A method of generating near real-time hourly all-weather land surface temperatures

Abstract

The application discloses a kind of generation near real-time hourly all-weather ground temperature method, belong to ground temperature acquisition technical field.The application will near real-time hourly all-weather ground estimation be divided into time fusion and spatial fusion two parts, realize the space-time fusion of two kinds of ground temperature.For the target moment of remote sensing ground temperature serious missing, in time scale, the more accurate remote sensing ground temperature of certain clear sky reference time before target time is initial quantity, using reanalysis data provides the ground temperature variation between target time and clear sky reference time to estimate the near real-time all-weather ground temperature of target time;Since it needs to meet many conditions to execute time fusion algorithm, part of pixel is difficult to meet, when using machine learning model to map the correlation between time fusion all-weather ground temperature and influence factor in space, and the relationship is used for missing pixel, obtains the near real-time hourly all-weather ground temperature of final spatial continuity.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of land surface temperature acquisition, and particularly relates to a method for generating near-real-time hourly all-weather land surface temperature. BACKGROUND

[0002] Thermal infrared remote sensing is the only effective means to obtain high-precision and large-area land surface temperature. The surface thermal radiation cannot penetrate the cloud layer, and the thermal infrared remote sensing land surface temperature has a large number of missing values under the cloud, which also seriously hinders the related application process of the land surface temperature. Filling the spatial missing values of the thermal infrared land surface temperature to obtain spatially continuous all-weather land surface temperature data is an effective method to solve this problem. The most effective method to obtain all-weather land surface temperature at present is multi-source data integration. The multi-source data includes thermal infrared data, passive microwave data and reanalysis data. However, due to the observation limitation of passive microwave, no satellite has realized high-frequency observation (time resolution of 1 hour or more) of passive microwave at present. Therefore, it is impossible to obtain hourly all-weather land surface temperature by integrating thermal infrared and passive microwave. The thermal infrared data based on geostationary satellite and the reanalysis data both have high time resolution properties, and the integration of the two can realize high time resolution all-weather land surface temperature acquisition.

[0003] Most of the studies on the integration of reanalysis data and thermal infrared land surface temperature data are based on polar orbit satellite data. The methods in these studies are difficult to be directly applied to high time resolution all-weather land surface temperature estimation based on geostationary satellite. In the latest research, a small number of studies involve the estimation of high time resolution all-weather land surface temperature, but these studies are all based on historical data to reconstruct historical (past) all-weather land surface temperature, and few studies involve the estimation of near-real-time all-weather land surface temperature.

[0004] The reasons why the existing all-weather land surface temperature estimation methods cannot be used for near-real-time hourly all-weather land surface temperature estimation are as follows:

[0005] (1) Most of the existing studies integrate reanalysis data and polar orbit satellite thermal infrared data, without considering the high time resolution property of geostationary orbit satellite thermal infrared land surface temperature. The thermal infrared remote sensing land surface temperature at different times of a day has different degrees of missing values. The evapotranspiration is strong at high temperature, the probability of cloud cover of the surface is high, and the missing value proportion is high; while at low temperature, the evapotranspiration of the surface is relatively weak, the probability of cloud cover of the surface is low, and the missing value proportion is relatively low. The land surface temperature estimation based on polar orbit satellite does not consider the missing value difference at different times.

[0006] (2) There is a significant difference in available data between historical reconstruction and near real-time estimation. Historical reconstruction is to estimate the all-weather surface temperature at past time, without time requirement, so historical reconstruction can use remote sensing observation data and reanalysis data before and after the target time and at the target time as input data, and usually takes a natural year as the time scale for historical reconstruction. While near real-time estimation has a higher time requirement, usually requiring to estimate the all-weather surface temperature at the current time as soon as possible. Therefore, the current time is considered as the latest time when data can be obtained, that is, as long as the data before and at the current time can be used to estimate the near real-time all-weather surface temperature at the current time. In the process of near real-time estimation, data after the current time cannot be used to constrain the surface temperature at the current time. SUMMARY

[0007] The present application provides a method for generating near real-time hourly all-weather surface temperature, which uses near real-time or real-time available reanalysis data and remote sensing data to obtain near real-time high temporal resolution all-weather surface temperature with higher accuracy and image quality through spatio-temporal fusion.

