High-temporal-spatial-resolution surface evapotranspiration remote sensing inversion method
Through super-resolution technology and convolutional neural network combined with SEBAL model, the problem of spatial and temporal resolution in the surface evaporation remote sensing inversion is solved, and the remote sensing inversion with high precision and high spatial and temporal resolution is achieved, which improves the inversion accuracy and resolution Rate.
Patent Information
- Application Number
- CN202411007044.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing surface evaporation remote sensing inversion method is difficult to obtain input data that takes into account both high timing and high spatial resolution, resulting in insufficient inversion accuracy.
Super-resolution technology is used to build a high-super-resolution model, combining SEBAL model and convolutional neural network, surface evaporation inversion is performed through high-spatial-temporal resolution surface reflectivity remote sensing data and meteorological data to improve spatial and temporal resolution.
It significantly improves the temporal and spatial resolution of remote sensing inversion on the surface evaporation, combines physical and data-driven methods to improve the inversion accuracy, solves the 'space-time contradiction' of satellite remote sensing data sources, and realizes high-precision and high-space-time remote sensing remote sensing. Inversion.
Smart Images

Figure CN118968332A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of remote sensing inversion of surface evapotranspiration, and in particular to a remote sensing inversion method of surface evapotranspiration with high temporal and spatial resolution. Background Art
[0002] The remote sensing inversion method of surface evapotranspiration is considered to be the most effective method to obtain regional-scale surface evapotranspiration and its temporal and spatial changes with high timeliness and precision. At present, a variety of remote sensing inversion methods of surface evapotranspiration have been developed. According to the driving force of the model, the inversion of surface evapotranspiration is divided into process-driven physical inversion methods and data-driven inversion methods.
[0003] The process-driven physical inversion methods include the energy balance method, the method based on the Penman-Monte i th algorithm, the method based on the SEBAL model, the complementary relationship method, the surface temperature-vegetation index triangle or trapezoidal characteristic space method, etc. These methods have clear physical meanings and can obtain higher surface evapotranspiration inversion accuracy under the premise of high-precision input data.
[0004] The data-driven method is driven by data such as observed flux or evapotranspiration products and their closely related characteristic parameters, and the inversion of evapotranspiration is achieved by establishing the relationship between the data. Compared with the process-driven physical inversion method, the data-driven method does not require the prediction of the physical mechanism of the surface evapotranspiration process, nor is it limited by influencing factors. It can achieve high-precision surface evapotranspiration inversion by simply building the relationship between the acquired remote sensing, meteorological, observed flux and other data.
[0005] Remote sensing inversion of evapotranspiration uses satellite remote sensing technology to obtain short-term, long-term, and large-scale surface characteristic parameters closely related to evapotranspiration, such as soil moisture, surface temperature, net radiation, water vapor pressure, atmospheric pressure, air temperature, wind speed, vegetation index, land use type, etc. However, due to the different application requirements and designs of various satellite sensors, the data scales of the acquired remote sensing data vary greatly, that is, the temporal resolution, spatial resolution, and spectral resolution are different. In addition, due to the limitations of hardware and economic conditions, it is difficult to directly obtain remote sensing images with both high temporal resolution and high spatial resolution. Whether it is a process-driven physical inversion method or a data-driven method, it is difficult to obtain input data with both high temporal and high spatial resolution. Therefore, these two methods are difficult to meet the needs of high-precision and high-temporal surface evapotranspiration inversion. Summary of the invention
[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0007] A high temporal and spatial resolution surface evapotranspiration remote sensing inversion method comprises the following steps:
[0008] Step 1: Prepare high and low spatial resolution surface reflectance remote sensing samples;
[0009] Step 2: Build a high super-resolution model;
[0010] Step 3: Select time series low-resolution surface reflectance remote sensing data;
[0011] Step 4: Use the super-resolution model to perform super-resolution calculations on the time series low-resolution surface reflectance remote sensing data;
[0012] Step 5: Using high temporal and spatial resolution surface reflectance remote sensing data as input, perform surface resolution inversion based on the SEBAL model;
[0013] Step 6: Using the inversion results of the SEBAL model and some meteorological data as input, the convolutional neural network is used to perform surface evapotranspiration inversion;
[0014] Step 7: Output the inversion results.
