Aurum
Updated Forecast·Hybrid · Prithvi + per-state routing
Hybrid Model · D.1 Gate PASS

Phase 2-D.1 · Hybrid Production Model

2024 Holdout — Out-of-Sample Predictions

Per-state routing across two regressors. Iowa & Colorado pull from a Ridge linear probe over the 1024-D Prithvi-EO-2.0-300M embedding (foundation-model linear probe + XGBoost). Missouri, Nebraska, Wisconsin use the engineered baseline. The hybrid was trained on 2013–2022, validated on 2023, and the 2024 yields shown below were never seen during training.

263

Counties forecast

5

States covered

170 bu/ac

5-state avg

18.65

EOS RMSE (val 2023)

+6.50%

Lift vs canonical Row A

2024

Holdout year

Routing rule: Row B (Prithvi-aware) for CO/IA · Row A (engineered) for MO/NE/WI

2024 Holdout — Hybrid Model State Forecasts

Per-state routing · Prithvi-aware (Row B) for CO, IA · engineered (Row A) elsewhere

Hybrid · Out-of-sample
IowaRow B

201bu/ac

Actual 2024: 211-10

Row A 202 · Row B 201

Top county: Woodbury, IA

220 bu/ac predicted

73 counties · 18 above actual

EOS RMSE — hybrid 18.15 · Row A 16.49

NebraskaRow A

180bu/ac

Actual 2024: 184-4

Row A 180 · Row B 181

Top county: Phelps, NE

215 bu/ac predicted

60 counties · 24 above actual

EOS RMSE — hybrid 14.78 · Row A 14.78

WisconsinRow A

159bu/ac

Actual 2024: 162-3

Row A 159 · Row B 158

Top county: Lafayette, WI

197 bu/ac predicted

59 counties · 26 above actual

EOS RMSE — hybrid 16.12 · Row A 16.12

MissouriRow A

161bu/ac

Actual 2024: 175-14

Row A 161 · Row B 158

Top county: Atchison, MO

195 bu/ac predicted

53 counties · 11 above actual

EOS RMSE — hybrid 24.64 · Row A 24.64

ColoradoRow B

148bu/ac

Actual 2024: 131+17

Row A 143 · Row B 148

Top county: Delta, CO

217 bu/ac predicted

18 counties · 13 above actual

EOS RMSE — hybrid 37.57 · Row A 32.6

Hybrid uses Row B (Ridge probe on 1024-D Prithvi + engineered) for CO, IA; Row A (engineered only) for the others. 2024 yields were withheld from training.

Model Comparison — Val 2023 EOS

RMSE in bu/acre (lower = better) · Hybrid is the production model

Phase C · (2005–2022, baseline)19.95 bu/ac

canonical reference

Row A · (engineered, 2013–2022)19.01 bu/ac

engineered only

Row B · (Ridge probe + XGB)19.42 bu/ac

Prithvi-aware

HYBRID · (per-state routing)18.65 bu/ac

production

Per-Forecast-Date Validation

Val 2023 RMSE per stage · n_train, n_val are county-level rows

StageHybridRow ARow Bn_trainn_val
Aug 117.8117.7818.113,069308
Sep 120.3120.9220.933,069308
Oct 120.7920.8220.503,069308
Final18.6519.0119.423,069308

Hybrid Predictions vs Actual — 2024

State-mean predicted vs USDA NASS final estimate · bu/acre

Iowa-10 bu/ac vs actual
201
Hybrid prediction (Row B)USDA 2024: 211
Nebraska-4 bu/ac vs actual
180
Hybrid prediction (Row A)USDA 2024: 184
Missouri-14 bu/ac vs actual
161
Hybrid prediction (Row A)USDA 2024: 175
Wisconsin-3 bu/ac vs actual
159
Hybrid prediction (Row A)USDA 2024: 162
Colorado+17 bu/ac vs actual
148
Hybrid prediction (Row B)USDA 2024: 131
Hybrid prediction
USDA 2024 Actual

2024 Holdout Predictions — County Level

263 counties

Hybrid model · 80% CI ≈ ± hybrid EOS RMSE · sorted by predicted yield · actuals shown when reported

#CountyStateHybridActualΔ80% CIRow A · Row B
1Woodbury, IAIowa220206+14202239216 · 220
2Marshall, IAIowa219222-3200237217 · 219
3Fremont, IAIowa218231-13199237211 · 218
4Delta, COColorado217188+29199236211 · 217
5Mesa, COColorado217202+15198236213 · 217
6Cherokee, IAIowa217207+10198236206 · 217
7Jones, IAIowa217233-16198235217 · 217
8Plymouth, IAIowa217196+21198235210 · 217
9Sac, IAIowa217229-12198236216 · 217
10Dubuque, IAIowa216222-6197234224 · 216
11Clinton, IAIowa215236-21196234216 · 215
12Phelps, NENebraska215199+16196233215 · 217

How the Hybrid Model Works

A frozen-feature-extractor pipeline with per-state routing

  1. CDL corn masks — annual binary corn rasters (5 states × 12 years, 2013–2024) sourced from USDA CropScape and reprojected to EPSG:5070 / 30 m. Annual masks avoid corn-soy rotation contamination.
  2. HLS chips on /mnt/d/ — 43,724 corn-richest 224×224 chips extracted from cap=48 cleanest cloud-cover granules per (state, year, phase). Stored on a 1.9 TB external SSD; ~1.4 TB cumulative download.
  3. Prithvi-EO-2.0-300M-TL frozen embedding — 1024-D mean-pooled feature vector per (county, year, forecast_date), 14,560 rows total.
  4. Ridge linear probe — α=10.0, fit on chip-bearing 2013–2022 rows. Compresses the 1024-D embedding into a single yield-prediction scalar that XGBoost consumes alongside engineered features. Every Prithvi dim contributes via the linear weights.
  5. Per-state routing — at predict time, route bystate_alpha: CO & IA → Row B (engineered + Ridge probe); MO, NE, WI → Row A (engineered only). Selected from the cap=48 ablation breakdown — Prithvi adds noise where corn fraction is uniform across years.
  6. Gate result — hybrid val 2023 EOS RMSE 18.65, vs canonical Row A 19.95 = +6.50% lift. Threshold +5%; PASS.
Aurum · Phase 2-D.1.e · Hybrid bundle at models/forecast_d1_hybrid/ · 2024 holdout · Generated by scripts/gen_hybrid_results.py