An Improved Spatiotemporal Filling Method for Remote Sensing Lake Surface Temperature

Abstract

The present invention discloses an improved spatio-temporal filling method for remote sensing lake surface temperature, which obtains remote sensing retrieved land surface temperature data, makes a buffer zone inward from the permanent water boundary of the lake to extract lake surface water temperature (LSWT) data; divides the LSWT data into high-quality images and low-quality images based on the number of valid pixels; fills the high-quality images using the DCT-PLS method; constructs the anomaly value of the LSWT every n days as the background field, and after superimposing it with the median value of the LSWT on the corresponding date of the low-quality image, a spatially filled low-quality image is obtained, and then linear interpolation is performed on the time series of all the filled images, thereby completing the spatio-temporal filling of all image data. This method does not depend on other environmental factors of the lake, provides a practical remote sensing method for improving the long-term lake temperature monitoring level, and helps to further understand the response of lake surface temperature changes to global climate change.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing science and technology, and particularly relates to an improved spatio-temporal filling method for remote sensing lake surface temperature. Background Art

[0002] Lakes are crucial for the water cycle and ecological balance (Verpoorter et al., 2014). Lake surface temperature (LSWT) is an important physical variable for understanding the impact of climate change and energy exchange on lakes (Doney et al., 2012; Pachauri et al., 2014). It can sensitively detect the environmental characteristics and reaction processes of lakes, as well as changes in biodiversity and hydrodynamics (Yang et al., 2019). Existing studies have shown that large lakes globally generally exhibit a significant warming trend (O'Reilly et al., 2015).

[0003] The traditional method for early monitoring of LSWT was to measure the temperature at specific sites using in-situ sensors (Ptak et al., 2019). However, although this method has high precision and flexible operation, it has various limitations, such as high human and material costs, and low temporal and spatial resolutions (Kumari et al., 2018). Remote sensing is an effective method to overcome these difficulties and obtain more comprehensive temperature information (Hong et al., 2021). In recent years, satellite data for land surface temperature inversion mainly include two types: microwave and thermal infrared (TIR). Due to the diversity of TIR data, this remote sensing method has been widely used in temperature monitoring. TIR data includes high-resolution radiometers (AVHRR), along-track scanning radiometers (ATSR), Landsat, and moderate-resolution imaging spectrometers (MODIS). Among them, the MODIS land surface temperature (LST) product is a relatively commonly used data set. It has relatively high temporal and spatial resolutions, a wide spectral range, high radiation sensitivity, and appropriate TIR channel settings. In previous studies, the MODIS land surface temperature (LST) data set has been widely used to monitor land and water surface temperatures (Hu et al., 2020). However, due to cloud effects and other atmospheric interferences, the MODIS LST product has a large area of missing or "noisy" pixels, resulting in serious spatio-temporal information gaps, which in turn limit the long-term monitoring of lake surface temperature. For this reason, researchers have proposed a series of MODIS LST product reconstruction methods to fill in the missing data and improve the effectiveness of remote sensing data.

[0004] Existing LSWT gap filling methods are mostly similar to LST methods and generally achieve LSWT prediction by fusing data from multiple satellite products or adding corresponding auxiliary datasets and external environmental factors (Layden et al., 2016). These methods rely on external environmental factors closely related to LST or LSWT, such as meteorological variables (e.g., temperature and precipitation) and digital elevation models, to predict missing pixels. However, in the absence of auxiliary datasets, the accuracy of the reconstructed LSWT may decrease significantly. These conditions complicate the filling methods and limit their application to some extent, especially for lakes with limited available auxiliary data.

