A 1-km Resolution Daily Land Surface Temperature Dataset for the Qinghai-Tibet Plateau (2000-2020)

Xu, X. P.^{1, 2, 3}  Zhang, Y.^{1, 2, 3}  Zhang, Y. C.⁴  Ji, L. Y.^1,2  Tang, H. R.^{1, 2, 3*}

1. Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China;

2. the Key Laboratory of Technology in Geo-Spatial information Processing and Application System, Chinese Academy of Sciences, Beijing 100190, China;

3. the School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 101408, China;

4. A military representation room of the PLA rocket forces, Beijing 100000, China

Abstract: Remote sensing data are strongly correlated with continuity in space and time, giving remote sensing time-series images low rank. This paper repairs images using low-rank tensor complementation by pre-processing the Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature (LST) data and employing spatiotemporal interpolation to initially fill in missing values caused by cloud cover. We then treat the LST time series data as a third-order spatiotemporal tensor and introduce a Fourier transform on the time dimension to convert it into a space-frequency tensor. Performing singular value decomposition and Gaussian low-pass filtering on this tensor followed by inverse a Fourier transform provides the space-time tensor. We further optimize the missing tensor using the alternating direction method of multipliers. Accuracy is validated through simulations, where artificial masks are added and subsequently recovered. The resulting mean absolute error (MAE) falls within the 2.1–4.9 K. This dataset includes the following daily data for the Tibetan Plateau for the years 2000–2020. (1) The optimized surface temperature data (MOD11A1_QTP_PART and MYD11A1_QTP_PART) for the cloud-shaded regions of the MOD11A1 and MYD11A1 products. (2) The optimized MOD11A1 for the cloud-shaded regions, and MYD11A1 products as the optimized surface temperature data (MOD11A1_QTP_Temp and MYD11A1_QTP_Temp). (3) Original MOD11A1 and MYD11A1 products (MOD11A1_QTP_ORIGIN and MOD11A1_QTP_ORIGIN). All data have a spatial resolution of 1 km and are archived in an integer data format. The image element values represent the thermodynamic temperature of the surface with a scale factor of 0.02 K The dataset is archived in .tif format, which can be directly opened and processed using remote sensing software such as ENVI and ArcGIS.

Keywords: Qinghai-Tibet Plateaut; daily land surface temperature; 1 km; 2000–2020; MODIS

DOI: https://doi.org/10.3974/geodp.2023.03.03

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2023.03.03

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2023.10.02.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2023.10.02.V1.

1 Introduction

Land surface temperature (LST) refers to the energy emitted and radiated by Earth’s surface in the near-infrared and thermal infrared bands. It is a significant indicator describing the thermal state of Earth’s surface. This parameter holds considerable research value in fields such as climate change, ecological environment, and agricultural production^[1–3]. Leveraging remote sensing data to acquire LST allows for rapidly, comprehensively, and accurately assessing surface temperature distributions over large areas. This capability effectively guides decision-making and planning across relevant domains. The resolution and accuracy of LST data have remarkably improved with the continuous advancement of remote sensing technology and ongoing updates to satellite data. This has led to an increased production of LST products available for researchers.

The Qinghai-Tibet Plateau region is pivotal due to its vast territory, abundant resources, unique geographical environment, and distinctive climatic conditions. Thus, it is consistently positioned as a focal point in geography, climate, and environmental sciences. In recent years, intensified global climate change has drawn widespread attention and research efforts toward abnormal variations in LSTs within the Qinghai-Tibet Plateau region. Therefore, studying land surface temperatures in this area has profound theoretical and practical significance as it is intricately linked to many issues spanning climate change, water resource management, and ecological environment protection. However, this region’s intricate topography and complex climate substantially challenge acquiring and processing remote sensing data. One notable challenge is cloud cover, which significantly impacts the accuracy and usability of land surface temperature data, diminishing its application value. Consequently, enhancing the quality and usability of LST data is a critical challenge demanding immediate attention.

The low-rank tensor completion method effectively applies to cloud restoration in remote sensing imagery^[4–7]. This method exploits low-rank data attributes, enabling tensor completion from incomplete observations and recovering information masked by cloud layers in remote sensing data. This technique elevates data quality and utility by preserving fine details. This paper uses the MOD11A1 V6 and MYD11A1 V6 products of day-by-day surface temperature data and joint spatiotemporal low-rank tensor complementation^[8] to replace missing data and cropping day-by-day surface temperature data in the Tibetan Plateau region^[9]. Finally, a daily cloud-free surface temperature dataset is produced for the Tibetan Plateau region from 2000–2022. This dataset has a wide range of research and application value and promotes the development of climate research, ecological and environmental assessments, and other related fields in the Qinghai-Tibet Plateau region.

