Technology · Pilot-ready
Manage The Climate Volatility
Hiding in Your Supply Chain
Year‑to‑year climate variability explains around one‑third of global yield swings, yet most of that risk is treated as unpredictable until harvest. Predictive Climate Intelligence (PCI) changes the timeline: screening hundreds of climate variables, isolating the few that predict outcomes months in advance, and validating every signal out-of-sample before it reaches your desk.
Materials
Anticipate climate-driven supply shifts for critical inputs before price signals reach the market.
Manufacturing
Forecast productivity and quality disruptions at key facilities before they impact fulfilment.
Retail
Model how climate and macro shifts affect category-level demand and inventory risk.
Materials Module · Case Study
Climate-Driven Cotton Yield Forecasting
How PCI identifies climate signals that predict cotton yield variation, and gives lead time to sourcing teams, insurers, and traders to act. Developed on 29 years of USDA data across three US states.
Lead time
Up to 9 months
Signals can emerge as early as January in some regions.
Directional accuracy
70–77%
Out-of-sample, predicting above/below yield anomaly.
Extreme-year accuracy
74–88%
Holdout testing on the most disruptive seasons.
Coverage
~60% of US crop
40 counties across GA, TX, CA.
Who this is for
Sourcing & procurement
PCI's county-level forecasts cover ~60% of US cotton production across Georgia, Texas, and California. Signals emerge 3–9 months before harvest, during the window when forward contracts, supplier allocations, and buffer stock levels are typically set. Each region surfaces different climate drivers, so a wet winter in California and a heat spike in Texas can be tracked independently.
Trading & hedging
PCI provides directional signals on regional yield anomalies before USDA's August crop production surveys, the market's primary benchmark, are released. Texas alone accounts for ~40% of US cotton and shows yield swings from −69% to +80% versus trend. PCI's walk-forward validated accuracy in Texas is 77%, based on 14 years of out-of-sample testing across 15 counties.
Insurance & risk
PCI's extreme-year detection reaches 74–88% across states, identifying the loss years that drive portfolio outcomes. County-level climate thresholds (e.g. 6°C dewpoint in Georgia, 35°C heat-hour threshold in Texas) quantify where yield risk accelerates non-linearly. These thresholds, tested against 29 years of USDA data, are ready-made candidates for parametric trigger design.
The question isn't whether climate affects yields. It's whether you can see it early enough to act. PCI isolates the climate variables, during specific growth phases, that predict whether a season will finish above or below trend. The output is a probability across three yield scenarios (below-normal, normal, above-normal) so you can think in risk buckets rather than point forecasts. PCI is designed to surface the drivers relevant to different regions.
Performance at a Glance
| Region | Counties | Period | Directional Accuracy | Extreme Year Accuracy | Top Signal |
|---|---|---|---|---|---|
| Georgia Southeast, humid | 18 | 1996–2024 | 70% | 74% | Dewpoint variability |
| Texas High Plains, semi-arid | 15 | 1996–2024 | 77% | 76% | Flowering heat stress |
| California San Joaquin, irrigated | 7 | 2000–2024 | 73% | 88% | Heat & rainfall stress |
How to read accuracy:
- Directional Accuracy measures how often the model correctly predicts whether yields finish above or below trend. A naive baseline (always predicting average yields) scores ~50%. PCI's 70–77% represents a consistent, validated edge.
- Extreme Year Accuracy is the same metric tested only on the most disruptive seasons (statistical outliers), held out from training. These are the years that matter most for risk management, and where PCI reaches 74–88%.
PCI vs USDA August Forecasts
USDA's August crop production surveys, the market's primary benchmark, achieve 55–64% directional accuracy across these regions using in-season field data. PCI achieves 70–77% using pre-season climate data alone, delivering signals 3 to 9 months before USDA's estimates are available. PCI is designed to be orthogonal to USDA: an independent, climate-anchored risk signal. When PCI and USDA agree, confidence is high. When they diverge, that's where early positioning and optionality pay off.
