Time series decomposition of remotely sensed land surface temperature and investigation of trends and seasonal variations in surface urban heat islands

Previous time series methods have difficulties in simultaneous characterization of seasonal, gradual, and abrupt changes of remotely sensed land surface temperature (LST). This study proposed a model to decompose LST time series into trend, seasonal, and noise components. The trend component indicates long-term climate change and land development and is described as a piecewise linear function with iterative breakpoint detection. The seasonal component illustrates annual insolation variations and is modeled as a sinusoidal function on the detrended data. This model is able to separate the seasonal variation in LST from the long-term (including gradual and abrupt) change. Model application to nighttime Moderate Resolution Imaging Spectroradiometer (MODIS)/LST time series during 2000-2012 over Beijing yielded an overall root-mean-square error of 1.62K between the combination of the decomposed trend and seasonal components and the actual MODIS/LSTs. LST decreased (similar to -0.086K/yr, p<0.1) in 53% of the study area, whereas it increased with breakpoints in 2009 (similar to 0.084K/yr before and similar to 0.245K/yr after 2009) between the fifth and sixth ring roads. The decreasing trend was stronger over croplands than over urban lands (p<0.05), resulting in an increasing trend in surface urban heat island intensity (SUHII, 0.0220.006K/yr). This was mainly attributed to the trends in urban-rural differences in rainfall and albedo. The SUHII demonstrated a concave seasonal variation primarily due to the seasonal variations of urban-rural differences in temperature cooling rate (related to canyon structure, vegetation, and soil moisture) and surface heat dissipation (affected by humidity and wind).