MACHINE LEARNING TOOLS FOR PREDICTING SOLAR ENERGETIC PARTICLE HAZARDS

SOLAR ENERGETIC PARTICLES (SEPS) ARE AMONG THE MOST HAZARDOUS TRANSIENT PHENOMENA OF THE SOLAR ACTIVITY. THE PRIMARY OBJECTIVE OF THE PROPOSAL IS TO ENHANCE PREDICTIONS OF SEPS BY IMPLEMENTING AUTOMATIC DATA CHARACTERIZATION AND MACHINE-LEARNING TOOLS. THE PROPOSAL PURSUITS TWO MAIN GOALS: 1) DEVELOPMENT OF THE ONLINE-ACCESSIBLE AUTOMATICALLY-UPDATED DATABASE THAT INTEGRATES THE SOLAR AND HELIOSPHERIC DATA METADATA AND DESCRIPTORS RELATED TO SOLAR PROTON EVENTS (SPE) 2) DEVELOPMENT OF ROBUST ALL-CLEAR FORECASTS OF SPES WITH LOW FALSE-ALARM RATES TARGETED AT DIFFERENT TEMPORAL SCALES DIFFERENT ENERGY AND PARTICLE FLUX THRESHOLDS OF SPES AND ADAPTED TO OPERATIONAL AVAILABILITY OF DATA SOURCES AND GAPS IN THE DATA. WE PROPOSE TO DEVELOP DATABASES AND TOOLS FOR QUANTITATIVE CHARACTERIZATION OF AR MAGNETOGRAMS AND IMAGES ESTIMATION OF FLARE ERUPTION PROBABILITIES AND CHARACTERIZATION OF SEP EVENTS BASED ON MACHINE-LEARNING IMAGE ANALYSIS AND PREDICTION TECHNIQUES. THE PROPOSED DATABASE WILL INTEGRATE MULTI-SPACECRAFT DATA AND MODERN DATABASE TECHNOLOGIES WITH API-BASED ONLINE ACCESS TO DATABASE ENTRIES AND INTEGRATION PRODUCTS. IT WILL BE MADE PUBLICLY AVAILABLE AND SERVE A FOUNDATION FOR DEVELOPMENT OF MACHINE-LEARNING SPE FORECASTS. THE DATABASE WILL AUTOMATICALLY UTILIZE AND PROCESS INCOMING OBSERVATIONAL DATA DERIVE ADDITIONAL EVENTS DESCRIPTORS AND WILL AUTOMATICALLY UPDATE. THE TEAM WILL DEVELOP HIGH-FIDELITY ALL-CLEAR FORECASTS BY EMPLOYING ADVANCED MACHINE-LEARNING TECHNIQUES AND A CUSTOMIZED SKILL SCORE WHICH TO OPTIMIZE TOLERANCE FOR MISSED SPE EVENTS WITH RESPECT TO THE FALSE ALARM RATE. IN ADDITION THE ADVANCED MACHINE LEARNING TECHNIQUES (SUCH AS MULTI-LAYER PERCEPTRONS AND RECURRENT NEURAL NETWORKS) WILL BE EMPLOYED FOR SOLVING CLASSIFICATION TASKS AND PREDICTION OF ACTIVE REGION EVOLUTION. THE PROPOSED RESEARCH IS DIRECTLY RELATED TO THE ESI APPENDIX TOPIC 4 GOALS. IN PARTICULAR THE DEVELOPED DATABASE WILL PROVIDE AVAILABILITY AND READINESS OF THE PREPARED AND INTEGRATED DATA FOR THE HELIOPHYSICS COMMUNITY AS WELL AS INCLUDE AUTOMATIC DETECTION MEASUREMENT AND CHARACTERIZATION OF SPE-RELATED PRECURSOR FEATURES. CORRESPONDINGLY IT WILL OPEN NEW PERSPECTIVES FOR THE DATA DISCOVERY AS WELL AS INITIATE A DATA PREPARATION PHASE FOR FORECASTING ATTEMPTS. THE DEVELOPED ROBUST ALL-CLEAR FORECASTS WILL ENABLE POSSIBILITIES FOR SAFE SPACE EXPLORATION AND ADVANCED SPACE MISSION PLANNING WHICH IS ESPECIALLY IMPORTANT WITH RESPECT TO THE NASA PLAN MOON-BY-2024 . IN LONGER TERM PROSPECTIVE WE EXPECT THAT DEVELOPED ATTEMPTS WILL SIGNIFICANTLY CONTRIBUTE TO THE REPLACEMENT OF CURRENT SEP OPERATIONAL FORECASTING TECHNIQUES BY MACHINE LEARNING-BASED APPROACHES.