[0015] A further improvement of the technical solution of the present invention is that: step one is obtained by selecting multispectral surface reflectance remote sensing data of different resolutions of the same date, the same region and the same range, and cutting them according to a specified size. The sample size can be determined according to actual conditions, and the sample selection should be diversified, that is, the types of land objects and seasonal distributions are diverse.
[0016] A further improvement of the technical solution of the present invention is that step two constructs a super-resolution model based on the number of bands, corresponding order and proportional relationship of the high-resolution and low-resolution surface reflectance remote sensing image samples in step one.
[0017] A further improvement of the technical solution of the present invention is that: the step three selects temporal low-resolution surface reflectance remote sensing data within the region and time range to be calculated.
[0018] A further improvement of the technical solution of the present invention is that step four utilizes the super-resolution model constructed in step two, and uses the data of step three as input to perform super-resolution calculations to obtain a high spatial resolution remote sensing image.
[0019] A further improvement of the technical solution of the present invention is that: the step five uses the calculation result of step four as input, combines with reanalysis meteorological data, and inverts high temporal and spatial resolution surface evapotranspiration based on the SEBAL model.
[0020] A further improvement of the technical solution of the present invention is that: the step six is to construct a convolutional neural network model with the calculation result of step five and the net radiation flux and topographic data as input, and use the model to perform high-precision surface evapotranspiration inversion.
[0021] A further improvement of the technical solution of the present invention is that the inversion system is composed of four sub-modules, including a super-resolution model construction module, a SEBAL inversion module, a machine learning model construction module and a machine learning inversion module.
[0022] A further improvement of the technical solution of the present invention is that: the super-resolution model construction module is based on high and low spatial resolution remote sensing samples, adopts super-resolution technology, constructs a multi-spectral remote sensing image super-resolution model, and exports the model after training; the SEBAL inversion module 11 constructs a SEBAL model based on a mathematical formula, and then calls the super-resolution model exported by the super-resolution module construction module to generate high temporal and spatial resolution remote sensing data, reads meteorological parameters at the same time, and then calculates high temporal and spatial resolution surface evapotranspiration based on the SEBAL model; the machine learning model construction module calls the calculation results of the SEBAL inversion module in combination with meteorological factors, builds an evapotranspiration inversion model based on a convolutional neural network, trains and exports the model; the machine learning inversion module calls the model exported by the machine learning model construction module, takes the SEBAL model inversion results and meteorological data as input, performs high-precision surface evapotranspiration inversion, and generates and exports result data.
[0023] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:
[0024] 1. The present invention provides a high-temporal-spatial-resolution surface evapotranspiration remote sensing inversion method. The spatial resolution of satellite images with high temporal resolution and low spatial resolution is greatly improved through super-resolution technology, so that the satellite images have high temporal and spatial resolutions at the same time. The problem of inconsistent temporal resolution and spatial resolution of input data required for continuous evapotranspiration remote sensing inversion is fundamentally solved, thereby greatly improving the temporal and spatial resolution of surface evapotranspiration remote sensing inversion products. At the same time, the inversion based on process-driven physical methods and data-driven methods are combined to further improve the surface evapotranspiration inversion accuracy. The method first constructs a super-resolution model based on deep learning based on the spectral characteristics of surface reflectance remote sensing satellite images to improve satellite remote sensing The spatial resolution of the data can be increased by up to 16 times, thereby solving the "temporal and spatial contradiction" of satellite remote sensing data sources in the remote sensing inversion of surface evapotranspiration; using this data as the main input data, the physical model SEBAL is used to achieve high-temporal and spatial resolution surface evapotranspiration inversion; in order to further improve the inversion accuracy and break through the limitations of the applicability of physical models on a large spatial scale, this study combines physical methods with data-driven methods, takes the inversion results of the SEBAL model as combined with some meteorological data as input features, and uses convolutional neural networks to construct a new surface evapotranspiration machine learning model to further improve the surface evapotranspiration inversion accuracy, thereby achieving high-precision and high-temporal and spatial resolution surface evapotranspiration remote sensing inversion. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A flow chart of the high temporal and spatial resolution surface evapotranspiration remote sensing inversion method of the present invention;
[0026] Figure 2 It is a structural schematic diagram of the inversion system of the present invention. DETAILED DESCRIPTION
[0027] The present invention is further described in detail below in conjunction with embodiments:
[0028] Example 1
[0029] like Figure 1 As shown, the present invention provides a high temporal and spatial resolution surface evapotranspiration remote sensing inversion method, comprising the following steps:
[0030] Step 1: Prepare high and low spatial resolution surface reflectance remote sensing samples;
[0031] Step 2: Build a high super-resolution model;
[0032] Step 3: Select time series low-resolution surface reflectance remote sensing data;
[0033] Step 4: Use the super-resolution model to perform super-resolution calculations on the time series low-resolution surface reflectance remote sensing data;
[0034] Step 5: Using high temporal and spatial resolution surface reflectance remote sensing data as input, perform surface resolution inversion based on the SEBAL model;
[0035] Step 6: Using the inversion results of the SEBAL model and some meteorological data as input, the convolutional neural network is used to perform surface evapotranspiration inversion;
[0036] Step 7: Output the inversion results.
