Abstract

Evident climate change has been observed and projected in observation records and General Circulation Models (GCMs), respectively. This change is expected to reshape current seasonal variability; the degree varies between regions. High-resolution climate projections are thereby necessary to support further regional impact assessment. In this study, a gated recurrent unit-based recurrent neural network statistical downscaling model is developed to project future temperature change (both daily maximum temperature and minimum temperature) over Metro Vancouver, Canada. Three indexes (i.e., coefficient of determinant, root mean square error, and correlation coefficient) are estimated for model validation, indicating the developed model’s competitive ability to simulate the regional climatology of Metro Vancouver. Monthly comparisons between simulation and observation also highlight the effectiveness of the proposed downscaling method. The projected results (under one model set-up, WRF-MPI-ESM-LR, RCP 8.5) show that both maximum and minimum temperature will consistently increase between 2,035 and 2,100 over the 12 selected meteorological stations. By the end of this century, the daily maximum temperature and minimum temperature are expected to increase by an average of 2.91°C and 2.98°C. Nevertheless, with trivial increases in summer and significant rises in winter and spring, the seasonal variability will be reduced substantially, which indicates less energy requirement over Metro Vancouver. This is quite favorable for Metro Vancouver to switch from fossil fuel-based energy sources to renewable and clean forms of energy. Further, the cold extremes’ frequency of minimum temperature will be reduced as expected; however, despite evident warming trend, the hot extremes of maximum temperature will become less frequent.

Highlights

  • Distinct impacts of climate change on Canada are being observed

  • Despite a few stations are not quite ideal and competitive with previous studies of other regions, compared to Regional Climate Models (RCMs) outputs, prominent improvements could be found in all the indexes after employing the developed gated recurrent unit (GRU)-based recurrent neural network (RNN) downscaling model

  • A GRU-based RNN downscaling approach was developed to tackle the spatial mismatch between coarse-scale climate simulation and regional climatology for improving the representation of local future climate across Metro Vancouver (MV)

Read more

Summary

INTRODUCTION

Distinct impacts of climate change on Canada are being observed. The increasing rate of temperature over Canada is near twice the global rate (Canada in a Changing Climate, 2019). Motivated by the success of RNN in capturing complex non-linear relationships between time-dependent data (LeCun et al, 2015), a GRU-based RNN statistical downscaling method followed by Tian et al will be developed to generate temperature projections (both daily maximum temperature and minimum temperature) for further impact assessment of MV. With the local-scale observations over MV, the GRU-based RNN statistical downscaling model (detailed information is displayed ) will be developed to correct/downscale gridded simulations (daily maximum/ minimum temperature) from the selected RCM. Considering the complicated non-linear relationship that exists between relatively coarse-scale simulation and realistic temperature observations, RNN statistical downscaling model followed by Tian et al (2021) is developed for generating high-resolution temperature projections for MV. Three model evaluation criteria, i.e., determinant coefficient (R2), root mean square error (RMSE), and correlation coefficient (r), are used to evaluate the GRU-based RNN statistical downscaling

Validation Results
CONCLUSION
DATA AVAILABILITY STATEMENT
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call