Satellite imagery and derived remote-sensing products are no longer exotic inputs for commodity traders — by mid‑2026 they are core signals for early warnings on crop progress, yield deviations and export logistics. For traders focused on Brazilian soy and corn, satellite data can shorten the informational lag between field conditions and market moves. This guide walks traders through a practical, end‑to‑end approach to build, validate and trade on a satellite‑driven forecasting system for Brazil’s soy and corn exports.

Why satellites matter for Brazil soy & corn trading in 2026

  • Higher revisit rates and commercial constellations (Planet, Maxar) plus free Copernicus (Sentinel‑1/2) provide near‑daily coverage at operationally useful resolution.
  • SAR (Sentinel‑1) penetrates cloud cover during rainy seasons in central‑west Brazil — essential for the November–March critical windows.
  • Cloud platforms (Google Earth Engine, AWS Open Data) and off‑the‑shelf vegetation indices let traders move from raw pixels to indicators quickly.
  • Combining satellite indicators with logistics metadata (port throughput, river levels) translates crop stress into export flow risk — the tradeable outcome.

Step 1 — Define your trading objective and horizons

Be explicit about what you want to forecast and how that maps to instruments:

  • Objective examples: mid‑season yield deviation in Mato Grosso, delayed harvest in southern Brazil reducing November–January soy shipments, or inland congestion at Paranaguá affecting prompt export loadings.
  • Time horizon: intra‑season (days–weeks) for freight and basis trades; seasonal (1–6 months) for futures and options; multi‑year for procurement and offtake planning.
  • Choose instruments aligned to horizon: CBOT soy/corn futures and options for price exposure; FOB swaps and basis contracts for origin risk; freight derivatives or physical freight hedges for logistics exposure.

Step 2 — Select datasets and providers

Mix free and commercial sources to balance cost, latency and resolution:

  • Free, high‑value: Sentinel‑2 (optical, 10–20m), Sentinel‑1 (SAR, 10m, all‑weather), MODIS (daily coarse‑scale), Landsat (30m). Use Copernicus for baseline monitoring.
  • Commercial for higher cadence/resolution: PlanetScope (3–5m, daily), SkySat/Maxar (sub‑meter, tasking for field checks), BlackSky for intraday alerts.
  • Derived products: Google Earth Engine collections (NDVI/EVI composites), NASA SMAP (soil moisture proxy), ECMWF/ERA5 weather reanalysis for rainfall and evapotranspiration.
  • Ancillary data: Siloed ground truth — CONAB and IBGE crop reports, port throughput (Santos, Paranaguá, Rio Grande), customs loadings, and freight rates (BSI, Baltic) to tie signals to flows.

Step 3 — Build a processing pipeline

Turn imagery into daily operational indicators:

  1. Ingest: Use GEE or AWS for bulk access. Automate ingestion of Sentinel‑1/2 and Planet feeds (commercial API).
  2. Preprocess: Apply cloud masking for optical (Sentinel‑2) using QA bands; use SAR backscatter calibration for Sentinel‑1 to detect canopy and soil moisture signals.
  3. Mosaic & aggregate: Create field‑ or municipality‑level composites (7‑ and 14‑day rolling composites reduce noise and latency issues).
  4. Store: Keep time‑series at the farm/muni grid cell level (GeoTIFFs, vector polygons). Use a time‑series DB (PostGIS or specialized TSDB) for fast queries.

Step 4 — Derive indicators and models

Design both rule‑based indicators and machine‑learning models.

  • Vegetation indices: NDVI & EVI for greenness and biomass accumulation; compute phenological metrics — greenup date, peak NDVI, senescence timing.
  • SAR signals: Use VV/VH backscatter ratios and change detection to identify lodging, flooding, or tillage status when optical is obscured.
  • Soil‑moisture proxies: Combine SAR and SMAP/ERA5 to infer moisture stress during flowering/grainfill.
  • Phenology → yield models: Two parallel approaches work best — (a) statistical regression (NDVI/phenology percentiles to yield deviations using historical CONAB/IBGE harvests), (b) ML models (random forest or gradient boosting) that ingest multi‑source time series and weather data.
  • Logistics indicators: Monitor river gauge proxies (Sentinel, local gauges), port backup from AIS vessel clustering, and road congestion signals where available to infer shipment delays.

Step 5 — Calibrate with ground truth and reports

Calibration and validation are where many satellite projects fail. Use multiple ground anchors:

  • Historical match: Align satellite‑derived yield estimates with CONAB and IBGE harvests over 5–10 seasons to compute model error and bias by state (Mato Grosso, Paraná, Rio Grande do Sul, Mato Grosso do Sul).
  • Near‑real‑time anchors: Use field surveys, local agribusiness reports, and port loading manifests to validate signals during the season.
  • Bias correction: Apply municipality‑level bias corrections (if your model consistently under/over‑estimates in certain regions or soils).

Step 6 — Convert crop signals into export-flow and price implications

Mapping a yield deviation to a tradeable price signal requires these steps:

  1. Estimate production shock: Multiply satellite‑implied yield deviation (delta tonnes/ha) by sown area in affected municipalities to get a production delta.
  2. Derive exportable supply change: Adjust production delta for domestic demand, crush margins (for soymeal/oil), and stock movements to estimate change in exportable tonnes.
  3. Map to export windows and ports: Determine which harvest and shipping windows are affected; a localized shock in Mato Grosso has different shipping timing than southern harvests and therefore different futures months/basis impacts.
  4. Translate to price: Use a calibrated price‑impact function (historical regression of price moves vs. exportable supply shocks) or run scenario analyses across price elasticities. Convert the forecasted export shortfall/surplus into expected futures price delta and basis change ranges.