2 Metadata of the Dataset

The metadata of the 1-km/Daily land surface temperature optimized dataset for the Qinghai-Tibet Plateau based on MODIS data (2000–2020)^[10] is summarized in Table 1. It includes the dataset full name, short name, authors, year, temporal resolution, spatial resolution, data format, data size, data files, data publisher, data sharing policy, etc.

Table 1  Metadata summary of the 1-km/Daily land surface temperature optimized dataset for the Qinghai-Tibet Plateau based on MODIS data (2000–2020)

3 Methods

3.1 Data Source

The utilized dataset originates from MOD11A1 V6[1] and MYD11A1 V6[2], which are the daily LST products based on MODIS. These products offer diurnal and nocturnal LSTs around the globe, including the Qinghai-Tibet Plateau region. Specifically, the MOD11A1 data are acquired from the Terra satellite, while the MYD11A1 data are acquired from the Aqua satellite. Due to the differing satellite overpass times, these sources provide remote sensing data for morning and afternoon observations.

This paper focuses on the LST_Day_1km band for the recovery process. This band is a pivotal component of the MOD11A1 V6 and MYD11A1 V6 products and is designed to capture daytime LST information. It offers a spatial resolution of 1 km and a temporal resolution of 1 day, with each scene’s dimensions measuring 1200×1200 pixels. Data within the LST_Day_1km band fall within the range of 7,500–65,535 with a scaling factor of 0.02. Any invalid values are designated as 0.

3.2 Data Pre-processing

The Qinghai-Tibet Plateau region is frequently characterized by cloud cover throughout the year, leading to considerable LST data being persistently obscured. Directly applying the joint time-domain fast Fourier transform (FFT) low-rank tensor completion algorithm could mistakenly interpret persistent cloud cover as low-frequency components, interfering with the intended ground feature recovery. This often results in the appearance of numerous black stripes in the restoration output, affecting data usability. To address this issue, a pre-processing step is essential for the LST data. Spatial and temporal interpolation is employed to predominantly recover missing values caused by cloud cover, as illustrated in Figure 1.

The LST data for multiple temporal instances are initially arranged along the time dimension, denoted as , where  and  represent the spatial dimensions of the data, and  indicates the time series length. Subsequently, the time series  is divided into numerous small windows sized at 100×100. The effective values within each window are averaged to form the downsampled time series . A linear relationship between the original time series X and the downsampled effective values  is determined pixel by pixel using the least squares method. Finally, utilizing the coefficients obtained through this solution and the downsampled time series  restores the data for cloud-covered areas and yields the preprocessed LST time series . This data pre-processing removes a significant portion of the cloud contamination, enabling smoother tensor completion operations in subsequent steps.

Figure 1  Schematic diagram of the data pre-processing

Given the extended length of our time series, employing spatiotemporal interpolation with cloud-free periods of LSTs enhances the stability and accuracy of the acquired image information. Downsampling using 100×100 windows ensures local spatial consistency and prevents abrupt data discontinuities. Applying the least squares method to establish a linear relationship between the original and downsampled time series effective values allows for more accurately predicting missing values, mitigating cumulative errors stemming from an excess of missing points.

3.3 Algorithmic Principle

Due to the strong spatial and temporal correlations and the continuity in remote sensing data, the time series of remote sensing images, denoted as X, possess a low-rank property. Leveraging low-rank tensor completion is aimed at achieving image restoration. This paper employs distinct decomposition methods to handle spatial and temporal dimensions. We introduce Fourier transformations to filter the temporal dimension, adaptively select weights based on the temporal frequency spectrum attributes, and apply them to the low-rank matrix completion in the spatial dimension. We then exploit the conjugate symmetry in the frequency domain to accelerate the computational speed. The proposed approach emphasizes the low-frequency components brought about by land cover changes in the temporal dimension while suppressing high-frequency noise induced by clouds. This process achieves a joint low-rank completion in both the temporal and spatial dimensions.

3.4 Technological Route

A roadmap of the dataset production techniques is shown in Figure 2. The pre-processing part was described in Section 3.2, and the recovery part is described in this section.

