On Monday, March 27th at 10:00 a.m. Ronald Reagan Okoth will present his M.Sc. theses “Fusing Remote Sensing Data: Application in Soil Temperature Prediction” in room 1.005 (Zentrales Hörsaal- und Seminargebäude).
From the abstract: Understanding soil temperature (ST) is crucial for studying land-atmosphere interactions and ecosystem responses to climate change. However, the lack of global in-situ soil temperature measurements has limited the production of high-resolution soil temperature databases that are spatially continuous. In this study, we developed a framework for estimating soil temperature using high-resolution synthetic images generated by fusing Landsat and MODIS land surface temperature (LST) bands using the Spatial Temporal Adaptive Reflectance Fusion Model (STARFM) algorithm. Our fusion process achieved an R2 of 0.64 and an RMSE of 3.80 when compared to the original MODIS. The resulting synthetic LST images, in combination with soil temperature measurements from DWD stations, were used to build a Random Forest regressor that extrapolates point measurements to a spatially continuous daily soil temperature dataset at 100 m resolution and achieved an R2 of 0.97 and RMSE of 2.71. Our results demonstrate that high-resolution synthetic LST images can be successfully applied in soil temperature estimation. We observed expected patterns in soil temperature maps, such as forested areas recording lower temperatures than open fields/agricultural areas in summer, and the differences being less pronounced in wet seasons. However, our fusion results deviated slightly from values in the literature, possibly due to different climatic conditions in study areas and uncertainties from the ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) instrument. While previous studies focused on simulating Normalized Difference Vegetation Index (NDVI) values or actual Landsat band reflectance (NIR and VIS) using Landsat and MODIS, our study simulated Landsat’s Land Surface Temperature (LST) band, which is computed from ASTER data. This may have introduced uncertainties associated with the ASTER data and amplified errors in our study. Despite this, our study highlights the potential of remote sensing data fusion for soil temperature estimation.
Supervisors: Dr. Almudena García-García, Hemholtz Centre for Environmental Research (UFZ), Prof. Dr. Tobias Ullmann, University of Würzburg