Application of Machine Learning in Stormwater Risk Management for the Johnson County Stormwater Management Program, Kansas

The Stormwater Management Program (SMP) is a Johnson County, Kansas program which partners with the 20 cities in the County to manage stormwater and is funded by a 1/10th of one percent, county-wide sales tax. It administers these funds on behalf of the Cities, historically by providing matching funds to Cities for eligible projects, including study, design, and construction projects. In 2016, SMP as part of new strategic asset management program implemented watershed-based approach to fund projects that incorporate flooding, water quality, and system management. Under 'System Management' program, SMP started funding inspection, rehabilitation, and replacement of stormwater asset projects. As part of this program, SMP developed a risk-based tool to prioritize stormwater assets. This tool is used to assign a prioritization score to all eligible assets contained in County-wide asset database. This prioritization score is calculated using Likelihood of Failure (LoF), Consequence of Failure (CoF), and total risk (Business Risk Score, BRE). The two fundamental building blocks for defining total risk (BRE) are LoF and CoF. LoF describes the chance of an asset failure occurring and CoF measures the severity of the impacts if an asset were to fail. Total Risk or BRE = LoF * CoF Currently, SMP employs a linear age-based degradation model and incorporates an adjustment factor for increased salt load in estimating LoF. to prioritize inspection of stormwater assets (hard assets). For rehab/replacement projects, field verified condition score is used. However, existing field verified condition rating systems like the National Association of Sewer Service Companies' (NASSCO) Pipeline Assessment and Certification Program (PACP) and Water Resource Commission (WRC) were initially developed for wastewater systems. These standard ratings do not capture the environment factors and other variables specific to stormwater pipes. Currently, no standardized methods exist for assessing the condition of stormwater pipes and structures in the U.S. Given these challenges, SMP engaged with NEER to utilize its cloud-based Machine Learning (ML) solution to identify the risk condition of the stormwater assets and implement a proactive data driven asset management program. As a part of this project, NEER developed a Machine Learning (ML) Model that is specific to Johnson County SMP to calculate LoF for all the hard assets such as inlets, junction boxes, bridges, culverts, enclosed gravity. All of these assets are represented either as Links or Nodes. During the ML model creation, all the data obtained from AIMS and local municipalities (physical, functional/operational) were standardized. The NEER team developed micro-ML models to populate several missing parameters for few nodes and links. In addition, several environmental parameters were also superimposed to the existing datasets. After the normalization of the datasets, the original datasets (113,124 links and 122,957 nodes) that had field verified conditions were selected for model training and validation. There were 39,814 links (35% of total links) and 44,600 nodes (36% of total nodes) that had field verified conditions. NEER was able to develop a best performing ML model using 80% of the data (field verified conditions data) for model training and the rest of the 20% of the data (field verified conditions data) for model validation. This ML model is able to predict LoF with an accuracy of 90% & 91% respectively for the existing nodes and links. This SMP specific LoF prediction ML model was configured to continuously train and optimize itself to improve accuracy over time. NEER also adopted the same methodology that is currently being used by SMP to calculate the CoF and Business Risk Exposure (BRE)/Total Risk score. This CoF and BRE/Total Risk score calculation was implemented in the NEER Platform, so that SMP can calculate CoF and BRE/Total Risk for each asset in Watershed Organization 1.