TL;DR

Sentinel-2 wins for operational carbon accounting at sub-10m resolution with free, frequent revisit; Landsat's thermal bands and 50-year archive dominate historical baseline work and emissions validation. Most carbon credit PMs need both, but build on Sentinel-2 if your project measures change since 2015. The PM who doesn't understand when to trade spatial for spectral resolution will ship the wrong verification pipeline and watch their credits fail third-party audit.

Who This Is For

You are a product manager at a carbon accounting startup, a climate fintech, or an offset verification platform. You have shipped features before, but you have never had to defend a satellite data choice to a Verra validator, explain band selection to your data science lead, or scope procurement when "free" isn't actually free because engineering time dominates. You earn $165,000 to $240,000 base, report to a Head of Product or CTO, and your roadmap is squeezed between science team ambition and commercial team promises. This article replaces the generic "remote sensing for carbon" tutorial you found with the decision framework that survives a hiring committee debrief at Planet, NCX, or Pachama.

How do Sentinel-2 and Landsat actually differ for forest carbon measurement?

Sentinel-2 provides 10-meter resolution across four visible and near-infrared bands, with 20-meter shortwave infrared, and revisits every 5 days with both satellites. Landsat 8 and 9 deliver 30-meter resolution across nine bands including two thermal, with a 16-day revisit. The difference is not "better or worse" but "what question you are answering."

In a February 2023 debrief for a Series B carbon accounting platform, the hiring manager killed a senior PM candidate who kept describing Sentinel-2 as "higher resolution therefore better." The candidate missed that 30-meter Landsat thermal bands detect canopy stress and evapotranspiration signals that 10-meter Sentinel-2 cannot see at all. The debrief note: "Doesn't understand the measurement, not the image." The PM role is not to process pixels but to choose which physical signal validates your carbon claim.

Sentinel-2's 10-meter pixels resolve individual forest gaps, smallholder plots, and selective logging scars. For REDD+ projects in mixed-use landscapes—where a 30-meter Landsat pixel averages forest, cropland, and settlement—you need Sentinel-2 to avoid material misclassification. But Sentinel-2 lacks thermal bands, so you cannot directly estimate evapotranspiration or surface energy balance. Landsat's thermal archive stretches to 1982, enabling pre-project baseline construction that predates Sentinel-2 launch in 2015. The counter-intuitive truth: older data with coarser resolution often carries more regulatory weight because it establishes historical condition precedent that auditors accept.

The revisit cadence difference is fivefold—Sentinel-2's 5 days versus Landsat's 16—but cloud cover in tropical forest regions means actual usable observations converge closer than raw orbit math suggests. The PM who builds a pipeline assuming clear-sky availability without cloud probability masking will discover their annual monitoring report has three valid observations for a 50,000-hectare concession.

What resolution and spectral bands matter for carbon stock estimation, not just pretty maps?

Carbon stock estimation requires aboveground biomass (AGB) proxy measurement, and different bands correlate with different biomass components. Near-infrared (NIR) and shortwave infrared (SWIR) bands penetrate canopy to interact with leaf and wood structure. Thermal bands estimate water stress that modulates growth. The PM's job is matching band availability to allometric model requirements.

In a 2022 hiring committee at a vertical ag forestry startup, the debated candidate had built a beautiful Sentinel-2 NDVI dashboard. When pressed, they could not explain why their biomass model used NDVI—a measure of greenness and photosynthetic activity—rather than a SWIR-based vegetation index that correlates with woody volume. NDVI saturates in dense canopy; SWIR continues scaling with biomass. The candidate's product measured "forest health" for investors, not carbon stock for verification. They were passed over for a PM who had shipped a Landsat SWIR + backscatter fusion model with ESA CCI biomass calibration data.

The first counter-intuitive truth is that more bands do not mean better models if your ground truth is sparse. A model with 13 Sentinel-2 bands and 20 field plots will overfit and fail cross-validation. A model with 4 Landsat bands and 200 field plots will generalize. The PM who specs data volume without ground truth density matching will ship a model that validates beautifully on training data and collapses at audit.

