[1] We present results from a new three-dimensional empirical magnetopause model based on 15,089 magnetopause crossings from 23 spacecraft. To construct the model, we introduce a Support Vector Regression Machine (SVRM) technique with a systematic approach that balances model smoothness with fitting accuracy to produce a model that reveals the manner in which the size and shape of the magnetopause depend upon various control parameters without any assumptions concerning the analytical shape of the magnetopause. The new model fits the data used in the modeling very accurately, and can guarantee a similar accuracy when predicting unseen observations within the applicable range of control parameters. We introduce a new error analysis technique based upon the SVRM that enables us to obtain model errors appropriate to different locations and control parameters. We find significant east-west elongations in the magnetopause shape for many combinations of control parameters. Variations in the Earth's dipole tilt can cause significant magnetopause north/south asymmetries and deviation of the magnetopause nose from the Sun-Earth line nonlinearly by as much as 5 Re. Subsolar magnetopause erosion effect under southward IMF is seen which is strongly affected by solar wind dynamic pressure. Further, we find significant shrinking of high-latitude magnetopause with decreased magnetopause flaring angle during northward IMF.

[1] A new data mining technique called MineTool-TS is introduced which captures the time-lapse information in multivariate time series data through extraction of global features and metafeatures. This technique is developed into a JAVA-based data mining software which automates all the steps in the model building to make it more accessible to nonexperts. As its first application in space sciences, MineTool-TS is used to develop a model for automated detection of flux transfer events (FTEs) at Earth's magnetopause in the Cluster spacecraft time series data. The model classifies a given time series into one of three categories of non-FTE, magnetosheath FTE, or magnetospheric FTE. One important feature of MineTool-TS is the ability to explore the importance of each variable or combination of variables as indicators of FTEs. FTEs have traditionally been identified on the basis of their magnetic field signatures, but here we find that some plasma variables can also be effective indicators of FTEs. For example, the perpendicular ion temperature yields a model accuracy of $93%, while a model based solely on the normal magnetic field B N yields an accuracy of $95%. This opens up the possibility of searching for more unusual FTEs that may, for example, have no clear B N signature and create a more comprehensive and less biased list of FTEs for statistical studies. We also find that models using GSM coordinates yield comparable accuracy to those using boundary normal coordinates. This is useful since there are regions where magnetopause models are not accurate. Another surprising result is the finding that the algorithm can largely detect FTEs, and even distinguish between magnetosheath and magnetospheric FTEs, solely on the basis of models built from single parameters, something that experts may not do so straightforwardly on the basis of short time series intervals. The most accurate models use a combination of plasma and magnetic field variables and achieve a very high accuracy of prediction of $99%. We explain the high detection accuracies both in terms of the existence of clear physical signatures of FTEs (for the majority of cases) and in terms of the capability of the data mining technique to explore the data set in a much more thorough fashion than expert human eyes.

Conclusively determining the states of the solar wind will aid in tracing the origins of those states to the Sun, and in the process help to find the wind’s origin and acceleration mechanism(s). Prior studies have characterized the various states of the wind, making lists that are only partially based on objective criteria; different approaches obtain substantially different results. To uncover the unbiased states of the solar wind, we use “k-means clustering”—an unsupervised machine learning method—including constructed multipoint variables. The method allows exploration of different descriptive state variables and numbers of fundamental states (clusters). We show that the clusters reveal structures similar to those found by more ad hoc means, including coronal hole wind, interplanetary coronal mass ejections, “slow wind” (better: noncoronal hole flow), “pseudostreamers,” and stream interaction regions, but with differences that should be useful in refining our previous ideas. These results demonstrate the viability of the approach and warrant further study to understand the origin of remaining discrepancies. Complexity in k-means characterization of the wind may ultimately point to complexity at the source; studies closer to the Sun with Parker Solar Probe will help. We confirm the utility of a set of variables that can serve as a proxy for composition measurements. This proxy permits studies at high time resolution and where composition is not available. This and our recently developed unsupervised multivariate clustering technique are expected to be beneficial in the automated identification of structures and events in a variety of studies.

No abstract

[1] A novel data mining method called MineTool is introduced which, by virtue of automating the modeling process and model evaluations, makes it more accessible to nonexperts. The technique aggregates the various stages of model building into a four-step process consisting of (1) data segmentation and sampling, (2) variable preselection and transform generation, (3) predictive model estimation and validation, and (4) final model testing. Optimal strategies are chosen for each modeling step. However, the modular design of the MineTool enables the substitution of alternative strategies in any of the four modeling steps. A notable feature of the technique is that the final model is always in closed analytical form rather than ''black box'' form of most other techniques. MineTool can be used for analysis of data (e.g., time series) as well as images. The utility of the technique is illustrated through several examples based on synthetic data. Application of the technique to analysis of spacecraft data will be presented in subsequent papers.