Chapter 9: Results — airSpring

9.1 Validation Summary

airSpring validates BarraCuda and the ecoPrimals infrastructure against precision agriculture: evapotranspiration modeling (FAO-56 Penman-Monteith, Priestley-Taylor, Hargreaves, Thornthwaite), soil moisture sensor calibration (Dong et al. 2020, 2024), and water balance scheduling. 3,123+ checks pass across four validation phases plus cross-validation, with negligible compute cost.

Table 9.1 — Phase Summary

9.2 FAO-56 Penman-Monteith Reproduction (64/64 checks)

Table 9.2 — FAO-56 Reference Case Validation

The FAO-56 Penman-Monteith equation is the international standard for evapotranspiration estimation. These are textbook-level checks: reproducing the exact numerical examples from Allen et al. (1998).

9.3 Soil Sensor Calibration (Dong et al. 2020) — 36/36 checks

Table 9.3 — Factory Calibration Reproduction

Acceptance criteria: MBE ≤ 0.02, RMSE ≤ 0.035, IA ≥ 0.8, R² ≥ 0.65.

Topp equation validation: 8 published points (ε = 3–40) reproduced to ±0.005 VWC.

9.4 IoT Irrigation (Dong et al. 2024) — 24/24 checks

9.5 Real Data Validation (Phase 0+) — 918 Station-Days

Table 9.4 — Station-Level Metrics (ET₀ vs Open-Meteo)

All 6 Michigan MAWN stations achieve R² > 0.96. The slight positive bias (+0.076 mm/day) is consistent with Open-Meteo’s higher-resolution radiation estimates compared to station-level sensors.

Michigan Crop Water Atlas: The pipeline now scales to 100 Michigan stations and 15,300 station-days. validate_atlas runs 1,393 checks (100 stations × 13 checks each) across ET₀, water balance, yield response, and mass conservation for 10 crops per station-year.

9.6 Cross-Validation: Python ↔ Rust (75/75 matches)

Table 9.5 — Rust Binary Validation (27 binaries, grouped by domain)

75 cross-validation pairs between Python and Rust implementations match within 1e-5 tolerance. The water balance mass conservation check achieves exact closure (0.0000 mm residual).

9.7 Water Balance Application

Table 9.6 — Smart Irrigation vs Naive Scheduling

ET₀-driven scheduling produces 53–72% water savings over calendar-based irrigation. This connects to Dong’s applied research program at MSU: the computational pipeline validated here is the same pipeline that informs real irrigation recommendations for Michigan growers.

9.8 Scholarly Reproduction Log

9.9 Connection to Constrained Evolution Thesis

airSpring demonstrates constraint-driven fitness at the simplest scale. The FAO-56 equation is deterministic — there is no room for creative solutions; the check either matches the textbook or doesn’t. The 75 Python ↔ Rust cross-validation matches prove that the Rust type system constraint does not sacrifice numerical precision. The water balance mass conservation (exactly 0.0000 mm residual) shows that Rust’s type system enforces physical conservation laws through type-theoretic guarantees that Python’s runtime does not.

The real-data validation (R² = 0.967 across 918 station-days) proves the pipeline works on messy, real-world data — not just textbook examples. This is the transition from in vitro to in vivo: the constrained system is fit for its deployed environment, not just for controlled conditions.

The scale of evolution is itself evidence for the thesis. airSpring grew from 326 checks to 3,123+ in approximately 12 days — a 10× expansion without relaxing constraints. The 100-station Michigan Crop Water Atlas proves the pipeline scales: 15,300 station-days, 1,393 atlas checks, all passing. The four-method ET₀ portfolio (Penman-Monteith, Priestley-Taylor, Hargreaves, Thornthwaite) demonstrates convergent fitness: four independent methods, validated on the same infrastructure, producing consistent results across the same real-world stations.

See also:

• Science: airSpring papers — the reproduced papers