PosterPDF Available

Machine learning algorithms for potential recharge areas mapping in high-altitude Andean mountains with scarce field information

Authors:

Abstract

The high-altitude wetlands in the Central Andes are unique ecosystems located above 4000 masl in the Bolivian Altiplano. Slight modifications in anthropogenic interventions and the threat of global climatic changes constitute a threat to their existence. To address the issue, wetland mapping provides valuable information for the conservation and management of natural wetland areas. In this context, machine learning algorithms are valuable in regions with scarce field information. The objective of this study was to test machine learning algorithms to map Andean wetlands, to promote its management of groundwater recharge zones.
Machine learning algorithms for potential recharge areas mapping in
high-altitude Andean mountains with scarce field information
Aliaga, E.1, Soria, F.1
1 Centro de Investigación en Agua, Energía y Sostenibilidad, Universidad Católica Boliviana
INTRODUCTION
The high-altitude wetlands in the Central Andes are unique ecosystems located above 4000 masl in the Bolivian Altiplano. Slight modifications in anthropogenic interven-
tions and the threat of global climatic changes constitute a threat to their existence. To address the issue, wetland mapping provides valuable information for the conserva-
tion and management of natural wetland areas. In this context, machine learning algorithms are valuable in regions with scarce field information. The objective of this
study was to test machine learning algorithms to map Andean wetlands, to promote its management of groundwater recharge zones.
Figure 1. Sensitivity analysis performed by LASSO
Figure 2. Visual adjustment of RFR results
METHOD
The first step consist on apllying applying the machine
learning algorithm Least Absolute Shrinkage and Se-
lection Operator LASSO for the sensitivity analysis of
recharge parameters. Then, there were compared the
Random Forest Regressor RFR, Support Vector Re-
gressor SVR, and Multivariate Adaptative Regression
Splines MARS regression supervised machine lear-
ning algorithms. Its performance was measured by
the Receiver Operating Characteristic ROC. The re-
sults obtained were validated by means of a visual
adjustment of the maps superimposed on Google
Earth satellite images.
RESULTS
The application of the LASSO (Fig.1.) algorithm obtai-
ned the following sensitivity analysis for the recharge
parameters introduced for the analysis area. The ap-
plication of the RFR (Fig.2.) algorithm showed good
results for areas with slopes of 44 - 86 degrees; the
SVR algorithm showed good performance for areas
with slopes of 0 - 32 degrees, while for areas with
Figure 3. Performance of the algorithms by ROC
slopes of 54 - 70 degrees its performance was inaccurate. The
application of the MARS algorithm showed trivial results. Since
some results were good for certain areas with slopes of 0 - 12
degrees and areas with slopes of 54 - 77 degrees were erro-
neously flagged. The application of the ROC (Fig.3.) algorithm
obtained the performance of the machine learning algorithms.
CONCLUSION
Based on the results obtained and its performance, the best
algorithm for wetland mapping of high-altitude Andean moun-
tains with scarce field information, is RFR.
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.