Exploratory data analysis of Burkina Faso’s Raw daily meteorological records (1981–2021): data quality insights and potential outlier screening

Abstract This study performs a national-scale quality screening of raw daily meteorological observations from Burkina Faso’s National Meteorological Agency, covering 10 synoptic stations over 1981–2021. In data-sparse settings, long archives are often affected by interruptions, non-uniform measurement practices, and heterogeneous digitization, which can propagate bias into trend, extremes, and impact studies. We implement a structured exploratory data analysis workflow as an operational first-line diagnostic that integrates: (i) station–variable completeness profiling and missingness mapping, (ii) distributional and range-consistency checks to flag candidate outliers and truncation artefacts, (iii) standardized anomalies relative to the 1991–2020 climate normal, (iv) station-wise variability and linear trend estimation, and (v) cross-station seasonal/interannual pattern comparisons. As a complementary diagnostic step, a preliminary homogeneity assessment based on breakpoint detection was conducted on selected monthly series to identify candidate non-climatic discontinuities that may affect the interpretation of apparent long-term trends; no correction was applied, and the results are reported as first-order indicators pending full homogenization. Rather than proposing a new statistical theory, the novelty of this work is the systematic, multi-variable and multi-station consistency assessment applied to a 41 year national archive, which identifies and documents variable-specific singularities that are rarely reported but critical for downstream use. The analysis reveals (a) non-negligible missingness at national level (≈10% overall), with very large gaps for some variables (e.g. global radiation > 60% in several stations), and (b) suspicious behaviour in wind direction distributions consistent with potential measurement/processing artefacts, alongside coherent temporal shifts in wind anomalies that warrant targeted follow-up testing. The proposed workflow may be useful as a first-line screening template for meteorological archives in data-sparse regions, although its applicability to other settings has not been evaluated and would require independent validation. Comprehensive homogenization further requires complementary quality-control tests and advanced methods, including recent machine-learning developments.