Alternative approaches for estimating missing climate data: application to monthly precipitation records in South-Central Chile

Abstract

Background: Over the last decades interest has grown on how climate change impacts forest resources. However, one of the main constraints is that meteorological stations are riddled with missing climatic data. This study compared five approaches for estimating monthly precipitation records: inverse distance weighting (IDW), a modification of IDW that includes elevation differences between target and neighboring stations (IDW m), correlation coefficient weighting (CCW), multiple linear regression (MLR) and artificial neural networks (ANN). Methods: A complete series of monthly precipitation records (1995-2012) from twenty meteorological stations located in central Chile were used. Two target stations were selected and their neighboring stations, located within a radius of 25 km (3 stations) and 50 km (9 stations), were identified. Cross-validation was used for evaluating the accuracy of the estimation approaches. The performance and predictive capability of the approaches were evaluated using the ratio of the root mean square error to the standard deviation of measured data (RSR), the percent bias (PBIAS), and the Nash-Sutcliffe efficiency (NSE). For testing the main and interactive effects of the radius of influence and estimation approaches, a two-level factorial design considering the target station as the blocking factor was used. Results: ANN and MLR showed the best statistics for all the stations and radius of influence. However, these approaches were not significantly different with IDW m. Inclusion of elevation differences into IDW significantly improved IDW m estimates. In terms of precision, similar estimates were obtained when applying ANN, MLR or IDW m , and the radius of influence had a significant influence on their estimates, we conclude that estimates based on nine neighboring stations located within a radius of 50 km are needed for completing missing monthly precipitation data in regions with complex topography. Conclusions: It is concluded that approaches based on ANN, MLR and IDW m had the best performance in two sectors located in south-central Chile with a complex topography. A radius of influence of 50 km (9 neighboring stations) is recommended for completing monthly precipitation data.

Analysing Trends and Spatio-Temporal Variability of Precipitation in the Main Central Rift Valley Lakes Basin, Ethiopia

“…But only two stations (Adamitulu and Langano) are with the missing values of less than 16%. In this paper, the missing values have been estimated by using Correlation Coefficient Weighting (CCW) [15]. In this approach, missing data from target station are determined from the values observed in neighbouring stations weighted by the Pearson's correlation coefficient between the target and the neighbouring stations.…”

Spatial interpolation methods for estimating monthly rainfall distribution in Thailand

“…Several studies focus on searching for the optimal approach for spatial precipitation interpolation. Tabios et al (1985) Barrios et al (2018) compared five different interpolation schemes under the consideration of the radius of influence. The overall best performance can be obtained from the ANN, MLR, and the modified version of the IDW method.…”

A Comparative Assessment of Geostatistical, Machine Learning, and Hybrid Approaches for Mapping Topsoil Organic Carbon Content

“…Climate data of monthly average values of September and October 2016, i.e., mean relative humidity (W, %), average temperature (T, • C), and precipitation (P, mm) were recorded at 60 weather stations of Jilin Province (http://data.cma.cn/). The inverse distance weighted method, a useful and common approach to acquiring spatially continuous soil and climate factors [56][57][58], with the power value of two in ArcGIS was applied to data of BD, W, T and P.…”