An automated satellite cloud classification scheme using self‐organizing maps: Alternative ISCCP weather states

Abstract

This study explores the application of the self‐organizing map (SOM) methodology to cloud classification. In particular, the SOM is applied to the joint frequency distribution of the cloud top pressure and optical depth from the International Satellite Cloud Climatology Project (ISCCP) D1 data set. We demonstrate that this scheme produces clusters which have geographical and seasonal patterns similar to those produced in previous studies using the k‐means clustering technique but potentially provides complementary information. For example, this study identifies a wider range of clusters representative of low cloud cover states with distinct geographic patterns. We also demonstrate that two rather similar clusters, which might be considered the same cloud regime in other classifications, are distinct based on the seasonal variation of their geographic distributions and their cloud radiative effect in the shortwave. Examination of the transitions between regimes at particular geographic positions between one day and the next also shows that the SOM produces an objective organization of the various cloud regimes that can aid in their interpretation. This is also supported by examination of the SOM's Sammon map and correlations between neighboring nodes geographic distributions. Ancillary ERA‐Interim reanalysis output also allows us to demonstrate that the clusters, identified based on the joint histograms, are related to an ordered continuum of vertical velocity profiles and two‐dimensional vertical velocity versus lower tropospheric stability histograms which have a clear structure within the SOM. The different nodes can also be separated by their longwave and shortwave cloud radiative effect at the top of the atmosphere.

Regional Regime‐Based Evaluation of Present‐Day General Circulation Model Cloud Simulations Using Self‐Organizing Maps

“…Leinonen et al () conclude that clusters show strong intracluster variance between different regions and as such attributions to given clusters should include some measure of the intracluster variability. McDonald et al () was able to achieve similar results to many of the above papers with the usage of the SOM clustering technique instead of the k ‐means approach used in those papers. Analyzing the identified clusters can also directly lead to the identification of subtle phenomena, such as the discrimination between clusters that correspond to closed versus open mesoscale convective cells demonstrated in McDonald et al ().…”

A New Method to Evaluate Reanalyses Using Synoptic Patterns: An Example Application in the Ross Sea/Ross Ice Shelf Region

“…associated with the fact that these three nodes represent three very distinct synoptic states relative to their closest neighbors. Less frequently occurring nodes (Nodes 5, 7, and 8) display less coincidences, as identified in McDonald et al (2016) these nodes may reflect transition states since the SOM classification scheme occasionally identifies transition nodes because the SOM creates a continuous gridded representation of the data space. We would therefore expect these transition states to span relatively empty portions of the physical data space, and because these states are temporary we might also expect lower coincidence.…”

Cluster‐Based Evaluation of Model Compensating Errors: A Case Study of Cloud Radiative Effect in the Southern Ocean

“…The average values of several key variables for each of these clusters are shown in Table 1. Detailed descriptions of the data preparation and clustering processes used here are given in McDonald et al (2016) and Schuddeboom et al (2018). The MODIS data are used to determine the CTP, COT, relative frequency of occurrence (RFO), and CF of each cluster.…”

“…The CERES data are used to calculate the shortwave CRE (SW CRE). Detailed descriptions of the data preparation and clustering processes used here are given in McDonald et al (2016) and Schuddeboom et al (2018).…”