AUTHOR=Juanico Drandreb Earl , Cayson Florwilyn , Patiño Chito TITLE=Noise-Resilience Horizon of Groundwater Potential Maps Generated by Frequency-Ratio Data Integration JOURNAL=Frontiers in Environmental Science VOLUME=8 YEAR=2020 URL=https://www.frontiersin.org/journals/environmental-science/articles/10.3389/fenvs.2020.523988 DOI=10.3389/fenvs.2020.523988 ISSN=2296-665X ABSTRACT=

The study of groundwater distribution is gaining importance due to the mounting pressures exerted by rapid urban growth on water supply, especially in small islands that could experience faster supply deterioration through saltwater intrusion. Understanding the interplay between the groundwater supply and demand dynamics requires seeing the resources beneath the surface. One typical visualization technique is groundwater potential (GWP) mapping, which predicts groundwater’s spatial distribution from measurable variables on or above the Earth’s surface. However, system errors and noise can affect the quality of the input variables, which can influence the reliability and explanatory power of the GWP maps. Herein, we analyzed the effect of noise on the GWP map accuracy for Cebu and Mactan islands, Philippines. We found that the GWP map retains the fidelity of the zonal structure information in the presence of noise in the input map layers. With a combination of two binary-classifier performance curves, we established the noise-resilience horizon. This horizon is the limit noise-level that the input maps may contain such that the GWP maps retain high accuracy. This horizon indicates that the input maps may carry as much as 20% to 25% error without significantly corrupting the GWP map’s predictive accuracy. Our findings contribute to the knowledge of GWP mapping’s accuracy limits, which is valuable as such diagrams comprise the core of decision-support systems in groundwater management. We also anticipate our dither approach as a foundation for the generic assessment of GWP map accuracy, regardless of a priori details of the map-generating model.