Reconstructing NDVI and land surface temperature from cloud cover pixels of Landsat-8 images and assessing vegetation health index in the Northeast region of Thailand

Critical applications of satellite data products include monitoring vegetation dynamics using vegetation indices such as Normalized Difference Vegetation Index (NDVI) and thermal characteristics such as Land Surface Temperature (LST) variations across different spatial and temporal scales. One of the major problems with passive remote sensing satellite data products is cloud and shadow cover that void the information from a specific image. The cloud cover problems further amplified during monsoon seasons as the formation of clouds are more frequent during these seasons. The present study proposes temporal aggregation of images and developing Harmonic Analysis of Time Series (HANTS) and Pixel-wise Multiple Linear Regression (PMLR) algorithms for retrieving cloud contaminated NDVI and LST information from Landsat-8 (L8) data products, respectively. The developed algorithms were applied in the northeastern part of Thailand to recover the missing NDVI and LST values from time series of L8 images acquired in 2018. The predicted NDVI and LST values at artificially clouded locations were compared with the corresponding clear pixel values. Additionally, the model predicted LST and NDVI values were also compared with MODIS LST and NDVI data sets. The calculated root mean square (RMSE) values were ranging from 0.004 to 0.05 and 1.05°C to 3.11°C for NDVI and LST variables, respectively. The validation statistics show that these models can be satisfactorily applied to retrieve NDVI and LST values from cloud-contaminated pixels of L8 images. Furthermore, a vegetation health index (VHI) calculated at province level show that most of the western provinces showed a healthy vegetation condition than other provinces in the northeast of Thailand.