Figure 2  The technical route for dataset production

3.4.1 Time Dimension FFT

The Fourier transform projects time domain signals onto a set of orthogonal trigonometric function bases, suitable for serial data decomposition and processing. We introduce the Fourier transform in the time dimension of the tensor to transfer it into the frequency domain for processing as:

                            (1)

where,  denotes the Fourier-transformed tensor, which we call the space-frequency tensor.

3.4.2 Time Dimension Filtering

After the time-dimensional Fourier transform, the time-series surface temperature data spectrum can be divided into low- and high-frequency components. The low-frequency component corresponds to slow changes or static conditions, while the high-frequency component corresponds to significant changes in time, such as clouds and noise. We apply a Gaussian filter  in the time-dimensional to preserve the main low-frequency components and weaken the effects of clouds, noise, etc. The Gaussian filter function is , where  is the number of image elements, and  is a defined constant.

3.4.3 Spatial Dimension Adaptive Low-Ranking

To jointly achieve spatiotemporal and spatial low rank for better selection and preservation of feature information in images, the tensor  is subjected to low-rank processing with adaptive weights for frequency domain modulation. The threshold  for the low-rank processing of each slice is determined for different slices  based on the importance of the matrix information as:

                               (2)

                           (3)

3.4.4 Time Dimension iFFT

After the above three steps, we performed the inverse Fourier transform to obtain the solution in the form .

Updating the values of the cloud-covered locations until they are less than the threshold provides the cloud-free LST time series.

4 Data Results and Validation

4.1 Data Composition

The dataset is divided into two directories based on the satellite, MOD11A1_QTP_Temp and MYD11A1_QTP_Temp. The naming convention of the data in each directory is YYYYYDDD.tif, where YYYYY is the year, and DDD is the first day of a particular year, e.g., 2020001.tif.

4.2 Data Products

This paper selects the data of 2020001.tif from the MOD11A1_QTP_Temp product as an illustration in Figure 3. The black part of the image is the region outside the Tibetan Plateau, represented by a 0 in the data. The gray part is the study area with a valid value range of 7,500–65,535.

We qualitatively analyzed the data near the Nam Co Lake region. Figures 4 and 5 show scatter plots of surface temperature data for 2000–2009 and 2010–2020, respectively. The blue portion indicates valid retained values in the original product, and the red portion indicates missing data in the original product that was recovered using the proposed algorithm. The recovered data match the surface temperature trends.

Figure 3  Map of presentation

Figure 4  Scatter plot of land surface temperature data in the Nam Co Lake region (2000–2009)

Figure 5  Scatter plot of land surface temperature data in the Nam Co Lake region (2010–2020)

Besides, we plotted a line graph of the annual mean surface temperature in the Nam Co Lake region in Figure 6 to demonstrate the trend over 20 years. Changes in the surface temperature are relatively smooth, with a variation of about 1 K between adjacent years.

4.3 Data Validation

Due to the lack of real data, we used simulations to verify the recovery accuracy of the dataset. We took the data from the MOD11A1 V6 product in 2020 as an example, randomly selected eight cloud-free regions with different dates and locations, manually added masks for these regions, used 0 values to replace the surface temperature information in the original product, and used the proposed method to recover the surface information. Finally, we evaluated the recovered values of the regions using metrics, as shown in Table 2.

Figure 6  Trends in annual mean surface temperature in the Nam Co Lake region (2000–2020)

Table 2  The effectiveness of land surface temperature recovery in regions with manually added masks (dimensionless)

We also compared the proposed product with other existing products, as shown in Table 3. Product 1 is the Landsat time-series surface temperature for the Tibetan Plateau region in 2020^[11,12], and Product 2 is the 1-km seamless surface temperature dataset for the Chinese region (2002–2020)^[13–16] .

Table 3  Recovery of land surface temperature in the Nam Co Lake region by 2020 in the context of existing relevant studies

The different times for satellites transits do not allow directly comparing products. This paper compares the reliability and accuracy of the proposed methodology by plotting the trends for recovering the relevant surface temperatures in Figure 7. The proposed product shows the same trend as other products in recovering surface temperatures and better balances temporal resolution with spatial accuracy.