[0008] The technical scheme adopted by the present application is as follows:

[0009] A method for generating near real-time hourly all-weather surface temperature, the method comprising the following steps:

[0010] Step 1, obtaining remote sensing surface temperature data and reanalysis data of a target area, and unifying the spatio-temporal range and resolution of the obtained data;

[0011] Step 2, determining similar pixels:

[0012] Determine all pixels contained in the moving window according to the position of the target pixel and the size of the moving window;

[0013] According to the formula , calculate the correlation coefficient r i0 between the remote sensing surface temperature sequence of the i-th pixel in the current moving window and the target pixel 0 , E[] represents mathematical expectation, D[] represents variance, represents the surface temperature in the remote sensing surface temperature data of the i-th pixel in the current moving window, represents the surface temperature in the remote sensing surface temperature data of the target pixel 0, that is,

[0014] If the correlation coefficient r i0 is greater than or equal to a specified correlation coefficient threshold (preferably set to 0.8), it means that the i-th pixel is a similar pixel of the target pixel;

[0015] And set the surface type of similar pixels to be consistent with the target pixel;

[0016] Step 3: Calculate the transformation coefficient 'a' for each similar pixel based on the remotely sensed land surface temperature before the target time t0 (the time corresponding to the target pixel), the land surface temperature in the reanalysis data, and the Normalized Difference Vegetation Index (NDVI). i and weight W i :

[0017] Based on the conversion relationship between remotely sensed land surface temperature and reanalysis data Calculate the conversion factor a i ;in, Let x represent the surface temperature from remote sensing data and reanalysis data, respectively. i ,y i ) represents the position of similar pixel i, d i This represents the time t corresponding to similar pixel i. i b i Indicates the second conversion factor;

[0018] Calculate weight W i :

[0019]

[0020] Q i =(1-r i )×S i0

[0021]

[0022]

[0023] Where, r i This represents the remotely sensed surface temperature of similar pixel i before the target time t0. Land surface temperature compared with reanalysis data The correlation coefficient between them, S i0 Q represents the distance weight between similar pixel i and target pixel 0, R represents the size of the moving window (the size of the moving window is R×R), and Q represents the distance weight between similar pixel i and target pixel 0. i These are intermediate calculation parameters;

[0024] Step 4, determine the clear sky reference time for the target time:

[0025] (1) The target pixel is in clear sky at the reference time, and d0 is defined as the number of days in a landmark temperature time series at that reference time;

[0026] (2) Define dp represents a day sequence value of the target time t0 in a landmark temperature time series;

[0027] traverse all d0, when the reference time d p and the difference between d0 is not more than the first threshold value, the current day sequence value is stored in the set Day_clr, wherein the initial value of Day_clr is an empty set;

[0028] (3) if the elements of the set Day_clr do not exceed the second threshold value, the reference time corresponding to the day sequence value with the minimum difference between the remote sensing land surface temperature and the land surface temperature of the reanalysis data in the set Day_clr is taken as the clear sky reference time of the target time;

[0029] if the elements of the set Day_clr exceed the second threshold value, the first K elements in the set Day_clr with the minimum time difference of the day sequence value d p and the reference time corresponding to the day sequence value with the minimum difference between the remote sensing land surface temperature and the land surface temperature of the reanalysis data in the new set are taken as the clear sky reference time of the target time;

[0030] wherein the preferred value of K is the same as the second threshold value;

[0031] Step 5, calculate the time fusion near real-time all-weather land surface temperature:

[0032]

[0033] wherein T s-AW-T (x0,y0,d0,t0) represents the time fusion near real-time all-weather land surface temperature of the target pixel at the target time t0, T s-AW-T (x0,y0,d0,t0) in the subscript "T" represents the number of pixels successfully performing time fusion; d0 is the day sequence value corresponding to the clear sky reference time determined in step 4;

[0034] Step 6, obtain spatially continuous all-weather land surface temperature based on spatial fusion:

[0035] based on the near real-time all-weather land surface temperature calculated in step 5 and the land surface temperature influencing factor Θ T train the first spatial mapping model T s-AW-T (d p ,t0) = RF T (Θ T ) to obtain the spatial fusion model of the target time;

[0036] when training, T s-AW-T (d pthe value of T (d, t0) is the near real-time all-weather land surface temperature calculated in step 5, and RF T () represents the mapping relationship between the near real-time all-weather land surface temperature and the land surface temperature influence factor.

