A geostatistical simulation algorithm for the homogenisation of climatic time series: a contribution to the homogenisation of monthly precipitation series

Research output: ThesisDoctoral Thesis


As defined by the Intergovernmental Panel on Climate Change (IPCC), climate change refers to a change in the state of the climate that can be identified by changes in the statistical characteristics of its properties and that persists for an extended period, typically decades or longer. In order to assess climate change and to develop impact studies, it is imperative that climate signals are clean from any external factors. However, non-natural irregularities are an inevitable part of long-time climate records. They are introduced during the process of measuring and collecting data from weather stations. Accordingly, it is essential to detect and correct those irregularities a priori, through a process called homogenisation. This process became a hot topic in the last decades and many researchers have focused on developing efficient methods. Still, some climatic variables are lacking homogenisation procedures due to their high variability and temporal resolution (e.g., monthly precipitation).
We propose the gsimcli (Geostatistical SIMulation for the homogenisation of CLImate data) homogenisation method, which is based on a geostatistical simulation method, namely the direct sequential simulation. The proposed approach considers simulated values of the candidate station’s neighbouring area, defined by the local radius parameter, aiming to account for local characteristics of its climatic zone. gsimcli has other modelling parameters, such as the candidates order in the homogenisation process, the detection parameter, and the correction parameter (also used to fill in missing data). A semi-automatic version of gsimcli is also proposed, where the homogenisation adjustments can be estimated from a comparison series. The efficiency of the gsimcli method is evaluated in the homogenisation of precipitation data. Several homogenisation exercises are presented in a sensitivity analysis of the parameters for two different data sets: real and artificial precipitation data. The assessment of the detection part of gsimcli is based on the comparison with other detection techniques using real data, and extends a previous study for the south of Portugal. Artificial monthly and annual data from a benchmark data set of the HOME project (ACTION COST-ES0601) is used to assess the performance of gsimcli. These results allow the comparison between gsimcli and state-of-the-art methods through the calculation of performance metrics.
This research allowed identifying gsimcli parameters that have a high influence in the homogenisation results: correction parameter, grid cell size and local radius parameter. The set of parameters providing the best values of performance metrics are recommended as the most suitable set of homogenisation parameters for monthly precipitation data. Results show gsimcli as a favourable homogenisation method for monthly precipitation data that outperformed a few well established procedures. The filling in of missing data is an advantage when compared to other methods. Taking advantage of its capability of filtering irregularities and providing comparison series, gsimcli can also be used as a pre-homogenisation tool followed by the use of a traditional homogenisation method (semi-automatic approach).
As future work, it is recommended the performance assessment of the gsimcli method with denser monitoring networks, and the inclusion of a multivariate geostatistical simulation algorithm in the homogenisation procedure.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • NOVA Information Management School (NOVA IMS)
  • Costa, Ana C., Supervisor
Award date5 May 2017
Publication statusPublished - 5 May 2017


  • geostatistical simulation
  • homogenisation
  • precipitation series


Dive into the research topics of 'A geostatistical simulation algorithm for the homogenisation of climatic time series: a contribution to the homogenisation of monthly precipitation series'. Together they form a unique fingerprint.

Cite this