Open Earth intelligence for everyone

Detect environmental trendsbefore they become headlines.

Turn 40+ years of NASA, NOAA, and ECMWF Earth observations into transparent, statistically defensible case files—without hiding the science.

Start investigating Explore workflows

Explore upcoming conditions

See what the historical record can tell us next.

Explore local machine-learning forecasts, prediction ranges, and model evaluations built from historical weather and Earth observation records at six configured benchmark points worldwide. Each forecast includes its full source data and measured baseline skill.

Explore a forecast

app.iris.earth

Case IR–042Temperature trend investigation

Analysis complete

Detective verdictSignificant warming trend detected

Surface temperature has risen by +0.42°C per decade.

p = 0.0014

Decadal rate+0.42°CHamed–Rao n_eff184MK p-value0.001495% CI0.31–0.53

Temperature over timeDeseasonalized (STL period 12)

Next clueInvestigate soil moisture

ConfidenceStatistically significant (Grade A)

Air temperature (MERRA-2)Soil moisture (SMAP / GLDAS)Vegetation index (MODIS NDVI)Precipitation (GPM IMERG)Land surface temp (MODIS LST)Snow cover fractionSolar radiation (CERES)Air temperature (MERRA-2)Soil moisture (SMAP / GLDAS)

Three investigative pathways

Choose the workflow that matches
your scientific question.

Whether you need live ad-hoc data discovery, planetary gridded trends, or history-only machine learning outlooks, Iris provides an auditable path.

Track A · Live Ad-Hoc Investigation

User-Directed Discovery from Any Question

Start with any plain-language inquiry or custom Earth observation product. Iris scopes parameters, verifies spatial-temporal coverage, and executes the full statistical pipeline on the fly.

Start an ad-hoc investigation

Frame with Super Intelligence

Describe an inquiry in natural language; AI parses domain, region, and temporal horizons.

Gap-Preserving QC

Inspect observational completeness; missing satellite values are excluded, never fabricated.

Deterministic Case File

Execute Hamed–Rao Mann–Kendall and Theil–Sen slope tests to assign cryptographic Evidence Grades.

Defensible methodology

Every clue you need.
One evidence trail.

Iris brings data discovery, robust statistics, visual analysis, and multi-level interpretation into one connected, auditable scientific workflow.

Non-Parametric Significance

Mann–Kendall rank tests avoid fragile Gaussian normality assumptions and handle non-linear environmental shifts without distortion.

p < 0.05Hypothesis rejection threshold

Autocorrelation Defense

Hamed–Rao (1998) variance inflation across lags 1–24 prevents false trend detections caused by persistent weather spells.

n_effEffective sample size correction

Outlier-Resistant Rates

Theil–Sen estimator computes median pairwise slopes resistant to up to 29% extreme anomalies, cloud dropouts, or sensor spikes.

95% CIRank-based confidence bounds

Cryptographic Evidence Grades

Objective A–D rubric evaluating completeness (≥90%), test concordance, and audit trails. 0% synthetic data in production.

Grade A–DDeterministic audit grading

Transparent by design

From raw environmental signal
to irrefutable evidence.

Iris never generates a mysterious confidence score. Every conclusion exposes its data provenance, transformation choices, and mathematical assumptions.

Data Ingestion & Provider Routing

Stream raw observation records directly from NASA POWER, CMR, AppEEARS, and ECMWF catalogs without alterations.

Signal Preparation & Gap Preservation

Apply robust STL seasonal decomposition while isolating missing intervals—never fabricating synthetic fillers.

Autocorrelation-Adjusted Testing

Compute serial correlation coefficients across 24 lags and adjust variance to eliminate false trend claims.

Evidence Synthesis & Case File

Calculate Theil–Sen decadal slopes, evaluate sensor caveats, and compile an immutable SHA-256 evidence package.

Near-surface temperature (MERRA-2)+0.42°C / decade · p = 0.0014

Observed signal Theil–Sen slope Flagged anomaly (|z| ≥ 2)

Validated case files

Significant investigations
grounded in real data.

Explore three hallmark environmental case files built with multi-decadal observational records and strict non-parametric verification.

Los Angeles County, USA

Urban Heat Island & Warming Trajectory

MERRA-2 near-surface temperature analysis reveals sustained multi-decadal warming with robust statistical significance across all 24 autocorrelation lags.

+0.42°Cper decade · p = 0.0014 (Hamed–Rao)

Central Valley, USA

Decadal Agricultural Precipitation Deficit

GPM IMERG and GLDAS soil moisture records uncover severe long-term precipitation declines with effective sample size correction removing seasonality artifacts.

−18.7mmper decade · n_eff = 114 (Corrected)

Amazon Basin, Brazil

Tropical Forest Canopy Stress & Resilience

MODIS vegetation index tracking demonstrates canopy moisture loss in deforested corridors, corroborated by optical sensor quality flags and high completeness.

−0.08NDVI per decade · Grade A Evidence

Contributors & Architecture

Team
and research contributors.

The multidisciplinary team behind Iris combining climate science, non-parametric statistics, geospatial engineering, and machine learning.

Navida Jayaweera

Backend Development, Team Lead, Research

FastAPI microservices and data pipelines, backend architecture leadership, and Earth observation research.

Oshadha Shiro

Frontend Dev, Forecasting, Documentation

Next.js 15 user interface, interactive predictive forecasting interfaces, and end-to-end platform documentation.

Vimukthi Prashan

ML Dev, Forecasting, Data Visualization

Empirical machine learning forecasting models (iris.ml-forecast.v2), quantile regressors, and 3D visual landscapes.

Dinodh Dassanayake

Data Analyzing, Research, Documentation

Multi-decadal satellite data analysis, Earth observation literature research, and scientific methodology documentation.

Shenon Kaushika

Data Analyzing, Research, Documentation

Observational time-series data analysis, environmental anomaly synthesis, and benchmark case documentation.

Nadula Nisith

Slept, Rage Bating

High-friction stress testing, unconventional peer review through debate, and spontaneous morale disruption.

Frequently asked questions

Got questions?
We have answers.

Everything you need to know about Iris’s statistical methodology, data providers, and AI governance.

Why use Mann–Kendall instead of ordinary linear regression?

Environmental and satellite time series frequently violate the core assumptions of Ordinary Least Squares (OLS) regression: residuals are rarely normally distributed, sensor noise creates extreme outliers, and weather anomalies introduce spikes. The Mann–Kendall test evaluates relative ranks rather than raw values, making it strictly monotonic, outlier-resistant, and free of Gaussian distribution assumptions.

How does Iris eliminate false alarms caused by serial autocorrelation?

What do the Evidence Grades (Grade A through Grade D) mean?

How does the Super Intelligence AI assistant work without hallucinating?

Your next investigation starts here

Follow the signal.
Find the story in the data.

Select any region, choose an environmental variable, and build a defensible evidence trail in minutes.

Open Iris