Scalable Methodology, Region-Specific Results
The same pipeline runs on every region: screen, validate, surface. Each geography naturally produces different climate drivers. Beyond US cotton, PCI has been validated on perennial fruit crops in Mediterranean climates with no structural changes to the methodology.
A Leading Signal: Georgia
This chart isolates one of the strongest climate-yield relationships for illustration. PCI identifies 6 predictors across 3 growth phases for Georgia; the three most actionable are shown in the timeline below.
What this means
When pre-planting dewpoint exceeds ~6°C, or dewpoint variability during boll development crosses 2.4°C, Georgia yields are more likely to fall below trend. The first signal is observable in January, 8–9 months before harvest.
What You Get From PCI: Georgia
From 400+ climate variables screened, 6 survive out-of-sample validation for Georgia. The three most actionable are shown below.
8–9 months before harvest
January: first signal
Pre-planting dewpoint
If dewpoints are running high, this is the earliest indication the season may underperform.
Procurement: review forward booking windows Insurance: flag for pre-season pricing review Trading: establish directional bias
Validated across 16 of 18 counties
3–4 months before harvest
June: confirmation
Flowering heat variability
Erratic temperatures during flowering are a reliable negative signal. Combined with the winter reading, confidence increases.
Procurement: adjust supplier mix and buffers Insurance: update portfolio exposure estimates Trading: refine hedge sizing
Validated across all 18 counties
1–2 months before harvest
August: final read
Boll-stage dewpoint variability
Beyond a dewpoint variability threshold, boll-fill yields deteriorate. At this point the forecast has maximum precision.
Procurement: finalise spot vs. contract allocation Insurance: refine loss estimates before settlement Trading: high-confidence directional call
Validated across all 18 counties
Climate Fingerprint: Good, Average, and Stress Years
The closer the current season tracks the red (stress) profile, the higher the risk. Green = historically good years. PCI monitors this in real time.
Reading the fingerprint
Good years: drier winters, stable flowering heat, moderate boll-stage dewpoint variability. Stress years: the opposite. PCI compares the current season against these archetypes in real time and alerts when conditions shift toward the red profile.
A Leading Signal: Texas
This chart isolates one of the strongest climate-yield relationships for illustration. PCI identifies 3 predictors across 2 growth phases for Texas; all three are shown in the timeline below.
What this means
The critical vulnerability window in Texas centres on flowering: erratic heat accumulation (hours above 32°C) is the strongest lead indicator, followed by extreme heat during boll fill. In 2011, Texas yields fell 62% below trend. ICE cotton futures had risen from ~80 to over 200 cents/lb. PCI's flowering heat signal would have flagged elevated risk by June, before the full damage was priced in.
What You Get From PCI: Texas
From 400+ variables screened, 3 survive validation for Texas, all heat-related. Yield swings of −69% to +80% from trend make directional accuracy here highly valuable.
3–4 months before harvest
June: heat stress signal
Flowering heat stress variability
Variability in heat hours above 32°C during flowering is the strongest leading indicator. Consistent warmth supports pollination; volatile spikes disrupt it.
Procurement: review forward commitments for Texas-origin cotton Insurance: update regional exposure assessment Trading: establish directional bias on ICE cotton
Strongest signal · R²=0.24
3–4 months before harvest
June: night temperature read
Flowering minimum temperature
When nights stay hot, plants can't recover from daytime heat stress. Combined with the heat hours signal, this differentiates moderate stress from severe damage.
Procurement: adjust buffer stock levels Insurance: refine loss probability estimates Trading: size positions with two confirming signals
Piecewise threshold at 31°C
1–2 months before harvest
August: the decisive read
Boll-stage extreme heat
When heat hours above 35°C cross the threshold during boll development, yield losses accelerate sharply. Highest R² of any feature.
