Keywords
class, dyresm, rcm, parallel computing
Start Date
1-7-2004 12:00 AM
Abstract
Canada has several of Earth's largest lakes and many small lakes. Heat storage and circulation are greatly affected by lakes. Currently the Canadian Regional climate model does not incorporate a lake component. Therefore, we are linking atmospheric and lake models for such applications as climate prediction and assessing changes in the lake water quality and quantity. We investigate use of highly parallel arrays of clustered processors, available through Canada's SHARCNET. The accuracy of lake, land and atmospheric models depends on grid spacing. Coarser grids adversely affect accuracy. Regional climate model inputs are required subhourly, placing a lower bound on the grid sizes that can be employed. We link a one-dimensional lake model such as the Dynamic Reservoir Model (DYRESM) to a Regional Climate model (RCM) to incorporate the effects of lake on the regional climate. The land model used is the Canadian Land Surface Scheme (CLASS). CLASS and DYRESM are vertical models, with no interaction between horizontally neighboring nodes. CLASS computes heat and moisture fluxes for bare ground (fractional coverage by ground, FG), snow-covered ground (fractional coverage by snow, FSN), ground with canopy (fractional coverage by ground, FC), and ground with both snow and canopy. These fractions are combined to calculate node characteristics. Lake flux values are provided by DYRESM, which are combined with land values according to the fractional lake coverage. Hybrid model is designed and implemented using a mix of both serial farm and task parallel approaches on Guelph SHARCNET high performance computing cluster.
A Spatially Parallel Implementation of a Lake and Land Surface Model Interaction with a Regional Climate Model
Canada has several of Earth's largest lakes and many small lakes. Heat storage and circulation are greatly affected by lakes. Currently the Canadian Regional climate model does not incorporate a lake component. Therefore, we are linking atmospheric and lake models for such applications as climate prediction and assessing changes in the lake water quality and quantity. We investigate use of highly parallel arrays of clustered processors, available through Canada's SHARCNET. The accuracy of lake, land and atmospheric models depends on grid spacing. Coarser grids adversely affect accuracy. Regional climate model inputs are required subhourly, placing a lower bound on the grid sizes that can be employed. We link a one-dimensional lake model such as the Dynamic Reservoir Model (DYRESM) to a Regional Climate model (RCM) to incorporate the effects of lake on the regional climate. The land model used is the Canadian Land Surface Scheme (CLASS). CLASS and DYRESM are vertical models, with no interaction between horizontally neighboring nodes. CLASS computes heat and moisture fluxes for bare ground (fractional coverage by ground, FG), snow-covered ground (fractional coverage by snow, FSN), ground with canopy (fractional coverage by ground, FC), and ground with both snow and canopy. These fractions are combined to calculate node characteristics. Lake flux values are provided by DYRESM, which are combined with land values according to the fractional lake coverage. Hybrid model is designed and implemented using a mix of both serial farm and task parallel approaches on Guelph SHARCNET high performance computing cluster.