The second counter-intuitive truth: spatial resolution trades with temporal stability. Sentinel-2's 10-meter pixels shift slightly between acquisitions due to parallax and topographic effects. Landsat's 30-meter pixels are more geometrically stable over decades. For long-term carbon stock trend analysis, coarser consistent pixels beat finer inconsistent ones. The product decision is not "which satellite" but "which uncertainty budget can we defend to Gold Standard."

When does revisit frequency actually change your product architecture?

Sentinel-2's 5-day revisit enables near-real-time deforestation alerts and rapid event detection. Landsat's 16-day revisit forces batch-oriented, state-based monitoring. The architectural choice ripples through data pipeline design, alert latency SLAs, and customer communication.

I sat in a Q3 2023 debrief where a PM candidate described building a "live deforestation monitoring system" on Landsat 8. The hiring manager—a former NASA JPL scientist turned VP Product—asked how live something could be with 16-day gaps. The candidate pivored to "we composite multiple acquisitions," which defeated the "live" claim. The deeper failure: the candidate had not understood that their customer promise (immediate alert) and their data choice (16-day revisit) were structurally incompatible. They were rejected before the committee even reached compensation discussion.

The product architecture implication is stark. Sentinel-2 enables streaming pipelines: each new acquisition triggers change detection, issues alerts within 24-48 hours of downlink. Landsat forces periodic batch jobs: composite 30-day or annual periods, detect change against previous period. The PM who promises "real-time" with Landsat will either fail technically or ship a product that alerts on month-old change and loses customers to competitors with Sentinel-2 pipelines.

But the counter-intuitive third truth: faster revisit creates more false positives. A single cloud-contaminated pixel in one Sentinel-2 acquisition triggers a change alert that a 16-day Landsat composite would suppress. The PM who optimizes for detection speed without false positive rate management will drown their operations team in verification tickets. I've seen this kill a $2.3M ARR contract when a major corporate buyer received 400 spurious alerts in a single month and invoked their data quality clause.

How do you cost and source imagery when "free" data isn't free engineering?

Both Sentinel-2 and Landsat are free to download. The PM who presents this as procurement savings misses the engineering cost iceberg. Processing, storage, atmospheric correction, cloud masking, and harmonization dominate total cost of ownership.

In a 2021 offer negotiation I advised on, a PM had accepted a role at 15% below market because the company emphasized "we use free public data." Eighteen months later, their AWS bill for Sentinel-2 processing exceeded $180,000 monthly, and their three-person data engineering team spent 70% of time on data preparation rather than model development. The "free" data had consumed a $2.1M annual burn. The PM left in the subsequent restructuring.

The real cost structure: Landsat's longer archive means larger total data volume for historical analysis, but simpler processing pipelines due to fewer bands and coarser resolution. Sentinel-2's higher resolution means more pixels per scene, more storage, more compute for identical geographic coverage. A single Sentinel-2 tile is approximately 100MB; a Landsat scene is 1-2GB but covers 9.6x the area at coarser resolution. The naive cost comparison—"Sentinel-2 smaller files therefore cheaper"—collapses when you need 9-16 Sentinel-2 tiles to cover one Landsat footprint.

Google Earth Engine changes this calculus by externalizing storage and providing pre-processed collections. But Earth Engine's computational model has quota limits, and its Python API introduces latency that real-time pipelines cannot tolerate. The PM who architects around "free" without modeling egress costs, API rate limits, and alternative platform pricing (AWS Open Data, Microsoft Planetary Computer, Sinergise Sentinel Hub) will underbudget by 3-5x.

The specific numbers that matter for carbon accounting PMs: Earth Engine computation quotas are 70 concurrent tasks and 33,000 requests per 100 seconds. Sentinel Hub commercial processing for full archive access runs approximately €500-2000 monthly for project-scale coverage. AWS storage for a continental Sentinel-2 archive exceeds $50,000 annually before egress. These are not abstract concerns; they appear in PM interviews when the engineering director asks "how would you scale this to Indonesia?"

What does an auditor actually accept for satellite-derived carbon claims?

Verra VCS, Gold Standard, and ART TREES each specify data requirements differently, and the PM who treats "satellite data" as a checkbox will face validation failure. The regulatory signal is not "use satellites" but "demonstrate measurement uncertainty below threshold with traceable data lineage."

