Spatial Neighborhood Constraint-Based Collaborative Normalization Method for Land Surface Temperature and Angle

Abstract

The present invention discloses a method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraints. Using multi-temporal multi-pixel multi-angle land surface temperature observation values as inputs, a time-angle collaborative normalization model that couples an angle effect kernel-driven model and a diurnal variation curve model of land surface temperature with time effect is adopted. The least squares mathematical method is used to determine the parameters of the time-angle collaborative normalization model, and spatial neighborhood information is added. Finally, the number of unknowns of the land surface temperature is reduced, the ill-conditioned problem is solved, and the collaborative normalization of the land surface temperature product is realized; moreover, the time-angle collaborative normalization of the land surface temperature product can be realized with a small number of observations.

Description

Technical Field

[0001] The present invention relates to a method for collaborative normalization of land surface temperature and angle, and particularly to a method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint. Background Art

[0002] Land surface temperature is an important physical quantity characterizing the interaction and energy exchange between the earth and the atmosphere, and has wide applications in fields such as regional climate change and ecological environment monitoring. The acquisition of high-precision land surface temperature depends on thermal infrared remote sensing technology. However, existing land surface temperature remote sensing products have uncertainties in time and directionality. On the one hand, due to the inherent observation and scanning characteristics of polar orbiting satellites, the local observation time differences between different polar orbiting satellite data and on the east-west scan lines of the same satellite data are relatively large, which in turn leads to the lack of consistency among different land surface temperature products; on the other hand, due to the complex and diverse land surface types and their three-dimensional structural characteristics, there may be significant differences in the land surface temperature products obtained at different observation angles. These uncertainties severely limit the wide application of land surface temperature products.

[0003] Existing time normalization methods mainly use temperature diurnal variation models or empirical statistical models to describe the time variation characteristics of thermal radiation; angle normalization is achieved through physical or parametric models of emissivity directionality or the change of component proportions within the field of view. However, these methods usually can only process time or angle factors separately, ignoring the synergistic effect between the two, resulting in limited accuracy of the normalization results. Summary of the Invention

[0004] In order to solve the deficiencies of the above technologies, the present invention provides a method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint, and the method for collaborative normalization of land surface temperature and angle is as follows:

[0006] Using the land surface temperature observation values of multi-temporal multi-pixel multi-angle as input, coupling the time-angle collaborative normalization model of the angle effect kernel-driven model and the land surface temperature diurnal variation curve model of the time effect, and using the least squares mathematical method to determine the parameters of the time-angle collaborative normalization model, adding spatial neighborhood information, and finally realizing the collaborative normalization of land surface temperature and time-angle to produce time-angle collaborative normalized land surface temperature products.

[0007] Preferably, it specifically includes the following steps:

[0008] Step 1: Multi-temporal, multi-angle and multi-pixel data extraction: According to the spatio-temporal range of the area to be obtained, extract the land surface temperature data, and the corresponding solar zenith angle, observation zenith angle, solar azimuth angle, and observation azimuth angle data; and extract the imaging time and convert the imaging time into decimal time;

[0009] Step 2: Parametric expression of the land surface temperature time-angle collaborative model: including parametric expression of angular effect, parametric expression of time effect, and parametric expression of time-angle collaboration;

[0010] Step 3: Introduce neighborhood information to calibrate the parameters of the land surface temperature time-angle collaborative model;

[0011] Step 4: After parameter solving, a normalized product of land surface temperature at any time and any observation angle can be obtained for each pixel.

[0012] Preferably, in Step 2, the parametric expression of the angular effect is for the land surface temperature measured by the sun-sensor geometry, expressed as a linear combination of kernels, as shown in formulas (1)-(3):

[0013]

[0014] K geo (θ v )=1 - cos(θ v ) (2)

[0015]

[0016] where θ s , θ v represent the solar zenith angle and the sensor observation zenith angle respectively, represents the difference between the solar azimuth angle and the observation azimuth angle; K geo (θ v ) is the geometric optics kernel, representing the influence of the three-dimensional structure; K solar is the solar kernel, representing the influence of the sun-sensor observation geometry, and K solar = 0 at night due to no solar illumination; f geo , f solar represent the weights of each kernel respectively; T N is the land surface temperature of nadir observation.

