Data-Driven Deman Forecasting to Enable Regenerative and Net-Positive Urban Water Systems: A District-Level Case Study

Abstract

Global efforts to move beyond conventional sustainability towards regenerative and net positive water systems require robust, spatially explicit intelligence on future water demand. This study develops a data-driven framework to support regenerative urban water planning by forecasting district- and customer-type-specific water consumption and identifying behaviourally district consumer clusters. Using multi-year billing records disaggregated by district, customer category, year and month, we first clean and harmonise subscriber labels and compute per-subscriber consumption. To capture seasonal dynamics, month information is encoded via sine-cosine transformations, while right-skewed distributions of total consumption, subscriber counts and per-subscriber use are stabilized. Log-transformed consumption variables and seasonality components are then used in a HDBSCAN-based behavioural clustering, with cluster characteristics visualized via standardized heatmaps. For forecasting, we construct monthly time series for each district–customer combination, assess temporal completeness and retain series with limited gaps, which are filled using time-based interpolation. Lagged consumption variables and rolling statistics are engineered to represent short-, medium- and long-term dependencies. A LightGBM regression model with recursive forecasting is trained on pre-2022 data and validated on 2022–2024, using seasonality, subscriber attributes, lags and moving-window features as predictors. The model generates monthly projections of per-subscriber use and subscriber counts through 2026, which are combined to obtain total water demand at district level and aggregated to annual indicators. The results reveal districts with rising or stabilising demand, spatial variation in model errors and behaviourally distinct consumer clusters, providing actionable intelligence for circular, regenerative and net-positive water and wastewater interventions at local and regional scales.