Exploring Continental Convection-Permitting Model Simulations for South America: Cross-correlation Dynamics between precipitation and temperature time series at São Paulo

Abstract

Increasing spatial resolution to kilometre scales allows the deactivation of deep convection parameterisation schemes. As a result of various global initiatives for the next generation of climate studies, continental convection-permitting model (CPM) simulations are now accessible. Nonstationary local extremes, like heatwaves and intense precipitation, are probabilistically linked to regional circulation through scaling relationships. However, these relationships have not been extensively explored in the new simulations available in the early 2020s. Hourly time series data were extracted from the UK Climate Science for Service Partnership (CSSP) and the US South America Affinity Group (SAAG) CPM simulations to compare extreme characteristics of precipitation and temperature for 39 stations in a region of São Paulo, Brazil. Compared to reanalysis and satellite data, which exhibit lower variance in hourly time series, these two sets of CPM simulations have precipitation that is more similar to station observations than the ERA5 data and the Integrated Multi-satellitE Retrievals for the Global Precipitation Measurement (IMERG) data.The cross-correlation structures of the time series are investigated to quantify temporal dependence and reveal patterns between temperature and precipitation at an hourly timescale. Within a higher-dimensional probability space for joint risk, the cross-correlation structures between temperature and precipitation at different lags demonstrate the "memory" of these variables, indicating the influence of past values on future behaviour across multiple time points. Their forecasting power for these two variable based on each other is also explored to offer insights into the physical processes within the evolving simulated dynamic system.Overall, the results underscore the added value of convection-permitting models in providing more realistic simulations of local dynamics of extremes. The identified cross-correlation structures from the CPMs are valuable for exploring opportunities to design AI engines based on weather generator algorithms that use stochastic differential equations. Using CPM simulations, these weather generators can be employed to develop AI approaches for rapid decision support tools aimed at stakeholders facing extreme weather events related to compound risks of temperature and precipitation.