Trends in surface temperature from new long–term homogenized thermal data by applying remote sensing techniques and its validation using in-situ data of five southern European lakes

Recent studies, based on a combination of long-term in-situ and satellite derived temperature data indicate that lakes are rapidly warming at the global scale. Since Lake Surface Water Temperature (LSWT) is highly responsive to long-term modifications in the thermal structure of lakes, it is a good indicator of changes in lake characteristics. There have not been done many studies at a regional scale to understand the lakes’ response to climate change, mainly due to lack of high spatio-temporal data. Therefore, further studies are needed to understand variation in trends, impacts and consequences at a regional scale. It is essential to have highly frequent spatially explicit data to understand the spatiotemporal thermal variations of LSWT. Continuous in-situ water temperature data measured at high temporal resolution from permanently installed stations are becoming increasingly available through GLEON (Global Lake Ecological Observatory Network: http://gleon.org/) or NetLake (Networking Lake Observatories in Europe). But these data are often heterogeneous with different sources and time line, point based, and not available for many lakes around the globe. To establish permanent weather stations for all the large lakes in the world is also not economically viable. As an alternative to direct measurements, remote sensing is considered as a promising approach to reconstruct complete time series of LSWT where direct measurements are missing. Temperature of land/water surfaces is one of the direct and accurate measurements using satellite data acquired in the thermal infra-red spectral region. Furthermore, the availability of daily satellite data since the 1980s at a moderate resolution of 1 km from multiple polar orbiting satellites is an opportunity not to be missed. But owing to the complexities related to earlier satellite missions, and the need of high level of processing, the potential of the historical satellite data in deriving a homogenised LSWT is still not explored well. There is a gap in the availability of long-term time series of LSWT from the satellite data which could be used in understanding the patterns and drivers of thermal variations in large lakes. This thesis aims to fill this gap by developing reproducible and extendable methods to derive homogenised daily LSWT for thirty years from 1986 to 2015. Hence, the main objectives of this thesis are i) to reconstruct thirty years (1986-2015) of daily satellite thermal data as a homogenised time series of LSWT for five large Italian lakes by combining thermal data from multiple satellites, ii) to assess the quality of the satellite derived LSWT using long-term in-situ data collected from the same lakes, iii) to report the seasonal and annual trends in LSWT using robust statistical tests. The first part of the thesis deals with the accurate processing of historical Advanced Along-Track Scanning Radiometer (AVHRR) sensor data to derive time series of LSWT. A new method to resolve the complex geometrical issues with the earlier AVHRR data obtained from National Oceanic and Atmospheric Administration (NOAA) satellites has been developed. The new method can accurately process historical AVHRR data and develop time series of geometrically aligned thermal channels in the spectral range of 10.5-12.5 µm. The validation procedure to check the accuracy of image to image co-registration using 2000 random images (from a total of 22,507 images) reported sub-pixel accuracy with an overall Root Mean Square Error (RMSE) of 755.65 m. The usability of newly derived time series of thermal channels to derive LSWT for lakes were tested and validated. Furthermore, crossplatform and inter-platform validations were performed using corresponding same day observations which reported an overall RMSE of less than 1.5 °C. In the second part of the thesis, a new method was developed to derive homogenised daily LSWT standardized at 12:00 UTC from thermal channels of thirteen different satellites. The new method is implemented for Lake Garda in Northern Italy developing time series of homogenised daily LSWT for last thirty years from 1986 to 2015. The sensors used in this study are the AVHRR from multiple NOAA satellites, Along Track Scanning Radiometer (ATSR) series from European Remote Sensing (ERS) satellites and Moderate Resolution Imaging Spectroradiometer (MODIS) from Aqua and Terra satellites. The LSWT time series are then validated using long-term in-situ data obtained from a deep and a shallow sampling location in the lake. Validation of LSWT from individual satellites against corresponding in-situ data reported an overall RMSE of 0.92 °C. The validation between final homogenised LSWT and the in-situ data reported a coefficient of determination (R2) of 0.98 and a RMSE of 0.79 °C. In the third part of the thesis, homogenised daily LSWT for the last thirty years (1986-2015) were developed for five large lakes in Italy using the newly developed methods. The LSWT time series was validated against the in-situ data collected from the respective lakes. Furthermore, long-term trend analysis to study the seasonal and annual variations in LSWT over thirty years was performed over the newly developed LSWT data. The validation procedure reported an average RMSE and Mean Absolute Error (MAE) of 1.2 °C and 0.98 °C, respectively, over all the lakes. The trend analysis reported an overall regional summer warming rate of 0.03 °C yr-1 and an annual warming rate of 0.017 °C yr-1. During summer, all studied sub-Alpine lakes showed high coherence in LSWT to each other. The summer mean LSWT of Lake Garda, located in the sub-Alpine region also exhibit high temporal coherence with that of central Italian Lake Trasimeno. Annually, mean LSWT of all subAlpine lakes were found to be highly coherent to each other, while mean LSWT of Lake Trasimeno resulted less coherent to the other lakes. Overall, the thesis aims at contributing to the accurate processing of the various historical satellite data and the development of a new method which allows to merge them into a unified, longest possible time series of LSWT. The newly developed methods used open source geospatial software tools, which ensure the reproducibility and also extensibility to any other geographic location given the availability of satellite data. Although this study is using LSWT as the primary physical variable, the developed methods can be used to derive any other time series of land and water based regional products from satellite data