Projects Funded for Mehdi Nemati

2020-2021

The Impacts of Wildfires on Water Utilities and Communities in California

  • Samane Zare
  • Mehdi Nemati

2019-2020

Unpacking Residential Water Consumption and the Impacts of Nudges: A Machine Learning Application

  • Mehdi Nemati

Abstract

Specific Objectives of the Project:
1. Disaggregate residential water consumption to indoor and outdoor usage using machine learning methods.
2. Estimate effect of HWURs on indoor and outdoor water consumption (obtained in the first objective).
3. Estimate the impact of HWURs on peak hour and day water consumption.
Analysis of the data revealed that we could not analyze objectives two and three using hourly data (the hourly data started around the same time the HWURs program launched). Instead, we use daily data to perform the analysis for objectives 2 and 3. Additional analysis is done to identify rebound effects and it is heterogeneity after the CA water mandate in 2015.

Project Report/Summary of Results:
Increased frequency and severity of droughts and rapidly growing populations increase the stress on water resources in many arid and semi-arid regions worldwide, such as the Western United States. In response to these evolving realities and their associated challenges, water providers often use demand-side management via conservation and efficiency to buffer against short-term water supply shortfalls. The implementation of a smart water metering system in the medium-size water utility in Northern California in 2014 allowed the water utility to record the hourly water consumption of all its customers. This data availability has enabled a large-scale research project to proceed with the aim to disaggregate residential daily water consumption to indoor (e.g., shower, washing machine, etc.) and outdoor (e.g., irrigation) components. Such information can guide the development of alternative tariff structures and other demand management initiatives to reduce peak demand which is a critical parameter for water infrastructure planning and design. We also contribute to the literature on social and economic patterns of water use rebound after the 2015-2016 CA water mandate.

We use hourly residential water consumption data (more than 500 million data points) between 2015-2019 medium-size water utility in Northern California to identify the peak/off-peak use hours and gain insights into how it changed once mandatory restrictions were lifted in June 2016.

Our results illustrate the peak use hours are between 1-5 am, but its distribution changed dramatically after the drought in 2017-2019. There was a shift in the peak hour of consumption from morning (6 am) to the early mornings (4-5 am). Water use distribution became narrower, with decreases in standard deviations while increases in means. Our results also indicate that the water use rebound from the mandate period was 2.8 gallons/hour, which equals 31% of average hourly water use during the mandate period (8.97 gallons/hour). The rebound varies considerably by the hour, season, and consumption, and income levels. This was the highest level of consumption of a day. The highest rebound at 5 am was 11.53 gallons/hour, followed by a rebound at 4 am of 11.40 gallons/hour. The rebounds were noticeably flat during 4 pm-1 am, with a range of 1.08-3.23 gallons/hour. Interestingly, we found that the rebounds were negative during 8 am-2 pm, with the largest rebound of -0.88 gallons/hour. The results showed that the rebound in summer was four times higher than that in Winter. Rebound in quintile five of consumption level was 6.34 gallons/hour while the rebound in quintile one was ten times lower than that.

The second part of our analysis is based on daily data from the same agency. This part estimates how web-based Home Water Use Reports (HWURs) affect household-level water consumption in a medium-size water utility in Northern California. The HWURs under the study share social comparisons, consumption analytics, and conservation information to residential accounts, primarily through digital communications. The data utilized in this part is a daily panel dataset that tracks single-family residential households from January-2013 to September-2019. We found that there is a 6.2% reduction in average daily household water consumption for a typical household who enrolled in the program. We estimate heterogeneous treatment effects by the day of the week, the content of push notifications, and baseline consumption quintile. For the latter, we provide an illustrative test to emphasize how mean reversion can severely bias a naïve panel data estimator for heterogeneous treatment effects when the source of heterogeneity is the outcome variable (e.g., consumption or expenditures). We find evidence that leak alerts are effective in reducing consumption immediately following the alert.