A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Identifying impacts of global climate teleconnection patterns on land water storage using machine learning
Study Region: The Zambezi River Basin in southern Africa Study focus: Modelling global freshwater systems is difficult because of the complexities of climate and anthropogenic influence on hydrological systems, especially at local scales. The rapid changes in hydrological systems caused by the influence of global climate trigger complex processes that make traditional machine learning algorithms limited in quantifying impacts of climate variability on freshwater. In this study, we developed a novel machine learning routine based on the Gaussian process regression (GPR) technique to improve understanding of the interaction of non-linear climatic variables with hydrological stores (includes surface water and terrestrial water storage-TWS). The GPR is built on the principle of the Gaussian process, which is a stochastic process that simplifies multivariate Gaussian distribution to infinite-dimensional space, such that the distributions over function values can be defined. The prediction of the GPR is tested using twenty-three independent climate variables against satellite observations of TWS between April 2002 and June 2017. We explored the use of a characteristic length scale for the kernels, which we tagged as ‘first order kernels’, and another set of kernels having a separate length scale for each discrete predictor. The latter was implemented using the automatic relevance determination (ARD), tagged as ‘higher order kernels’. The first and higher order kernels of the GPR technique were further examined using multivariate statistical indices to reveal the close relationship among hydro-climatic similarity and predictability. New hydrological insights for the region: Our results indicate that the large fluctuations of TWS in our tentative test-bed (the Zambezi Basin) for the GPR technique are largely caused by strong changes in sea surface temperature and global teleconnection patterns of the nearby oceans. The GPR introduced in this study provides an improved modelling framework to keep track on the influence of these global climate teleconnection patterns on major hydrological systems, like the Zambezi Basin, which contributes to global hydro-climatology but currently showing large amplitudes of land water storage loss.
Identifying impacts of global climate teleconnection patterns on land water storage using machine learning
Study Region: The Zambezi River Basin in southern Africa Study focus: Modelling global freshwater systems is difficult because of the complexities of climate and anthropogenic influence on hydrological systems, especially at local scales. The rapid changes in hydrological systems caused by the influence of global climate trigger complex processes that make traditional machine learning algorithms limited in quantifying impacts of climate variability on freshwater. In this study, we developed a novel machine learning routine based on the Gaussian process regression (GPR) technique to improve understanding of the interaction of non-linear climatic variables with hydrological stores (includes surface water and terrestrial water storage-TWS). The GPR is built on the principle of the Gaussian process, which is a stochastic process that simplifies multivariate Gaussian distribution to infinite-dimensional space, such that the distributions over function values can be defined. The prediction of the GPR is tested using twenty-three independent climate variables against satellite observations of TWS between April 2002 and June 2017. We explored the use of a characteristic length scale for the kernels, which we tagged as ‘first order kernels’, and another set of kernels having a separate length scale for each discrete predictor. The latter was implemented using the automatic relevance determination (ARD), tagged as ‘higher order kernels’. The first and higher order kernels of the GPR technique were further examined using multivariate statistical indices to reveal the close relationship among hydro-climatic similarity and predictability. New hydrological insights for the region: Our results indicate that the large fluctuations of TWS in our tentative test-bed (the Zambezi Basin) for the GPR technique are largely caused by strong changes in sea surface temperature and global teleconnection patterns of the nearby oceans. The GPR introduced in this study provides an improved modelling framework to keep track on the influence of these global climate teleconnection patterns on major hydrological systems, like the Zambezi Basin, which contributes to global hydro-climatology but currently showing large amplitudes of land water storage loss.
Identifying impacts of global climate teleconnection patterns on land water storage using machine learning
Ikechukwu Kalu (author) / Christopher E. Ndehedehe (author) / Onuwa Okwuashi (author) / Aniekan E. Eyoh (author) / Vagner G. Ferreira (author)
2023
Article (Journal)
Electronic Resource
Unknown
Metadata by DOAJ is licensed under CC BY-SA 1.0
| Elsevier | 2023
| British Library Conference Proceedings | 2013
Impacts of the Arctic-midlatitude teleconnection on wintertime seasonal climate forecasts
| DOAJ | 2020
Projected ENSO teleconnection on the Southeast Asian climate under global warming
| DOAJ | 2023
Atmospheric teleconnection influence on North American land surface phenology
| DOAJ | 2018