HydroDeep -- A Knowledge Guided Deep Neural Network for Geo-Spatiotemporal Data Analysis

TL;DR

HydroDeep is a hybrid deep neural network that integrates process-based models with CNN and LSTM to improve geo-spatiotemporal data analysis for regional climate change and flood prediction, demonstrating superior performance and transfer learning capabilities.

Contribution

The paper introduces HydroDeep, a novel hybrid neural network architecture combining process-based models with deep learning for enhanced geo-spatiotemporal analysis and transfer learning in climate prediction.

Findings

HydroDeep outperforms standalone CNN and LSTM models in efficiency.

Pre-trained HydroDeep effectively transfers knowledge across regions.

HydroDeep reduces training time for new regions through transfer learning.

Abstract

Due to limited evidence and complex causes of regional climate change, the confidence in predicting fluvial floods remains low. Understanding the fundamental mechanisms intrinsic to geo-spatiotemporal information is crucial to improve the prediction accuracy. This paper demonstrates a hybrid neural network architecture - HydroDeep, that couples a process-based hydro-ecological model with a combination of Deep Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) Network. HydroDeep outperforms the independent CNN's and LSTM's performance by 1.6% and 10.5% respectively in Nash-Sutcliffe efficiency. Also, we show that HydroDeep pre-trained in one region is adept at passing on its knowledge to distant places via unique transfer learning approaches that minimize HydroDeep's training duration for a new region by learning its regional geo-spatiotemporal features in a reduced number of iterations.