Tutorials overview

From a first simulation to a full course manual: sensor convolution, sensitivity analysis, and trait inversion, in R and Python. Every tutorial linked below is verified by actually running it end-to-end — not just written. This page groups all 31 tutorials by what they're about, with a real generated figure from each area, as a five-minute visual map before you dive into any one of them.

Tip: use your browser's Print → Save as PDF to keep an offline copy of this page.

ToolsRTMLeaf, canopy, soil, atmosphere models

Five leaf models (PROSPECT-D, PROSPECT-PRO, Fluspect-B, Fluspect-B-Cx, Liberty) × three canopy models (fourSAIL, foursail2, INFORM) × two soil models (BSM, MARMIT) × SPART for the full soil–canopy–atmosphere chain to top-of-atmosphere. All 15 leaf×canopy combinations agree with their Python port to R²=1.0 (see R vs Python).

All 15 leaf × canopy combinations, one example spectrum each (Tutorial 04)

SPART: top-of-canopy vs. top-of-atmosphere reflectance (Tutorial 03)

MARMIT: soil reflectance across a moisture gradient (Tutorial 16)

SCOPEinR / scopeinpythonSCOPE, in R and Python

The full coupled model: radiative transfer + energy balance + photosynthesis + chlorophyll fluorescence, in one simulation. Ported line-for-line to Python (scopeinpython) and cross-checked against the R original.

TOC reflectance components and the fluorescence spectrum (How-in-R/Python — SCOPEinR)

SIF emission spectrum, PSI/PSII contributions (Tutorial 04)

Full SCOPE 2.1: R vs. Python agreement, 600 simulations (R vs Python)

Hybrid inversionMachine-learning inversion: spectra → traits

Going backwards: from a simulated or real reflectance spectrum to the vegetation trait that produced it. LUT-trained Random Forest, SVM, PLSR and ensemble models, applied to both synthetic and real Sentinel-2 data.

Cab retrieved from Sentinel-2A bands, Random Forest (How-in-R/Python)

LAI retrieved from SCOPE apparent reflectance (How-in-R/Python — SCOPEinR)

Net photosynthesis mapped from real Sentinel-2, Speulderbos (Tutorial 11)

ToolsRTM Tutorial 13, R + PythonDeep learning for RTM inversion

Dense networks and a 1D-CNN reading a PRISMA-resolution spectrum directly (TensorFlow/Keras, R and Python both), compared head-to-head against Random Forest from Tutorial 12 -- and a real, documented bug hunt (a train/predict scaling mismatch) that took both architectures from looking broken to genuinely competing with, and sometimes beating, RF.

Dense vs. RF on 9 well-chosen features — Cab (top) and LAI (bottom), where dense edges out RF (R, Tutorial 13)

1D-CNN on 234 PRISMA bands, true vs. predicted Cab (R, Tutorial 13)

Same 1D-CNN architecture, Python (toolsrtm.deep_learning)

Global & local sensitivityWhich traits actually move the signal

One-at-a-time sweeps, Sobol variance-based indices, and Johnson relative-weights regression — three complementary ways to answer "which input trait matters, and where in the spectrum." Ported to Python (toolsrtm.sensitivity) and verified against R to 8 decimal places.

Canopy reflectance sensitivity to Cab and EWT (How-in-R/Python)

Which trait dominates at each wavelength — Cab (VIS), LAI (red-edge/NIR), EWT (SWIR) (Tutorial 10)

Reflectance sensitivity to Vcmax25 — direct vs. indirect effect (SCOPEinR Tutorial 07)

Real Sentinel-2 & FLEXSatellite & spatial applications

Same models, real pixels: pulling actual Sentinel-2 scenes via STAC, inverting trait maps pixel-by-pixel, and scaling up to FLEX's coarser footprint with sub-pixel heterogeneity. This is where the LUTs, ML inversion, and sensitivity results above turn into an actual map.

REIP1 index and retrieved Cab, real Sentinel-2 scene (Tutorial 18)

Cab, LAI, Car trait maps at 10m, FLEX ESU (Tutorial 19)

NDVI from a real Sentinel-2 scene, Speulderbos (SCOPEinR Tutorial 11)

This is a curated visual subset — the full numbered list (31 tutorials) is on the homepage, and every package's complete article index is under its own Documentation (ToolsRTM / SCOPEinR).