39HYDRO Quantitative Estimation Precipitation Using Conditional Merge for Make Merging of Radar, Satellite and Surface Gauge Precipitation in Paraná State

In many fields, including hydrology, meteorology, agriculture, and power generation, a growing need for high-resolution and highly reliable spatial precipitation products has been growing. Quantitative Precipitation Estimation (QPE) models that integrate multiple data sources have been proposed and evaluated in research and operational contexts to meet this demand. The accurate spatial precipitation estimates from the optimal combination of diverse data sources (surface stations, weather radar, satellite) still represent a significant challenge for atmospheric sciences. Modern weather radar and satellite data can reasonably measure the spatial distribution of precipitation in high temporal and spatial resolution. However, when the estimated accumulations are compared with those recorded at surface stations, the precipitation indirectly measured by weather radar or satellite exhibits significant errors. On the other hand, values obtained from surface stations are commonly considered accurate rainfall due to their direct observation. Despite surface records being as precise as possible, their primary disadvantage lies in their limitation to local representation.

Several methodologies have been developed to merge available sources into a single product, aiming to leverage the advantages of each and minimize their disadvantages. Existing methods can be divided into two main groups: adjustment and integration. Within these classifications are numerous approaches, and the choice of approach depends on various factors such as the intended use, available data sources, data quality, and temporal scale, among others. The objective of this work is to evaluate the adjustment method known as Conditional Merge (CM), used operationally in the SIPREC project of the Environmental Technology and Monitoring System of Paraná (SIMEPAR) to generate hourly precipitation fields, utilizing weather radar, satellite, and surface station data in the state of Paraná, located in southern Brazil. The method is based on the premise that satellite and weather radar products reasonably measure the distribution and contrast of rainfall but have a bias relative to actual precipitation values. Considering this premise, precipitation fields need bias correction while preserving the distribution and contrast of rainfall. The CM applies a bias correction process based on surface measurements from rain gauges, maintaining the spatial structure and variance observed in satellite and weather radar data. The method was chosen for its relevance to operational use, such as easy data integration, implementation flexibility, and simplified, fast calculation process.

Two satellite products were utilized in this study. The precipitation rate was obtained from GOES 16 (NASA/NOAA), with a spatial resolution of 2 km and a frequency of 10 minutes. Subsequently, the sub-hourly GOES data were converted to hourly accumulations. Secondly, the hourly precipitation rate from the Global Satellite Mapping of Precipitation (GSMap-JAXA) was used, with a resolution of 10 km and an hourly frequency. In addition to satellite products, precipitation estimates from the reflectivity mosaic generated by SIMEPAR were also used. The precipitation is derived from reflectivity using a specific distrometer adjustment for the state of Paraná. For model adjustment and evaluation, 40 surface stations from SIMEPAR's telemetric network were used. The stations are spread throughout the state and provide measurements at sub-hourly intervals of 15 minutes. The analysis period covers the months from September to November 2023. During this period, the state of Paraná experienced historical precipitation records, making this segment of the historical series ideal for experiments and precipitation product analysis. For a fairer and more realistic product analysis, precipitation intervals were defined using discrete metrics instead of continuous metrics such as correlation and error magnitude. The metric employed was the F1-Score (F1S). The estimated data are converted to binary data (0 or 1) using the defined precipitation intervals. Additionally, a variation of the Fraction Skill Score (FSS) metric calculation method was used to convert the estimated data to binary. Finally, an F1S value was calculated for each precipitation interval. The estimates generated by CM were contrasted with estimates from GSMap, GOES, and weather radar. This study aims to obtain a rainfall estimation model with reasonable accuracy that can be used operationally, generating accumulated rainfall in almost real-time. The model must be flexible to meet different demands, such as hydrology, nowcasting, and atmospheric modeling, among others.