Domain Deep Dive — airSpring

Michigan Crop Water Atlas (100 stations × 80 years), seasonal GPU pipeline, and the path to Penny Irrigation — sovereign compute on consumer hardware.

Data sources: experiment_catalog.json, benchmark_timing.json, composition_validation.json

Reproduce: cargo run --release --bin validate_atlas (1354/1354)

For other springs: This notebook covers the domain-specific “crown jewel” experiment. Replace with your flagship validation story. The frozen data pattern captures the result without requiring live hardware.

import json
from pathlib import Path

import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

RESULTS = Path('..') / 'experiments' / 'results'

def load(name):
    with open(RESULTS / name) as f:
        return json.load(f)

catalog = load('experiment_catalog.json')
bench = load('benchmark_timing.json')
comp = load('composition_validation.json')

atlas = next(e for e in catalog['key_experiments'] if e['id'] == 18)
pipeline = next(a for a in bench['algorithms'] if a['name'] == 'seasonal_pipeline')

print(f"Michigan Crop Water Atlas: {atlas['checks']:,} checks ({atlas['status']})")
print(f"Atlas R²: {atlas.get('note', 'N/A')}")
print(f"Seasonal pipeline speedup: {pipeline['speedup']}× (Python {pipeline['python_us']}µs → Rust {pipeline['rust_us']}µs)")
print(f"Atlas-scale throughput: {bench['atlas_scale']['throughput_et0_per_sec']:,} ET₀/s")

Seasonal Pipeline: ET₀ → Kc → Water Balance → Yield

The seasonal pipeline chains four stages: evapotranspiration (FAO-56 PM) → crop coefficient (dual Kc with cover crops) → water balance (FAO-56 Ch 8) → yield response (Stewart 1977). Each stage can run on CPU or GPU.

pipeline_methods = [
    ('ET₀ (FAO-56 PM)', 'fao56_pm_et0'),
    ('Dual Kc', 'dual_kc_daily'),
    ('Water Balance', 'water_balance_step'),
    ('Yield Response', 'yield_response_stewart'),
]

algos = {a['name']: a for a in bench['algorithms']}
stages = [algos[m] for _, m in pipeline_methods]
stage_names = [n for n, _ in pipeline_methods]

fig, ax = plt.subplots(figsize=(10, 5))
x = range(len(stage_names))
width = 0.25

py = [s['python_us'] for s in stages]
rs = [s['rust_us'] for s in stages]
gpu = [s['gpu_us'] for s in stages]

ax.bar([i - width for i in x], py, width, label='Python', color='#e74c3c', alpha=0.8)
ax.bar(list(x), rs, width, label='Rust CPU', color='#3498db', alpha=0.8)
ax.bar([i + width for i in x], gpu, width, label='GPU', color='#2ecc71', alpha=0.8)
ax.set_xticks(list(x))
ax.set_xticklabels(stage_names)
ax.set_ylabel('µs per call')
ax.set_title('Seasonal Pipeline: 4-Stage Timing (Python vs Rust vs GPU)')
ax.legend()
plt.tight_layout()
plt.savefig('/tmp/airspring_05_pipeline.png', dpi=150)
plt.show()

ET₀ Method Comparison (8 methods)

airSpring validates 8 evapotranspiration methods against peer-reviewed baselines.

et0_methods = [
    'fao56_pm_et0', 'hargreaves_et0', 'priestley_taylor',
    'thornthwaite', 'makkink_et0', 'turc_et0', 'hamon_et0', 'blaney_criddle_et0'
]
et0_data = [algos[m] for m in et0_methods]
et0_names = [a['name'].replace('_et0', '').replace('_', ' ').title() for a in et0_data]
et0_speedups = [a['speedup'] for a in et0_data]

fig, ax = plt.subplots(figsize=(10, 5))
bars = ax.bar(et0_names, et0_speedups, color='#2ecc71', edgecolor='white')
ax.set_ylabel('Speedup (×)')
ax.set_title('8 ET₀ Methods: Rust vs Python Speedup')
for bar, val in zip(bars, et0_speedups):
    ax.text(bar.get_x() + bar.get_width()/2, bar.get_height() + 0.1,
            f'{val}×', ha='center', fontsize=9)
plt.xticks(rotation=30, ha='right')
plt.tight_layout()
plt.savefig('/tmp/airspring_05_et0_methods.png', dpi=150)
plt.show()

Key Experiments

key_exps = catalog['key_experiments']
print(f"{'ID':>4s}  {'Status':8s}  {'Checks':>8s}  Name")
print('-' * 60)
for e in key_exps:
    note = f" ({e['note']})" if 'note' in e else ''
    print(f"{e['id']:4d}  {e['status']:8s}  {e['checks']:8,d}  {e['name']}{note}")

The Path to Penny Irrigation

Penny Irrigation is the Garden-level product vision: sovereign irrigation scheduling on consumer hardware ($600 GPU + $99 NPU). The pipeline:

Open-Meteo weather → FAO-56 PM ET₀ → Dual Kc (cover crops) →
    Water balance → Yield prediction → Scheduling recommendation

All stages validated through 87 experiments. GPU pipeline delivers 6.8M field-days/s on consumer hardware (RTX 4070 + AKD1000). The full NUCLEUS composition deploys via biomeOS from pre-built plasmidBin binaries.

Current state: Science validated (L2), primal composition readiness (L0→L1). Next: guideStone scaffold, then Tier 2 IPC validation against live NUCLEUS.

Summary

Provenance: airSpring v0.10.0 · MSU BAE (Dong lab) · primals.eco