Documentation

Documentation Accuracy and comparisons

Validation of albedo data

To evaluate the accuracy of Solargis albedo data products, a comparison between Solargis estimates and ground measurements of albedo has been made. Details of the ground measurements used for validation purpose are given in the table below.

Station name Latitude, Longitude Data source Installation height Field of view Period of evaluation*
Metolius Mature Pine 44.452, -121.557 AMERIFLUX 32 m 0.16 km2 2011 to 2015
Morgan Monroe State Forest 39.323, -86.413 AMERIFLUX 46 m 0.33 km2 2011 to 2015
Willow Creek 45.806, -90.08 AMERIFLUX 29.6 m 0.14 km2 2011 to 2015
Walnut Gulch Kendall Grasslands 31.737, -109.942 AMERIFLUX 3 m 0.0014 km2 2011 to 2015
Fort Peck 48.308, -105.102 SURFRAD 10 m 0.015 km2 2011 to 2015
Desert Rock 36.624, -116.019 SURFRAD 10 m 0.015 km2

2011 to 2015

Table Mountain 40.125, -105.237 SURFRAD 10 m 0.015 km2 2011 to 2015
Bondville 40.052, -88.373 SURFRAD 10 m 0.015 km2 2011 to 2015

* The averaging period of the Solargis monthly average database [2006 to 2015] is different from the period of albedo measurements [2011 to 2015].

Although this validation exercise is limited to a relatively small set of locations, they represent a reasonably broad climatic diversity.

Satellite vs albedometer field of view

It is important to highlight that the field of view of an albedometer is typically smaller than the area covered by a pixel of satellite imagery. The spatial resolution of Solargis albedo data products is 0.25 km2 (for time series product) or 1 km2 (for the monthly-average product). Whereas the surface area covered by the FOV (field of view) of the albedometer varies as a function of the installation height from 0.001 km2 (for albedometer at 1.5m) to 0.14 km2 (for albedometer at height of 30m)

When looking at satellite pictures of the area surrounding the albedometers, we see that land cover is not always homogenous. Figure 1 show pictures of satellite views of some locations. In case of Bondville, there is strong heterogeneity in the pixel in relation to the FOV of the instrument at 10m. Similar heterogeneity is seen for Desert Rock, although less pronounced. The Morgan Monroe location seems to present the greatest homogeneity among the four locations.


Figure 1. Satellite views of the locations of Bondville, Desert Rock, Morgan Monroe State Forest and Walnut Gulch Kendall Grasslands. Pictures show areas that approximately have the same pixel size as of the satellite (500m)

The evaluation of the accuracy of Solargis albedo products is thus affected by the difference in areas covered by the FOV of albedometers and the satellite pixel.

Estimation of uncertainty of Solargis albedo data

To estimate the uncertainty of Solargis albedo values, an approach based on statistical deviations between Solargis estimates and ground observations has been considered. This uncertainty also includes uncertainty due to the differences in FOV of satellite-based estimates and ground-based observations.

Two percentile thresholds have been selected for the statistical significance of the uncertainty: P68 (± 1 standard deviation of the normal distribution) and P95.

 

Time-series product

Monthly average product (Prospect database)

Conditions

U at 1STDEV [68%]

U at 2STDEV [95%]

U at 1STDEV [68%]

U at 2STDEV [95%]

All conditions

±0.029

±0.081

±0.026

±0.045

No snow (<=0.4)

±0.026

±0.071

±0.0194

±0.039

Snow (>0.4)

±0.159

±0.218

Nan

Nan

(Nan values are due to lack of data to infer statistics)

Uncertainty values have been provided separately for no-snow and snow when possible (e.g. for desert or tropical areas) because snow is a specific event that doesn't occur everywhere.

An extension of the validation study is of great interest to confirm and reinforce these results.