SoilGrids250m: Global gridded soil information based on machine learning

PLoS ONE, Journal Year: 2017, Volume and Issue: 12(2), P. e0169748 - e0169748

Published: Feb. 16, 2017

This paper describes the technical development and accuracy assessment of most recent improved version SoilGrids system at 250m resolution (June 2016 update). provides global predictions for standard numeric soil properties (organic carbon, bulk density, Cation Exchange Capacity (CEC), pH, texture fractions coarse fragments) seven depths (0, 5, 15, 30, 60, 100 200 cm), in addition to depth bedrock distribution classes based on World Reference Base (WRB) USDA classification systems (ca. 280 raster layers total). Predictions were ca. 150,000 profiles used training a stack 158 remote sensing-based covariates (primarily derived from MODIS land products, SRTM DEM derivatives, climatic images landform lithology maps), which fit an ensemble machine learning methods—random forest gradient boosting and/or multinomial logistic regression—as implemented R packages ranger, xgboost, nnet caret. The results 10–fold cross-validation show that models explain between 56% (coarse 83% (pH) variation with overall average 61%. Improvements relative considering amount explained, comparison previous 1 km spatial resolution, range 60 230%. can be attributed to: (1) use instead linear regression, (2) considerable investments preparing finer covariate (3) insertion additional profiles. Further could include refinement methods incorporate input uncertainties derivation posterior probability distributions (per pixel), further automation modeling so maps generated potentially hundreds variables. Another area future research is multiscale merging local national gridded products (e.g. up 50 m resolution) increasingly more accurate, complete consistent information produced. are available under Open Data License.

Language: Английский

---

The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019

Earth system science data, Journal Year: 2021, Volume and Issue: 13(8), P. 3907 - 3925

Published: Aug. 11, 2021

Abstract. Land cover (LC) determines the energy exchange, water and carbon cycle between Earth's spheres. Accurate LC information is a fundamental parameter for environment climate studies. Considering that in China has been altered dramatically with economic development past few decades, sequential fine-scale monitoring urgent need. However, currently, fine-resolution annual dataset produced by observational images generally unavailable due to lack of sufficient training samples computational capabilities. To deal this issue, we first Landsat-derived land (CLCD) on Google Earth Engine (GEE) platform, which contains 30 m its dynamics from 1990 2019. We collected combining stable extracted China's land-use/cover datasets (CLUDs) visually interpreted satellite time-series data, Maps. Using 335 709 Landsat GEE, several temporal metrics were constructed fed random forest classifier obtain classification results. then proposed post-processing method incorporating spatial–temporal filtering logical reasoning further improve consistency CLCD. Finally, overall accuracy CLCD reached 79.31 % based 5463 samples. A assessment 5131 third-party test showed outperforms MCD12Q1, ESACCI_LC, FROM_GLC GlobeLand30. Besides, intercompared thematic products, exhibited good consistencies Global Forest Change, Surface Water, three impervious surface products. Based CLCD, trends patterns changes during 1985 2019 revealed, such as expansion (+148.71 %) (+18.39 %), decrease cropland (−4.85 grassland (−3.29 increase (+4.34 %). In general, reflected rapid urbanization series ecological projects (e.g. Gain Green) revealed anthropogenic implications under condition change, signifying potential application global change research. The introduced article freely available at https://doi.org/10.5281/zenodo.4417810 (Yang Huang, 2021).

Language: Английский

---

SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty

SOIL, Journal Year: 2021, Volume and Issue: 7(1), P. 217 - 240

Published: June 14, 2021

Abstract. SoilGrids produces maps of soil properties for the entire globe at medium spatial resolution (250 m cell size) using state-of-the-art machine learning methods to generate necessary models. It takes as inputs observations from about 240 000 locations worldwide and over 400 global environmental covariates describing vegetation, terrain morphology, climate, geology hydrology. The aim this work was production properties, with cross-validation, hyper-parameter selection quantification spatially explicit uncertainty, implemented in version 2.0 product incorporating practices adapting them digital mapping legacy data. paper presents evaluation predictions produced organic carbon content, total nitrogen, coarse fragments, pH (water), cation exchange capacity, bulk density texture fractions six standard depths (up 200 cm). quantitative showed metrics line previous global, continental large-region studies. qualitative that coarse-scale patterns are well reproduced. uncertainty scale highlighted need more observations, especially high-latitude regions.

Language: Английский

---

Reconstructing Three Decades of Land Use and Land Cover Changes in Brazilian Biomes with Landsat Archive and Earth Engine

Remote Sensing, Journal Year: 2020, Volume and Issue: 12(17), P. 2735 - 2735

Published: Aug. 25, 2020

Brazil has a monitoring system to track annual forest conversion in the Amazon and most recently monitor Cerrado biome. However, there is still gap of land use cover (LULC) information all Brazilian biomes country. Existing countrywide efforts map lack regularly updates high spatial resolution time-series data better understand historical dynamics, subsequent impacts country biomes. In this study, we described novel approach results achieved by multi-disciplinary network called MapBiomas reconstruct between 1985 2017 for Brazil, based on random applied Landsat archive using Google Earth Engine. We mapped five major classes: forest, non-forest natural formation, farming, non-vegetated areas, water. These classes were broken into two sub-classification levels leading comprehensive detailed mapping at 30 m pixel resolution. The average overall accuracy time-series, stratified sample 75,000 locations, was 89% ranging from 73 95% 33 years LULC change series revealed that lost 71 Mha vegetation, mostly cattle ranching agriculture activities. Pasture expanded 46% 2017, 172%, replacing old pasture fields. also identified 86 converted native vegetation undergoing some level regrowth. Several applications dataset are underway, suggesting reconstructing maps useful advancing science guide social, economic environmental policy decision-making processes Brazil.

