Document Type : علمی - پژوهشی
Authors
1
PhD Student of Range Management , University of Tehran
2
Professor, Department of Arid and Mountainous Regions Reclamation, Faculty of Natural Resources, University of Tehran, Karaj, Iran.
3
Assistant Professor, Department of Geography, Yazd University, Yazd, Iran.
4
Associate Professor, Department of Rehabilitation of Arid and Mountainous, Faculty of Natural Resources, University of Tehran, Karaj, Iran.
5
Assistant Professor, Department of Arid and Mountainous Regions Reclamation, Faculty of Natural Resources, University of Tehran, Karaj, Iran.
Abstract
Introduction and purpose: Generating Land Surface Temperature (LST) data with temporal and spatial continuity is in great demand for hydrology, meteorology, ecology, environment, and, etc. studies. Approximately, 60 to 75 percent of the Earth is covered by clouds at any given moment. Therefore, clouds, by creating an obstacle, absorb part of the thermal energy emitted from the earth by affecting thermal infrared energy, creating gaps and outliers in LST time series data. Removing the effect of cloud cover is always a common problem in the field of using satellite images. The purpose of this research is to evaluate the performance of Multi-channel Singular Spectrum Analysis (M-SSA) in order to reconstruct gaps and remove outliers data due to the cloud coverage in the hourly LST time series of the Meteosat-9 satellite.
Materials and methods: The study area in the present research was whole Iran. Also, the hourly LST time series of the SEVIRI sensor from the Meteosat-9 geostationary satellite in 2022 was used. At first, using SSA software and the Monte Carlo test, the window size and the number of significant components of an hourly LST time series were determined. Then, using the identified significant components, LST time series were reconstructed using M-SSA algorithm. Reconstruction error in clear sky conditions with available time series data and reconstruction error in cloudy sky conditions by creating artificial missing data (artificial cloud) with intensities of 10, 20, 30, ..., 90% in time series were evaluated using root mean square error (RMSE) and coefficient of determination (R2) statistics.
Results: On average, in Iran, 25.5% of the hourly LST time series in 2022 was lost due to cloud cover, and the highest percentage of lost data was observed at the edge of the Caspian Sea. The results of analyzing the annual hourly LST time series in a window size of 96 hours with the Monte Carlo test showed that components 1 to 5 are significant components of this time series. These components control 97.5% of the LST time series variance. The frequency of the first,, second-third, and fourth-fifth components are respectively 0, 0.042 and 0.083 cycles per image. The first component indicates annual periodic changes, the second and third components indicate 24-hour or daily temperature changes, and the fourth and fifth components indicate 12-hour periodic temperature changes. Based on the results, the RMSE and the R2 between the original and the reconstructed data in clear sky conditions were 1.38 and 0.99 Kelvin, respectively. Also, in cloudy sky conditions, the RMSE error up to the level of 80% of randomly lost data (artificial cloud) was always less than 2.1 Kelvin.
Discussion and conclusion: The main key to reconstructing time series with periodic behavior is to identify significant periodic components and trends. In hourly LST time series, annual, 24-, 12- and 8-hour periods are the most important components of the time series. These components are formed due to the rotation of the earth around itself and the sun and the deviation of its axis. Therefore, these components are generally the same for the reconstruction of hourly LST time series in the major part of the globe. Based on the findings, M-SSA algorithm can be effective in reconstructing lost data with large distance in LST time series due to consideration of periodic components and trends as well as using temporal and spatial correlation. One of the significant cases in reconstructing the effect of cloud cover in the present study and many other studies is the reconstruction of LST with the clear sky condition. Therefore, reconstruction of LST under cloud cover can be a challenge and suggestion for further studies in the future.
Keywords
)