[0017] Preferably, in Step 2, for the parametric expression of the time effect, a cosine function based on the heat diffusion equation is used to model the diurnal temperature change, and an exponential function based on Newton's cooling is used to model the nocturnal temperature decay, which is called the DTC model, and the functional relationship is shown in formula (4):

[0018]

[0019] where t is time, T0 is the residual temperature around sunrise, T a is the temperature amplitude, ω is the half-period width of the cosine term and can be regarded as a constant; t m is the time when the temperature reaches the maximum value, t s is the starting time of free decay, δT is the temperature difference between T0 and T(t→∞), and k is the decay constant;

[0020] T day (t) is the diurnal variation of the surface temperature, and T night (t) is the nocturnal variation of the surface temperature.

[0021] Preferably, in step 2, for the time-angle co-parameterization expression, the surface temperature T N observed at the nadir can be described by the DTC model in time. Substituting formula (4) into formula (1), the time-angle co-model of the surface temperature is expressed as shown in formula (5):

[0022]

[0023] where T N (t) represents the time dynamic variation of the surface temperature observed at the nadir and is expressed as:

[0024]

[0025] In the time-angle co-model, the unknown parameters are: five DTC model parameters, namely T0, T a , t m , t s and δT; and two kernel-driven parameters, namely f geo and f solar .

[0026] Preferably, the specific processing process of step 3 is: setting that the shapes of the DTC models are the same, that is, sharing a set of T0, t m , t s and δT, but the temperature amplitudes T a are different;

[0027] The kernel-driven parameters f geo and f solar of each pixel are different;

[0028] Therefore, the time-angle co-model of the surface temperature is further parameterized as:

[0029]

[0030] where i and j are the row and column numbers of the pixel respectively.

[0031] Preferably, after introducing the neighborhood information, the initial values of the parameters are first determined, and then the optimal solution of the surface temperature time-angle collaborative model parameters is iteratively obtained through a non-linear algorithm. The objective function can be written as the sum of the squares of the observed values and the model predicted values, as shown in Equation (8):

[0032]

[0033] where T obs,i,j,t is the observed value, and T model,i,j,t is the model predicted value of Equation (7);

[0034] By minimizing the objective function, the unknown parameters are iterated, including T0, T a , t m , t s and δT, as well as the pixel-related parameters T a,i,j , f geo,i,j and f solar,i,j , and finally the optimal model parameters are obtained.

[0035] Preferably, in Step 4, the production of the time-angle collaborative normalized surface temperature product is achieved through Equation (9):

[0036]

[0037] The present invention discloses a method for spatial neighborhood constraint and surface temperature and angle collaborative normalization. Through the spatio-temporal and angular information of remote sensing data, the surface temperature at the local observation time is normalized to the same time, and the angular effect is quantified to normalize the surface temperatures at different observation angles to a unified angle. This method will effectively improve the accuracy and consistency of surface temperature products and provide unified standard data support for research such as climate change and vegetation monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic diagram of the overall technical process of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0039] The present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0040] The method for spatial neighborhood constraint and surface temperature and angle collaborative normalization proposed by the present invention uses the surface temperature observation values of multi-temporal, multi-pixel, and multi-angle as inputs, couples the angular effect kernel-driven model and the surface temperature diurnal variation curve model of the time effect to form a time-angle collaborative normalization model, adds spatial neighborhood information, and uses the least square mathematical method to determine the parameters of the time-angle collaborative normalization model, and finally realizes the collaborative normalization of surface temperature and time-angle to produce a time-angle collaborative normalized surface temperature product.

[0041] The overall technical flow chart of the present invention is as Figure 1 shown, and it is mainly divided into the following four steps:

[0042] 1) Multi-temporal, multi-angle and multi-pixel data extraction;

[0043] 2) Parametric expression of the surface temperature time-angle collaborative model;

[0044] 3) Parameter calibration of the surface temperature time-angle collaborative model;

[0045] 4) Production of surface temperature and time-angle normalized products.