[0037] For the number of pixels that are not successfully executed time fusion, the corresponding near real-time all-weather land surface temperature is obtained based on RF T () to obtain the corresponding near real-time all-weather land surface temperature, thereby obtaining the spatially continuous near real-time all-weather land surface temperature.

[0038] Further, step 6 is replaced by:

[0039] For the remote sensing land surface temperature data of the target region, if the proportion of clear sky pixels in the target region exceeds a specified value (preferably 70%), the second spatial mapping model T s-AW-NRT (d p ,t0) is trained based on the remote sensing land surface temperature data obtained in step 1 and the corresponding land surface temperature influence factor Θ to obtain the spatial fusion model at the target time;

[0040] During training, the value of T s-AW-T (d p ,t0) is the land surface temperature corresponding to the remote sensing land surface temperature data, and RF() represents the mapping relationship between the near real-time all-weather land surface temperature and the land surface temperature influence factor.

[0041] The all-weather land surface temperature influence factor is taken as the input of RF() to obtain the spatially continuous near real-time all-weather land surface temperature.

[0042] The technical solution provided by the present application at least brings the following beneficial effects:

[0043] (1) The present application can realize the fusion of static satellite remote sensing land surface temperature and reanalysis data land surface temperature under the requirement of near real-time;

[0044] (2) The present application realizes the fusion of the two kinds of data from the two-dimensional space-time perspective, has strong operability, and can take into account the spatial missing differences of the target region at different times;

[0045] (3) The present application can obtain high temporal resolution all-weather land surface temperature with precision comparable to remote sensing land surface temperature in near real-time (delaying no more than one hour). BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0047] Figure 1 A processing process schematic diagram of a method for generating near real-time hourly all-weather surface temperature provided by the embodiment of the present application is shown in

[0048] Figure 2 In the embodiment of the present application, the spatial mode comparison of the near real-time all-weather surface temperature estimation result is shown in Figure 2 The left column in FIG. 1 is the FY-4A surface temperature image at 0:00 and 06:00 (UTC) on the 218th and 336th day of 2020, Figure 2 The middle column in FIG. 1 is the corresponding time fusion result, Figure 2 The right column in FIG. 1 is the final near real-time all-weather surface temperature after spatial fusion.

[0049] Figure 3 In the embodiment of the present application, the near real-time hourly all-weather surface temperature verification based on the measured data is shown in FIG. 3, wherein (a) is the measured station verification result of Aru, and (b) is the measured station verification result of Damann. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0051] The method for generating near real-time hourly all-weather surface temperature provided by the embodiment of the present application is a near real-time hourly all-weather surface temperature estimation method based on the integration of the near real-time obtainable geostationary satellite thermal infrared surface temperature and the reanalysis data. The present application is applied to the FY-4A thermal infrared surface temperature product and the CLDAS surface temperature product, and the near real-time (the all-weather surface temperature at the current time is obtained within one hour after the input data is obtained) hourly all-weather surface temperature is reconstructed. The method realized by the present application can be divided into two parts of time fusion and spatial fusion, as shown in FIG. 1. Under the premise that the geostationary satellite remote sensing surface temperature data and the reanalysis data have a stable error relationship in a short time and a small area, the time fusion method can estimate the near real-time all-weather surface temperature of more than 70% of the pixels in the study area without requiring any observation in the study area at the current time. The spatial fusion method can further obtain the spatially completely continuous near real-time all-weather surface temperature on the basis of the time fusion result using a machine learning algorithm. In addition, the spatial fusion algorithm can be independently run, and when the effective value proportion of the thermal infrared remote sensing data in the study area is more than 70%, the spatial fusion algorithm can be directly used to obtain the completely spatially continuous near real-time all-weather surface temperature. Figure 1 The method for generating near real-time hourly all-weather surface temperature provided by the embodiment of the present application can be divided into two parts:

[0052]

[0053] ​I. Time fusion algorithm.

