IEEE Geoscience and Remote Sensing Magazine - March 2022 - 310

level of accuracy in the instruments employed to capture
the data. These limitations mean the presence points better
represent the realized distribution of a species instead of
their true potential distribution [4]. Pseudoabsences created
based on the values of the explanatory variables at
recorded presence points are subject to how well the environmental
gradients for a species are exemplified in the
presence samples. When the samples are less representative,
model performance may suffer when making predictions
with unseen testing data [30], [109]. This can be particularly
evident when developing absences using remotely
sensed variables that have their own inherent accuracy limitations.
Some of these effects can be mitigated by using a
geographic exclusion method to generate pseudoabsences.
Geographic separation also has inherent environmental
separation in most study regions, and therefore pseudoabsences
may more likely represent true absences that
can help models distinguish between the two classes [4].
However, the effectiveness of this approach may be limited
by the level of spatial aggregation in the presence samples
[30], [110]. For this study, we achieved good results with
the environmental normalization and geographic exclusion
approaches. The success across multiple models seems
to indicate that the presence samples were representative
enough of the spatial and environmental distribution of
the locusts at a regional scale for the models to establish
distinct correlations between explanatory variables and
class labels. Whether these results could be duplicated
at local and global scales [111], which may shift the environmental
gradients of the features [112], warrants further
study. The randomly generated absences had a high
number of false negatives in both study areas. The models
were less successful because they were defining relationships
based on the correlation between the species and the
explanatory variables in environmental space. Randomly
generated absences points can be interpreted by a model
as " false absences " because the environmental variables
match those of the present samples [3]. If we had been using
regression-based models as opposed to classification
models, these results may have been different [113].
LIMITATIONS OF COARSE SOIL MOISTURE
The spatial mismatch between the ground sampling distance
(GSD) of a pixel and an observed geographic phenomenon
is one of the primary limitations of satellite remote sensing,
particularly for thermal remote sensing applications [114].
Jackknife of Test Gain for Present
landcover_annual_sudan
LST_summer_sudan
ndvi_summer_sudan
rainfall_summer_sudan
sm_summer_sudan
1
0.92
0.85
0.77
0.69
0.62
0.54
0.45
0.38
0.31
0.23
0.15
0.08
0.2 0.4 0.6 0.8 1 1.2 1.4
Test Gain
Without Variable
With Only Variable
With All Variables
(a)
landcover_annual_sudan
LST_winter_sudan
ndvi_winter_sudan
rainfall_winter_sudan
sm_winter_sudan
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Test Gain
1
0.92
0.85
0.77
0.69
0.62
0.54
0.45
0.38
0.31
0.23
0.15
0.08
(c)
Without Variable
With Only Variable
With All Variables
(d)
FIGURE 8. The MaxEnt average results for (a) the summer predicted probability of prevalence, (b) the summer model feature importance,
(c) the winter predicted probability of prevalence, and (d) the winter model feature importance for the Sudan study site [95].
310
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE MARCH 2022
Jackknife of Test Gain for Present
(b)
Environmental Variable
Environmental Variable

IEEE Geoscience and Remote Sensing Magazine - March 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2022