Propuesta metodológica para completar series de tiempo mensuales de temperatura cuando no existen estaciones adyacentes

Abstract

El uso de series temporales sin valores faltantes resulta fundamental para la veracidad de los análisis científico-geográficos. Dado que en el caso de la temperatura (variable objeto de estudio), es posible encontrar vacíos de información en sitios con escasas estaciones meteorológicas, el objetivo de investigación fue desarrollar una metodología para completar series de tiempo mensuales cuando no se dispone de estaciones meteorológicas adyacentes. El método planteado se vincula al uso de temperaturas anuales, medias mensuales y promedio seriales, de modo que su proporción permita calcular los datos carentes de información. Tomando como base una estación meteorológica localizada al centrooccidente árido argentino, la investigación se centró en cuatro fases vinculadas a la aplicación de la propuesta. Inicialmente, se aplicó la metodología planteada para calcular valores faltantes mensuales sobre la estación San Juan Aero, con N= 46 años. Adicionalmente, se comparó la propuesta con otras metodologías que utilizan estaciones adyacentes y su aplicación sobre otra estación fuera del área de estudio. Todo lo anterior fue verificado mediante nueve índices de precisión. Los resultados indican que la metodología propuesta presenta óptimos resultados en estaciones con N ≥ 30 años, 10 % de datos faltantes y con precisión superior a otras metodologías de naturaleza similar. Se utilizó el método propuesto en series de tiempo pertenecientes a estaciones fuera del área de estudio y los resultados mostraron una alta fiabilidad en el cálculo de datos faltantes. Por lo tanto, se recomienda su uso para completar series temporales de temperaturas mensuales en estaciones sin pares adyacentes.

Complete time series with no missing values are essential for reliable scientific-geographic analyses. Temperature time series commonly show data gaps, particularly in meteorological stations located in regions with few scattered stations. Scarce meteorological stations exist in the arid central-western region of Argentina, where vast, sparsely populated, or unproductive areas far away from major urban centers and oases may have restrained the installation of sufficient stations. Thus, climate data records from existing stations, especially those in rural areas, often lack temporal continuity, and the data gaps have to be filled in based on data from adjacent stations. However, this is not possible in the absence of nearby stations with reliable and sufficiently long records that can be used for estimating the missing data. This study aimed to develop an easy-to-apply, highly accurate operational method to fill data gaps in monthly temperature time series, which is particularly suitable for locations with no nearby meteorological stations. The method developed herein is based on the use of annual and monthly means and the overall time series average. The method was tested on the 46-yr time series of monthly temperature data recorded at the meteorological station of San Juan Aero (base station), located in the Province of San Juan, central-western Argentina. The base station is close to two other weather stations whose data were used to validate the results of one of the phases of the method. The study included: a) the application of the method proposed to the San Juan Aero station (base station) using subsets of varying lengths of the entire time series data set, and comparing the accuracy of the estimates thus obtained by means of ad hoc indices; b) the application of the same procedure used in a), but with varying percentages of missing data; c) the comparison of the missing values estimated by the method developed herein versus those estimated using conventional methods based on data from adjacent stations, and d) the application of the method developed herein to a meteorological station located outside the study area under different climate and environmental conditions. The method proposed (Ti) estimates missing monthly temperature values as the product of the average of the mean annual temperatures of the years immediately before and after the year with missing monthly data multiplied by the mean temperature of the target month, divided by twice the mean annual temperature of the data series. The method was used to estimate missing monthly temperature values for the 46-yr time series recorded in the San Juan Aero station. Tests were run to determine the percentage of missing data (5%, 10%, and 15%) with which the method yields the best fit. The efficiency of the Ti method was compared versus three traditional methods (arithmetic mean, normal proportion, and inverse distance weighting) that impute the missing values from data recorded at nearby stations. Finally, based on the results from the previous stages, the Ti method was applied to a test station located some 150 km from the baseline station to determine whether it can also be applied to meteorological stations located outside the study region, under different physical environmental characteristics. The results showed that the Ti method works better on meteorological stations having at least 30-yr records and no more than 10% missing data; under these conditions, its estimates are more accurate that those yielded by the three traditional methods tested and can be reliably applied to stations located outside the study region under different physical and environmental conditions. The limitations of the Ti method are worth mentioning: it cannot be used when the baseline station records have a data gap longer than one full year, or when data for the same month are missing for two consecutive years. Given the results yielded by this method and taking into account the limitations mentioned above, compared to other methods that use data from nearby stations, we recommend using the Ti method to estimate missing monthly temperature values for meteorological stations lacking nearby stations. Compared to the traditional methods tested, the Ti method seems highly valuable as a tool to fill missing data in temperature time series from isolated weather stations, which then could be used for climate analyses of remote zones.