5 Discussion and Conclusion

This study developed a daily land surface temperature dataset for the Qinghai-Tibet Plateau region from 2000 to 2022 using a joint spatiotemporal low-rank approach, followed by accuracy validation. This dataset holds significant research and application value, contributing to the advancement of various fields, such as climate research and ecological environment assessment in the Qinghai-Tibet Plateau region.

Figure 7  Trends in surface temperature recovery in the Nam Co Lake region by 2020 based on relevant studies

Author Contributions

Xu, X. P., Zhang, Y., Ji, L. Y., Tang, H. R. made the overall design for the development of the dataset. Zhang, Y., Ji, L. Y. contributed to the data processing and analysis. Xu, X. P., Tang, H. R. designed the models and algorithms. Zhang, Y., Ji, L. Y., and Zhang, Y. C. did the data validation. Xu, X. P. wrote the data paper. Zhang, Y. C. embellished the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]    Wang, A. H., Zeng, X. B. Development of global hourly 0.5° land surface air temperature datasets [J]. Journal of Climate, 2013, 26(19): 7676–7691.

[2]     Mostovoy, G. V., King, R. L., Reddy, K. R., et al. Statistical estimation of daily maximum and minimum air temperatures from MODIS LST data over the state of Mississippi [J]. GIScience & Remote Sensing, 2006, 43(1): 78–110.

[3]     Xu, Y. M., Qin, Z. H., Shen, Y. Study on the estimation of near-surface air temperature from MODIS data by statistical methods [J]. International Journal of Remote Sensing, 2012, 33: 7629–7643.

[4]     Ng, M. K-P., Yuan, Q. Q., Yan, L., et al. An adaptive weighted tensor completion method for the recovery of remote sensing images with missing data [J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(6): 3367–3381.

[5]     Ji, T. Y., Yokoya, N., Zhu, X. X., et al. Nonlocal tensor completion for multitemporal remotely sensed images’ inpainting [J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(6): 3047–3061.

[6]     Chen, Y., He, W., Yokoya, N., et al. Blind cloud and cloud shadow removal of multitemporal images based on total variation regularized low-rank sparsity decomposition [J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 157: 93–107.

[7]     Lin, J., Huang, T. Z., Zhao, X. L., et al. Robust thick cloud removal for multitemporal remote sensing images using coupled tensor factorization [J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1–16.

[8]     Chen, Z. H., Zhang, P., Zhang, Y., et al. Thick cloud removal in multi-temporal remote sensing images via frequency spectrum-modulated tensor completion [J]. Remote Sensing, 2023, 15(5): 1230.

[9]     Zhang, Y. L., Li, B. Y., Liu, L. S., et al. Redetermine the region and boundaries of Tibetan Plateau [J]. Geographical Research, 2021, 40(6): 1543–1553.

[10]   Xu, X. P., Zhang, Y., Ji L.Y., et al. 1-km/Daily land surface temperature optimized dataset for the Qinghai-Tibet Plateau based on MODIS data (2000–2020) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2023.  https://doi.org/10.3974/geodb.2023.10.02.V1. https://cstr.escience.org.cn/CSTR:

20146.11.2023.10.02.V1.

[11]   GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[12]   Zhang, Z., M. Landsat surface temperature products over the Tibetan Plateau (2020) [Z]. National Tibetan Plateau / Third Pole Environment Data Center, 2022. DOI: 10.11888/Terre.tpdc.272304.

[13]   Wang, M. M., Zhang, Z. J., Hu, T., et al. A practical single-channel algorithm for land surface temperature retrieval: application to Landsat series data [J]. Journal of Geophysical Research: Atmospheres, 2019, 124: 299–316.

[14]   Cheng, J., Dong, S. Y., Shi, J. C. 1km seamless land surface temperature dataset of China (2002–2020) [Z]. National Tibetan Plateau/Third Pole Environment Data Center, 2021, DOI: 10.11888/Meteoro.tpdc.271657.

[15]   Xu, S., Cheng, J. A new land surface temperature fusion strategy based on cumulative distribution function matching and multiresolution Kalman filtering [J]. Remote Sensing of Environment, 2021, 254: 112256.

[16]   Zhang, Q., Wang, N. L., Cheng, J., et al. A stepwise downscaling method for generating high-resolution land surface temperature from AMSR-E data [J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, 13: 5669–5681.

[17]   Zhang, Q., Cheng, J. An empirical algorithm for retrieving land surface temperature From AMSR-E data considering the comprehensive effects of environmental variables [J]. Earth and Space Science, 2020, 7: e2019EA001006.