[0054] Remote sensing land surface temperature can provide clear-sky historical land surface temperature before the target time; reanalysis data can provide land surface temperature change information of the target pixel on the time scale. Under the premise that remote sensing land surface temperature and reanalysis land surface temperature have certain comparability and correlation, it is considered that remote sensing land surface temperature and reanalysis land surface temperature have stable error relationship in short period and small area. The relationship between the two kinds of land surface temperature between different days can be expressed as follows:

[0055]

[0056] In the formula, T s is the land surface temperature; (x0, y0) is the target pixel position; d p and d0 are the day sequence values of the reference time and the target time in a land surface temperature time sequence respectively; (x0, y0) is clear-sky in (d0, t0), and needs to meet the series conditions; t0 is a certain whole point time; a and b are the conversion coefficients of remote sensing land surface temperature and reanalysis land surface temperature at the target pixel; the subscripts "-FY4A" and "-CLDAS" respectively represent remote sensing land surface temperature (Fengyun 4A land surface temperature) and reanalysis land surface temperature (China Land Data Assimilation System data). The parameter a can be calculated according to the historical sequence of the two kinds of land surface temperature before the target time, and b is irrelevant to the subsequent estimation process. (b) minus (a) of formula (1) can obtain the following formula:

[0057]

[0058] The second term on the right side of formula (2) has instability on a single pixel. Therefore, the similar pixels around the target pixel are used to improve the stability of formula (2), and formula (2) can be further expressed as:

[0059]

[0060] Wherein, W i and a i are the weight and conversion coefficient of the similar pixel; (x i ,y i ) is the spatial position of the similar pixel; m is the number of similar pixels; (d0, t0) is the target time; T s-AW-T is the near real-time all-weather land surface temperature estimated based on the time and space fusion algorithm; the subscript "T" represents the pixel which successfully executes the time fusion algorithm.

[0061] From formula (3), once the similar pixels of the target pixel are determined, d0 and W iThen the near real-time all-weather land surface temperature of the target pixel can be estimated based on the time fusion algorithm. In order to ensure that the similar pixels are not too far away from the target pixel and meet the certain number requirement, the similar pixels are searched in a moving window with the target pixel as the center, and the size of the moving window is R x R pixels (R can be determined according to the spatial resolution).

[0062] The determination rule of the similar pixels is as follows:

[0063] Firstly, the ground is divided into four types according to the normalized vegetation index (NDVI): dense vegetation (NDVI≥0.5, LC-1); sparse vegetation (0.2≤NDVI<0.5, LC-2); bare land (NDVI<0.2, LC-3); water body (LC-4). The determination rule of the similar pixels is as follows:

[0064]

[0065] In the formula, r i0 is the correlation coefficient between the similar pixel and the target pixel historical FY-4A land surface temperature sequence; T s-FY4A-0 and T s-FY4A-i are the FY-4A land surface temperature vectors of (x0, y0) and (x i , y i ) before the target time; E and D are the expectation and variance functions respectively; LC-k is the ground type. In order to ensure the estimation accuracy and stability, only when the number of similar pixels searched based on formula (4) is greater than 10 (m>10), the T s-AW-T of the target pixel is estimated by using formula (3). When the number of similar pixels does not meet the requirement, the near real-time land surface temperature of the corresponding target pixel will be estimated in the next step.

[0066] After the similar pixels are determined, the determination method of the weight W i of the similar pixels is as follows:

[0067]

[0068] In the formula, r i is the correlation coefficient between the similar pixel and the FY-4A and CLDAS land surface temperatures before the target time; S i0 is the distance weight between the similar pixel and the target pixel; and R is the size of the moving window. After the weight W i is determined, the clear sky reference time (d0, t0) of the target time (d p , t0) needs to be further determined, and d0 needs to meet the following conditions:

[0069] (1) the target pixel is in clear sky at the reference time;

[0070] (2) dp The time difference between d0 and d is not more than 30 days, and the day sequence value satisfying conditions (1) and (2) is recorded as d0 clr ;

[0071] (3) When the number of samples in d clr is not more than 5, select the day in d clr with the minimum FY-4A and CLDAS land surface temperature difference as d0;

[0072] (4) When the number of samples in d clr is more than 5, select the 5 samples in d clr with the minimum time difference from d p as new d clr , and then select d0 according to condition (3).

