Machine Learning Improvement of Streamflow Simulation by Utilizing Remote Sensing Data and Potential Application in Guiding Reservoir Operation

Abstract

Hydro-meteorological datasets are key components for understanding physical hydrological processes, but the scarcity of observational data hinders their potential application in poorly gauged regions. Satellite-retrieved and atmospheric reanalysis products exhibit considerable advantages in filling the spatial gaps in in-situ gauging networks and are thus forced to drive the physically lumped hydrological models for long-term streamflow simulation in data-sparse regions. As machine learning (ML)-based techniques can capture the relationship between different elements, they may have potential in further exploring meteorological predictors and hydrological responses. To examine the application prospects of a physically constrained ML algorithm using earth observation data, we used a short-series hydrological observation of the Hanjiang River basin in China as a case study. In this study, the prevalent modèle du Génie Rural à 9 paramètres Journalier (GR4J-9) hydrological model was used to initially simulate streamflow, and then, the simulated series and remote sensing data were used to train the long short-term memory (LSTM) method. The results demonstrated that the advanced GR4J9–LSTM model chain effectively improves the performance of the streamflow simulation by using more remote sensing data related to the hydrological response variables. Additionally, we derived a reservoir operation model by feeding the LSTM-based simulation outputs, which further revealed the potential application of our proposed technique.

Satellite Data Applications for Sustainable Energy Transitions

“…For example, machine learning has been combined with near-real-time rainfall data from NASA's Tropical Rainfall Measuring Mission (TRMM) and soil moisture data from the National Oceanic and Atmospheric Administration's (NOAA's) Advanced Scatterometer (ASCAT) to simulate streamflow in India (Kumar et al, 2021). Machine learning with various satellite-derived hydrometeorological variables has also been used to calculate streamflow in the Hanjiang River in China (He et al, 2021). Data assimilation, another approach to data fusion, has also improved land surface model predictions of water storage, particularly when multiple satellite data products are combined (Khaki et al, 2020).…”

Modelling and forecasting daily streamflow with reservoir operationin the upper Chao Phraya River basin, Thailand

“…However, it should be noted that the reservoir water balance in reality is more complex due to the contribution of precipitation and evaporation over the reservoir surface (as previously indicated in Chapter 1; Equation (1.1)). Some previous studies found that the meteorological variables played a more important role in ML model performance than timing information (e.g., Zhang et al, 2019;He et al, 2021). Other studies stated that meteorological variables had an insignificant influence on ML model performance, as they were already captured by the timing information, and thus could be neglected (e.g., Yang et al, 2016).…”

“…On the other hand, it can be argued that the addition of Q t−1 can essentially account for the operators' decisions on a daily basis, which have been the main issue in real-time reservoir operation modelling for a long time. Several studies have applied the Q t−1 input and achieved excellent ML model performance (e.g., Chen et al, 2018;Zhang et al, 2018aZhang et al, , 2019He et al, 2021;Hong et al, 2021). In reality, however, using the Q t−1 input is only logical and useful when the Q t−1 data are obtained from prediction instead of observations, considering that the observed Q t−1 data are not available in the case of multi-day outflow forecasting.…”

“…Despite the extensive applications of RNN, LSTM and GRU in real-time reservoir operation modelling, previous studies have primarily focused on historical reservoir outflow simulations. Some studies have implemented the aforementioned algorithms to simulate real-time outflows for small or single-purpose reservoirs (e.g., Zhang et al, 2019;He et al, 2021;Hong et al, 2021;Yang et al, 2021;Gangrade et al, 2022). Some other studies have investigated the algorithms' performance in simulating real-time outflows for large or multi-purpose reservoirs (e.g., Chen et al, 2018;Zhang et al, 2018a;Zheng et al, 2022).…”

“…To account for the real-time judgments and decisions of the operators, one may make use of the previous Q values as the input to simulate the current Q value. This addition has been proven very successful to achieve excellent accuracy of reservoir outflow simulations in several ML model studies (Chen et al, 2018;Zhang et al, 2018aZhang et al, , 2019He et al, 2021;Hong et al, 2021). This is due to the increasing linear relationship between the inputs and output.…”