Biosphere
Canopy vigour relative to what this time of year normally looks like. A negative NDVI anomaly marks vegetation under stress before it is visible from the ground.
Inputs in-distribution, members in agreement, interval tight — the Trust Core stands behind this number.
Stress basis: NDVI anomaly vs the same composite in prior years: +0.077 NDVI versus the 8-year seasonal mean of 0.722.
Ensemble spread
Six heterogeneous members — a GRU, an MLP, a ridge autoregression, Holt smoothing, a seasonal baseline and persistence — each vote on the next value. Their weighted scatter relative to the recent one-step error is the disagreement signal; wide scatter means the members have genuinely split.
Trust signals
Spread of members vs expected one-step error.
Standardised distance of today's 8-feature input row from the training set. Furthest feature: Clear-pixel fraction (-2.63σ).
The combined trust the Core assigns this output.
What drove this prediction
Input×Gradient attribution: the gradient of the neural members' forecast with respect to every input, multiplied by the input, summed over the window — so each named driver carries a sign and a magnitude. The AR member's coefficients fold into the target's own history.
Yesterday → today difference
Mean temperature over the 16-day composite
The series' own recent values (autoregressive memory)
Day-of-year phase
Share of the 9×9 MODIS window passing QA
Mean shortwave radiation over the composite
Day-of-year phase
Rain over the 16-day composite period
This stream, audited
2-step path
The ensemble is rolled forward recursively; each horizon carries its own conformal quantile from the calibration window, scaled by the adaptive factor, so the band widens honestly with lead time. Drivers are held at their last value and only the season advances.
Rolling backtest, last 90 steps
Every point is a real one-step forecast made from the data available at the time, against what the feed then reported. Misses outside the band are marked. Adaptive conformal inference nudges α each step toward the 90 % target — currently α=0.100 (band ×1.00 vs static).
Model card
trained 2026-10-02A heterogeneous ensemble: two neural networks trained by backpropagation on the archive, two classical forecasters, and two baselines every member must beat. Weights are set by validation error, never by the test window.
| Member | Kind | Weight | Test MAE* | Vote |
|---|---|---|---|---|
| GRU recurrent net1,217 params | neural · recurrent | 43% | 0.021 | 0.790 |
| MLP feed-forward net809 params | neural · feed-forward | 27% | 0.020 | 0.831 |
| AR(4) ridge | statistical | 10% | 0.033 | 0.831 |
| Holt smoothing | statistical | 1% | 0.085 | 0.828 |
| Seasonal naive | baseline | 15% | 0.054 | 0.731 |
| Persistence | baseline | 4% | 0.043 | 0.837 |
* held-out test window, native units
Canopy context
The latest 16-day composite against its own seasonal climatology, with the QA fraction that says how much of the window was cloud-free.
Provenance
Governed quantity: Next 16-day composite corridor-mean NDVI.
archive pulled 2026-10-02 10:02 UTC · live tail merged at request time · all sources publicly available · nothing on this page is simulated