Climate data homogenisation is of major importance in monitoring climate change and in validating weather forecasts, general circulation and regional atmospheric models, modelling of erosion and drought monitoring, among other impact studies. Discontinuities in the time series, also named inhomogeneities, may lead to biased conclusions in such studies, so they should be detected and corrected. Previous studies have suggested a geostatistical stochastic approach, which uses Direct Sequential Simulation (DSS), as a promising methodology for the homogenisation of precipitation data series. Based on the spatial and temporal correlation between the neighbouring stations, DSS calculates local probability density functions at a candidate station to detect inhomogeneities. Here, we present a new method named gsimcli (Geostatistical SIMulation for the homogenisation of CLImate data), which is an improved and extended version of that approach. This technique is novel in its incorporation of spatial correlation metrics for the homogenisation of climate time series. The method's performance is assessed with annual and monthly precipitation, and monthly temperature data from two regions of the COST-HOME benchmark data set, and the results are compared using performance metrics. We also evaluate a semi-automatic version of the gsimcli method, which performs additional adjustments for sudden shifts. Both gsimcli versions provided similar results in the homogenisation of annual series. The gsimcli method was more efficient in the homogenisation of the benchmark's precipitation series than the original geostatistical approach. The gsimcli approach performed more closely to state-of-the-art procedures in the homogenisation of monthly data than in the homogenisation of annual data. We expect that the proposed procedure will open new perspectives for the development of techniques that detect and correct inhomogeneities in climate data with monthly and sub-monthly resolution.
- Climate data
- Data quality