Quality assessment of HelioClim3

A problem of units ? of terminology ? Consult the Education service.

• Framework

We remind that the HelioClim3 database were globally calibrated using the 29 stations listed on this page. As a consequence, important discrepancies might be observed between satellite estimations and ground stations measurements. The origins of these differences are exposed in this page (to be developed). The SoDa team proposes a method to calibrate the HelioClim3 data onto ground station data using a least square regression.

Framework

The HelioClim3 databases are globally calibrated using the 29 ground stations. The protocol of this calibration is described here. As a consequence, discrepancies might be locally observed between ground station values and satellite estimations. The local calibration of HelioClim3 with ground stations values turns out to return more accurate irradiation values.

The ground station irradiation values can come from:
- our own ground stations or little solar device associated to your solar panels,
- free nearby ground stations,
- or finally from nearby ground stations which provide data for pay.

Actually, the question of the precision is always very tricky for several reasons:

• Because our routines are improved very often by our center of research MINES ParisTech, leading to different versions of the same base in the time, as well as some slight enhancements inside the same version.
• Because the access to ground station data of “good enough” quality and covering the same time period as HC3 is almost never free of charge.
• Because data over the tilted of normal surfaces require the use of empirical models of decomposition of the radiation components that add errors to the irradiation values over the horizontal plane. As the ground station data over inclined surfaces are even more rare, the quality assessment of the HelioClim data on this plane orientation is still more complicated.

The statistical results that are given on the website only concerns the second version of HelioClim3 (hc3v2) over the horizontal plane. The results are the following:
GHI month: relative bias: approx 5 % and RMSE: approx 5 %
GHI day: relative bias: approx 5 % and RMSE: approx 10 %
GHI hour: relative bias: 5% and RMSE: approx 15 %
According to our experience, the RMSE are at least doubled after the computation over an inclined or normal surface.

The results turns out to be better on the few points we studied for hc3v3. We definitely need to carry out the same assessment of hc3v2 on hc3v3 on the values over the horizontal surface and publish the results. Concerning the hc3v3 data over the normal and titled surface, we are worked on several sets of ground station data, in the framework of the calibration of long term HelioClim3 time series. However, most of the time these data are confidential.

I hope that these elements might partially answer your need for quality information.

 

Top of page