[0073] The determination process of d0 can be parameterized as follows:

[0074]

[0075] In the formula, d i-clr is the day sequence satisfying conditions (1) and (2); N clr is the number of days satisfying equation (6a). If d0 cannot be determined according to equation (6), the time fusion algorithm is not performed for this target pixel, and the near real-time all-weather land surface temperature of this pixel will be determined in the spatial fusion process.

[0076] II. Spatial fusion algorithm

[0077] The time fusion algorithm can successfully estimate the near real-time all-weather land surface temperature of most pixels, but part of the pixels cannot perform the time fusion algorithm due to not meeting the conditions. In order to obtain a spatially completely continuous all-weather land surface temperature, the spatial fusion algorithm needs to be further performed on the time fusion result or the original remote sensing land surface temperature. The land surface temperature is affected by many influencing factors, such as meteorological factors, surface characteristics, and spatial position, etc. Therefore, these influencing factors can be used to estimate the near real-time all-weather land surface temperature of the pixels that failed to successfully perform the time fusion algorithm. The relationship between the land surface temperature and its influencing factors is mapped using a machine learning algorithm (such as random forest, lightGBM, neural network, convolutional neural network, etc.), and here a random forest is taken as an example for description.

[0078] First, the near real-time all-weather land surface temperature obtained by time fusion and the corresponding influencing factors are used as input parameters to train the spatial mapping model of the near real-time all-weather land surface temperature. The mapping process is as follows:

[0079]

[0080] where lat, lon, DEM, NDVI, T a , NDVI, q, p, a, S, and T s-CLDAS-T are latitude, longitude, elevation, normalized vegetation index, near-surface air temperature, normalized leaf area index, humidity, pressure, surface albedo, wind speed, and land surface temperature in reanalysis data (here, CLDAS is taken as an example), respectively; subscript "T" represents a pixel for which the time fusion algorithm is successfully executed.

[0081] Then, the near-real-time land surface temperature of an arbitrary target pixel can be estimated based on the trained model:

[0082]

[0083] where T s-AW-NRT is the near-real-time all-weather land surface temperature.

[0084] When the proportion of clear-sky pixels of the original geostationary satellite remote sensing land surface temperature is more than 70%, the missing land surface temperature can be directly estimated using the spatial fusion algorithm. At this time, formulas (7) and (8) are replaced by the following forms:

[0085] T s-FY4A (d p ,t0)=RF FY4A (lat FY4A ,lon FY4A ,DEM FY4A ,NDVI FY4A , (9)T a-FY4A ,LAI FY4A ,q FY4A ,p FY4A ,α FY4A ,S FY4A ,T s-CLDAS-FY4A )

[0086] T s-AW-NRT (d p ,t0)=RF FY-4A (lat,lon,DEM,NDVI, (10)T a ,LAI,q,p,α,S,T s-CLDAS )

[0087] where subscript "-FY4A" represents a clear-sky pixel of remote sensing land surface temperature from a geostationary satellite (here, FY-4A land surface temperature data is taken as an example); RF FY4A is a spatial fusion model trained directly based on remote sensing land surface temperature (FY-4A land surface temperature).

[0088] As a possible implementation manner, a specific implementation process of the method for generating near real-time hourly all-weather surface temperature provided by the embodiment comprises the following steps:

[0089] Firstly, data is acquired, and the data acquired in the embodiment mainly comprises FY-4A, reanalysis data produced by CLDAS, main parameters including surface temperature, wind, temperature, humidity, pressure and other surface temperature influencing factors, mainly including longitude, latitude, elevation, surface albedo, normalized vegetation index and the like. Based on the acquired data, the following steps are executed:

[0090] (1) Spatiotemporal range and resolution unification:

[0091] All related data is unified to the same spatial resolution (0.04°) as the remote sensing surface temperature by using weighted average, and then the spatial range is also unified to the same range.

[0092] (2) Similar pixel determination:

[0093] Firstly, all pixels contained in the moving window are determined according to the target pixel position and the moving window size, and then the similar pixels in the moving window are determined according to formula (4).

[0094] (3) Weight and conversion coefficient calculation:

[0095] The conversion coefficient (a i ) and the weight (W i ) of each similar pixel are calculated by using the target time point preceding stationary satellite remote sensing surface temperature (FY-4A surface temperature), reanalysis data surface temperature (CLDAS surface temperature) historical data and NDVI data.

