Jonas Bhend, Irina Mahlstein, Christoph Spirig, Jacopo Riboldi, Mark - PowerPoint PPT Presentation

Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Seasonal forecasting of climate indices Jonas Bhend, Irina Mahlstein, Christoph Spirig, Jacopo Riboldi, Mark Liniger, Christof Appenzeller

Indices: Example Heating Degree Days (HDD) HDD: Index to reflect demand of energy to heat a building. Defined as number of degrees that day’s average temperature is below 18°C. Characteristics of Indices: • Nonlinear transfer function of basic climate variable • scalar • Often based on daily data • Aggregated or counts over time • Threshold based, absolute or relative to climatology DJF, Nov Initialization, ECMWF System4 Forecasting climate indices, Euporias workshop, Santander 2

Summary Threshold-based climate indices require some form of daily bias correction of seasonal forecasts. De-biasing is non-trivial due to small sample of re-forecast / observation pairs, auto-correlation and bias in higher-order moments. Smoothing of observed daily climatology improves de-biasing and skill of forecasts. Forecasting climate indices, Euporias workshop, Santander 3

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 4

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 5

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 6

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 7

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 8

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 9

Example: Debiasing daily series Forecasting climate indices, Euporias workshop, Santander 10

Bias correction of daily data: daily mean temperature Problem: 30 years of observations not enough to calculate daily climatology Explore smoothing in model world by perfect model approach K average of 30 years with 1 member Loess smoothing of 30 years with 1 member average of 30 years with 51 members Forecasting climate indices, Euporias workshop, Santander 11

Bias correction of daily data: daily mean temperature MAE between full hindcast daily mean and of hindcast climatology estimated from 1 member through fit daily data DJF, Nov Initialization, System4 Forecasting climate indices, Euporias workshop, Santander 12

Bias correction of daily data: daily mean temperature Mean hindcast spread MAE for daily data Forecasting climate indices, Euporias workshop, Santander 13

Bias correction of daily data: 5% percentile of daily temperature MAE between full hindcast daily mean and of estimated from 1 member through fit 5day window Forecasting climate indices, Euporias workshop, Santander 14

Bias correction of daily data: 90% percentile of daily precipitation MAE between full hindcast daily mean and of estimated from 1 member through fit 5-day window Forecasting climate indices, Euporias workshop, Santander 15

Skill of debiased vs. raw metrics: Heating degree days climatology Forecasting climate indices, Euporias workshop, Santander 16

Skill of debiased vs. raw metrics: HDD correlation HDD correlation (raw) HDD correlation (debiased) Forecasting climate indices, Euporias workshop, Santander 17

Skill of debiased vs. raw metrics: HDD continuos ranked probability skill score HDD crpss (raw) HDD crpss (debiased) Good Bad Forecasting climate indices, Euporias workshop, Santander 18

HDD crpss (raw) HDD crpss (debiased) Good Bad Temperature crpss (raw) Temperature crpss (debiased) Forecasting climate indices, Euporias workshop, Santander 19

Skill of debiased vs. raw metrics: Frost day climatology Forecasting climate indices, Euporias workshop, Santander 20

Skill of debiased vs. raw metrics: Frost days correlation FD correlation (raw) FD correlation (debiased) Forecasting climate indices, Euporias workshop, Santander 21

Skill of debiased vs. raw metrics: FD continuous ranked probability skill score FD crpss (raw) FD crpss (debiased) Good Bad Forecasting climate indices, Euporias workshop, Santander 22

Summary Threshold-based climate indices require some form of daily bias correction of seasonal forecasts. Daily de-biasing is non-trivial due to small sample of re- forecast / observation pairs, auto-correlation and biases in higher-order moments (yet to be addressed). Smoothing of observed daily climatology improves de-biasing and skill of forecasts. Forecasting climate indices, Euporias workshop, Santander 23

Deliverable WP22 Please send whatever you have as soon as possible to christoph.spirig@meteoswiss.ch - Climatologies of climate indices - Forecasts of climate indices - Skill metrics of climate indices Forecasting climate indices, Euporias workshop, Santander 24