About

This folder contains the scripts to classify and post-process the Pampa Biome.

We recommend that you read the MapBiomas Appendix of the Algorithm Theoretical Basis Document (ATBD).

How to use

First, you need to copy these scripts (including those in utils folder) to your Google Earth Engine (GEE) account.

Pre-processing - preparing training data

• Step001A_stable_samples_coll41.js: build stable pixels from Colleciton 4.1 and save a new asset;
• Step001B_stable_samples_agriculture.js;
• Step001C_estable_samples_weetland_coll41.js;
• Step001C_merge_estable41_agriculture.js;
• Step001D_merge_estable41_referenceMap.js;
• Step002_proportion_classes_2000.js: calculate area proportion for each class to each region that will be used to generate training samples;
• Step003_export_stable_sample_geometry.js: export geometries (points) for training samples;
• Step004_export_anual_sample.js: export training samples for each year;

Classification - run the random forest classification

• Step005_class_run_Pampa_R01_col05.js: run the randonforest and export classification for region1;
• Step005_class_run_Pampa_R02_col05.js: run the randonforest and export classification for region2;
• Step005_class_run_Pampa_R03_col05.js: run the randonforest and export classification for region3;
• Step005_class_run_Pampa_R04_col05.js: run the randonforest and export classification for region4;
• Step005_class_run_Pampa_R05_col05.js: run the randonforest and export classification for region5;
• Step005_class_run_Pampa_R06_col05.js: run the randonforest and export classification for region6;
• Step005_class_run_Pampa_R07_col05.js: run the randonforest and export classification for region7;

Post-processing - apply filters

• Step006_Filter_01_gagfill.js: filter tho replace pixels classified as Non Observed;
• Step006_Filter_02_espatial.js: this filter uses a mask to change only those patches with pixels connected to five or less pixels of the same class;
• Step006_Filter_03_temporal.js: filter uses the information from the previous year and the year later to identify and correct a pixel misclassification;
• Step006_Filter_04_frequency_multiple.js: this filter were applied to use the temporal information available for each pixel to correct cases of false positives;
• Step006_Filter_05a_pre_incidence_forest.js: prepare data to run incident filter;
• Step006_Filter_05b_pre_incidence_others.js: prepare data to run incident filter;
• Step006_Filter_05c_incidence.js: this filter were applied to correct the classification of pixels considered with an excessive amount of changes along the 35 years;
• Step006_Filter_06_espatial_pos_inci.j: this filter uses a mask to change only those patches with pixels connected to five or less pixels of the same class, pos incidence filter corrections;
• Step006_Filter_07_temporal_pos_incidence.js: filter uses the information from the previous year and the year later to identify and correct a pixel misclassification, pos incidence filter corrections;
• Step006_Filter_09_frequency_water_R1.js: the fifth frequency filter corrected false positives of water in shaded relief covered with forest which appeared at region 1;
• Step006_Filter_10_frequency_rocky.js: tho resolve the confusion in rocky outcrop;
• Step006_Filter_11_frequency_wetland.js: to resolve the confusion in wetlands particularly with false positives of forest.