Hybrid Multi Models Ensemble Framework Based on Clustering Algorithms for Runoff Reconstruction

Runoff is the key hydrological process, which is vital to the sustaining of human life on earth in examining
the climate change scenario. There are a lot of hydrological models available to simulate the runoff, but
these models’ outputs have biases due to uncertainty. Most machine learning algorithms cannot capture
the runoff generated by the real-world complex hydrological system accurately. The hybrid model combines
the efficiency of hydrological, machine learning, and ensemble modeling to minimize the bias of output [1],
[2]. The recent development of evolutionary computation in hybrid modelling frameworks combines the
efficiency of different components such as hydrological models, spatial autocorrelation, machine learning,
and machine learning ensemble to estimate robust and less biased runoff [1]. However, these components
need to significantly capture the heterogeneity and similarity of the catchment properties, which are highly
linked with the spatial variation of various hydrological patterns. Clustering is a technique that can group
similar types of hydrological patterns, which can be integrated within a hybrid modeling framework.
However, there is rarely found literature on the hybrid framework, which consists of different clustering
techniques and their ensemble. These clustering algorithms are based on different categories. We proposed
the hybrid ensemble framework based on extended input data, hydrological models, different clustering
algorithms, deep learning, and an ensemble of deep learning to reconstruct the minimum biased surface
runoff. We tested our proposed hybrid framework, which is robust compared to previously developed
frameworks. This proposed hybrid framework methodology will help to develop a new hybrid algorithm
to estimate the less biased surface runoff using various available climate data to understand the dynamics
of surface runoff for different spatial-temporal scales and climates.

 

[1] U. Singh, P. Maca, M. Hanel, et al., “Hybrid multi-model ensemble learning for reconstructing gridded
runoff of europe for 500 years,” vol. Available at SSRN: doi: 10 . 2139 / ssrn . 4188518. [Online].
Available: http://dx.doi.org/10.2139/ssrn.4188518.
[2] S. M. Hauswirth, M. F. Bierkens, V. Beijk, and N. Wanders, “The suitability of a hybrid framework
including data driven approaches for hydrological forecasting,” Hydrology and Earth System Sciences
Discussions, pp. 1–20, 2022.