Cloud-free Observations Research Articles

Invariant object recognition with the adaptive resonance (ART) network

The cloud-gap-filled (CGF) method developed by Hall et al. (2010) is a novel method for efficiently mitigating cloud obscuration using the most recently available cloud-free observations from prior days at each pixel. In this paper, we extend this method not only using prior observations but also using subsequent observations. Four CGF using 3 or 5 days prior or subsequent observations are applied to the standard MODerate resolution Imaging Spectroradiometer (MODIS) snow cover product and are evaluated against 3-years, in situ observations at 244 SNOwpack TELemetry (SNOTEL) stations. Results indicate that daily CGF snow cover maps using prior or subsequent observations are obviously different due to daily cloud shifting. However, the monthly and annual cloud reductions are similar. Although the overall accuracies under all-sky conditions of CGF using prior or subsequent observations are very similar, CGF snow cover maps using subsequent images have less underestimation errors, except during the snow melting period, and higher snow accuracies. When the observation interval is expanded from 3 to 5 days, this method is able to fill more cloud obscurations and increase the overall accuracies by ∼30% and ∼37%, respectively; however, they introduce slightly more uncertainty. Monthly and regional analyses indicate that this method is more efficient during the snow-covered period or in high-elevation zones.