[0037] Example 2
[0038] like Figure 1 As shown, on the basis of Example 1, the present invention provides a technical solution: Preferably, the step 1 is to select multispectral surface reflectance remote sensing data of different resolutions of the same date, the same region and the same range, and cut them according to the specified size. The sample size can be determined according to the actual situation, and the sample selection should be diversified, that is, the types of land objects are diverse and the seasonal distribution is diverse;
[0039] The second step is to construct a super-resolution model based on the number of bands, corresponding order and proportional relationship of the high-resolution and low-resolution surface reflectance remote sensing image samples in the first step;
[0040] The step 3 selects temporal low-resolution surface reflectance remote sensing data within the region and time range to be calculated;
[0041] The fourth step is to use the super-resolution model constructed in the second step to perform super-resolution calculations with the data in the third step as input to obtain a high spatial resolution remote sensing image;
[0042] The step five uses the calculation result of step four as input, combines the reanalysis meteorological data, and inverts the high temporal and spatial resolution surface evapotranspiration based on the SEBAL model;
[0043] The sixth step is to construct a convolutional neural network model using the calculation results of the fifth step, the net radiation flux, and the topographic data as input, and use the model to perform high-precision surface evapotranspiration inversion.
[0044] In this embodiment, the spatial resolution of satellite images with high temporal resolution and low spatial resolution is greatly improved through super-resolution technology, so that they have high temporal and spatial resolutions at the same time, which fundamentally solves the problem of inconsistent temporal resolution and spatial resolution of input data required for continuous evapotranspiration remote sensing inversion, thereby greatly improving the temporal and spatial resolution of surface evapotranspiration remote sensing inversion products, and at the same time, combining the inversion based on process-driven physical methods and data-driven methods to further improve the inversion accuracy of surface evapotranspiration. This method first constructs a super-resolution model based on deep learning based on the spectral characteristics of surface reflectance remote sensing satellite images, and improves the spatial resolution (spatial resolution) of satellite remote sensing data. The spatial resolution can be increased by up to 16 times), thereby solving the "temporal and spatial contradiction" of satellite remote sensing data sources in the remote sensing inversion of surface evapotranspiration; using this data as the main input data, the physical model SEBAL is used to achieve high-temporal and spatial resolution surface evapotranspiration inversion; in order to further improve the inversion accuracy and break through the limitations of the applicability of physical models on a large spatial scale, this study combines physical methods with data-driven methods, takes the inversion results of the SEBAL model as combined with some meteorological data as input features, and uses convolutional neural networks to construct a new surface evapotranspiration machine learning model to further improve the surface evapotranspiration inversion accuracy, thereby achieving high-precision and high-temporal and spatial resolution surface evapotranspiration remote sensing inversion.
