Evaluating patterns of solar irradiance errors over an area of complex topography

Abstract:

In a number of recent studies, errors in the downwelling shortwave irradiance (Qsi) have been shown to hinder the ability of physically-based land-surface models to simulate fluxes of energy and water and can lead to erroneous parameter values. However, the uncertainty in the forcing data, especially Qsi, is rarely considered, at least in part because the uncertainties in Qsi are not well-characterized over complex terrain.

In this study we provide a novel evaluation and diagnosis of error sources for 4 different methods for estimating Qsi: NLDAS, MTCLIM, CERES Synoptic Fluxes and Clouds, and output from a regional reanalysis. These methods are evaluated over a study area in California spanning the Central Valley and Sierra Nevada, which contains 73 sites that observe Qsi over the period of 2002-2012. This environment represents a particular challenge for estimating Qsi due to cloud patterns over fine temporal (e.g., afternoon thunderstorms) and spatial (e.g., valley fog) scales. Since observations are made with lower-quality instruments, limiting absolute accuracy at any individual site, we instead use the spatial coherence in bias magnitude and sign to characterize errors.

The absolute biases at all stations in the study domain over the study period are small (<20Wm-2). However, at the monthly scale, large biases between irradiance methods and groups of stations occur. Mountain stations west of the Sierra crest exhibit large (>40Wm-2) monthly biases while the stations nearby, but to the east of the crest, exhibit small, negative biases (<-20Wm-2). Valley stations have overall small errors during the winter, suggesting all methods capture the behavior of the persistent radiation fog that forms in the valley. In contrast, during late spring and summer, biases in Qsi increase dramatically (>80Wm-2), correlating with increasing aerosol optical depth. At high elevations the four methods disagree on the monthly mean of Qsi with a spread greater than 100Wm-2. Finally, we show how the choice of Qsi method impacts gradients of Qsi with elevation, mostly as a consequence of the method’s spatial resolution. These errors are attributable to differences in how each method estimates Qsi.