evandroct/mdpi-nrt-forest-monitoring

How evandroct/mdpi-nrt-forest-monitoring works

This repository reproduces selective-logging monitoring from source-data preparation through model training, inference, temporal filtering, and evaluation. The documented main route starts with the Zenodo corpora, pretrains an ST-MAE, fits leave-one-monitoring-unit-out soft-label regressors, then evaluates detection filters and their outputs. Separate scripts rebuild the upstream corpora from DETER-derived ROIs and HLS imagery. The graph retains README-documented evaluation branches; unsampled scripts are represented by their exact paths, with relationships included only where the README or shown source supports them.

Main entry points: build_dense_pred_variant.py Dense predictions · model.py ST-MAE model · train_classifier.py Regressor training · train.py Pretraining

Components 33

Corpus rebuilding

ROI candidates 01_rois/build_roi_candidates.py
Eligibility scan 02_download/scan_eligibility.py
HLS acquisition 02_download/hls_stac_download.py
GeoZarr assembly 03_cubes/build_geozarr_datacubes.py
Shared eligibility common/eligibility.py
Record download 00_data/download_record.py

Foundation model

Pretraining dataset 04_pretraining/dataset.py
ST-MAE model 04_pretraining/model.py
Pretraining 04_pretraining/train.py
Training curriculum 04_pretraining/recipes/train_extend.py

Downstream regression

Crop caches 06_regressor/build_caches.py
Soft-label dataset common/dataset_classifier.py
Classifier head common/classifier_head.py
Regressor training 06_regressor/train_classifier.py
LOO-CV orchestration 06_regressor/loocv_chain.py
OOF aggregation 06_regressor/agg_oof.py

Prediction products

Raster inference 05_inference/predict_geotiffs.py
Index VRTs 05_inference/make_indices_vrts.py

Detection evaluation

Dense predictions 07_detector/build_dense_pred_variant.py
Kernel sweep 07_detector/sweep_kernels.py
Nested evaluation 07_detector/nested_sweep.py
Nested results 07_detector/make_nested_tables.py
Negative test 07_detector/negative_test/evaluate_targets.py
Target selection 07_detector/negative_test/select_targets.py
Target scoring 07_detector/negative_test/score_targets.py
DETER survey 07_detector/deter_alerts/deter_survey.py
Alert tables 07_detector/deter_alerts/make_deter_tables.py
MU numbering 07_detector/paper_mu_numbers.py
Kernel summaries 07_detector/paper_mu_numbers.py

Other components

Pipeline operator
Zenodo record
DETER alerts
HLS imagery

Connections 45

Pipeline operator → Record download: fetch record README.md
Zenodo record → Record download: provides archives README.md
Pipeline operator → ROI candidates: rebuild corpora README.md
DETER alerts → ROI candidates: derives ROIs README.md
ROI candidates → Eligibility scan: screens candidates README.md
Eligibility scan → Shared eligibility: uses rules 02_download/scan_eligibility.py
Pipeline operator → HLS acquisition: acquire imagery README.md
HLS imagery → HLS acquisition: supplies scenes README.md
HLS acquisition → GeoZarr assembly: assembles imagery README.md
Pipeline operator → Pretraining: train foundation README.md
Pretraining dataset → Shared eligibility: uses rules 04_pretraining/dataset.py
Pretraining → Pretraining dataset: loads samples 04_pretraining/train.py
Pretraining → ST-MAE model: trains model 04_pretraining/train.py
Training curriculum → Pretraining: launches stages README.md
Pipeline operator → LOO-CV orchestration: run cross-validation README.md
LOO-CV orchestration → Regressor training: runs folds README.md
Pipeline operator → Crop caches: build caches README.md
Crop caches → Soft-label dataset: prepares variants 06_regressor/build_caches.py
Regressor training → Soft-label dataset: loads labels 06_regressor/train_classifier.py
Regressor training → ST-MAE model: loads trunk 06_regressor/train_classifier.py
Regressor training → Classifier head: fits head 06_regressor/train_classifier.py
Pipeline operator → OOF aggregation: aggregate metrics README.md
Pipeline operator → Raster inference: run inference README.md
Raster inference → ST-MAE model: loads checkpoint model 05_inference/predict_geotiffs.py
Pipeline operator → Index VRTs: create views 05_inference/make_indices_vrts.py
Index VRTs → Raster inference: reads prediction rasters 05_inference/make_indices_vrts.py
Pipeline operator → Dense predictions: build dense output README.md
Dense predictions → ST-MAE model: uses trunk 07_detector/build_dense_pred_variant.py
Dense predictions → Classifier head: uses classifier 07_detector/build_dense_pred_variant.py
Dense predictions → Soft-label dataset: uses dataset logic 07_detector/build_dense_pred_variant.py
Pipeline operator → Kernel sweep: search filters README.md
Kernel sweep → Dense predictions: reads predictions 07_detector/sweep_kernels.py
Pipeline operator → Nested evaluation: evaluate nested folds README.md
Nested evaluation → Kernel sweep: uses kernels 07_detector/nested_sweep.py
Pipeline operator → Nested results: summarize nested results README.md
Nested results → MU numbering: imports numbering 07_detector/make_nested_tables.py
Pipeline operator → Target selection: select test targets README.md
Pipeline operator → Negative test: evaluate negatives README.md
Negative test → Kernel sweep: uses filter kernels 07_detector/negative_test/evaluate_targets.py
Target scoring → ST-MAE model: runs trunk 07_detector/negative_test/score_targets.py
Target scoring → Classifier head: scores with head 07_detector/negative_test/score_targets.py
Pipeline operator → DETER survey: survey alert periods README.md
Pipeline operator → Alert tables: build alert tables README.md
Alert tables → MU numbering: imports numbering 07_detector/deter_alerts/make_deter_tables.py
Pipeline operator → Kernel summaries: report kernel metrics README.md

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·