Fusion of landsat and modis thermal imagery: a model based satellite image fusion approach
Student name: Mr Vinay Hudda
Guide: Dr Nithiyanandam Yogeswaran
Year of completion: 2018
Host Organisation: TERI School of Advanced Studies
Abstract: Remote sensing data acquired by satellite sensors comes with a trade off between spatial and
temporal resolution. Satellite sensors that provide high spatial resolution images have low
temporal resolution. Due to this, the satellite images of the Earth with high spatial resolution are
not available on a daily basis. This problem can be addressed by taking advantage of image
fusion methods to merge datasets from different sensors and obtain fused images with high
spatial and high temporal resolution. This study examines the application of regression variable
substitution image fusion model on MODIS Terra thermal imagery to enhance its spatial
resolution from 1km to 30m by fusing it with Landsat – 8 data. The linear regression method
uses the variance and covariance statistical tools to predict the fused image using input data
collected for the study area Delhi, India. The results depict the correlation coefficient (R) values
between fused and original images that ranged from 0.55 to 0.80 for MOD11A1 (daily) product
and 0.53 to 0.78 for MOD11A2 (8 – day composite) product. It was depicted by the results that
the developed model performed marginally better for MOD11A1dataset. The model was tested
for accuracy over different land use land cover classes. The LULC class water showed higher
values of deviation while the vegetation class had lowest deviation in values of temperature. The
developed model produces fused thermal imagery from MODIS TIR dataset at the spatial
resolution of Landsat – 8 TIR for up to five succeeding days and five days prior to the availability
of Landsat – 8 image.
Keywords: Image fusion, Landsat, MODIS, upscale, spatial resolution
