*3.1. Weather Data Comparison*

For the analysis and comparison of the weather data, the first step was the data gathering from the *on-site* and *third-party* sources for the three locations for the period of study, which is the whole year 2019. *On-site* weather data were provided by weather stations installed in the buildings' surroundings. In Pamplona and Gedved, the weather station was installed on the buildings' roofs. In the case of Lavrion, the weather station was placed in the Technological and Cultural Park where the two test sites were located, near H2SusBuild. Table 1 shows the range, resolution, and accuracy of the sensors that formed part of each weather station. In general, the sensors of Pamplona's weather station had the best accuracy. In the case of Lavrion, the diffuse solar radiation had a manual shadow bar that required readjustment every two days.

The time period of the measured data gathered from the three weather stations is the whole year 2019. Despite the good quality of the measured weather data, the raw data contained small gaps, usually a few hours, so interpolation was performed in order to fill in the missing data. On the other hand, the Weather Converter tool, which is used to generate the weather files, allows undertaking a complementary validation of the data since it produces a warning if data out of the range are used in the weather files' generation process.

The *third-party actual* weather data for the year 2019 and for three locations were provided by meteoblue [65]. They are simulated historic data for a specific place and time calculated with models based on the NMM (nonhydrostatic meso-scale modeling) or NEMS (NOAA Environment Monitoring System) technology, which enables the inclusion of the detailed topography, ground cover, and surface cover. Further information about the computation of the weather data provided by meteoblue is available in [66].

The results of the weather parameter comparison between data from *on-site* weather stations and *third-party* (meteoblue) are shown using the Taylor diagram described in Section 2.1. Figure 3 summarizes all the results: showing the three statistics (*R*, *RMSdi f f erence*, and *σ*) for the whole period of study (2019) calculated on an hourly basis, for the six weather parameters analyzed (*Temp*, *RH*, *DNI*, *DHI*, *WD*, and *WS*), and for the three locations (each one represented in a different color).

For Pamplona weather (represented in blue), the diagram shows that *Temp* provided the weather parameter for this location that better agreed with the *on-site* observations as it had the highest correlation *R* of around 0.95, the smallest *RMSdi f f* (±0.3), and a very close *σ<sup>f</sup>* (standard deviation) to the reference (±0.95). *RH*, *DNI*, and *DHI* provided similar results with a correlation around 0.7–0.8, an *RMSdi f f* between 0.5–0.6, and a good standard deviation. The parameters that correlated worse with the observed values were the wind parameters, especially *WD* (*R* = 0.09).

For Gedved weather (red color), the Taylor diagram shows that *Temp* was the *third-party* weather parameter that agreed best with the *on-site* observations, with an *R* of around 0.97. As in Pamplona, the wind parameters delivered the results furthest from the reference point. *WS* had an acceptable correlation, but a very high standard deviation, and *WD* performed better for *σ<sup>f</sup>* , but had a low correlation (less than 0.5). For the third location, Lavrion (green color), *Temp* also had a good correlation *R* (higher than 0.95). *RH*, *DHI*, and *WS* had a medium *R* for the observed data (around 0.8), but they presented differences in the other metrics. *RH* had a better standard deviation than the other two, and *RH* and *DHI* had a lower *RMSdi f f* than *WS*. In this location, the parameter that provided the worst results was *WD*, which had the worst *R* (−0.22) and the highest *RMSdi f f* (1.5).


**Table 1.** Technical specifications of the sensors of the weather stations installed in the office building in Pamplona (Spain), in Gedved School (Denmark), and

 in the

**Figure 3.** The normalized Taylor diagram for the three weather locations (Pamplona (Spain), Gedved (Denmark), and Lavrio (Greece)) compared on an hourly basis for *on-site* and *third-party* weather data for the year 2019. This shows the dry bulb temperature, relative humidity, direct normal irradiation, diffuse horizontal irradiation, wind speed, and wind direction.

Comparing the statistical results for the three locations, the performance of data provided by the *third-party* varied for each location. Gedved provided the best results for four of the six weather parameters (*Temp*, *RH*, *DHI*, and *WD*). In the three cases, *Temp* was the parameter that best matched the reference (*R* around 0.95, *RMSdi f f* lower than 0.4, and *σ<sup>f</sup>* near one), and *WD* was the worst (correlations lower than 0.5 and *RMSdi f f* higher than 0.9). *WS* also provided poor correspondence with the observations, especially for the Gedved location. The rest of the parameters were scattered in the medium part of the diagram.

In Appendix A, a deeper analysis is shown where the statistical indexes for the monthly and seasonal data are represented in order to analyze their homogeneity. Figures A1–A3 show that the *Temp*, *RH*, *DNI*, *DHI*, and *WD* parameters for the three weathers were quite homogeneous, with the seasonal and monthly indexes quite concentrated, providing similar *R*, *RMSdi f f* , and *σ<sup>f</sup>* . There are some exceptions, such as *DNI* for November in Pamplona and January in Gedved, which agreed worse with the observations than for the rest of the months. On the other hand, *WS* had more heterogeneous monthly and seasonal results since more scattered points were seen in the diagrams. In general, the winter and autumn months correlated the worst with the observed data.

Since the wind parameters produced higher variations when comparing *on-site* and *third-party* weather datasets, a deeper comparison analysis was performed using wind rose diagrams (see Figure 4). This diagram shows the distribution of the wind speed and wind direction. The rays point to the direction the wind is coming from, and their length indicates the frequency in percentage. The color depends on the wind speed, growing from blue to red colors. Pamplona's wind rose shows that *WS* from the *third-party* data was much higher than observed, and although the prevailing direction was north in both cases, there were important differences in the frequency percentages. For Gedved, the *third-party* data provided much higher wind speed (yellow to red colors in the wind rose) than the observed data (blue colors) and a different prevailing wind direction. Finally, the wind roses for Lavrion show differences in the prevailing wind direction and very different wind speeds, being higher in the *third-party* wind rose.

**Figure 4.** Wind rose comparison between the weather station and *third-party* data of the Pamplona (Spain), Gedved (Denmark), and Lavrion (Greece) locations.

S