[0005] In summary, for the acquisition of existing lake temperature data, the surface temperature products are often severely lacking, and there are still obstacles in filling the spatio-temporal gaps of LSWT products by existing methods, making it impossible to obtain complete long-term lake temperature data. Against this background, conducting research on the spatio-temporal gap filling method of lake temperature can provide important data guarantee and method support for exploring scientific issues related to hydrological environment changes, and has important scientific significance. Summary of the Invention

[0006] To fill the spatio-temporal gaps of the original remotely sensed surface temperature images and achieve long-term monitoring of lake surface temperature. The present invention proposes an improved gap filling method that combines the penalty least squares method based on discrete cosine transform (DCT-PLS) in the spatial domain with the linear interpolation (LI) algorithm in the time domain. Using publicly available and free-to-obtain MODIS remotely sensed image data and the Yearly Water Classification History dataset, and by means of the method combining spatio-temporal domain filling, the reconstruction of long-term lake surface temperature information is realized.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0008] An improved spatio-temporal filling method for remotely sensed lake surface temperature, comprising:

[0009] Obtain remotely sensed retrieved surface temperature data, and make a buffer zone inward from the permanent water boundary of the lake to extract lake surface temperature LSWT data;

[0010] Divide the LSWT data into high-quality images and low-quality images based on the number of valid pixels;

[0011] Use the DCT-PLS method to fill the high-quality images;

[0012] Construct every nThe anomaly value of the LSWT for the day is used as the background field, which is superimposed on the median value of the LSWT for the corresponding date of the low-quality image to obtain a spatially filled low-quality image. Then, interpolation is performed on the time series of all the filled images to complete the spatio-temporal filling of all the image data.

[0013] As a preferred implementation, the method further includes using quality control QC to control the quality of the land surface temperature data and extracting the LSWT data based on the pixels that pass the quality control.

[0014] As a preferred implementation, the water area boundaries of a lake over the years are compared, and the water area boundary of the smallest permanent water body over the years is selected as the permanent water area boundary of the lake.

[0015] As a preferred implementation, the method further includes, for the extracted LSWT data, using the method of random blanking to select a validation data set for data validation.

[0016] As a preferred implementation, a number of pixels are randomly selected on each LSWT image, the corresponding original temperature values are replaced with null values, the LSWT data after replacing the null values is filled, and the filling accuracy is verified using the number of pixels.

[0017] As a preferred implementation, a threshold value of the number of valid pixels is preset. An image with the number of valid pixels higher than the threshold is a high-quality image, and an image with the number of valid pixels lower than the threshold is a low-quality image.

[0018] As a preferred implementation, the linear interpolation method is used to interpolate the time series of all the filled images.

[0019] As a preferred implementation, after completing the spatio-temporal filling of all the image data, the accuracy of the LSWT data filling is evaluated through 4 error metrics of R 2 , MAE, MSE, and RMSE.

[0020] As a preferred implementation, the remotely sensed retrieved land surface temperature data selects the MOD11A2 land surface temperature data product.

[0021] As a preferred implementation, the water area boundary data of a lake over the years is obtained based on the Yearly Water Classification History data set.

[0022] The present invention has the following two advantages:

[0023] (1) The present invention has relatively low requirements for data. Based on publicly available remotely sensed image data and surface water body data, without relying on other lake environmental factors, it can complete the filling of the spatio-temporal gaps of the lake surface temperature, and further construct long-term sequence lake temperature data.

[0024] (2) The present invention proposes an improved spatio-temporal gap filling method for lake surface water temperature (LSWT), which is not only not limited to the research area and the size of the lake area, but also can carry out lake temperature change analysis on long time series and multiple time scales, providing important basic data and technical support for further understanding the response of lake hydrological environment changes to global climate change. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in each figure may be denoted by the same reference numeral. For clarity, not every component is labeled in each figure. Now, each step of the present invention will be described through research cases and with reference to the drawings, where:

[0026] Figure 1 is the topographic map of Hulun Lake and its surrounding areas in Embodiment 1 of the present invention.

[0027] Figure 2 is the flowchart of the method of the present invention.

[0028] Figure 3 is the multi-year average LSWT image of Hulun Lake after spatial filling based on the DCT-PLS method in Embodiment 1: (a) daytime; (b) nighttime.

