Land surface temperature (LST) is one of the most important and widely used parameters for studying land surface processes. The surface temperature products of the Moderate Resolution Imaging Spectroradiometer (MODIS), such as MOD11A1 and MYD11A1, can provide surface temperature information with medium spatiotemporal resolution covering the globe. However, the application of these data has been hindered by missing values caused by factors such as cloud pollution, indicating the need to create a seamless global LST dataset similar to MODIS, which currently does not exist. In this study, we utilized a spatiotemporal imputation framework to generate a seamless global 1 km daily (mid day and mid night) MODIS LST dataset based on standard MODIS LST products from 2003 to 2020.
The research area covers almost the entire global land surface, including 178 MODIS tiles. The main data used in this study is the 6th edition of the MODIS LST product, which covers a daily distance of 1 kilometer from 2003 to 2020. This product is based on NASA's Earth Observation System (EOS) satellites Terra Aqua (MOD11A1 and MYD11A1). These two satellites conduct four observations per day (i.e. 10:30 and 22:30 local time, Terra: T1 and T3; 13:30 and 01:30, Aqua: T2 and T4): T2 and T4. The other two auxiliary datasets used are the annual MODIS land cover product (MCD12Q1) and the urban range derived from nighttime light observation data and its surrounding rural areas. Our analysis excluded water pixels in the MCD12Q1 product.
This method consists of two steps: (1) data preprocessing and (2) spatiotemporal fitting. During the data preprocessing process, we filtered out pixels with lower data quality and filled in the gaps using LSTs observed at three other time points on the same day. In the process of spatiotemporal fitting, we first use a smooth spline function to fit the time trend (overall average) of the observed data based on the year, month, day (independent variable) of each pixel point. Then, we perform spatiotemporal interpolation on the residuals between the daily observations and the overall average. Finally, we add the overall average value to the interpolated residual value to estimate the missing value of LST.
The results indicate that the missing values in the original MODIS LST have been effectively and efficiently filled, and the computational cost has been reduced. Moreover, there is no significant block effect caused by large missing values (especially near tile boundaries) that may exist in other seamless LST datasets. The cross validation results of different missing rates globally indicate that LST data with gap filling has high accuracy, with average root mean square errors (RMSE) of 1.88 and 1.33 for intraday (13:30) and nighttime (01:30), respectively. The seamless global daily (daytime and nighttime) LST dataset with a spatial resolution of 1 kilometer has important applications in global urban system research, climate research and modeling, and terrestrial ecosystem research.
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.IASTATE.C.DB6465.2024
10.25380/iastate.c.5078492.v3