Adam
Zachry

~300km · Ionosphere

Hysell Lab · Cornell

Investigating whether sporadic-E layers in the E-region (~100km) are electrodynamically coupled to medium-scale travelling ionospheric disturbances in the F-region (~300km). Built a Madrigal line-of-sight TEC pipeline implementing Zhang et al. SGolay detrending on raw los_tec data, achieving a noise floor of 0.19 TECu.

• ▸foEs (total Es plasma conductance) correlates with MSTID activity at ρ = +0.62
• ▸Layer patchiness (dfo-b) does not correlate — diverging from Otsuka Japan result
• ▸Suggests conductance-driven rather than gradient-driven coupling at 43°N magnetic dip
• ▸Building RTI overlays from Hysell's Ithaca radar against TEC keograms

~10km · Troposphere

Independent

Predicted next-day PJM electricity price volatility and spike probability using GEFS 31-member ensemble spread as a proxy for forecast uncertainty. When meteorologists disagree about tomorrow's temperature, power traders should price that uncertainty.

• Shannon entropy of ensemble spread as primary volatility feature◂
• Forecast evolution signal — temperature forecast shift over 7 days pre-delivery◂
• Gas-power coupling interaction term for demand-supply dynamics◂
• Quantile regression for tail risk · walk-forward validation with HAC errors◂

~0km · Ocean Surface

Independent

Building a system to detect vessels engaging in deceptive shipping practices — AIS spoofing, transponder blackouts, and ship-to-ship transfers — using anomaly detection on historical AIS data. Dark fleet activity creates measurable signals in commodity flow intelligence.

• ▸AIS gap detection and trajectory reconstruction for vessels going dark
• ▸Spoofing fingerprinting via position/speed physics consistency checks
• ▸Flow aggregation to construct dark-fleet supply indices for LNG and crude

~0km · Ground

CU GeoData

Ground sensor network analysis on soil moisture data across four field sites. Built correlation pipeline with first-differencing and rolling precipitation windows. Next: LSTM model for soybean yield prediction — sequence modeling applied to agricultural remote sensing.

• Pearson correlation and first-differencing across CAMPS, GLITZ, GRASP, NINJA sensors◂
• GRASP identified as consistent anomaly across all analysis windows◂
• LSTM soybean yield model in development — next semester◂