[0029] Figure 4 is the 8-day scale LSWT time series of Hulun Lake before and after the linear interpolation method in Embodiment 1.

[0030] Figure 5 is the accuracy evaluation of the linear interpolation result of the lake temperature time series of Hulun Lake in Embodiment 1.

[0031] Figure 6 is the accuracy evaluation of the spatial filling result of the lake temperature of Hulun Lake in Embodiment 1. MODE OF IMPLEMENTATION

[0032] The following further describes in detail the specific implementation manners of the present invention in conjunction with the drawings and Embodiment 1. The following implementation cases are used to illustrate the present invention, but are not used to limit the scope of the present invention. EXAMPLE

[0033] Example 1 of this application takes Hulun Lake, the largest lake on the Inner Mongolia Plateau, as the research object. The Inner Mongolia Plateau has a vast area and a high altitude, and its surrounding areas play an important role in global climate change through unique atmospheric interactions. The Inner Mongolia Plateau has a typical temperate continental climate, and the regional average temperature difference is very large. There are many large lakes on the Inner Mongolia Plateau, mostly distributed in the northwest. These lakes play an important role in the water, cryosphere, and energy cycles. As the fourth largest freshwater lake in China and the largest lake on the Inner Mongolia Plateau, Hulun Lake has a strong response to climate change and is more suitable as a research object.

[0034] As Figure 2 shown, it is the flowchart of Example 1. This Example 1 includes the following steps:

[0035] Step 1: Obtain remote sensing image data and preprocess the MODIS remote sensing image data. First, download all MODIS LST remote sensing images covering Hulun Lake since 2000 from the EARTHDATA (https: / / modis.ornl.gov / globalsubset / ) data website, including two sets of data, Day and Night. For the preprocessing of MODIS land surface temperature data, two data layers, quality control QC and land surface temperature LST in the MOD11A2 land surface temperature data product, are selected. Use the quality control data QC to control the quality of MODIS land surface temperature and retain the better-quality pixels (pixels with a QC value of 1). Compare the water boundaries of the lake over the years, select the smallest perennial water body over the years as the final boundary, and make an inner buffer to extract the Hulun Lake LSWT data. For lakes with an area greater than 50 km 2 , set an inner buffer of 1000 m, and for lakes less than 50 km 2 , set an inner buffer of 500 m. The permanent water boundary of the lake is derived from the Yearly Water Classification History dataset.

[0036] Step 2: Construct a validation dataset and evaluate the accuracy of spatially filled LSWT.

[0037] In this example, the "random blanking" method is used to select the validation dataset to evaluate the accuracy of filled LSWT;

[0038] Specifically, randomly sample 10% of the pixels on the spatial image and time series of the LSWT data extracted in Step 1, and assume the corresponding original LSWT values as null values. These pixels are used as the validation dataset.

[0039] Step 3. This step is the core of the LSWT gap filling process. It involves filling the images with relatively good filling quality based on the DCT-PLS model, and then constructing a background field to fill the images with relatively poor quality.

[0040] (1) Image classification: First, the images (the original LSWT data with some pixel values set as null values) are divided into high-quality images and low-quality images according to the number of valid pixels. In this embodiment, the threshold of the number of valid pixels is set to 30%, that is, the images with the number of valid pixels greater than 30% are high-quality images, and vice versa.

[0041] (2) Application of the DCT-PLS model: The DCT-PLS method is used to fill the high-quality images.

[0042] (3) Construction of the LSWT image background field: Based on the high-quality filled images, a multi-temporal average LSWT background field is constructed to assist in filling the low-quality images. It is assumed that most lakes have certain spatial variation rules in different seasons or on different dates. Therefore, on the basis of the existing complete LSWT images, the anomaly value of LSWT every 8 days is constructed as the background field, and adding the median value of LSWT of the corresponding low-quality image gives the filled low-quality LSWT image. As Figure 3 shown, the daytime LSWT of Hulun Lake all shows that the temperature gradually decreases from the lakeshore to the center of the lake ( Figure 3 a). The changing trends of daytime and nighttime LSWT are opposite. At night, Hulun Lake shows a warming trend from the lakeshore to the center of the lake ( Figure 3 b).

