Pyranometer calibration drift caused by radiation exposure: effects on sensor response and its detection in historical time series

Accurate measurements of surface solar radiation fluxes are needed for validating and forcing climate models, and surface energy balance studies. The most common sensors used to measure hemispherical, broadband incoming (K↓) and outgoing (K↑) solar radiation fluxes are all-black thermopile pyranometers. In the absence of a high-class reference pyranometer that can be used in cross-comparisons with field-deployed sensors, it is difficult to determine when calibrations are required. Therefore, the objectives of this research were to examine changes in the responsivities of continuously deployed pyranometers, and to develop a method that can detect sensor drift and when recalibration is needed. To meet our objectives we analyzed 5-y of historical data and a sensor cross-comparison. Historical data included measurements of K↓ and K↑ from four all-black thermopile pyranometers (model PSP, The Eppley Lab Inc., Newport RI), and one black and white (BW) pyranometer (model 8-48, The Eppley Lab Inc.) that were initially deployed in agricultural systems at Rosemount, Minnesota in 2001. During the cross-comparison, the four PSPs were co-located adjacent to a previously un-deployed BW (model 8-48) to measure K↓ over a 10 d period.

We found that PSP factory calibrations can be stable from 1–5 y, after which the responsivity degrades linearly as a function of radiation exposure at a rate of -0.38 ± 0.054% (GJ m^-2)^-1, which equates to a decay rate of -1.5 ± 0.19% y^-1 for pyranometers deployed facing upwards in mid-latitude, continental regions. As the decay worsened, the sensing elements visibly changed to a green color. This discoloration indicated that the spectral responses of the pyranometers had changed over time. An important implication of this effect will be manifested when attempting a recalibration, because a field pyranometer with a discolored sensing element (and altered spectral response) will only be accurate during subsequent deployment when incident radiation has the same spectral characteristics as the source radiation during the calibration procedure. Watching for discoloration of the sensing element, and having it reconditioned when necessary should therefore be part of pyranometer maintenance protocols. It is, however, a more difficult task to determine when a calibration is required in the absence of a high-class reference instrument, given that the initial stable period may last from 1–5 y.

Here, we propose a method to determine when a calibration is necessary based on an analysis of K↓, photosynthetically active radiation (PAR), and total solar irradiance (TSI) data that are obtained from the SORCE satellite mission. This involves examining the temporal trends in three ratios: (i) PAR:K↓ (expected to increase if pyranometer responsivity decays), (ii) K↓:TSI (expected to decrease if the pyranometer responsivity decays), and (iii) PAR:TSI (expected to mirror the K↓:TSI trend if the pyranometer is stable). The use of these three ratios overcomes the challenges of detecting a change in sensor responsivity that may be masked by natural variations in solar intensity and atmospheric optical depth. Results from our agricultural systems as well as other AmeriFlux and Fluxnet Canada sites will be presented. The ability to obtain a quantitative estimate of when calibrations are necessary will improve the accuracy of long-term solar radiation measurements, and avoid costly unnecessary calibrations. This approach could also be used to perform QA/QC on historical datasets that are to be included in site-synthesis analyses.