Step 7 — Choose instruments and execution strategy

Match signal nature to instruments and execution timelines:

  • Prompt basis trades and FOB swaps: Best for short, port‑level disruptions flagged by satellites + AIS port clustering.
  • Futures & options: Use futures for directional exposure and options for convex protection when uncertainty is high (e.g., late‑season drought signals with large model error).
  • Calendar spreads: Use when satellite signals indicate a timing shift in shipments (e.g., delayed southern harvest increases prompt premium over deferred months).
  • Freight: Combine signals with freight derivatives or forward charters when inland constraints imply longer vessel dwell times.

Step 8 — Risk management, position sizing and operational rules

Operationalize the model with explicit risk rules:

  • Signal confidence bands: Only scale into positions when model uncertainty (ensemble spread or historical RMSE) falls below a threshold.
  • Hedge cross‑risks: Correlation between soy and corn, crush margins and currency (BRL) exposure should be hedged or accounted for in sizing.
  • Liquidity checks: Ensure instrument liquidity (CBOT front months, FOB swaps) supports position size; use options for asymmetric risk if liquidity is limited.
  • Stops and exits: Define exit rules for both model invalidation (contrary ground reports) and profit targets.

Step 9 — Backtest and measure economic value

Backtesting must measure economic P&L, not just signal accuracy:

  • Walk‑forward testing: Re‑train models on rolling windows to simulate realistic deployment.
  • Include transaction costs, slippage and margin/collateral costs in the P&L simulation.
  • Performance metrics: Sharpe, hit rate, average drawdown, and information ratio against benchmarks (e.g., long CBOT soybean futures).
  • Stress tests: Run scenarios for extreme events (multiple municipalities underperforming, port closures) to understand tail exposure.

Step 10 — Operationalize alerts, monitoring and governance

Turn research into an operational workflow:

  • Automated alerts: Trigger signals when NDVI percentile breaches thresholds, SAR backscatter shows abrupt change, or port AIS clustering passes defined limits.
  • Human‑in‑the‑loop: Require trader sign‑off for large positions — provide dashboards with imagery, indicator time series, and model confidence.
  • Audit trail and compliance: Log data sources, model versions, and decision rationale for internal audit and regulatory compliance.
  • Cost control: Monitor data subscription vs. alpha produced — combine free Copernicus for baseline monitoring with occasional commercial tasking when confirmation is needed.

Worked (hypothetical) example — Turning an NDVI shock into a trade

Imagine mid‑season NDVI in a cluster of Mato Grosso municipalities falls to the 10th historical percentile during key grainfill. Your model estimates a 4% yield shortfall for those muni areas, equating to a 2% hit to national soy production in a hypothetical season. After adjusting for likely domestic crush demand and carryover stocks, you estimate a 1.2m tonne reduction in exportable soy between November–January.

Using a calibrated price‑impact rule (derived from your historical regression), the 1.2m tonne exportable shortfall implies an estimated 4–6% spot price uplift for prompt CBOT soy futures over the next 6 weeks. Execution cascade: 1) Buy prompt CBOT futures sized to risk budget; 2) Buy a small amount of call options to cap margin risk; 3) Hedge a portion with a short nearby physical FOB swap if you have counterparty access to lock premium. Monitor subsequent satellite passes and port AIS; if NDVI recovers or port throughput increases, scale down per exit rules.

Common pitfalls and how to avoid them

  • Overfitting: Avoid models that perform brilliantly in-sample but poorly out-of-sample. Favor simpler, interpretable features and ensemble approaches.
  • Ignoring logistics: Crop shocks do not equal immediate export shortfalls — incorporate port and transport constraints explicitly.
  • Latency mismatch: Planet daily resolution is powerful but costly; free Sentinel data may be delayed by mosaicking and processing. Match data latency to your trading horizon.
  • Signal confirmation bias: Always seek independent confirmation (ports, on‑the‑ground intel) before large position commits.

Checklist to deploy in 90 days

  1. Define target regions, harvest windows and instruments (week 1).
  2. Set up data feeds: Sentinel and GEE access (free) plus negotiate a 3‑month Planet trial or tasking credits (weeks 1–3).
  3. Build preprocessing pipeline and 14‑day composite workflow (weeks 2–6).
  4. Train baseline NDVI→yield regression using 5 past seasons (weeks 4–8).
  5. Backtest trading rules on 3 seasons and produce P&L simulation (weeks 6–10).
  6. Deploy alerting and live dashboard; run parallel paper trading for 30–60 days (weeks 10–12).

Final advice

Satellite‑driven forecasting is a force multiplier, not a silver bullet. The highest value comes from integrating imagery with logistics, market microstructure and rigorous risk controls. Start small, prove economic value in paper trading, and scale data spend to alpha. In Brazil’s fast-moving soy and corn complex, a disciplined satellite system can buy you crucial days of foresight — and those days can make the difference between a well‑timed trade and being late to the market.