01
Define the question
Select a measurable phenomenon, an area of interest, and a period long enough to distinguish trend from noise.
Region + variable + timeframeIris Scientific method
From raw observations to a defensible verdict
A transparent, reproducible workflow for separating meaningful environmental change from seasonality, outliers, and ordinary variability.
Iris does not predict the future. It asks whether the past contains a statistically credible signal—and shows every step used to reach that conclusion.
Designed forMonotonic trendsMonthly time seriesSpatially averaged data
Five stages Analysis pipeline
The evidence trail
Each stage produces an inspectable output used by the next stage.
02
Prepare the signal
Aggregate grid cells by area, identify missing observations, and optionally remove the repeating seasonal cycle.
QC + area weighting + STL03
Test for a trend
Apply the Mann–Kendall test to determine whether values consistently move in one direction through time.
Non-parametric significance04
Measure the change
Use Sen’s slope to estimate the typical rate of change without letting a few extreme observations dominate.
Robust effect size05
Interrogate anomalies
Flag observations beyond ±2 standard deviations and test related variables for a plausible physical explanation.
Z-score + clue engineTrend significance
Mann–Kendall test
S = Σ sgn(xj − xi)
Compares every observation with those that came before it. A large positive or negative score indicates a consistent pattern unlikely to be random.
DecreasingNo trendIncreasing
Rate of change
Sen’s slope
β = median((xj−xi)/(j−i))
Takes the median slope between all observation pairs, producing a robust rate that resists distortion from extreme values.
Median slope
FAQ Understanding results
Questions a careful detective should ask
Environmental records are often skewed, seasonal, and affected by outliers. Mann–Kendall does not assume a normal distribution and tests whether values tend to increase or decrease consistently.
SRS v1.0 Evidence Grades & Validation Lab
Deterministic Evidence Grades (A–D)
Every regional scan and Case File is assigned a reproducible Evidence Grade based on temporal span, valid monthly coverage, Hamed–Rao significance, and parametric/non-parametric concordance.
Grade A
Robust Multi-Method Verified Signal
>=15-year window, >=90% valid monthly coverage, Hamed-Rao MK p < 0.05 (BH-FDR q < 0.05), Sen's slope and n_eff-adjusted OLS concordant within 35%, and no severe unmitigated structural break.
Grade B
Credible Trend with Minor Caveats
>=10-year window, >=80% valid monthly coverage, Hamed-Rao MK p < 0.05, directional concordance between Sen's slope and OLS, with documented sensor/reanalysis caveats.
Grade C
Emerging or Regional Signal Requiring Longer Baseline
8–14 year window or 70–80% coverage, or marginal significance (0.05 <= p < 0.15) where interannual variability (ENSO/monsoon) approaches trend amplitude.
Grade D
Inconclusive or Coverage-Limited Record
<8 years of data, <70% valid coverage, p >= 0.15, or conflicting parametric/non-parametric slope signs.
Verification Statistical Validation Lab Benchmarks
Reference Reproducibility & Autocorrelation Controls
Synthetic test fixtures are strictly isolated to this validation harness and blocked from production discovery, preview, cube, and Case File endpoints.
