EESS talk on "Closing the survey-modelling loop: an analysis of consumer preferences for load-shifting via a model-driven adaptive survey design"
Abstract:
Most climate models show a precipitation increase with warming that is smaller than the increase in moisture, which requires a weakening of the convective mass flux or a slowing of the overturning circulation. We analyze simulations of global-storm resolving models (DYAMOND models) in which deep convection is explicitly resolved to better understand the role of large- and small-scale drivers on precipitation. First, δ-MAPS technique infers coherent regions of relative humidity in the high-resolution outputs, which allows to both sample the large datasets and control the large-scale environment. We train different Machine Learning approaches in these thermodynamically controlled regions to identify the nature and persistence of the dependencies, before proceeding to a process-oriented comparison in a causal framework. The analysis of causal effects reveals a strong link between small-scale dynamics and quantiles of convective precipitation. We also show that the variability of the convective area is driven by the large-scale relative humidity. Finally, the results are compared across the DYAMOND models to evaluate the intermodel spread. By doing so, we evaluate to what extent the spread in precipitation in model ensemble may arise from the differences in representation of the large-scale relative humidity or the vertical velocity in convective areas.
Short Biography:
I am last year Phd student in the LAPI lab, and part of the Marie Curie cohort iMiracli. I started studying physical oceanography in France, and notably the dynamics of geophysical fluids. Then I arrive at EPFL to start my PhD thesis under the supervison of Pr. Athanasios Nenes to apply ML algorithms on climate model outputs, and better understand some important climate processes. My focus is to better understand the contribution of thermodynamics and dynamics on the convective rainfall in the tropics, which is part of the globe where 50% of the global population is expected to live by 2050.
Most climate models show a precipitation increase with warming that is smaller than the increase in moisture, which requires a weakening of the convective mass flux or a slowing of the overturning circulation. We analyze simulations of global-storm resolving models (DYAMOND models) in which deep convection is explicitly resolved to better understand the role of large- and small-scale drivers on precipitation. First, δ-MAPS technique infers coherent regions of relative humidity in the high-resolution outputs, which allows to both sample the large datasets and control the large-scale environment. We train different Machine Learning approaches in these thermodynamically controlled regions to identify the nature and persistence of the dependencies, before proceeding to a process-oriented comparison in a causal framework. The analysis of causal effects reveals a strong link between small-scale dynamics and quantiles of convective precipitation. We also show that the variability of the convective area is driven by the large-scale relative humidity. Finally, the results are compared across the DYAMOND models to evaluate the intermodel spread. By doing so, we evaluate to what extent the spread in precipitation in model ensemble may arise from the differences in representation of the large-scale relative humidity or the vertical velocity in convective areas.
Short Biography:
I am last year Phd student in the LAPI lab, and part of the Marie Curie cohort iMiracli. I started studying physical oceanography in France, and notably the dynamics of geophysical fluids. Then I arrive at EPFL to start my PhD thesis under the supervison of Pr. Athanasios Nenes to apply ML algorithms on climate model outputs, and better understand some important climate processes. My focus is to better understand the contribution of thermodynamics and dynamics on the convective rainfall in the tropics, which is part of the globe where 50% of the global population is expected to live by 2050.
Practical information
- General public
- Free
- This event is internal
Organizer
- EESS - IIE
Contact
- Prof. Athanasios Nenes, LAPI