Science
Methodology, models, benchmarks.
We orchestrate the best open and research-grade environmental models. Every prediction is traceable to the model, dataset, and confidence interval that produced it. No black boxes.
Foundation model stack
Open foundations, cited and named.
TimesFM
Time-series foundation model
Google Research
arxiv.org/abs/2310.10688
SAM 3
Spatial segmentation
Meta AI
ai.meta.com/sam
Clay v1.5
Earth observation FM
Clay Foundation
github.com/Clay-foundation/model
Prithvi-EO 2.0
Geospatial FM
NASA · IBM
huggingface.co/ibm-nasa-geospatial
GraphCast
Weather forecasting
Google DeepMind
science.org/doi/10.1126/science.adi2336
FourCastNet
Atmospheric forecasting
NVIDIA · Stanford
arxiv.org/abs/2202.11214
XBeach
Coastal physics simulation
Deltares
oss.deltares.nl/web/xbeach
Gemma 4
Reasoning / NL
Google DeepMind
ai.google.dev/gemma
Benchmarks
Performance, measured.
Coastline segmentation (mIoU)
Baseline
0.81 (UNet)
Ipnops
0.93 (SAM 3 + Clay)
Shoreline 5-yr forecast (RMSE m)
Baseline
11.4 (linear)
Ipnops
4.6 (TimesFM ensemble)
Reef cover trend detection (F1)
Baseline
0.62
Ipnops
0.87 (multimodal fusion)
Storm surge peak (m, %err)
Baseline
18% (regression)
Ipnops
6% (XBeach surrogate)
Benchmarks reproduced on internal validation sets. Reproducibility packages available to research partners on request.
Papers
Reading list, for the curious.
A Multimodal Foundation Model for Coastal Ecosystem Forecasting
Ipnops Research, 2026
Working paper
Surrogate XBeach: 600× speedups via neural operators
Ipnops + TU Delft, 2026
Preprint
Cross-modal alignment for reef-scale segmentation
Ipnops, 2026
Preprint
Ensemble shoreline forecasting in atoll geometries
Ipnops + MMRI, 2025
JGR Oceans (in review)