Analysing the detection and correction parameters in the homogenisation of climate data series using gsimcli

Homogenisation of climate data series is the process of detection and correction of non-natural irregularities present in the data. Such process is extremely important due to the use of climate data in many hydrological and environmental projects. Several homogenisation methods have been developed in the last decades. In the geostatistical field, studies already showed an approach based on the direct sequential simulation algorithm as a very promising technique for the detection and correction of irregularities. This approach, called gsimcli, uses the probability distribution function (estimated from simulated values) to identify the presence of irregularities, with a specific probability p. The correction
of the identified irregularity can be done through the replacement of that value by a given percentile value of the probability distribution function. The present work depicts an analysis undertaken in order to assess two parameters, the probability p of detection and the percentile for correction in the homogenisation using gsimcli. Two networks of the HOME benchmark data set were used and the performance metrics were calculated to compare this analysis with other homogenisation methods. Results show gsimcli as a favourable homogenisation method for monthly precipitation data, and reveal the most efficient detection and correction parameters for the homogenisation procedure.