Prediction in Data Scarce Regions: A Novel Clustering and Classification Method for Simulation of Earth and Environmental Systems

Abstract:

Prediction in ungauged basins (PUB) has been one of the main challenges of hydrologist in the last decades. Regionalization techniques are widely used to address this problem where the focus is to estimate the time series or signatures of streamflow in ungauged areas. Here, we show that regionalization can be generalized for a wide range of purposes. We first introduce a generic Machine Learning framework, named BHC-CA that combines Behavioral Hierarchical Clustering (BHC), Classification (C) and Aggregation (A) algorithms for the regionalization of different environmental models. Then, the efficacy and reliability of this framework is demonstrated for two different case studies at catchment scale: In the first study, we regionalize a DEM-based statistical model to generate probabilistic floodplain maps in data-scarce regions (Jafarzadegan et al., 2020). In the second study, we regionalize streamflow rating curves to estimate rating curves at ungauged basins and detect homogeneous regions for hydrodynamic modeling (Jafarzadegan and Moradkhani, 2020). The result of both case studies confirm the high capability of the BHC-CA framework for predicting environmental variables in data-scarce regions.

Jafarzadegan, K., Merwade, V., Moradkhani, H., 2020. Combining clustering and classification for the regionalization of environmental model parameters: Application to floodplain mapping in data-scarce regions. Environ. Model. Softw. 125, 104613. https://doi.org/10.1016/j.envsoft.2019.104613

Jafarzadegan, K., Moradkhani, H., 2020. Regionalization of stage-discharge rating curves for hydrodynamic modeling in ungauged basins. J. Hydrol. 589, 125165. https://doi.org/10.1016/j.jhydrol.2020.125165