University of Twente Student Theses


Statistical downscaling of methane emissions from abrupt permafrost thawing over Alaska

Chaturvedi, Vasudha (2021) Statistical downscaling of methane emissions from abrupt permafrost thawing over Alaska.

[img] PDF
Abstract:Permafrost regions are areas below the Earth's soil surface that stays permanently frozen for two consecutive years. The permafrost layer is highly rich in organic matter; with global warming and uprising temperatures, there is a severe threat of abrupt permafrost thawing leading to rapid bacterial degradation of organic carbon. As a result, a disproportionate amount of greenhouse gases such as CH4 and CO2 are released into the atmosphere. Methane, amongst all greenhouse gases, contributes more to global warming. UNFCCC recognizes various sources of methane emissions causing positive climate feedback, but emissions from abrupt permafrost thawing are the least understood feedback to the warming climate. In order to monitor, understand and record the changing emissions at the regional level, it is essential to obtain fine-resolution data. The development of spaceborne satellite-derived remote sensing methane data brought a rapid increase in homogeneous and global level data. However, these datasets are coarse in resolution limiting the information gained from these images. Therefore, there is a need to downscale these datasets to obtain maximum information and conduct regional-level studies. There is hardly any study on regional level monitoring of methane emissions from abrupt permafrost thawing using TROPOMI S5P downscaled data. Thus, this study firstly aims at recognizing variables contributing to rapid permafrost thawing leading to increased methane emissions. Secondly, comparing and estimate the downscaling technique which gives the best estimates of methane data. Thirdly, to monitor changes over methane emissions using the selected method for downscaling. The study was conducted in four counties of Alaska, The United States of America. Estimating variables affecting the magnitude of methane emissions were obtained using the AIC index, stepwise regression, correlation coefficient, and multicollinearity. Due to the non-linear and linear relationship between auxiliary variables and dependent variables, various regression models were performed to estimate the model that explains the maximum variance. Three downscaling techniques of RK, ATAK, and ATARK were used for comparison. The three techniques accuracy was evaluated against the coarse resolution dataset using leave-one-out cross-validation (LOOCV) and six indices (mean, SD, variance, MAE, CC, CH). Lastly, to monitor changes in methane emissions, the selected technique is used for downscaling the recorded dataset over the years The statistical analysis showed MODIS1 LSTDAY, MODIS2 LSTNIGHT, VIIRS NDVI, GMTED 2010 DEM, and VIIRS EVI as the chosen predictors against the TROPOMI S5P methane product as the dependent variable. These predictors showed a weak and statistically significant correlation values. The linear regression model was selected with the lowest RMSE (11.47), MAE (9.04), and highest R squares value (0.29). Amongst the three techniques, ATAK showed a slightly better performance with a CH value of 0.99, 0.99 for tile 17 and 21, and CC value of 0.98,0.99. Monitoring the downscaled images from August 2018-May 2021 did not produce much information about the rate of changes in methane emissions. However, what might be inferred is a spatial shift in methane emissions from South to North to South again which is due to seasonal changes.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page