An insight from homogeneity testing of long-term rainfall datasets over East Java, Indonesia

Abstract

Robust, reliable, and trustworthy ground observation datasets are the preliminary requirement for assessing the impact of climate change over regions. Principal testing to assess the quality of ground observation rely on the missing data and homogeneity result. The study used 40 years of monthly rainfall documented from different topographical features in the monsoonal region of East Java, Indonesia. The test included annual rainfall, early rainy season (October-November-December), and primary rain season (January-February-March). The homogeneity of rainfall determined by absolute technique: Pettitt’s test, the Standard Normal Homogeneity Test, the Buishand Rank Test, and the von Neumann Ratio. Among the time series, October-November-December observation results in better homogeneity. However, the rainfall datasets during primary rainy season showed the worst homogeneity. By performing annual and seasonal homogeneity test from 67 rainfall stations: 5 stations out of data length required, 5% stations ‘rejected’, 11% ‘suspect’, 11% ‘doubtful’, and 73% were ‘trusted’. Therefore, a total of 45 stations can be used as metadata for relative comparison and 7 stations can be considered to be useful for analysis despite ‘doubtful’. The remaining 10 stations need careful consideration to be used for future water management.  Change point detected particularly between the year of 1997 through 2000. Pettitt’s test has outstanding results in the case of extreme climatic anomaly, but less sensitive of continuous abrupt change. The von Neumann test could detect abnormal data, but was not suitable for datasets containing few extreme values. The insights from homogeneity testing were: a) it is important to remove any outliers in the datasets before conducting homogeneity testing, b) both parametric and nonparametric homogeneity tests should be performed, and c) comparisons should be made with surrounding rainfall stations. Comparison with trusted long-term rainfall data is valuable for stations labeled as ‘doubtful’ or ‘suspect’ to mitigate false detections in individual homogeneity tests. The identified ‘useful’ rainfall data can then serve as reference stations for relative homogeneity tests. These findings suggest that reference stations should be assessed within similar rainfall zones.