96
P. J. Deshpande et al.
4 Results and Discussion
The monthly homogeneity maps are presented in the Fig. 3. A temporally averaged
map for the study period is also provided at the end for comparison. The threshold
value [1] is set as 10%.
The homogeneity maps for different months show homogeneous areas for
different months. Same cluster ID belongs to same prediction variable, i.e., soil
moisture. These clusters, i.e., homogeneous areas, are spread in different locations,
i.e., the prediction variables may be same for different locations. Moreover, across
the different months, it is observed that some of the clusters are changing while
some are not changing for the same location. The clusters which are not changing
across the months are the locations where we do not expect much temporal as well
as spatial variablity. On the other hand, in the clusters which are changing across the
months, a higher temporal variability is observed. The locations of less variability
requires lesser samples as compared to the locations of higher variability. In the areas
with higher variability, more samples (in terms of frequency and spatial distribution)
are needed for a good representation. Finally, the temporally lumped map has more
clusters as the accumulated variability over the time increases leading to generation
of more clusters, i.e., homogeneous areas.
5 Conclusion
The homogeneity maps generated at a monthly scale can help to understand the
variability of the prediction variable using auxiliary variables. Here, the number of
clusters shows the minimum number of required sampling points. The pixels with
same cluster ID are the locations with similar soil moisture. These maps show the
representative area of different sampling locations at different times, thus helping in
deciding the number and locations for samplings. Sampling in the same homogeneous
area leads to redundancy, as the expected value of soil moisture is same. On the other
hand, the sample can be taken in a different location. The proposed methodology
avoids redundancy and thus presents a cost-effective way for sampling.
The framework can be applied in an online setting where any new input satellite
data can be used to update the homogeneity maps. Also, the patchy nature of the
homogeneity maps can be removed, i.e., the small partches of different cluster IDs
can be merged together to make a single large cluster, by spatial aggregation.
P. J. Deshpande et al.
4 Results and Discussion
The monthly homogeneity maps are presented in the Fig. 3. A temporally averaged
map for the study period is also provided at the end for comparison. The threshold
value [1] is set as 10%.
The homogeneity maps for different months show homogeneous areas for
different months. Same cluster ID belongs to same prediction variable, i.e., soil
moisture. These clusters, i.e., homogeneous areas, are spread in different locations,
i.e., the prediction variables may be same for different locations. Moreover, across
the different months, it is observed that some of the clusters are changing while
some are not changing for the same location. The clusters which are not changing
across the months are the locations where we do not expect much temporal as well
as spatial variablity. On the other hand, in the clusters which are changing across the
months, a higher temporal variability is observed. The locations of less variability
requires lesser samples as compared to the locations of higher variability. In the areas
with higher variability, more samples (in terms of frequency and spatial distribution)
are needed for a good representation. Finally, the temporally lumped map has more
clusters as the accumulated variability over the time increases leading to generation
of more clusters, i.e., homogeneous areas.
5 Conclusion
The homogeneity maps generated at a monthly scale can help to understand the
variability of the prediction variable using auxiliary variables. Here, the number of
clusters shows the minimum number of required sampling points. The pixels with
same cluster ID are the locations with similar soil moisture. These maps show the
representative area of different sampling locations at different times, thus helping in
deciding the number and locations for samplings. Sampling in the same homogeneous
area leads to redundancy, as the expected value of soil moisture is same. On the other
hand, the sample can be taken in a different location. The proposed methodology
avoids redundancy and thus presents a cost-effective way for sampling.
The framework can be applied in an online setting where any new input satellite
data can be used to update the homogeneity maps. Also, the patchy nature of the
homogeneity maps can be removed, i.e., the small partches of different cluster IDs
can be merged together to make a single large cluster, by spatial aggregation.