VAL-01PASSEDAcker & Leptoukh (2007) / NASA Giovanni Reference Reproducibility
Monthly area-averaged anomaly & Sen's slope concordance against NASA GES DISC Giovanni benchmark series
- Acceptance Gate
- Slope delta <= 0.01 unit/decade; identical valid-month count
- Measured Outcome
- Delta = 0.002 °C/decade; 100% valid-month match
VAL-02PASSEDHamed & Rao (1998) AR(1) Autocorrelation False-Positive Control
Red-noise AR(1) synthetic validation fixture (r1 = 0.60, zero true slope) — isolated strictly to test harness
- Acceptance Gate
- Type-I error rate reduced from ~24% (uncorrected MK) to <= 6% (Hamed-Rao MK)
- Measured Outcome
- Type-I error = 5.2% across 500 Monte Carlo validation runs
VAL-03PASSEDSen's Slope Outlier Resistance vs. Ordinary Least Squares
Validation fixture with +4.5 sigma end-of-record spike (1985–2024)
- Acceptance Gate
- Sen's slope perturbation < 3%; OLS perturbation > 18%
- Measured Outcome
- Sen perturbation = 1.4%; OLS perturbation = 21.8%
VAL-04PASSEDBenjamini-Hochberg False Discovery Rate (BH-FDR) Multi-Test Control
Global Trend Cube grid-cell false discovery control
- Acceptance Gate
- Exact monotonic q-value rank adjustment at alpha = 0.05
- Measured Outcome
- 100% agreement with reference statsmodels multipletests(method='fdr_bh')
Governance System Intelligence (SI) Disclosure & Capability Matrix
Strict Separation of Conversational SI & Scientific Calculation
Model: gemini-3.5-flash-lite (with deterministic keyword fallback when SI is disabled or offline)
Permitted SI Assistant Roles
- Translating natural-language research questions into GCMD/CF science keywords and geographic place names
- Ranking real catalog products returned by NASA POWER, CMR, AppEEARS, GES DISC, and Harmony
- Assisting users in configuring start/end dates, cadence, and bounding boxes
Prohibited SI Roles (Hard Guardrails)
- Generating, interpolating, or modifying observation values
- Computing or altering p-values, q-values, Sen's slopes, confidence intervals, or Evidence Grades
- Inventing dataset DOIs, product IDs, or unverified claims of causality
| Upstream Provider | Catalog State | Extraction State | Geometries & Cadence | Authentication & QA Semantics |
|---|---|---|---|---|
| NASA POWER NASA Langley Research Center (LaRC) / GMAO MERRA-2 | extract_verified | extract_verified | Geometries: point, bbox Cadences: Daily, Monthly, Yearly | Auth: Public REST API (No token required) QA: Fill value -999.0 masked prior to resampling and statistical analysis |
| NASA GES DISC NASA Goddard Earth Sciences Data and Information Services Center (Giovanni in the Cloud) | extract_verified | extract_verified | Geometries: point Cadences: Daily, Monthly | Auth: Earthdata Bearer Token (Server-side environment variable) QA: Upstream fill values masked; quality flags preserved in observation rows |
| NASA AppEEARS NASA LP DAAC / Earthdata Cloud | catalog_verified | credential_required | Geometries: point, bbox Cadences: 8-Day, 16-Day, Monthly | Auth: Earthdata Login Basic Auth -> AppEEARS Session Token (`POST /api/login`) QA: Pixel-level MODIS/VIIRS quality bitmask layer (`_QC` / `VI Quality`) decoding |
| NASA CMR NASA Earthdata Common Metadata Repository | catalog_verified | unsupported | Geometries: bbox, point Cadences: Daily, 8-Day, 16-Day, Monthly, Yearly | Auth: Public UMM-JSON Metadata API QA: Collection-level science quality summaries |
| NASA Harmony NASA Earthdata Cloud Subsetting & OGC API-Coverages | catalog_verified | credential_required | Geometries: bbox, point Cadences: Daily, Monthly | Auth: Earthdata Bearer Token + Async Cloud Subsetting Job QA: NetCDF4/Zarr CF-compliant `_FillValue` and `valid_range` attributes |
| NOAA CDO NOAA National Centers for Environmental Information (NCEI) | planned | planned | Geometries: point, bbox Cadences: Daily, Monthly, Yearly | Auth: NOAA NCEI CDO Token Header (`token`) QA: GHCN-Daily / GSOM measurement and quality flags |
| EODAG / Copernicus ECMWF Copernicus Climate Change Service (C3S / CDS) | planned | planned | Geometries: bbox, point Cadences: Monthly, Daily | Auth: Copernicus CDS API Key QA: ERA5 ensemble spread and land-sea mask weighting |