In a hiring committee for a carbon MRV (Measurement, Reporting, Verification) platform, the winning candidate described how they had maintained a data provenance log that tracked every satellite scene from download through processing to final emission factor calculation, with DOI references, processing version hashes, and cloud cover thresholds. The losing candidate described "using Sentinel for monitoring." The committee's judgment: one understood the audit as product constraint; the other treated it as external imposition.

The specific requirements: Verra VCS Methodology VM0033 requires demonstration that remote sensing data are "appropriately calibrated and validated." Gold Standard's A/R requirements specify minimum 30-meter resolution for forest cover change, which technically permits Landsat but challenges Sentinel-2's advantage. ART TREES, used by jurisdictional REDD+ programs, requires integration with national forest monitoring systems that often predate Sentinel-2 and thus rely on Landsat continuity.

The PM's judgment moment: when your science team wants to switch from Landsat to Sentinel-2 for improved change detection, but your Gold Standard validation is built on a 2018 Landsat baseline, do you absorb the re-baseline cost and risk, or maintain legacy data for compliance while building parallel? There is no universal answer, but the PM who cannot frame this trade-off in audit-specific terms will defer to science or commercial teams and ship the wrong priority.

Preparation Checklist

  • Map your carbon accounting use case to required physical measurement: deforestation detection, biomass estimation, or degradation monitoring, then match to band and resolution requirements
  • Audit your ground truth density against model complexity; if ratio falls below 10:1 (plots to variables), simplify before adding spectral bands
  • Model total cost of ownership for 3-year horizon including storage, compute, personnel, and platform fees, not just data acquisition cost
  • Build a data provenance architecture that captures scene ID, processing version, atmospheric correction method, and cloud mask threshold for every output
  • Work through a structured preparation system for carbon accounting PM interviews; the PM Interview Playbook covers satellite data product cases with real debrief examples where candidates failed on technical depth
  • Script the specific satellite data choice justification you would present to a Verra validator, including uncertainty quantification
  • Benchmark your pipeline latency and false positive rate against competitor offerings; know whether your "real-time" claim survives technical due diligence

Mistakes to Avoid

BAD: "We use Sentinel-2 because it has better resolution."

GOOD: "We use Sentinel-2 at 10m for disturbance detection in fragmented landscapes, with Landsat thermal for historical baseline back to 2000, because our validation against 847 field plots showed SWIR-based indices reduced AGB uncertainty by 12%."

BAD: "The data is free so our COGS stays low."

GOOD: "Our imagery COGS is $14,000 monthly at current scale, driven by Earth Engine compute and AWS storage for Indonesia-wide coverage, with a path to $8,000 through scene subsetting and compressed manifest formats."

BAD: "Our science team handles the satellite selection."

GOOD: "I led the trade-off analysis between Sentinel-2 revisit frequency and Landsat archive depth, presented three options to the commercial team with audit risk scoring, and we committed to dual-source with Sentinel-2 primary for 2024 forward and Landsat for pre-2015 baseline per Gold Standard requirements."

FAQ

Will Sentinel-2 replace Landsat for carbon accounting?

No. Sentinel-2's temporal and spatial advantages are real, but its 2015 launch date and lack of thermal bands create irreplaceable gaps that Landsat fills. The PM who bets on single-satellite strategy will face baseline construction or emissions validation failures that dual-source architecture prevents.

How do I answer "which satellite would you choose" in a PM interview?

With a structured trade-off, not a single answer. State the measurement requirement first, then match to sensor characteristics, then discuss uncertainty and cost. The candidate who asks clarifying questions about project vintage, geographic scale, and audit standard before answering demonstrates product judgment; the candidate who answers immediately demonstrates shallow pattern matching.

What salary range validates this level of technical product decision-making?

Carbon accounting PMs with satellite data pipeline experience command $175,000 to $225,000 base in Series A-C climate tech, with $240,000+ at public remote sensing companies and top-tier offsets platforms. Equity varies dramatically: 0.15-0.35% at Series B, lower at later stages. The premium over generic climate PM is 20-35%, reflecting scarce overlap between product craft and remote sensing fluency.

The 0→1 PM Interview Playbook (2026 Edition) — view on Amazon →