[0046] For the method of spatial neighborhood constraint and surface temperature and angle collaborative normalization proposed by the present invention, it mainly aims at the MODIS or VIIRS sensor data that can form 4 observations within a day. The specific implementation steps are as follows:

[0047] Step 1: Multi-temporal, multi-angle and multi-pixel data extraction:

[0048] According to the spatio-temporal range of the area to be obtained, extract the surface temperature data, and its corresponding solar zenith angle, observation zenith angle, solar azimuth angle, and observation azimuth angle data; and extract the imaging time and convert the imaging time into decimal time.

[0049] Taking MODIS as an example, the specific extraction process of multi-temporal, multi-angle and multi-pixel data is as follows: According to the spatio-temporal range of the area to be obtained, download the L2-level surface temperature products MOD11A1 and MYD11A1 of the Moderate Resolution Imaging Spectroradiometer (MODIS) carried on the Terra or Aqua satellite accordingly (on the same day). Extract the surface temperature data of 4 moments of MODIS from the MOD11A1 and MYD11A1 products in HDF format, and its corresponding solar zenith angle, observation zenith angle, solar azimuth angle, and observation azimuth angle data. In addition, read the accurate imaging time from the HDF file and convert it into decimal time. The 4 observation times are around 1:30 am, 10:30 am, 13:30 pm, and 22:30 pm respectively.

[0050] Step 2: Parametric expression of the surface temperature time-angle collaborative model:

[0051] It includes parametric expression of angle effect, parametric expression of time effect, and parametric expression of time-angle collaboration.

[0052] Among them:

[0053] (1) Parametric expression of angle effect

[0054] Under natural conditions, the land surface at the satellite pixel scale is always heterogeneous and consists of multiple homogeneous components with different temperatures. The land surface temperature measured for a specific sun-sensor geometry can be expressed as a linear combination of kernels, as shown in equations (1)-(3):

[0055]

[0056] K geo (θ v )=1 - cos(θ v ) (2)

[0057]

[0058] where θ s , θ v represent the solar zenith angle and the sensor viewing zenith angle, respectively, represents the difference between the solar azimuth angle and the viewing azimuth angle; K geo (θ v ) is the geometric optics kernel, representing the influence of the three-dimensional structure; K solar is the solar kernel, representing the influence of the sun-sensor viewing geometry, and K solar = 0 at night due to the absence of solar illumination; f geo , f solar represent the weights of each kernel respectively; T N is the land surface temperature of nadir observation.

[0059] There are three free parameters for modeling the angular effect, including the weights of each kernel (f geo and f solar ) and the land surface temperature T N of nadir observation.

[0060] (2) Parametric expression of the time effect

[0061] For the diurnal variation of the land surface temperature under clear-sky conditions, the cosine function based on the heat diffusion equation can be used to model the diurnal temperature variation, and the exponential function based on Newton's cooling can be used to model the decay of the temperature at night. This is called the DTC model, and the functional relationship is shown in equation (4):

[0062]

[0063] where t is time, T0 is the residual temperature around sunrise, T a is the temperature amplitude, ω is the half-period width of the cosine term, which can be regarded as a constant; t m is the time when the temperature reaches the maximum value, t s is the starting time of free decay, δT is the temperature difference between T0 and T(t→∞), and k is the decay constant;

[0064] T day (t) is the diurnal variation of the surface temperature, T night (t) is the nocturnal variation of the surface temperature.

[0065] Therefore, five free parameters (i.e., T0 、T a 、t m 、t s and δT) are used to describe the diurnal and nocturnal variations of the surface temperature.

