Urban water and power systems are usually studied separately, but in reality they are linked: watering lawns, doing laundry, taking a hot shower, and running the dishwasher all require both water and electricity at the same time. This link becomes especially important under climate change, when heat and drought simultaneously boost demand for air conditioning and irrigation. The work sits alongside studies of climate sensitivity of individual urban systems, but unlike them, it treats the water–electricity pair as a single object. Neighboring works in our archive deal with El Niño prediction, precipitation regime shifts, and solar cycles, but do not connect them directly to joint urban resource demand.
The authors used a multivariate extension of gradient-boosted decision trees, called multivariate tree boosting. This method predicts two dependent variables at once (water and electricity consumption), simultaneously minimizing error for each and maximizing explained covariance between them. Data were collected from 2007 to 2016 for six cities: Chicago, Columbus, Indianapolis, Minneapolis, Cleveland, and Madison. Predictors included temperature (dry-bulb and dew point), relative humidity, wind speed, precipitation, and the El Niño index. Accuracy was assessed with five-fold cross-validation, and periodograms were used to identify seasonality. The seasonal component was removed from the time series to separate the climate signal from the calendar effect.
Before removing seasonality, the model predicted electricity better than water (R² up to 0.87 for electricity in Indianapolis versus 0.83 for water). However, after removing seasonality, the picture changed: water became better predicted than electricity in all cities except Chicago and Cleveland. For example, in Indianapolis, R² for water was 0.71, while for electricity it was 0.41. This indicates that water is more sensitive to climatic variables than electricity. The key predictor turned out to be the El Niño index, which explained a large part of the covariance between water and electricity in each city. Strong El Niño events led to an overall increase in consumption of both resources. Comparison with a univariate model showed that the multivariate model predicts better in most cities, confirming the importance of accounting for interdependence.
This work differs from neighboring studies in that it treats water and electricity demand as a single system rather than as two independent variables. This allows capturing covariance that is lost when modeling separately. The results show that the El Niño index can serve as a simple and reliable indicator for forecasting peak loads on both resources simultaneously. For utilities, this means the ability to prepare in advance for periods of high demand, for example, during strong El Niño events.
The model could be extended by including socioeconomic variables such as household income, family size, or the presence of pools. It would also be interesting to test the approach in other climate zones, such as arid regions where the water–electricity link might be even stronger due to desalination or water pumping. Further development could include forecasting based on seasonal El Niño predictions, enabling early warnings for utilities.
The results can be used by municipal utilities for planning peak loads and developing targeted resource conservation programs. They are also useful for researchers studying the water–energy nexus in urban settings.
The next step could be to test the model on cities with different climates and different water consumption patterns, for example, in the southwestern United States. It would also be worth investigating why in Cleveland and Madison the univariate model predicted water better than the multivariate one, possibly due to the smaller size of those cities.
This work links climate variability to joint water and electricity demand, which is a key issue for the resilience of urban infrastructures. It shows that ignoring interdependence can lead to underestimating peak loads and ineffective conservation measures.
In Chicago and Cleveland, no seasonality was found in water consumption, unlike in the other cities. This may be because both cities are located on the shores of the Great Lakes, which smooths out climatic fluctuations.