2:00 PM - 2:15 PM
[ACG42-02] Development of the emulator of Integrated Land Simulator for global hydrologic simulations
Keywords:land surface model, emulator, MATSIRO, CaMa-Flood, Integrated Land Simulator
ILS (Integrated Land Simulator) is a framework for coupling geoscientific models, developed at the University of Tokyo. In its original form it couples land surface model MATSIRO to the hydrodynamic model CaMa-Flood. The ILS has been used in a number of studies, and it is able to model not just the river discharge but also policy-relevant information such as land inundation fraction.
Unfortunately, the ILS is still computationally expensive or a number of applications. In order to use the ILS effectively for tasks such as uncertainty quantification, parameter estimation, and large ensemble or multi-model experiments, speeding up of the model is required. Here, we create a fast analogue (emulator) of the ILS to make progress on this task.
Within the ILS emulator, MATSIRO is replaced by a simple physically-guided land surface model. The model is similar to the one in Olson et al. (2024), with a number of changes. First, the updated model incorporates subsurface flow and replaces surface soil moisture with total soil moisture. Second, a local parameter estimation method has been replaced with differential evolution -- a global parameter optimization method. Finally, a new implicit scheme with time linearization has been formulated and implemented. This ensures model stability. The ILS emulator takes as input five atmospheric forcings at the daily scale: snowfall, surface air temperature, rainfall, specific humidity, and surface air pressure. With respect to CaMa-Flood, it has been replaced by a mixed-precision version. Some changes have been made to the ILS coupler to accommodate the MATSIRO emulator and the mixed-precision version of the CaMa-Flood. The conversion of precipitation to rainfall and snowfall follows the original ILS whenever separate rainfall and snowfall inputs are not available. The emulator can, in theory, be integrated at a range of scales and horizontal resolutions. However, here we make use of decadal global simulations at 0.5 degree horizontal resolution.
We show that the ILS emulator well represents many aspects of the original ILS, including both MATSIRO and CaMa-Flood variables. Notably, river discharge for major world river basins is well simulated, expect a sizeable overestimation of discharge over the Congo river basin. Moreover, our results suggest that snowmelt-driven flooding well matches the original ILS output. In terms of the computational aspect, the emulator is an order of magnitude faster than the original ILS. Moreover, if the metric of node-hours is considered, the speedup of thirty times is attained.
We present ongoing applications of the ILS emulator (including high-resolution simulations), and outline plans and ideas for future use. Overall, we conclude that the ILS emulator is a fast and efficient tool for hydrologic modelling that can replace the ILS for a range of applications.
Reference:
R. Olson, K. Yoshimura and T. Nitta (2024): A fast physically-guided emulator of MATSIRO land surface model. Journal of Hydrology, 634, 131093, doi: 10.1016/j.jhydrol.2024.131093
Unfortunately, the ILS is still computationally expensive or a number of applications. In order to use the ILS effectively for tasks such as uncertainty quantification, parameter estimation, and large ensemble or multi-model experiments, speeding up of the model is required. Here, we create a fast analogue (emulator) of the ILS to make progress on this task.
Within the ILS emulator, MATSIRO is replaced by a simple physically-guided land surface model. The model is similar to the one in Olson et al. (2024), with a number of changes. First, the updated model incorporates subsurface flow and replaces surface soil moisture with total soil moisture. Second, a local parameter estimation method has been replaced with differential evolution -- a global parameter optimization method. Finally, a new implicit scheme with time linearization has been formulated and implemented. This ensures model stability. The ILS emulator takes as input five atmospheric forcings at the daily scale: snowfall, surface air temperature, rainfall, specific humidity, and surface air pressure. With respect to CaMa-Flood, it has been replaced by a mixed-precision version. Some changes have been made to the ILS coupler to accommodate the MATSIRO emulator and the mixed-precision version of the CaMa-Flood. The conversion of precipitation to rainfall and snowfall follows the original ILS whenever separate rainfall and snowfall inputs are not available. The emulator can, in theory, be integrated at a range of scales and horizontal resolutions. However, here we make use of decadal global simulations at 0.5 degree horizontal resolution.
We show that the ILS emulator well represents many aspects of the original ILS, including both MATSIRO and CaMa-Flood variables. Notably, river discharge for major world river basins is well simulated, expect a sizeable overestimation of discharge over the Congo river basin. Moreover, our results suggest that snowmelt-driven flooding well matches the original ILS output. In terms of the computational aspect, the emulator is an order of magnitude faster than the original ILS. Moreover, if the metric of node-hours is considered, the speedup of thirty times is attained.
We present ongoing applications of the ILS emulator (including high-resolution simulations), and outline plans and ideas for future use. Overall, we conclude that the ILS emulator is a fast and efficient tool for hydrologic modelling that can replace the ILS for a range of applications.
Reference:
R. Olson, K. Yoshimura and T. Nitta (2024): A fast physically-guided emulator of MATSIRO land surface model. Journal of Hydrology, 634, 131093, doi: 10.1016/j.jhydrol.2024.131093