Land Surface Temperature Time Series in the Upper Mekong Basin

Abstract

Land surface temperature (LST) is an important indicator for climate change and can be sensed remotely by satellites with a high temporal resolution on a broad spatial scale. In this research MODIS LST are used to derive an 11 year time series from the upper Mekong delta to analyse the development of LST. The data shows the regular annual curve of surface temperature with maximum values in summer and minimum values in winter. Average temperatures in Southern parts of the basin are higher than in the Northern part. To assess temporal variations, maps of monthly anomalies are created. In a selected area in the province Qinghai for example the daytime monthly anomalies range from -25 K to +19 K, what are exceptional values however. Average deviations are found in the range between -6 K and +5 K. Daytime monthly anomalies were also compared with monthly anomalies of NDVI and a correlation coefficient r2 = 0.39 was found for the test area: In warmer month also the NDVI tends to be higher. Nighttime anomalies correlated with r2 = 0.18 only. Some inter-annual variations occur mainly during summer: in some years a two peak distribution is found, which could be traced back to the low number of observations in the respective months. A main challenge of optical satellite data is the cloud contamination over the area in the summer months, where peak rainfall occurs. In the test area of the province Qinghai for example, the average number of available daytime observations of MODIS LST in July ranges between 9 and 16 observations per month. It can be assumed that any climate statistics calculated from such data might be biased and an appropriate gap filling method would be helpful. In this context, an algorithm was developed to overcome this problem. It estimates missing LST from existing LST data from a given environment in the same scene using stable neighbourhood relations and is to create a set of clear-sky LSTs. Modelled daytime LST correlated well with the original data for the selected test area (r2 = 0.75) but showed quite a high RMS (= 5.6 K). The newly created set is analysed in terms of its suitability to fill the gaps in the LST time series. Another issue with MODIS LST data are day-to-day differences in the acquisition time. A temporal homogenisation was applied to all LST data, converting them to one fixed acquisition time using adjusted ECMWF ERA Interim skin temperature data. The daytime original and the homogenised dataset are still well correlated (r2 = 0.97 for the selected test area) and the effect on the daily/monthly mean is found to be small (RMS = 1.6 K/0.35 K).