Quantitative Estimation of Causality and Predictive Modeling for Precipitation Observation Sites and River Gage Sensors

Abstract

This project seeks to investigate two questions: correlations from precipitation measurement sensors to river gage sensors, and predictive modeling of peak river gage heights during precipitation events. First, if correlations can be quantified, then a predictive model can be explored to predict peak water levels at river gage sensors, in response to precipitation inputs. Answering both research questions can provide early flood detection benefits and provide quantitative time assessments for flood risks. An extensive data-driven study was conducted across a geographical area of the U.S, spanning the time period 2008-2016 to identify river gage sensors that are closely correlated to nearby rainfall events. More than 1000 precipitation observation sites were identified and for each precipitation site, nearby river gage stations/sensors were ranked using a cross correlation measure. The cross correlation measures provide information such as which river gage sensors are most sensitive to nearby precipitation inputs. Predictive machine learning models were also developed around each rainfall-river gage pair to learn from historical rainfall and river gage levels, and then predict peak river gage heights. The predictive models generated were accurate and verified a strong causality between precipitation events and river gages that were sensitive to such events. A web-based and map-based decision support and visualization tool was also developed to depict the causality between precipitation and river gage sites and to graphically display the results of the predictive models. This study found about 3500 strongly correlated rain station and river gage pairs. Machine Learning models for these pairs yield high accuracy - 80 percent and above.