[0043] Step 4. Since there is no validation of image loss in winter, after spatial filling, there are still gaps in the time series of LSWT of Hulun Lake. This method further uses the linear interpolation method to fill the gaps in the time series of LSWT data. Figure 4 a and Figure 4 b respectively show the 8-day scale LSWT time series of Hulun Lake before and after using the linear interpolation method.

[0044] Step 5. Evaluation of the spatio-temporal LSWT filling results. Based on the simulated missing data and the corresponding original LSWT, the accuracy of the spatio-temporal filling of LSWT data of Hulun Lake is evaluated through 4 error indexes of R 2 , MAE, MSE and RMSE. For the spatio-temporal filling results of lake temperature ( Figure 5 ), the reconstructed LSWT of Hulun Lake during the day and at night is in good agreement with the measured LSWT values. The values of R 2 , MAE, MSE, RMSE are 0.99, 0.21 °C, 0.17 °C, 0.42 °C and 0.99, 0.18 °C, 0.10 °C, 0.32 °C respectively. Combining Figure 4 andFigure 5 , it can be seen that the effect of linear interpolation of the lake temperature time series is also very good.

[0045] Through the above method, the long-term lake temperature of Hulun Lake can be obtained. It is not only not limited to the size of the study area and lake range, but also can carry out the analysis of lake temperature changes on long time series and multiple time scales, providing important basic data and technical support for further understanding the response of lake hydrological environment changes to global climate change.

Claims

1. An improved spatio-temporal filling method for remote sensing lake surface temperature, characterized in that, Including: Obtain remotely sensed retrieved land surface temperature data, and create a buffer inward from the permanent water boundary of the lake to extract lake surface water temperature (LSWT) data; Divide the LSWT data into high-quality images and low-quality images based on the number of valid pixels; Use the DCT-PLS method to fill the high-quality images; Construct the anomaly value of LSWT every n day as the background field. After superimposing it with the median value of LSWT on the corresponding date of the low-quality image, a spatially filled low-quality image is obtained. Then, interpolation is performed on the time series of all the filled images to complete the spatio-temporal filling of all image data.

2. The method according to claim 1, characterized in that, It also includes using quality control (QC) to control the quality of the land surface temperature data, and extracting LSWT data based on the pixels that pass the quality control.

3. The method according to claim 1, characterized in that Compare the water boundaries of the lake over the years, and select the water boundary of the smallest permanent water body over the years as the permanent water boundary of the lake.

4. The method according to claim 1, wherein It also includes, for the extracted LSWT data, using the method of random blanking to select a validation data set for data validation.

5. The method according to claim 4, wherein Randomly select a number of pixels on each LSWT image, replace the corresponding original temperature values with null values, fill the LSWT data after replacing the null values, and use the said number of pixels to verify the filling accuracy.

6. The method according to claim 1, wherein Preset a threshold for the number of valid pixels. Images with the number of valid pixels higher than the threshold are high-quality images, and images lower than the threshold are low-quality images.

7. The method according to claim 1, wherein Use the linear interpolation method to fill the time series of all filled images.

8. The method according to claim 4, characterized in that, After completing the spatio-temporal filling of all image data, the accuracy of LSWT data filling is evaluated by four error metrics: R 2 , MAE, MSE, and RMSE.

9. The method according to claim 1, wherein The remotely sensed retrieved land surface temperature data uses the MOD11A2 land surface temperature data product.

10. The method according to claim 3, wherein Obtain the water boundary data of the lake over the years based on the Yearly Water ClassificationHistory dataset.

Images