[0096] (4) Determination of clear sky reference time point (d0, t0):

[0097] The clear sky reference time point (d0, t0) of the target time point is determined based on formula (6) by using the historical surface temperature data.

[0098] (5) Calculation of time fusion near real-time all-weather surface temperature:

[0099] The time fusion all-weather surface temperature (T s-AW-T ) of the target pixel is calculated based on formula (3) by using the conversion coefficient (a i ) and the weight (W i ) in step (3), the FY-4A surface temperature and the CLDAS surface temperature in step (4). In the embodiment, the time fusion near real-time all-weather surface temperature is shown in the second column (the middle column) of FIG. 2. Figure 2 Figure 2 ​The first column of the table is the FY-4A land surface temperature image at 0:00 and 6:00 (UTC) on the 218th and 336th day of 2020.

[0100] (6) Spatial fusion to obtain spatially continuous all-weather land surface temperature:

[0101] T s-AW-T or T s-FY4A The target time spatial fusion model is trained by inputting T s-AW-NRT and the land surface temperature influencing factor into formula (7) or (9); then the all-weather land surface temperature influencing factor is input into the spatial fusion model to obtain the spatially continuous near real-time all-weather land surface temperature (T s-AW-NRT ). That is, the all-weather land surface temperature of the pixels that failed to successfully perform the time fusion algorithm (missing pixels in the time fusion stage) is obtained based on the target time spatial fusion model. In this embodiment, the near real-time all-weather land surface temperature is shown in the second column (middle column) of Figure 2 .

[0102] To further verify the performance of the method for generating near real-time hourly all-weather land surface temperature provided in the embodiments of the present application, the measured data of the Arou and Daman regions and the near real-time all-weather land surface temperature obtained based on the present application are verified, and the verification results are shown in Figure 3 . The present application can obtain high temporal resolution all-weather land surface temperature with precision comparable to remote sensing land surface temperature in near real time (delay of no more than one hour).

[0103] The present application aims at the fact that most of the existing historical all-weather ground temperature reconstruction methods need data after the target time as input (usually a complete natural year scale), and do not consider the missing differences of remote sensing ground temperature at different times under high time resolution, and cannot be used to estimate near real-time all-weather ground temperature. After considering the correlation and error between the ground temperature in the reanalysis data and the stationary satellite remote sensing ground temperature, the present application divides the near real-time hourly all-weather ground estimation into two parts of time fusion and space fusion, realizes the spatio-temporal fusion of the two kinds of ground temperature. For the target time with serious missing of remote sensing ground temperature (more than 30% missing in the region), firstly, the relatively accurate remote sensing ground temperature at a certain clear sky reference time before the target time is used as the initial quantity on the time scale, and the reanalysis data is used to provide the ground temperature change between the target time and the clear sky reference time to estimate the near real-time all-weather ground temperature at the target time; since many conditions need to be met to perform the time fusion algorithm, part of the pixels cannot meet the conditions, so the all-weather ground temperature cannot be successfully estimated at the time fusion stage, at this time, the machine learning model is used to map the correlation between the all-weather ground temperature of time fusion (when the clear sky pixels in the target region are more than 70%, the remote sensing ground temperature is used as the input parameter) and the influencing factors, and the relationship is used for the missing pixels to obtain the final spatially continuous near real-time hourly all-weather ground temperature.

[0104] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

[0105] The above only describes some embodiments of the present application. Those skilled in the art can make several modifications and improvements without departing from the inventive concept, and these all belong to the protection scope of the present application.

