Energy system modelling on an international scale requires meteorological spatial input data on hourly resolution. With an additional requirement of keeping the input data size low, typically two compromises are made: either the spatial resolution is drastically reduced, for example by aggregating to NUTS regions or using coarse models like ERA5-Land with a resolution of 0.1°, or a selection of ’representative’ weather years is taken, compromising on the variety of weather phenomena possibly impacting the system.
Choosing coarse resolution bears the risk of losing valuable information, especially in mountainous regions. Therefore, how spatial resolution is reduced is also contributing to the quality of data. Simply averaging over space or the use of coarse resolution data results in mean meteorological variables, which do not necessarily describe conditions by which most of the population is affected. Especially in mountainous countries, where most of our research team is based, elevation has an high impact on parameters like wind speed, radiation and temperature, drastically influencing electricity relevant parameters like power production, efficiency and demand. Fig. 1 shows such an example in two cities surrounded by mountains. With a resolution of 0.1°, ERA5-Land represents a coarse model or a simple spatial averaging. The dots show the number of people living at a specific elevation according to a 1 km population density data set within the area of one ERA5-Land grid cell, in total 100 dots for Innsbruck in Austria (blue) and Andorra (red). The dotted vertical lines indicate the average elevation of the ERA5-Land grid cell. These lines clearly show, that the average elevation is far away from where the people are actually living. The full vertical lines indicate the average elevation of the grid cell, when using the population weighting approach from the SECURES-Met data set. For Innsbruck the simple spatial averaging leads to an elevation of ∼1300 m but the city centre of Innsbruck, where the people are living, is at approximately 600 m. A similar effect can be seen in Andorra (red) where population weighting leads to an average elevation of ∼1350 m and simple spatial averaging to approximately 1900 m. As temperature has a strong relation to elevation with an average vertical lapse rate of approximately 6.5°C/km, differences in the mean elevation of 700 m as in Innsbruck, correspond with an underestimation ofthe temperature by more than 4.5°C, which leads to implausible values for heating and cooling demands. The use of the direct output of ERA5-Land and/or a simple spatial averaging to NUTS-levels therefore leads to systematic biases in mountainous regions, which highly impacts especially the energy demand for heating and cooling. In general the temperatures are too cool, which leads to an overestimation of heating demand in winter and and underestimation of cooling demand in summer.
During our research, we therefore aimed to provide historical and future scenario meteorological input data, which is as small and handy as possible for energy modellers but also represents the variability weather and climate change has to offer, specifically in areas where people who usually consume energy live. An aggregation to NUTS regions was therefore aimed at to reduce data size while conserving meteorological accuracy. Fig. 2 shows the population density of the two mountainous NUTS2 regions across Europe which contain the cities from Fig. 1, on a 1 km digital elevation model (DEM). It is again observable, that people tend to live in lower elevated areas. For energy system modelling, this is our location of interest. Taking the spatial mean would still result in an average elevation far from the band where the majority of the population is located, visible by the dashed lines. That means even when enhancing the resolution to 1 km, the spatial mean of meteorological variables without any additional weightings would still result in non-representative conditions. Temperatures would be lower on average, indicating higher heating demand during energy modelling and more cold days.
That is why we decided to use population weighting for the temperature and radiation in our SECURES-Met data set. Fig. 3 shows the impact of such population weighted temperature compared to the conventional mean temperature in the very same areas. In between 1981 and 2020, the number of summer days with temperatures above 25°C occur multiple times more often when looking at places where people actually live, resulting in higher cooling demands than expected. The number of frost days with temperatures below freezing level tends to sink, as expected.
Fig. 4 shows the spread of elevation - meaning the difference between the maximum and minimum elevation - within a single NUTS2 region based on the 1 km DEM. The Alpine region and the Pyrenees for example show a spread of several thousand metres per NUTS2 region, indicating the desperate need for using population weighting when wanting to observe temperature dependent impacts on population like heating and cooling demand in mountainous regions.
Other variables of course require different weighting methods. Wind power is not dependent on the elevation but rather where it would be possible and legally allowed to build wind farms. Therefore, our data set provides different aggregation methods for hydro power and wind power, which are optimised for every indicator. With these aggregation methods, we ended up with a data set useful and small enough for energy modelling while still covering meteorological precision, focused on locations where people life, roofs are decorated with solar panels and heating and cooling is required. Being a result from interdisciplinary science, we are quite proud of the produced data and hope it will serve the energy system modelling community well.
(Cover-image: Innsbruck and surrounding mountains. Image rights belong to David Leidinger.)