A Statistical Interpolation of Satellite Data with Rain Gauge Data over Papua New Guinea

Abstract

Abstract Satellites provide a useful way of estimating rainfall where the availability of in situ data is low but their indirect nature of estimation means there can be substantial biases. Consequently, the assimilation of in situ data is an important step in improving the accuracy of the satellite rainfall analysis. The effectiveness of this step varies with gauge density, and this study investigated the effectiveness of statistical interpolation (SI), also known as optimal interpolation (OI), on a monthly time scale when gauge density is extremely low using Papua New Guinea (PNG) as a study region. The topography of the region presented an additional challenge to the algorithm. An open-source implementation of SI was developed on Python 3 and confirmed to be consistent with an existing implementation, addressing a lack of open-source implementation for this classical algorithm. The effectiveness of the analysis produced by this algorithm was then compared to the pure satellite analysis over PNG from 2001 to 2014. When performance over the entire study domain was considered, the improvement from using SI was close to imperceptible because of the small number of stations available for assimilation and the small radius of influence of each station (imposed by the topography present in the domain). However, there was still value in using OI as performance around each of the stations was noticeably improved, with the error consistently being reduced along with a general increase in the correlation metric. Furthermore, in an operational context, the use of OI provides an important function of ensuring consistency between in situ data and the gridded analysis. Significance Statement The blending of satellite and gauge rainfall data through a process known as statistical interpolation (SI) is known to be capable of producing a more accurate dataset that facilitates better estimation of rainfall. However, the performance of this algorithm over a domain such as Papua New Guinea, where gauge density is extremely low, is not often explored. This study reveals that, although an improvement over the entire Papua New Guinea domain was slight, the algorithm is still valuable as there was a consistent improvement around the stations. Additionally, an adaptable and open-source version of the algorithm is provided, allowing users to blend their own satellite and gauge data and create better geospatial datasets for their own purposes.