MODIS Regimes of Regimes

The RORs are derived from the 1° daily MODIS CR dataset of Cho et al. (2021). The salient characteristics of these CRs are expressed via: (1) their centroids, namely mean joint histograms of the clusters representing the CRs obtained from applying a clustering algorithm on CTP-COT joint histograms; and (2) the multi-year mean maps of the relative frequency of occurrence (RFO) of the CRs assigned to each grid cell over a 20-year period. We built “rolling” histograms of CR frequency of occurrence (FOC) over 3-month periods for all one-degree grid cells containing CRs. The main reason for employing such overlapping 3-month windows is to achieve sample sizes that lessen noise in one-degree CR FOC histograms. Both Terra and Aqua CRs are used in the creation of these histograms, so each of them is built from ~3 × 30 × 2 ≈ 180 values. A large number of histograms of CR ROC (~15M) were furnished to a k-means clustering algorithm. We settled on K=8 as the “best” number of clusters to compactly describe the planet’s cloudiness in terms of seasonal one-degree RORs. Staying with K=8, has also the additional advantage of avoiding extreme disparities in global RFO among RORs: global RFOs range between ~7% and ~20% indicating that no ROR is exceedingly rare or overly frequent. While refinement can always be achieved in future efforts, here our first order goal is to demonstrate the physical foundation and viability of the ROR concept.