Procurement: finalise spot allocation and supplier diversification Insurance: prepare for potential loss year; refine reserves Trading: high-confidence directional call, maximum precision
Validated across all 15 counties
Heat is everything
All three Texas drivers relate to heat stress across two growth phases. Flowering variability provides the early warning; boll-stage extreme heat delivers the decisive read. Three variables, 77% directional accuracy, in a region where yield swings routinely exceed ±50%.
Climate Fingerprint: Good, Average, and Stress Years
The closer the current season tracks the red (stress) profile, the higher the risk. Green = historically good years. PCI monitors this in real time.
Reading the fingerprint
Good years: stable flowering heat, mild nights, limited boll-stage extremes. The 2011 drought sits squarely in the stress zone on all three axes. PCI tracks the current season against these profiles and alerts on shifts toward the red pattern.
A Leading Signal: California
This chart isolates one of the strongest climate-yield relationships for illustration. PCI identifies 5 predictors across 4 growth phases for California; the three most actionable are shown in the timeline below.
What this means
Irrigation buffers routine variation in the San Joaquin Valley, but cannot offset waterlogged soils or the cascading effects of an abnormally wet winter. Pre-planting rainfall remains a strong negative signal, observable in January. Combined with mid-season heat stress signals, PCI achieves 88% accuracy on extreme seasons. For Pima cotton buyers (California ≈ 95% of US supply), this is the earliest available warning.
What You Get From PCI: California
From 400+ variables screened, 5 survive validation for California. The three most actionable are shown below, with a reliable signal as early as January.
8–9 months before harvest
January: earliest warning
Pre-planting rainfall
Each additional mm of winter rain correlates with lower yields across all 7 counties. A wet January is a red flag for the season ahead.
Procurement: review Pima and Upland forward commitments Insurance: flag CA portfolio for pre-season pricing review Trading: establish directional bias on CA-origin supply
Validated across all 7 counties
5–6 months before harvest
April–May: planting and vegetative read
Temperature & heat variability
When planting-period temperature variability crosses the 2.8°C threshold, yield risk increases non-linearly. Combined with the winter signal, confidence rises.
Procurement: adjust supplier mix if two signals confirm Insurance: update loss probability estimates Trading: refine position sizing with second signal
Non-linear threshold at 2.8°C
2–3 months before harvest
July: boll-stage confirmation
Boll-stage heat & pre-planting evaporation
In irrigated systems, extreme heat is the constraint water can't offset. Combined with earlier signals, this completes the picture, particularly for the stress years that reach 88% detection.
Procurement: finalise sourcing allocation with full signal set Insurance: refine reserves with three confirming signals Trading: high-confidence directional call for CA supply
Validated across all 7 counties
Climate Fingerprint: Good, Average, and Stress Years
The closer the current season tracks the red (stress) profile, the higher the risk. Green = historically good years. PCI monitors this in real time.
Reading the fingerprint
Good years: dry winters, stable planting temperatures, limited boll-stage heat. The 2017 season (record winter rainfall) sits at the stress extreme. A shift toward the red profile in January is an early trigger to reassess sourcing exposure.
Selective disclosure
PCI is an active R&D platform. This page shows representative signals, thresholds, and lead times.
Feature engineering, calibration logic, and full variable lists are reserved for pilot engagements.
How PCI works
Every signal shown above passed the same gauntlet: out-of-sample testing across multiple counties and years, with no manual tuning. Models are recalibrated each season as new data arrives. When PCI is deployed on a new commodity or geography, the pipeline runs identically; only the data changes.
Validation methodology and diagnostics
Every pilot delivers a full diagnostic package alongside the forecast: walk-forward R², directional accuracy, calibrated prediction intervals (PICP), stress recall, and false-alarm rates. Models are selected by walk-forward directional accuracy on an expanding window with no data leakage. Feature budgets, sign-agreement checks, and residual diagnostics are logged transparently so you can evaluate robustness before acting.
Next step
See PCI applied to your commodities and regions
In 8 to 10 weeks, we run the pipeline on your commodities or regions, validate against your data, and deliver monitoring-ready output you can test against the next season in real time.