[0045] Example 3
[0046] like Figure 2As shown, on the basis of Example 1, the present invention provides a technical solution: preferably, the inversion system is composed of four submodules, including a super-resolution model construction module, a SEBAL inversion module, a machine learning model construction module and a machine learning inversion module. The super-resolution model construction module is based on high and low spatial resolution remote sensing samples, adopts super-resolution technology, constructs a multi-spectral remote sensing image super-resolution model, and exports the model after training. The SEBAL inversion module 11 constructs a SEBAL model based on a mathematical formula, and then calls the super-resolution model exported by the super-resolution module construction module to generate high spatiotemporal resolution remote sensing data, reads meteorological parameters at the same time, and then calculates high spatiotemporal resolution surface evapotranspiration based on the SEBAL model. The machine learning model construction module calls the calculation results of the SEBAL inversion module in combination with meteorological factors, constructs an evapotranspiration inversion model based on a convolutional neural network, trains and exports the model, and the machine learning inversion module calls the model exported by the machine learning model construction module, takes the SEBAL model inversion results and meteorological data as input, performs high-precision surface evapotranspiration inversion, and generates and exports result data.
[0047] The above generally describes the present invention in detail, but it is obvious to a person skilled in the art that some modifications or improvements can be made to the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.
Claims
1. A high temporal and spatial resolution surface evapotranspiration remote sensing inversion method, characterized by: The following steps are involved: Step 1: Prepare high and low spatial resolution surface reflectance remote sensing samples; Step 2: Build a high super-resolution model; Step 3: Select time series low-resolution surface reflectance remote sensing data; Step 4: Use the super-resolution model to perform super-resolution calculations on the time series low-resolution surface reflectance remote sensing data; Step 5: Using high temporal and spatial resolution surface reflectance remote sensing data as input, perform surface resolution inversion based on the SEBAL model; Step 6: Using the inversion results of the SEBAL model and some meteorological data as input, the convolutional neural network is used to perform surface evapotranspiration inversion; Step 7: Output the inversion results.
2. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 1, characterized in that: The first step is to select multispectral surface reflectance remote sensing data of different resolutions of the same date, the same area and the same range, and cut them according to the specified size. The sample size can be determined according to the actual situation, and the sample selection should be diversified, that is, the types of land objects and seasonal distributions are diverse.
3. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 2, characterized in that: The second step is to construct a super-resolution model based on the number of bands, corresponding order and proportional relationship of the high-resolution and low-resolution surface reflectance remote sensing image samples in the first step.
4. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 1, characterized in that: The step three selects temporal low-resolution surface reflectance remote sensing data within the region and time range to be calculated.
5. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 1, characterized in that: The fourth step is to use the super-resolution model constructed in the second step, and to carry out super-resolution calculation with the data in the third step as input to obtain a high spatial resolution remote sensing image.
6. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 5, characterized in that: The step five uses the calculation result of step four as input, combines with the reanalysis meteorological data, and inverts the high temporal and spatial resolution surface evapotranspiration based on the SEBAL model.
7. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 6, characterized in that: The sixth step is to construct a convolutional neural network model using the calculation results of the fifth step, the net radiation flux, and the topographic data as input, and use the model to perform high-precision surface evapotranspiration inversion.
8. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 1, characterized in that: The inversion system is composed of four submodules, including a super-resolution model building module, a SEBAL inversion module, a machine learning model building module and a machine learning inversion module.
9. The high temporal and spatial resolution surface evapotranspiration remote sensing inversion method according to claim 8, characterized in that: The super-resolution model construction module is based on high and low spatial resolution remote sensing samples, adopts super-resolution technology, constructs a multi-spectral remote sensing image super-resolution model, and exports the model after training. The SEBAL inversion module 11 constructs the SEBAL model based on a mathematical formula, and then calls the super-resolution module to construct the super-resolution model exported by the module to generate high spatiotemporal resolution remote sensing data, reads meteorological parameters at the same time, and then calculates the high spatiotemporal resolution surface evapotranspiration based on the SEBAL model. The machine learning model construction module calls the calculation results of the SEBAL inversion module and combines meteorological factors to build an evapotranspiration inversion model based on a convolutional neural network, trains and exports the model. The machine learning inversion module calls the model exported by the machine learning model construction module, takes the SEBAL model inversion results and meteorological data as input, performs high-precision surface evapotranspiration inversion, and generates and exports result data.
Citation Information
Cited By
Evaporation simulation method and system based on PINN neural network
CN119759989A
