THIS PROJECT ADDRESSES SCIENCE CHALLENGES FROM THE NASA EARTH SURFACE AND INTERIOR (ESI) WORKSHOP WHAT ARE THE DYNAMICS OF THE EARTH S MAGNETIC FIELD AND ITS INTERACTIONS WITH THE REST OF THE EARTH S SYSTEM? . THE EARTH S MAGNETIC FIELD IS PRIMARILY GENERATED BY THE GEODYNAMO A DYNAMIC PROCESS INVOLVING CONVECTIVE MOTION IN THE ELECTRICALLY-CONDUCTING FLUID OUTER CORE. THE STRUCTURE OF THE FIELD EVOLVES ON TIME SCALES THAT RANGE FROM DECADES TO MILLIONS OF YEARS. KNOWLEDGE OF THESE VARIATIONS COME FROM AMONG OTHER SOURCES THE PALEOMAGNETIC RECORD TERRESTRIAL OBSERVATORIES AND SATELLITE BASED OBSERVATIONS. DATA FROM SUCH SOURCES HAS BEEN COMBINED WITH NUMERICAL GEODYNAMO MODELS DEVELOPED AT NASA GSFC IN EFFORTS TO PRODUCE IMPROVED FORECASTS OF THE MAGNETIC FIELD. THIS PROCESS OF COMBINING DATA WITH NUMERICAL MODELS IS KNOWN AS DATA ASSIMILATION (DA). PRODUCING THE BEST POSSIBLE FORECAST OF THE GEOMAGNETIC FIELD ON DECADAL TIME SCALES REQUIRES EFFECTIVE ASSIMILATION OF FIELD OBSERVATIONS. WHILE DA HAS GAINED SOME ATTENTION IN THE GEOMAGNETISM COMMUNITY SEVERAL FUNDAMENTAL QUESTIONS IN GEOMAGNETIC DA REMAIN UNEXPLORED. MY RESEARCH HAS SO FAR FOCUSED ON ANSWERING TWO OF THESE QUESTIONS. 1. LOCALIZATION IN GEOMAGNETIC DA: LOCALIZATION IS A NUMERICAL TECHNIQUE THAT IS ESSENTIAL FOR MAKING HIGHDIMENSIONAL DA PRACTICAL. LOCALIZATION TYPICALLY MAKES USE OF THE FACT THAT OBSERVATIONS HAVE ONLY A LOCAL EFFECT OBSERVATIONS WITHIN A PART OF THE SPATIAL DOMAIN ONLY AFFECT STATE ESTIMATES NEAR THE OBSERVATION. IN GEOMAGNETIC DA GLOBAL OBSERVATIONS COME FROM FIELD MODELS IN THE FORM OF SPHERICAL HARMONIC COEFFICIENTS MAKING THE VALUE OF TRADITIONAL APPROACHES TO LOCALIZATION UNKNOWN. MY GOAL IS TO INVESTIGATE HOW TO EFFECTIVELY LOCALIZE GEOMAGNETIC DA FOR DECADAL-SCALE FORECASTS. 2. INFLATION AND BIAS CORRECTION IN GEOMAGNETIC DA: OPERATIONAL NUMERICAL GEODYNAMO MODELS OPERATE IN A PARAMETER REGIME THAT IS UNLIKE THE EARTH. ADDITIONALLY THE EXACT RELATION BETWEEN MODEL TIME AND GEOPHYSICAL TIME IS NOT FULLY UNDERSTOOD LEADING TO AN INCORRECT TIMING OF OBSERVATIONS. SUCH SYSTEMATIC MODEL ERROR MUST BE ACCOUNTED FOR DURING DA. NUMERICAL TECHNIQUES FOR ACCOUNTING FOR SYSTEMATIC MODEL ERROR INCLUDE (COVARIANCE) INFLATION AND BIAS-CORRECTION SCHEMES. MY GOAL IS TO INVESTIGATE HOW TO EFFECTIVELY ACCOUNT FOR SYSTEMATIC MODEL ERROR IN GEOMAGNETIC DA FOR DECADAL-SCALE FORECASTS. I ACKNOWLEDGE THAT LOCALIZATION INFLATION AND BIAS-CORRECTION ARE WELL-UNDERSTOOD IN NUMERICAL WEATHER PREDICTION (NWP). IN FACT THE NWP COMMUNITY AGREES THAT DA AT AN OPERATIONAL SCALE IS IMPOSSIBLE WITHOUT LOCALIZATION INFLATION AND BIAS-CORRECTION. ATMOSPHERIC MODELS AND OBSERVATIONS HOWEVER BEHAVE VERY DIFFERENTLY FROM GEODYNAMO MODELS. EFFECTIVELY USING LOCALIZATION INFLATION AND BIAS-CORRECTION IN GEOMAGNETIC DA REQUIRES A DEEP UNDERSTANDING OF HOW THESE TECHNIQUES FUNCTION ON GEODYNAMO MODELS. TO THE BEST OF MY KNOWLEDGE SUCH ISSUES ARE THUS FAR ENTIRELY UNEXPLORED. I PLAN TO ANSWER THE ABOVE RESEARCH QUESTIONS USING A SUITE OF RIGOROUS NUMERICAL TESTS TO EXPLORE THE EFFECTIVE USE OF LOCALIZATION INFLATION AND BIAS-CORRECTION. THESE TESTS WILL BE CARRIED OUT ON A REDUCED-SCALE MODEL. THE REASONS ARE THAT (I) NUMERICAL MODELS WHICH PRODUCE HIGH-RESOLUTION REPRESENTATIONS OF THE GEODYNAMO WITH AN EARTH-LIKE MAGNETIC FIELD (SUCH AS NASA S MOSST-DAS) ARE COMPUTATIONALLY EXPENSIVE AND (II) SYSTEMATIC DA EXPERIMENTS REQUIRE REPEATEDLY PERFORMING DA USING DIFFERENT LOCALIZATION INFLATION AND BIAS-CORRECTION WHICH DRIVES THE COMPUTATIONAL REQUIREMENTS BEYOND WHAT IS FEASIBLE EVEN ON LARGE SUPER COMPUTERS. NATURALLY THE REDUCED-SCALE MODEL SHOULD BE SELECTED VERY CAREFULLY AND DEVELOPING TESTING AND ULTIMATELY SELECTING A USEFUL REDUCED-SCALE MODEL IS A SECOND IMPORTANT ASPECT OF MY PROJECT.