Combining thin-plate spline interpolation with a lapse rate model to produce daily air temperature estimates in a data-sparse alpine catchment

Abstract

Insufficient availability of weather stations recording air temperature is a common problem in many alpine regions. The low station density combined with the high variability of air temperature means that interpolated fields based on simple or more complex interpolation techniques are unlikely to be representative of the real patterns of air temperature. In this study, a novel method was developed to tackle this problem, following initial investigation of lapse rate variability in the study domain: the alpine Clutha catchment, New Zealand. Owing to a series of complexities in lapse rate variability, a multi‐layer approach was adopted to produce 1 km2 daily fields of maximum (Tmax) and minimum air temperature (Tmin). The first layer of the Tmax and Tmin models was calculated using a trivariate thin‐plate spline, which was constrained to the elevations of the continuous network to avoid unrealistic extrapolation. To compensate for missing continuous high elevation records, two lapse rate models were implemented to scale air temperature above the first layer. The two lapse rate models were based on the dominant processes driving lapse rate variation, which were found to be cold air ponding (Tmin) and spatial differences in relative humidity (Tmax). Independent station records were used to assess accuracy and compare the resultant fields to an existing product (the Virtual Climate Station Network) and a more conventional method based on a bivariate spline and a constant lapse rate. The validation revealed that the new methods developed here have led to a substantial improvement in producing spatial estimates of Tmax, with a mean root mean square error (RMSE) of 2.38 °C, while progress in regard to Tmin was more limited (mean RMSE of 2.93 °C). As such, this work demonstrates that inclusion of the driving processes controlling lapse rates in interpolation routines can lead to improvements in accuracy.

Intercomparison of different uncertainty sources in hydrological climate change projections for an alpine catchment (upper Clutha River, New Zealand)

“…Inaccuracies in the meteorological fields (METEO OBS ) that were used for the bias correction could also have caused some of the seasonal over-and underestimations in the hydrological regime. As discussed in Jobst et al (2017) the climate network in the upper Clutha is sparse, with very few sites located in medium to high elevations. Notwithstanding the improved representation of temperature provided by the Jobst et al (2017) data set compared to other products, the remaining biases in this temperature field would have also propagated into the bias-corrected RCM fields and the corresponding hydrological baseline simulations.…”

“…Station-based meteorological observations of mean daily air temperature (T mean ), precipitation, solar radiation, relative humidity and wind speed were interpolated (Jobst, 2017;Jobst et al, 2017) and served as input to WaSiM during the calibration (2008-2012) and validation (1992-2008) periods. The last four hydrological years of the reference period were chosen for calibration because of the higher density of weather stations compared to previous years and a better consistency of the streamflow records.…”

“…The downscaling of precipi- tation (and the remaining three variables) is based on the topographical scaling approach of Frueh et al (2006), while maximum and minimum air temperatures are scaled via monthly lapse rate models (as described in Jobst et al, 2017 but excluding the thin plate spline layer). As part of the downscaling, additional processing steps were adopted from Marke (2011) to ensure the conservation of mass and energy when transforming the RCM data between the model scales.…”

Intercomparison of different uncertainty sources in hydrological climate change projections for an alpine catchment (Clutha River, New Zealand)

“…Two versions of this WaSiM implementation were set up, one with a simple temperature index (Tindex) snow melt routine (Schulla, 2012) and the other with the conceptual energy balance model of Anderson (1973). Station-based meteorological observations of air temperature, precipitation, solar radiation, relative humidity and wind speed were interpolated (Jobst, 2017;Jobst et al, 2017) and served as input to WaSiM during the calibration (2008-2012) and validation (1992-2008) periods. The last four years of the reference period were chosen for calibration because of the higher density of 10 weather stations compared to previous years and a better consistency of the streamflow records.…”

“…The individual submodels of WaSiM (unsaturated zone, groundwater, snow and glacier model) were calibrated iteratively using a combination of auto-calibration and manual parameter optimization (see Jobst (2017) for a detailed description of the calibration process). Particle swarm optimization (Kennedy and Eberhart, 1995) was used for auto-calibration due to its effective performance during the first iterations and fast operation (Jiang et al, 2010), allowing for an adequate compromise 15 between processing time and efficiency.…”