Using Big-Data to Develop Catchment-Scale Hydrological Models for Chile

Abstract

Streamflow prediction is very important to the economic and human development of a country. For example, it is used in the quantification and distribution of the water resource, and in the design of new hydraulic infrastructure, risk quantification, rapid response to mitigate flooding, etc. For this reason, learning how to improve our estimation of streamflow must be one of the aspirations of any surface hydrologist. Chile has an extensive stream gauge network, which is part of the new CAMELS-CL database. This database also includes data about several static attributes for each of the 516 catchments represented within it, which provides us with a valuable database that can be used to develop process-based and data-based models with the ultimate goal of implementing a national hydrological model.Recent studies have shown that Machine Learning (ML) can provide better predictive performance than traditional process-based (PB) models. In hydrology, Kratzert et al. (2019), Nearing et al. (2020a), and others have reported similar results when comparing an ML-based model with the extensively studied and calibrated SAC-SMA and other benchmark models over the USA. This finding creates the opportunity to bridge the gap between ML-based and PB models by transferring insights gained via the process of developing a ML model into improvements of the PB model(s). With this in mind, we implemented the GR4J process-based catchment model as a baseline, and two ML-based models, Random Forest (RF) decision tree approach, and the Long-Short Term Memory (LSTM) dynamic state variable approach, on 322 selected Chilean catchments. The three models were compared in detail to examine their strengths and weakness, and to determine the best candidate for a national model. Our results showed that none of the three models performed “best” across the entire country, and all of them had problems in the north of Chile, indicating that additional informative attributes and variables must be incorporated into the database. Furthermore, the models showed complementary performance abilities, which opens the opportunity to develop an ensemble of the three or more models in the future to merge their respective strengths. Overall, the model performance results were found to be related to the meteorological forcings, but also with certain climatic conditions such as aridity, which emerges as an important variable to characterize the behaviors of different catchments.