Evaluation of extreme rainfall indices from CHIRPS precipitation estimates over the Brazilian Amazonia

Highlights

• Use of the rainfall dataset from CHIRPS for different purposes in Brazilian Amazon.

• CHIRPS data tend to underestimate the rainfall for the rainiest months.

• The correlation was greater for the rainy season in eastern Amazonia.

• CHIRPS data underestimate the extreme rainfall indices.

Abstract

Since several datasets are available with marked differences, the assessment of precipitation data is a key aspect to support the choice of the most adequate precipitation product for a certain research or operational application. In the present study, we evaluated the use of the daily rainfall dataset from CHIRPS with spatial resolution of 0.05° for different purposes in the states of the Brazilian Legal Amazon. We compared monthly rainfall, annual rainfall indices and their trends calculated using CHIRPS data and rain gauge observations with a point-to-pixel analysis. The use of daily CHIRPS data provided mean monthly rainfall similar to that obtained using data from the rain gauge stations, but CHIRPS data tend to underestimate the values for the rainiest months. The correlation was usually lower in the western Amazon, especially during its rainy season. The same underestimation was observed for extreme rainfall indices. CHIRPS product produces more similar results to rain gauge data for the indices PRCPTOT, nP, and R95pad, while strong underestimate the most extreme rainfall indices (R50mm, Rx1day, Rx5days). For the 45 stations and 15 rainfall indices analysed, 63 significant trends were detected using rain gauge data, of which only 13 were detected using CHIRPS product. Therefore, the use of CHIRPS data does not well represent the trends in rainfall indices.

Introduction

Rainfall information is an important input for different applications, such as agriculture, hydrology, climate, management of water resources, design of hydraulic structures, drought and flood risk assessment and forecasting, and ecological modelling. In recent decades, several gridded precipitation datasets have been developed with different design objectives, data sources, spatial resolutions, spatial coverage, published temporal resolutions, temporal spans, and latencies (Beck et al., 2017; Sun et al., 2018). Exclusively gauge-based datasets tend to show lower performances in regions with sparse rain gauge networks since precipitation features high spatial variation (Sun et al., 2018). The advancement of remote sensing instruments and precipitation retrieval algorithms has made available a series of satellite-based precipitation products. These products use information from visible/infrared sensors on geostationary and low Earth orbit satellites, passive or active microwave sensors on low Earth orbit satellites, and combinations of both (Kidd and Levizzani, 2011; Sun et al., 2018). The satellite-based precipitation estimates have significant uncertainty (Tian and Peters-Lidard, 2010) because none of the satellite sensors detect rainfall as such, and further merging or blending with rain gauge information can improve precipitation products.

Since several datasets are available, the assessment of precipitation data is a key aspect to support the choice of the most adequate precipitation product for a certain research or operational application (Paredes Trejo et al., 2016). Many studies evaluate the advantages and limitations of the available datasets and compare their results (see review by Karimi and Bastiaanssen, 2015), and marked differences have been found even among datasets employing the same data sources (Beck et al., 2017).

The Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) provides blended gauge-satellite precipitation spanning 50°S-50°N from 1981 to near the present. CHIRPS are provided at spatial resolutions of 0.25° and 0.05° and daily to annual temporal resolutions with a short latency (Funk et al., 2015). According to the authors, this dataset can be used for trend analysis and seasonal drought monitoring. Validation studies have been conducted in different regions of the globe, e.g., Cyprus (Katsanos et al., 2016), Argentina (Rivera et al., 2018), Nepal (Shrestha et al., 2017), Italy (Duan et al., 2016), Chile (Zambrano et al., 2017), China (Bai et al., 2018; Gao et al., 2018; Zhong et al., 2019; Lai et al., 2019), Burkina Faso (Dembélé and Zwart, 2016), Mozambique (Toté et al., 2015), the Tibetan Plateau (Wu et al., 2019), India (Prakash, 2019), Ethiopia (Duan et al., 2019; Bayissa et al., 2017), Myanmar (Sirisena et al., 2018), Venezuela (Paredes Trejo et al., 2016), and in other regions of South America (Baez-Villanueva et al., 2018; Funk et al., 2015). In these studies, the evaluation of CHIRPS products is usually accomplished by comparing rainfall indices with other datasets and with in situ observations at different temporal scales using different performance metrics. The most common purposes are drought monitoring, rainfall detection, and streamflow simulation. The results show that CHIRPS performs satisfactorily in almost all studies and better than other datasets in many cases.

In Brazil, Beck et al. (2017), using hydrological modelling to evaluate the accuracy of precipitation datasets, found that CHIRPS V2.0 tended to perform better than the other tested datasets in central and eastern Brazil and is a viable choice for daily temporal resolution in tropical regions if the peak magnitude underestimation and spurious drizzle are not critical. Funk et al. (2015) showed that the wet season correlations between the CHIRPS and the Global Precipitation Climatology Center (GPCC) are lower in western Amazonia than in other regions of Brazil. For northeastern Brazil, CHIRPS correlates well with monthly rain gauge data but tends to overestimate low and underestimate high rainfall values, and rain detection is poor in semiarid areas (Paredes-Trejo et al., 2017). (Costa et al., 2019), when comparing monthly precipitation data from CHIRPS 2.0 with rain station data from 1998 to 2010, found the largest spatial differences in precipitation in northwestern Amazonas and southwestern Pará. Correa et al. (2017) used hydrological modelling and rainfall datasets as input data to a hydrological model in order to analyse past floods and droughts in the Amazon River Basin. Using different performance metrics, the authors showed that CHIRPS V2.0 is one of the best rainfall datasets for the region, probably because this dataset not only is based on atmospheric models but also uses in situ rainfall information.

The aim of this study is to evaluate the use of a rainfall dataset from CHIRPS for different purposes in the states of the Brazilian Legal Amazon. We compared monthly rainfall, annual rainfall indices and their trends calculated using CHIRPS and gauge observations. Although Beck et al. (2017) criticize studies that re-use gauge observations already incorporated into rainfall datasets to determine their accuracy, in the case of CHIRPS data, even in the presence of co-located stations, the CHIRPS have some influence from the CHIRP (exclusively satellite-based data) (Funk et al., 2015). The results of this study will provide bases for decision makers on the applicability of these data products for different meteorological studies and the probability of occurrence of extreme events.