Figure 5. Predictive
accuracy of AOD retrievals from Landsat imagery against ground-based
measurements at the individual-site scale from 2013 to 2022 using the
spatiotemporally independent validation approach (i.e., data samples
from the years 2015 to 2020 for training and other years for testing):
(a) correlation coefficient (R), (b) median bias, (c) mean absolute
error (MAE), (d) root-mean-square error (RMSE), and the percentages of
retrievals falling within the (e) expected error envelope (= EE) of the
MODIS Deep Blue product and (f) the goal uncertainty defined by the
Global Climate Observing System (GCOS).
Table 2. Predictive
accuracy of AOD retrievals from Landsat imagery in specific continental
regions of interest and on a global scale using the spatiotemporally
independent validation approach, i.e., data samples from the years 2015
to 2020 for training and other years for testing.