EVALUATION OF M5P DECISION TREE MODEL IN DOWNSCALING CMIP6 CLIMATE OUTPUT FOR ERBIL PLAIN

ABBAS  YEGANEH-BAKHTIARY; HOSSEIN  EYVAZOGHLI; SOORKEU A.  ATROOSHI; SARFRAZ  MUNIR; ILYAS  MASIH; ANNE VAN  DAM

doi:10.26682/csjuod.2023.26.2.70

ABBAS YEGANEH-BAKHTIARY *School of Science & Engineering, University of Kurdistan Hewler, Kurdistan Region- Iraq
HOSSEIN EYVAZOGHLI **School of Civil Engineering, Iran University of Science and Technology, Iran
SOORKEU A. ATROOSHI *School of Science & Engineering, University of Kurdistan Hewler, Kurdistan Region- Iraq
SARFRAZ MUNIR *School of Science & Engineering, University of Kurdistan Hewler, Kurdistan Region- Iraq
ILYAS MASIH *** Dept. of Water Resources and Ecosystems, IHE Delft Institute, Netherland
ANNE VAN DAM *** Dept. of Water Resources and Ecosystems, IHE Delft Institute, Netherland

DOI: https://doi.org/10.26682/csjuod.2023.26.2.70

Keywords: Machine learning, regression, climate change, trend, climatic parameters

Abstract

Downscaling the effective parameters of Global Climate Models is crucially required to project climate conditions in the future. The capability of machine learning approaches in downscaling Global Climate Models is getting interesting these days. The present study evaluates the M5p Decision Tree (DT) skill in reproducing high resolution monthly precipitation and temperature (predictands) data. To this end, the significant climate-related parameters (predictors) were derived from the General Climate Model of the Coupled Model Intercomparison Project Phase 6 (CMIP6) for the Erbil Plain. Initially, the effective parameters were carefully chosen for developing a downscaling model. Subsequently, multiple models were formulated to accommodate various maximum depths of decision trees (DT). Results obtained from the training process revealed a notably higher correlation between precipitation and temperature predictors in contrast to wind speed and direction. The evaluation of skill indicated enhanced accuracy in downscaling when increasing the maximum depth (MD) of M5p models up to an optimal threshold, with MD = 5 identified as the optimal depth for generating predictive DTs. Finally, it was proved that the M5p model serves as a highly effective tool for downscaling the hydroclimatic parameters in climate change studies.

Downloads

Download data is not yet available.

References

Downscaling the effective parameters of Global Climate Models is crucially required to project climate conditions in the future. The capability of machine learning approaches in downscaling Global Climate Models is getting interesting these days. The present study evaluates the M5p Decision Tree (DT) skill in reproducing high resolution monthly precipitation and temperature (predictands) data. To this end, the significant climate-related parameters (predictors) were derived from the General Climate Model of the Coupled Model Intercomparison Project Phase 6 (CMIP6) for the Erbil Plain. Initially, the effective parameters were carefully chosen for developing a downscaling model. Subsequently, multiple models were formulated to accommodate various maximum depths of decision trees (DT). Results obtained from the training process revealed a notably higher correlation between precipitation and temperature predictors in contrast to wind speed and direction. The evaluation of skill indicated enhanced accuracy in downscaling when increasing the maximum depth (MD) of M5p models up to an optimal threshold, with MD = 5 identified as the optimal depth for generating predictive DTs. Finally, it was proved that the M5p model serves as a highly effective tool for downscaling the hydroclimatic parameters in climate change studies.