FROM-GLC (Finer Resolution Observation and Monitoring of Global Land Cover)



Global land cover data are key sources of information for understanding the complex interactions between human activities and global change. FROM-GLC (Finer Resolution Observation and Monitoring of Global Land Cover) is the first 30 m resolution global land cover maps produced using Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) data. Our long-term goal in FROM-GLC is to develop a multiple stage approach to mapping global land cover so that the results can better meet the needs of land process modeling and can be easily cross-walked to existing global land cover classification schemes.


FROM-GLC (Gong et al., 2013) was produced using 91433 training samples and 38664 test samples collected via human interpretation of TM/ETM+ images. The interpretation was carried out based primarily on a color composite of TM images of Bands 4, 3, and 2 displayed with the red, green and blue color guns respectively. In addition, the spectral curve based on the 6 optical bands of TM/ETM+, the MODIS time series during the whole year of 2010, and high resolution images and field photos found in Google Earth were used as reference.
Four sets of global land cover maps were produced based respectively on four types of supervised classifiers including the conventional maximum likelihood classifier (MLC), the J4.8 decision tree classifier, the random forests ensemble classifier (RF) and the support vector machine (SVM). The SVM produced the highest overall classification accuracy of approximately 64.9% that was assessed with a set of test samples independently collected. The random forests produced the second highest classification accuracy of 59.8%, with J4.8 and the MLC ranked the third to the fourth.

[2] FROM-GLC-seg
FROM-GLC-seg (Yu et al., 2013) is an improved version of FROM-GLC. A segmentation approach was used in FROM-GLC-seg to integrate multi-resolution datasets, including Landsat TM/ETM+ (30 meter), MODIS EVI time series (250 meter), Bioclimatic variables (1km) (Hijmans et al., 2005), global DEM (1km) (Hijmans et al., 2005), Soil-water variables (1km) (Zomer et al., 2007; 2008; Trabucco & Zomer, 2010). FROM-GLC-seg used the same training/test samples as FROM-GLC, and followed the same classification system with slight modification (The impervious land cover type was not mapped, due to severe spectral mixing effects and its small coverage. In addition, the clouds, which temporally exist on Landsat TM/ETM+, were removed as well). The Random Forest (RF) classifier was used and achieved better overall accuracy. Accuracies for vegetation land cover types (i.e. cropland, forest) and bareland were improved. However, mapping accuracies for water bodies, snow/ice land cover types are slightly lower because coarser resolution MODIS (250 meter) and Bioclimatic, DEM, Soil-Water variables (1km) are not ideal for recognizing small scale objects.

[3] FROM-GLC-agg
FROM-GLC-agg (Yu et al., 2014) is a further improvement by aggregating FROM-GLC and FROM-GLC-seg, together with two coarse resolution impervious maps, i.e. Nighttime Light Impervious Surface Area (Elvidge et al., 2007) and MODIS urban extent (Schneider et al., 2009; 2010). FROM-GLC-agg has an overall accuracy of 65.51%, which is significantly better than FROM-GLC (63.69%) and FROM-GLC-seg (64.42%). Accuracies for individual land cover types in FROM-GLC-agg have been increased or better balanced compared to FROM-GLC and FROM-GLC-seg.

FROM-GC (Yu et al., 2013) is a 30-m spatial resolution global cropland extent (with other land cover types) product developed with two 30-m global land cover maps (i.e. FROM-GLC, Finer Resolution Observation and Monitoring, Global Land Cover; FROM-GLC-agg) and a 250-m cropland probability map (Pittman et al., 2010). A common land cover validation sample database (Zhao et al., 2014) was used to determine optimal thresholds of cropland probability in different parts of the world to generate a cropland/noncropland mask according to the classification accuracies for cropland samples. A decision tree was then applied to combine two 250-m cropland masks: one existing mask from the literature and the other produced in this study, with the 30-m global land cover map FROM-GLC-agg. For the smallest difference with country-level cropland area in Food and Agriculture Organization Corporate Statistical (FAOSTAT) database, a final global cropland extent map was composited from the FROM-GLC, FROM-GLC-agg, and two masked cropland layers. From this map FROM-GC (Global Cropland), we estimated the global cropland areas to be 1533.83 million hectares (Mha) in 2010, which is 6.95 Mha (0.45%) less than the area reported by the Food and Agriculture Organization (FAO) of the United Nations for the year 2010. A country-by-country comparison between the map and the FAOSTAT data showed a linear relationship (FROM-GC = 1.05*FAOSTAT ?1.2 (Mha) with R2=?0.97). Africa, South America, Southeastern Asia, and Oceania are the regions with large discrepancies with the FAO survey.

[5] FROM-GLC-Hierarchy
FROM-GLC-Hierarchy (Yu et al., 2014) is land cover dataset with multi-resolution (i.e. 30 m, 250 m, 500 m, 1 km, 5 km, 10 km, 25 km, 50 km, 100 km) to meet requirements for different resolutions from different applications. The 30 m base map was improved from FROM-GLC-agg with additional coarse resolution datasets (i.e., MCD12Q1 (Friedl et al., 2010), GlobCover2009 (Bontemps et al., 2010) etc.) to reduce land cover type confusion. Around 1.1% pixels were replaced by coarse resolution products. Validation based assessments indicate the accuracy for land cover maps at 30 m, 250 m, 500 m, 1 km resolutions are 69.50%, 76.65%, 74.65%, and 73.47%, respectively. Further analysis of area-estimation biases for different land cover types at different resolutions suggests that maps at coarser than 5 km resolution contain at least 5% area estimation error for most land cover types. Proportion layers, which contain precise information on land cover percentage, are suggested for use when coarser resolution land cover data are required.