

[0056] T air,低 =f1(LST,DEM,NDVI,T GFS,a ,LAT,LON,JD,hour) (1)

[0057] Where T air,低 Low spatial resolution temperature, LST, DEM, NDVI, T GFS,a , LAT, LON, JD, and hour are low-resolution surface temperature, elevation, vegetation index, numerical model predicted temperature, latitude, longitude, Julian day, and hour, respectively. f1() is a nonlinear temperature estimation function implemented using a neural network.

[0058] Furthermore, in the high spatial resolution near-surface temperature estimation method under clear sky conditions as described above, the temperature difference model between the high spatial resolution pixel and the low spatial resolution pixel in step 4 includes the following contents:

[0059] ΔT air =T air,低 -T air,高 =f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (2)

[0060] Where, ΔT air is the temperature difference between the high spatial resolution pixel and the low spatial resolution pixel, T air,低 and T air,高 Low and high resolution temperatures, DEM 低 and DEM 高 They are low and high resolution elevation, NDVI 低 and NDVI 高 They are low and high resolution vegetation indices respectively. f2() is a nonlinear estimation function implemented using a neural network.

[0061] Further, in the method for estimating near-surface temperature with high spatial resolution under clear sky conditions as described above, step 5 deriving a model for estimating temperature with high spatial resolution under clear sky conditions includes the following contents:

[0062] The high spatial resolution temperature estimation model includes: According to formula (2),

[0063] T air,高 =T air,低 -f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (3)

[0064] Substituting formula (1) into formula (3), the high spatial resolution temperature can be expressed as:

[0065] T air,高 =f1(LST,DEM,NDVI,T GFS,a ,LAT,LON,JD,hour)-f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 )(4)

[0066] Considering that both f1() and f2() are nonlinear estimation functions and the errors of the two models will be superimposed, the high spatial resolution temperature estimation model can be expressed as:

[0067] T air,高 =f(LST,DEM 低 DEM 高 ,NDVI 低 ,NDVI 高 ,T GFS,a ,LAT,LON,JD,horu) (5)

[0068] Where f() is a nonlinear estimation function implemented using a neural network.

[0069] like Figure 2 As shown in Figure 2, the root mean square error of the estimated high spatial resolution temperature and the meteorological station is less than 1.5°C, which is better than the accuracy reported in the existing literature. Figure 3 As shown, it can be seen that the RMS error is mainly distributed between 1.0-1.75°C, which shows that the algorithm has good applicability in different regions.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating near-surface temperature with high spatial resolution under clear sky conditions, characterized in that: The method comprises: Step 1. Collect the measured temperature at meteorological stations, low spatial resolution surface temperature from satellite remote sensing, temperature predicted by numerical models, high spatial resolution elevation, vegetation index, and other auxiliary data; The other auxiliary data include: latitude, longitude, Julian day and hour time information of the satellite observation point; Step 2. Temporally and spatially match the data collected in step 1 according to the location and observation time of the meteorological station, and obtain the measured temperature of the meteorological station with temporal and spatial matching, the low spatial resolution surface temperature of satellite remote sensing, the temperature predicted by the numerical model, the elevation of two spatial resolutions, the vegetation index, and other auxiliary data; The time and space matching includes: selecting the satellite observation data closest to the observation time of the meteorological station according to the observation time, extracting the surface temperature and longitude and latitude information of the satellite observation pixel closest to the observation time by using the nearest neighbor method, averaging the high spatial resolution elevation and vegetation index for the satellite observation pixel, and extracting the low spatial resolution elevation and vegetation index corresponding to the nearest satellite pixel; Step 3. Using low spatial resolution surface temperature, numerical model predicted temperature, elevation, vegetation index, latitude, longitude and time information, a low spatial resolution temperature model is constructed based on a neural network model; Step 4. Assuming that the difference in near-surface temperature between the high spatial resolution sub-pixel and the low spatial resolution pixel is mainly caused by the difference in elevation and vegetation index between the two, a temperature difference model between the high spatial resolution and the low spatial resolution pixel is constructed; Step 5. Based on the low spatial resolution temperature model and the temperature difference model between high spatial resolution and low spatial resolution pixels, a high spatial resolution temperature estimation model under clear sky conditions is derived; Step 6. Based on the derived high spatial resolution temperature model and the spatiotemporally matched meteorological stations, satellite surface temperature, numerical model predicted temperature, elevation at two spatial resolutions, vegetation index and other auxiliary historical data, a high spatial resolution near-surface temperature estimation model under clear sky conditions is constructed based on a multi-layer forward neural network.

2. The method for estimating near-surface temperature with high spatial resolution under clear sky conditions according to claim 1, characterized in that: The high spatial resolution elevation and vegetation index of step 1 are respectively derived from the 30m elevation of SRTM and the 250m vegetation index of MODIS.

3. The method for estimating near-surface temperature with high spatial resolution under clear sky conditions according to claim 1, characterized in that: The two spatial resolution elevations and vegetation indices in step 2 are respectively: low spatial resolution elevation and vegetation index corresponding to the nearest satellite pixel; High spatial resolution elevation and vegetation index closest to the meteorological station; Considering that the near-surface temperature within the high spatial resolution pixel range usually does not change significantly, it is reasonable to use the temperature of the meteorological station to represent the temperature at the 250 m grid point when constructing a high spatial resolution temperature estimation model.

4. The method for estimating near-surface temperature with high spatial resolution under clear sky conditions according to claim 1, characterized in that: The low spatial resolution temperature model in step 3 is: T air,低 =f1(LST,DEM,NDVI,T GFS,a ,LAT,LON,JD,hour) (1) Where T air,低 Low spatial resolution temperature, LST, DEM, NDVI, T GFS,a , LAT, LON, JD, and hour are low-resolution surface temperature, elevation, vegetation index, numerical model predicted temperature, latitude, longitude, Julian day, and hour, respectively. f1() is a nonlinear temperature estimation function implemented using a neural network.

5. The method for estimating near-surface temperature with high spatial resolution under clear sky conditions according to claim 4, characterized in that: The temperature difference model between the high spatial resolution pixel and the low spatial resolution pixel in step 4 is: ΔT air =T air,低 -T air,高 =f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (2) Where, ΔT air is the temperature difference between the high spatial resolution pixel and the low spatial resolution pixel, T air,低 and T air,高 Low and high resolution temperatures, DEM 低 and DEM 高 They are low and high resolution elevation, NDVI 低 and NDVI 高 are low and high resolution vegetation indices respectively. f2() is a nonlinear estimation function implemented using a neural network.

6. The method for estimating near-surface air temperature with high spatial resolution under clear sky conditions according to claim 5, characterized in that: The high spatial resolution temperature estimation model derived in step 5 under clear sky conditions is: The high spatial resolution temperature estimation model is: According to formula (2), T air,高 =T air,低 -f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 ) (3) Substituting formula (1) into formula (3), the high spatial resolution temperature is expressed as: T air,高 =f1(LST,DEM,NDVI,T GFS,a ,LAT,LON,JD,hour)-f2(DEM 低 ,DEM 高 ,NDVI 低 ,NDVI 高 )(4) Considering that both f1() and f2() are nonlinear estimation functions and the errors of the two models will be superimposed, the high spatial resolution temperature estimation model is expressed as: T air,高 =f(LST,DEN 低 DEN 高 ,NDVI 低 ,NDVI 高 ,T GFS,a ,LAT,LON,JD,hour) (5) Where f() is a nonlinear estimation function implemented using a neural network.

Images