[0066] (3) Parametric expression of time-angle coordination

[0067] For the kernel-driven model describing the angular effect, the surface temperature T observed at the nadir N is a time-varying quantity. Therefore, T N can be described in time by the DTC model. Substituting Equation (4) into Equation (1), the time-angle coordination model of the surface temperature can be written as:

[0068]

[0069] where, T N (t) represents the time dynamic variation of the surface temperature observed at the nadir and is expressed as:

[0070]

[0071] In the time-angle coordination model, there are a total of five DTC model parameters (i.e., T0, T a 、t m 、t s and δT) and two kernel-driven parameters (f geo and f solar ) that are unknown.

[0072] Step 3: Introduce neighborhood information to calibrate the parameters of the surface temperature time-angle coordination model:

[0073] For MODIS with only 4 observations per day, it is impossible to solve the 7 parameters of the time-angle coordination model (i.e., T0, T a 、t m 、t s and δT, as well as f geo and f solar ) for each pixel. Therefore, it is necessary to reduce the number of unknowns by introducing neighborhood information.

[0074] Specifically, within a 3×3 window, it is reasonably assumed that the shape of the DTC model is the same, i.e., sharing a set of T0, t m 、t sand δT, but the energy balance states of each pixel are different, so the maximum temperatures that can be reached are different, that is, the temperature amplitude T a is different.

[0075] In addition, there are certain differences in the three-dimensional structures of each pixel, so the nuclear driving parameters (f geo and f solar ) of each pixel are different. Therefore, the surface temperature time-angle synergistic model can be further parameterized as:

[0076]

[0077] where i and j are the row and column numbers of the pixel respectively.

[0078] After introducing neighborhood information, there are a total of 4 + 3×3×3 unknowns, but there are 3×3×4 observations, overcoming the problem of ill-conditioned solution. These unknowns can be solved by non-linear optimization methods, but non-linear solution methods are often sensitive to the initial values, so the initial values of the parameters need to be determined first.

[0079] For the DTC model, the initial value of T0 is set to the average surface temperature observed at 1:30 am in step 1; the initial value of T a for each pixel is set to the difference between the surface temperature observed at 1:30 pm and 1:30 am; the initial values of t m and t s are set to 12.5 and 17 respectively; δT is set to 0.5. For the nuclear driving model, the initial values of the nuclear weights f geo and f solar for each pixel are both set to 0.

[0080] After determining the initial values of the parameters, the parameters in the surface temperature time-angle synergistic normalization model can be iteratively obtained by non-linear algorithms to find the optimal solution of the model parameters. The objective function can be written as the sum of the squares of the observed values and the model predicted values, as shown in formula (8):

[0081]

[0082] where T obs,i,j,t is the observed value, and T model,i,j,t is the model predicted value of formula (7).

[0083] By minimizing the objective function and iterating the unknown parameters, including the shared parameters T0, T a , t m , t s and δT within the window, and the pixel-related parameters T a,i,j , f geo,i,j and f solar,i,j , the optimal model parameters are finally obtained.

[0084] Step 4: Production of land surface temperature time-angle normalization product:

[0085] After parameter solution, the land surface temperature normalization product at any time and any observation angle can be obtained for each pixel, realizing the collaborative normalization of time-angle:

[0086]

[0087] It can be seen from this that the proposed method for collaborative normalization of land surface temperature and time-angle based on spatial neighborhood information constraint couples the kernel-driven model describing the angular effect of land surface temperature and the diurnal variation curve model of land surface temperature describing time variation, collaboratively models the time-angle, introduces spatial neighborhood information, reduces the number of unknowns, solves the ill-conditioned problem, and realizes the collaborative normalization of land surface temperature products. Using the normalization method provided by the present invention, the time-angle collaborative normalization of land surface temperature products can be realized with a small number of observations.

[0088] The above embodiments are not limitations on the present invention, and the present invention is not limited to the above examples either. Changes, modifications, additions or substitutions made by those skilled in the art within the scope of the technical solution of the present invention also fall within the protection scope of the present invention.

