Improvement in the skill of CMIP6 decadal hindcasts for extreme rainfall events over the Indian summer monsoon region

Abstract

Decadal climate predictions have been widely used to predict the near-term climate information relevant for decision-making at multi-year timescales. In the present study, we evaluate the quality of the Coupled Model Intercomparison Project phase-6 (CMIP6) Decadal Climate Prediction Project (DCPP) hindcasts in capturing the extreme rainfall events (EREs) over the monsoon core region during Indian summer monsoon season (June-September) up to lead years 1-10. For the first time, in this study, we have used quantile mapping approach to downscale and bias correct the DCPP CMIP6 simulation/hindcast rainfall for the better representation of EREs. Detailed analysis suggests that the models in general strongly underestimate the rainfall variability over the summer monsoon region. However, after the downscaling and bias correction, the representation of rainfall variability and intensity improved multifold. The bias-corrected decadal hindcasts in fact show ~ 80% improvement in capturing the frequency, intensity, and spatial distribution of rainfall associated with the EREs. Present study brought out a downscaled DCPP product, with potential prediction skill for EREs over India. It is important to highlight that the models predict an increase in the small and medium-area EREs as compared to the large-area EREs over the monsoon core region for the decade 2019-2028.

Figure 1

Annual cycle of monthly climatology of precipitation (mm) averaged over the Indian summer monsoon region for before (solid lines) and after (dashed lines) downscaling and bias correction (DBC) at lead year-1, black line shows observed (IMD) climatology of precipitation.

Figure 2

Summer monsoon seasonal mean bias of precipitation (mm/day) with lead years (from 1 to 10 top to bottom) for (a) before, (b) after DBC. Values in each panel represents the area averaged (over India) absolute mean bias (blue) and mean bias (red).

Figure 3

(a) Taylor metric for summer monsoon mean precipitation for the DCPP models (listed in Table S1). Different markers in alphabets (numerics) are for before (after) DBC with lead 1 to 10 years. The Correlation, normalized standard deviation, and normalized root mean square error are presented in blue, black and red lines respectively. Marker in black color for IMD (observations). (b) Difference (%) in the magnitude of JJAS [(hindcast-observation)/observation] *100 mean precipitation and its associated mean square error. Statistics before DBC lies on vertical grid lines. Statistics after DBC represented right of the vertical grid lines. (c) Improvement (in %) of seasonal mean rainfall over Indian summer monsoon region after DBC.

Figure 4

Frequency of extreme rainfall days with lead years (from 1 to 10 top to bottom) for (a) before, (b) after DBC. Values in each panel represents the percentage change in frequency distribution of extreme rainfall days (red) and pattern correlation (black).

Figure 5

(a) Taylor metric for the R95 thresholds over India for DCPP models (listed in Table S1). The Correlation and, normalized standard deviation are presented in blue and black lines respectively. Different markers in alphabets (numerics) are for before (after) DBC with lead 1 to 10 years. Marker in black color for IMD (observations). (b) Improvement (in %) of R95 threshold distribution over Indian summer monsoon region after DBC.

Figure 6

Composite of precipitation anomalies (mm/day) for large area extreme rainfall events with lead years (from 1 to 10 top to bottom) in DCPP models (a) before, (b) after DBC, and IMD. Values in each panel represents the pattern correlation (black).

Figure 7

Composite of precipitation anomalies (mm/day) for small area extreme rainfall events with lead years (from 1 to 10 top to bottom) in DCPP models (a) before, (b) after DBC, and IMD. Values in each panel represents the pattern correlation (black).

Figure 8

Radial distribution of precipitation (mm/day) for large area extreme rainfall events with lead years (from 1 to 10 top to bottom) in DCPP models (a) before, (b) after DBC. Values in each panel represents the pattern correlation (black).

Figure 9

Radial distribution of precipitation (mm/day) for small area extreme rainfall events with lead years (from 1 to 10 top to bottom) in DCPP models (a) before, (b) after DBC. Values in each panel represents the pattern correlation (black).

Figure 10

Pattern correlation of radial distribution of precipitation during large area EREs: before (red circles) and after (blue diamond) DBC in the models and MME for lead-1 year. Bar diagram inside of each panel represents the % of years having high pattern correlation after bias correction (x-axis represents the lead years and y-axis represents the % of years).

Figure 11

Number of extreme days over the monsoon core region for 10-year-average from 1965 to 2023 in MME, (a) before DBC and (b) after DBC, blue bars for large area, red for medium area, and black for small area extreme days. Similarly, lines represent the extreme days in the observations.