A global gridded municipal water withdrawal estimation method using aggregated data and artificial neural network

Area of Science:

• Environmental science
• Hydrology
• Data science

Background:

• Municipal water withdrawal (MWW) data is vital for water supply planning and management.
• Current MWW data is often spatially aggregated or lacking, hindering detailed applications.
• There is a growing demand for spatially explicit, high-resolution MWW data.

Purpose of the Study:

• To develop and evaluate models for estimating global MWW using aggregated data and gridded covariates.
• To identify the most accurate model for global MWW estimation.
• To generate a high-resolution global gridded MWW dataset for 2015.

Main Methods:

• Construction and evaluation of six different models for global MWW estimation.
• Utilizing aggregated MWW data and gridded raster covariates as inputs.
• Employing an artificial neural network-based indirect model (NNM) for its superior performance.

Main Results:

• The artificial neural network-based indirect model (NNM) demonstrated the highest accuracy (R², NMAE, NRMSE) across various spatial scales.
• NNM estimates showed consistency with census and survey data.
• The NNM model outperformed existing global gridded MWW datasets.

Conclusions:

• The NNM model is a reliable method for estimating global gridded MWW at high resolution (0.1 × 0.1°).
• The developed high-resolution MWW data can significantly benefit hydrological models and water resource management.
• The methodology is applicable to other aggregated output learning problems.