Claims

1. A method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraints, characterized in that: The method for collaborative normalization of land surface temperature and angle is as follows: Using the multi-temporal, multi-pixel, multi-angle land surface temperature observations as input, a time-angle collaborative normalization model that couples the angle effect kernel-driven model and the diurnal variation curve model of land surface temperature with time effect is adopted. The least squares mathematical method is used to determine the parameters of the time-angle collaborative normalization model, and spatial neighborhood information is added to finally achieve the collaborative normalization of land surface temperature and time-angle, so as to produce the time-angle collaborative normalized land surface temperature product; Specifically, it includes the following steps: Step 1: Extraction of multi-temporal, multi-angle, multi-pixel data: According to the spatio-temporal range of the area to be obtained, extract the land surface temperature data, and its corresponding solar zenith angle, observation zenith angle, solar azimuth angle, and observation azimuth angle data; and extract the imaging time and convert the imaging time into decimal time; Step 2: Parametric expression of the land surface temperature time-angle collaborative model: It includes the parametric expression of the angle effect, the parametric expression of the time effect, and the time-angle collaborative parametric expression; Step 3: Introduce neighborhood information to calibrate the parameters of the land surface temperature time-angle collaborative model; Step 4: After the parameters are solved, the normalized land surface temperature product at any time and any observation angle can be obtained for each pixel; In Step 2, for the parametric expression of the time-angle coordination, the surface temperature of nadir observation can be described by the DTC model in terms of time. Substituting the parametric expression formula of the time effect into the parametric expression formula of the angle effect, the time-angle coordination model of the surface temperature is shown in Formula (5) as follows: （5）， Among them, represents the temporal dynamic variation of the surface temperature observed at the nadir and is expressed as: （6）， In the time-angle coordination model, the unknown parameters are: five DTC model parameters, namely , , , and ; and two core drive parameters, namely and ; The specific processing procedure of Step 3 is as follows: It is assumed that the shapes of the DTC models are the same, that is, a group of , , and are shared, but the temperature amplitudes are different. The kernel driving parameter of each pixel and are different; Therefore, the land surface temperature time-angle collaborative model is further parameterized as: （7）， Among them, i , j are the pixel row and column numbers respectively.

2. The method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint according to claim 1, wherein: In Step 2, the parametric expression of the angle effect is for the land surface temperature measured by the sun-sensor geometry, expressed as a linear combination of kernels. The parametric expression formula of the angle effect is shown in Formulas (1)-(3): （1）， （2）， （3）， Among them, and represent the solar zenith angle and the sensor viewing zenith angle respectively, represents the difference between the solar azimuth angle and the viewing azimuth angle; is the geometric optics kernel, representing the influence of the three-dimensional structure; is the solar kernel, representing the influence of the solar-sensor viewing geometry. At night, due to no solar illumination ; and represent the weights of each kernel respectively; is the surface temperature of nadir observation.

3. The method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint according to claim 2, wherein: In Step 2, for the parametric expression of the time effect, the cosine function based on the heat diffusion equation is used to model the diurnal temperature change, and the exponential function based on Newton's cooling is used to model the decay of the night temperature. It is called the DTC model. The parametric expression formula of the time effect is shown in Formula (4): （4）， Among them, t is the time, is the residual temperature around sunrise, is the temperature amplitude, is the half-period width of the cosine term and can be regarded as a constant; is the time when the temperature reaches the maximum value, is the starting time of free decay, is and the temperature difference between, is the decay constant; is the diurnal variation of the surface temperature, is the nocturnal variation of the surface temperature.

4. The method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint according to Claim 3, wherein: After introducing the neighborhood information, first determine the initial value of the parameters, and then use the non-linear algorithm to iteratively obtain the optimal solution of the land surface temperature time-angle collaborative model parameters. The objective function can be written as the sum of the squares of the observed values and the model predicted values, as shown in Formula (8): （8）， wherein, is the observed value, is the model predicted value of Equation (7); By minimizing the objective function, iterating the unknown parameters, including , , , and , as well as the cell-related parameters , and , the optimal model parameters are finally obtained.

5. The method for collaborative normalization of land surface temperature and angle based on spatial neighborhood constraint according to claim 4, wherein: In Step 4, the production of the time-angle collaborative normalized land surface temperature product is achieved through Formula (9): （9）。

Images