Claims

1. A method for generating near real-time hourly, all-weather land surface temperature, characterized in that, Includes the following steps: Step 1: Acquire remote sensing surface temperature data and reanalysis data of the target area, and unify the spatiotemporal range and resolution of the acquired data; Step 2, determine similar pixels: Determine all pixels contained in the moving window based on the target pixel location and the moving window size; According to the formula Calculate the correlation coefficient between the i-th pixel in the current moving window and the remotely sensed surface temperature sequence of the target pixel 0. E[ ] represents the expected value, and D[ ] represents the variance. This represents the land surface temperature in the remote sensing land surface temperature data of the i-th pixel within the currently moving window. This represents the surface temperature in the remotely sensed surface temperature data for target pixel 0. If the correlation coefficient If the correlation coefficient is greater than or equal to the specified threshold, it means that the i-th pixel is a similar pixel to the target pixel. And set the surface type of similar pixels to be consistent with the target pixel; Step 3: Calculate the transformation coefficient 'a' for each similar pixel. i and weight W i : Based on the conversion relationship between remotely sensed land surface temperature and reanalysis data Calculate the conversion factor ;in, , These represent the surface temperatures from remotely sensed surface temperature data and reanalysis data, respectively. Indicates the position of similar pixel i. Represents the time corresponding to similar pixel i Day sequence values ​​in a surface temperature time series Indicates the second conversion factor; Calculate weight W i : ； in, Indicates that similar pixel i is at the target time. Pre-remote sensing of surface temperature Land surface temperature compared with reanalysis data The correlation coefficient between them The distance weights represent the distance between similar pixel i and target pixel 0. Indicates the size of the movable window. These are intermediate calculation parameters; Step 4, determine the clear sky reference time for the target time: (1) The target pixel is in clear sky at the reference time, and is defined as follows: This represents the number of days in a local temperature time series for that reference time. (2) Definition Indicates the target time Day sequence values ​​in a landmark temperature time series; Traverse all When the reference time and When the difference between them does not exceed the first threshold, the current day sequence value is stored in the set Day_clr, where the initial value of Day_clr is an empty set; (3) If the elements of the set Day_clr do not exceed the second threshold, the reference time corresponding to the day sequence value with the smallest difference between the remote sensing surface temperature and the surface temperature of the reanalysis data in the set Day_clr shall be used as the clear sky reference time for the target time. If the number of elements in the set Day_clr exceeds the second threshold, then select the Day_clr set whose values ​​match the day sequence. The K elements with the smallest time difference are taken as a new set, and the reference time corresponding to the day sequence value with the smallest difference between the remote sensing surface temperature and the surface temperature of the reanalysis data in the new set is taken as the clear sky reference time of the target time. Step 5: Calculate near real-time, all-weather surface temperature using time-fusion: ； in, Indicates the target pixel at the target time. The time-based fusion of near real-time, all-weather surface temperature The "T" in the subscript indicates the number of pixels successfully fused over time; The number of days corresponding to the clear sky reference time determined in step 4; Step 6: Obtain continuous, all-weather surface temperature based on spatial fusion: Based on the near-real-time all-weather surface temperature and corresponding surface temperature influencing factors calculated in step 5 For the first spatial mapping model Training is performed to obtain a spatial fusion model at the target time. During training The value is the near-real-time all-weather surface temperature calculated in step 5. This represents the mapping relationship between near-real-time all-weather land surface temperature and land surface temperature influencing factors. For the number of pixels for which time fusion was not successfully performed, based on The corresponding near-real-time all-weather surface temperature is obtained, thus obtaining the spatially continuous near-real-time all-weather surface temperature.

2. The method as described in claim 1, characterized in that, Replace step 6 with: For remote sensing surface temperature data of the target area, if the proportion of clear-sky pixels in the target area exceeds a specified value, then based on the remote sensing surface temperature data obtained in step 1 and the corresponding surface temperature influence factor... For the second space mapping model Training is performed to obtain a spatial fusion model at the target time. During training The value is the land surface temperature corresponding to the remotely sensed land surface temperature data. This represents the mapping relationship between near-real-time all-weather land surface temperature and land surface temperature influencing factors. Using the all-weather surface temperature influencing factor as The input is used to obtain the spatially continuous near real-time all-weather surface temperature.

3. The method as described in claim 1 or 2, characterized in that, In step 4, the value of K is the same as the second threshold.

4. The method as described in claim 1 or 2, characterized in that, The land surface types include four categories: Category 1: Dense vegetation with a normalized vegetation index greater than or equal to 0.5; The second category is sparse vegetation, with a normalized vegetation index of less than 0.5 and greater than or equal to 0.

2. The third category is bare land, with a normalized vegetation index of less than 0.2; The fourth category is water bodies.

5. The method as described in claim 1 or 2, characterized in that, Factors influencing land surface temperature include: latitude, longitude, elevation, normalized vegetation index, near-surface air temperature, normalized leaf area index, humidity, pressure, surface albedo, wind speed, and land surface temperature in reanalysis data.

Images