Impacto da advecção nas estimativas de precipitação por radar

Abstract

Extreme precipitation events have become increasingly frequent due to current climate changes. These episodes, characterized by heavy rainfall, bring various consequences such as floods, landslides, and harm to human health. In this context, rainfall estimates derived from radar imagery can be valuable tools in mitigating these impacts, especially with the use of nowcasting tools, which predict these extreme events in advance and aid in rapid decision-making by public authorities. However, rainfall estimates often have certain limitations, and some techniques can be applied to optimize results, such as advection, which uses temporal interpolation to reduce errors associated with the temporal limitations of rainfall estimates. Thus, this project aims to evaluate potential improvements achieved by incorporating advection techniques (specifically the Lucas-Kanade/LK and Variational Echo Tracking/VET methods) into radar-based rainfall estimates, using rain gauge measurements as a reference and validating the results with the R², Nash-Sutcliffe, and Mean Absolute Error (MAE) indices. The data treated with advection showed a slight improvement in relation to the rain gauge data, with average R² values ​​(all hourly groupings) of 0.2573 for the VET method and 0.2533 for the LK method, while the untreated data presented an R² of 0.2237. On the other hand, the MAE values ​​worsened in relation to the original data, indicating a possible offset in the data after the adjustments. The VET method proved more efficient than the LK method (in terms of R²) for all temporal data groupings, although its processing was slower than the LK method, indicating a potential limitation for use with large datasets. Moreover, the LK method showed a closer approximation to the average accumulated precipitation (15.31 mm/h) compared to the same parameter obtained from in situ rainfall data (23.08 mm/h). The results demonstrated that advection can be considered useful in precipitation estimates, but its cost-benefit in terms of computational processing may be a factor to consider in practical applications.