Decadal Predictions and Natural Hazard Risks

Validating Skill of Decadal Predictions in Resolving Natural Hazard Index

Sust Global, Climate Data Science Internship

Validating decadal predictions. A) The general workflow of validation. Decadal predictions are 5-10 year predictions initialized every year from 1960s to the present day. For our validation, the first 5 years of every initialized simulation is used along with its corresponding CMIP6 historical simulations (“Climate Projection”) and ERA5 Land (“Observation”) data. B) The skill and the consistency of skillful prediction of drought index SPEI12 in the Contiguous United States from 2000 to 2009. The color represents consistency: how many years of the sad ten-year span does decadal prediction outperforms its climate projection counterpart. The marker size represents the measure of the absolute skill (Mean Square Skill Score, or MSSS). This map only plots data from the best model, MIROC6, of the three models available in the decadal prediction forecast phase. C) Mean Square Skill Score over the ten different initialized years. The grey area with gridlines represents negative skill compared to the climate projections. The height of the bar indicates the absolute skill averaged over the Contiguous United States.

The emergence of climate products that bridge the gap between these two vastly different timescales – known as Decadal Prediction – opens up a host of opportunities. Climate modeling centers and groups that used to run global climate projections now join an international collaboration called The Decadal Climate Prediction Project (DCPP), as a part of the latest climate model intercomparison effort (CMIP6). These decadal climate predictions combine the practices from both the weather and climate worlds: they ingest the current state of climate from satellite and ground observations via data-assimilation and simulate the evolution for the next 5-10 years. In contrast to the climate projections, decadal predictions do not carry any significant climate change signal; instead, they aim to capture the inter-annual variability that arises naturally from the earth system.

My internship in the summer of 2022 at Sust Global was to build an extensible workflow that validates the performance of decadal prediction during the historical period (“hindcast”) in resolving multiple hazards, including drought index, wildfire potential, extreme precipitation, and heatwave.

Drought index, or specifically the Standardized Precipitation-Evapotranspirtaion Index (SPEI)
Wildfire potential index, or specifically Keetch-Byram Drought Index (KBDI)
Extreme precipitation day count
Heatwave day count

My work has shown encouraging results for some hazards. Here in the figure I demonstrate the reliability of 5-year drought index prediction, or Standardized Precipitation Evapotranspiration Index (SPEI), with DCPP data. The top figure shows that over the ten years between 2000 and 2010, all three DCPP models are generally more skillful than their climate projection counterparts. The map at the bottom shows the spatial variability of performance across the contiguous United States. Decadal predictions of drought in the majority of the contiguous United States are skillful, and are consistently more skillful in the north and the west coast.