The increased occurrence of extreme weather events due to climate change has heightened the need to develop support decision systems that can help farmers to mitigate losses in agriculture. Environmental phenomenon, such as frost and hail, have a relevant economy impact on crops since they may cause several damages.
Currently, the climate services market uptake is strongly limited by the resolution of the provided information by climate information suppliers. Seasonal predictions are usually provided with a spatial resolution of 80 x 80 km. European wide surveys conducted with end-users (farmers) have found evidences to support the conclusion that higher horizontal resolution climate services are needed to have a chance of market uptake of climate services. In order to fill this market gap, climate simulations need to be run at a finer grid resolution, requiring the use of HPC.
Cybele pilot demonstrator #3 aims to integrate and compare the estimated stage of fruit bud development models with temperature and air humidity forecasts at 2.2 Km resolution for risk probability mapping in order to establish an early warning system that can help farms to prevent damage effects through the use of protection methods for frost and hail.
The hail prediction model is built based on automatic learning techniques, using as input the weather forecasts data and the climate instability indices and performing validations against data collected in the field (hail event recordings).
This demonstrator is very demanding in terms of computational resources both in the input preparation and in the model building. The input data, that are provided by ECMWF at 5 Km resolution, need to be downsized to a 2.2 Km resolution through small scale modelling simulations that require 640 cores (20 nodes) on an HPC machine, consuming 128 core hours for each forecast. Two forecasts are run each day (48 hours and 24 hours in advance) for 2100 days (historical data starting from 2015). The simulation software, Cosmo, is adapted to the area of interest, which is the area around Valencia.
To overcome the scarcity of recorded hail events, while the application will be focused on Valencia region, a wider area has been selected for training purposes, as can be seen in the image (where the red surface identifies the simulation grid and the blue points identify hail events). It’s an area of around 200.000 square kilometres that has been mapped in 68.000 grid points.
Input data consist of 73 fields describing the weather conditions for each grid point for each hour of the day and at 64 different levels above the ground. This amounts to 3 GB of data for each day and 1.2TB of data for each year.