Using a BP Neural Network for Rapid Assessment of Populations with Difficulties Accessing Drinking Water Because of Drought

ABSTRACT

Accurately predicting the populations with difficulties accessing drinking water because of drought and taking appropriate mitigation measures can minimize economic loss and personal injury. Taking the 2013 Guizhou extreme summer drought as an example, on the basis of collecting meteorological, basic geographic information, socioeconomic data, and disaster effect data of the study area, a rapid assessment model based on a backpropagation (BP) neural network was constructed. Six factors were chosen for the input of the network: the average monthly precipitation, Digital Elevation Model (DEM), river density, population density, road density, and gross domestic product (GDP). The population affected by drought was the model's output. Using samples from 50 drought-affected counties in Guizhou Province for network training, the model's parameters were optimized. Using the trained model, the populations in need were predicted using the other 74 drought-affected counties. The accuracy of the prediction model, represented by the coefficient of determination (R²) and the normalized root mean square error (N-RMSE), yielded 0.7736 for R² and 0.0070 for N-RMSE. The method may provide an effective reference for rapid assessment of the population in need and disaster effect verification.

Key Words:

• geographical factor
• backpropagation (BP) neural network
• populations with difficulties accessing drinking water because of drought
• rapid assessment
• 2013 Guizhou Extreme Drought of China