Language: Английский

---

Optical remotely sensed time series data for land cover classification: A review

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2016, Volume and Issue: 116, P. 55 - 72

Published: March 23, 2016

Accurate land cover information is required for science, monitoring, and reporting. Land changes naturally over time, as well a result of anthropogenic activities. Monitoring mapping change in consistent robust manner large areas made possible with Earth Observation (EO) data. products satisfying range science policy needs are currently produced periodically at different spatial temporal scales. The increased availability EO data—particularly from the Landsat archive (and soon to be augmented Sentinel-2 data)—coupled improved computing storage capacity novel image compositing approaches, have resulted annual, large-area, gap-free, surface reflectance data products. In turn, these support development annual that can both informed constrained by detection outputs. inclusion time series process provides on class stability informs logical transitions (both temporally categorically). this review, we present issues opportunities associated generating validating time-series products, identify methods suited incorporating other inputs characterization.

Language: Английский

---

Annual maps of global artificial impervious area (GAIA) between 1985 and 2018

Remote Sensing of Environment, Journal Year: 2019, Volume and Issue: 236, P. 111510 - 111510

Published: Nov. 19, 2019

Language: Английский

---

A new generation of the United States National Land Cover Database: Requirements, research priorities, design, and implementation strategies

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2018, Volume and Issue: 146, P. 108 - 123

Published: Sept. 12, 2018

Language: Английский

---

PEATMAP: Refining estimates of global peatland distribution based on a meta-analysis

CATENA, Journal Year: 2017, Volume and Issue: 160, P. 134 - 140

Published: Sept. 23, 2017

Language: Английский

---

GLC_FCS30: global land-cover product with fine classification system at 30 m using time-series Landsat imagery

Earth system science data, Journal Year: 2021, Volume and Issue: 13(6), P. 2753 - 2776

Published: June 15, 2021

Abstract. Over past decades, a lot of global land-cover products have been released; however, these still lack map with fine classification system and spatial resolution simultaneously. In this study, novel 30 m for the year 2015 (GLC_FCS30-2015) was produced by combining time series Landsat imagery high-quality training data from GSPECLib (Global Spatial Temporal Spectra Library) on Google Earth Engine computing platform. First, were developed applying rigorous filters to CCI_LC (Climate Change Initiative Global Land Cover) MCD43A4 NBAR (MODIS Nadir Bidirectional Reflectance Distribution Function-Adjusted Reflectance). Secondly, local adaptive random forest model built each 5∘×5∘ geographical tile using multi-temporal spectral texture features corresponding data, GLC_FCS30-2015 product containing types generated tile. Lastly, validated three different validation systems (containing details) 44 043 samples. The results indicated that achieved an overall accuracy 82.5 % kappa coefficient 0.784 level-0 (9 basic types), 71.4 0.686 UN-LCCS (United Nations Cover Classification System) level-1 (16 LCCS 68.7 0.662 level-2 (24 types). comparisons against other (CCI_LC, MCD12Q1, FROM_GLC, GlobeLand30) provides more details than CCI_LC-2015 MCD12Q1-2015 greater diversity FROM_GLC-2015 GlobeLand30-2010. They also showed best 59.1 GlobeLand30-2010 75.9 %. Therefore, it is concluded first dataset 16 as well 14 detailed regional types) high at m. in paper are free access https://doi.org/10.5281/zenodo.3986872 (Liu et al., 2020).

Language: Английский

---

Sentinel-2 Data for Land Cover/Use Mapping: A Review

Remote Sensing, Journal Year: 2020, Volume and Issue: 12(14), P. 2291 - 2291

Published: July 16, 2020

The advancement in satellite remote sensing technology has revolutionised the approaches to monitoring Earth’s surface. development of Copernicus Programme by European Space Agency (ESA) and Union (EU) contributed effective surface producing Sentinel-2 multispectral products. satellites are second constellation ESA Sentinel missions carry onboard scanners. primary objective mission is provide high resolution data for land cover/use monitoring, climate change disaster as well complementing other such Landsat. Since launch instruments 2015, there have been many studies on classification which use images. However, no review dedicated application monitoring. Therefore, this focuses two aspects: (1) assessing contribution classification, (2) exploring performance different applications (e.g., forest, urban area natural hazard monitoring). present shows that a positive impact specifically crop, forests, areas, water resources. contemporary adoption can be attributed higher spatial (10 m) than medium images, temporal 5 days availability red-edge bands with multiple applications. ability integrate remotely sensed data, part analysis, improves overall accuracy (OA) when working free access policy drives increasing especially developing countries where financial resources acquisition limited. literature also produces accuracies (>80%) machine-learning classifiers support vector machine (SVM) Random forest (RF). maximum likelihood analysis common. Although offers opportunities challenges include mismatching Landsat OLI-8 lack thermal bands, differences among Sentinel-2. show promise potential contribute significantly towards

Language: Английский

---