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

SOIL, Год журнала: 2021, Номер 7(1), С. 217 - 240

Опубликована: Июнь 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.

Язык: Английский

---

Machine learning for digital soil mapping: Applications, challenges and suggested solutions

Earth-Science Reviews, Год журнала: 2020, Номер 210, С. 103359 - 103359

Опубликована: Сен. 11, 2020

Язык: Английский

---

Soil organic carbon stocks in European croplands and grasslands: How much have we lost in the past decade?

Global Change Biology, Год журнала: 2023, Номер 30(1)

Опубликована: Окт. 30, 2023

Abstract The EU Soil Strategy 2030 aims to increase soil organic carbon (SOC) in agricultural land enhance health and support biodiversity as well offset greenhouse gas emissions through sequestration. Therefore, the quantification of current SOC stocks spatial identification main drivers changes is paramount preparation policies aimed at enhancing resilience systems EU. In this context, (Δ SOCs) for + UK between 2009 2018 were estimated by fitting a quantile generalized additive model (qGAM) on data obtained from revisited points Land Use/Land Cover Area Frame Survey (LUCAS) performed 2009, 2015 2018. analysis partial effects derived fitted qGAM shows that use change observed LUCAS campaigns (i.e. continuous grassland [GGG] or cropland [CCC], conversion (GGC GCC) vice versa [CGG CCG]) was one changes. CCC factor contributed lowest negative Δ with an effect −0.04 ± 0.01 g C kg −1 year , while GGG highest positive 0.49 0.02 . This confirms sequestration potential converting grassland. However, it important consider local environmental conditions may either diminish grassland's storage. UK, (2018) topsoil (0–20 cm) stock below 1000 m a.s.l 9.3 Gt, −0.75% period 2009–2018. losses concentrated central‐northern countries, marginal southeast.

Язык: Английский

---

Mapping soil organic carbon stocks and trends with satellite-driven high resolution maps over South Africa

The Science of The Total Environment, Год журнала: 2021, Номер 771, С. 145384 - 145384

Опубликована: Янв. 27, 2021

Estimation and monitoring of soil organic carbon (SOC) stocks is important for maintaining productivity meeting climate change mitigation targets. Current global SOC maps do not provide enough detail landscape-scale decision making, allow tracking sequestration or loss over time. Using an optical satellite-driven machine learning workflow, we mapped (topsoil; 0 to 30 cm) under natural vegetation (86% land area) South Africa at m spatial resolution between 1984 2019. We estimate a total topsoil stock 5.6 Pg C with median density 6 kg m−2 (IQR: interquartile range 2.9 m−2). Over 35 years, predicted underwent net increase 0.3% (relative long-term mean) the greatest increases (1.7%) decreases (−0.6%) occurring in Grassland Nama Karoo biomes, respectively. At landscape scale, changes up 25% were evident some locations, as evidenced from fence-line contrasts, likely due local management effects (e.g. woody encroachment associated increased overgrazing decreased SOC). Our mapping approach exhibited lower uncertainty (R2 = 0.64; RMSE 2.5 m−2) less bias compared previous low-resolution (250–1000 m) national efforts (average R2 0.24; 3.7 trend map remains estimate, pending repeated measures samples same location (time-series); priority changes. While high can inform decisions aimed (natural solutions), potential are limited by soils. It also that such planting trees balance trade-offs carbon, biodiversity overall ecosystem function.

Язык: Английский

---

Estimating soil organic carbon stock change at multiple scales using machine learning and multivariate geostatistics

Geoderma, Год журнала: 2021, Номер 403, С. 115356 - 115356

Опубликована: Авг. 5, 2021

Many national and international initiatives rely on spatially explicit information soil organic carbon (SOC) stock change at multiple scales to support policies aiming land degradation neutrality climate mitigation. In this study, we used regression cokriging with random forest spatial stochastic cosimulation predict the SOC between two years (i.e. 1992 2010) in Hungary aggregation levels point support, 1 × km, 10 km square blocks, Hungarian counties entire Hungary). We also quantified uncertainty associated these predictions order identify delimit areas statistically significant change. Our study highlighted that prediction of totals averages requires a geostatistical approach cannot be solved by machine learning alone, because it does not account for correlation errors. The total topsoil was predicted increase 2010 14.9 Tg, lower upper limits 90% interval equal 11.2 Tg 18.2 respectively. Results showed both uncertainties average were smaller larger supports, while made easier obtain changes. latter is important accounting studies need prove Measurement, Reporting Verification protocols observed changes are only practically but significant.

Язык: Английский

---

A CNN-LSTM Model for Soil Organic Carbon Content Prediction with Long Time Series of MODIS-Based Phenological Variables

Remote Sensing, Год журнала: 2022, Номер 14(18), С. 4441 - 4441

Опубликована: Сен. 6, 2022

The spatial distribution of soil organic carbon (SOC) serves as critical geographic information for assessing ecosystem services, climate change mitigation, and optimal agriculture management. Digital mapping SOC is challenging due to the complex relationships between its environment. Except well-known terrain environmental covariates, vegetation that interacts with soils influences significantly over long periods. Although several remote-sensing-based indices have been widely adopted in digital mapping, variables indicating term growth less used. Vegetation phenology, an indicator characteristics, can be used a potential time series covariate prediction. A CNN-LSTM model was developed prediction inputs static dynamic Xuancheng City, China. spatially contextual features (e.g., topographic variables) were extracted by convolutional neural network (CNN), while temporal phenology period time) short-term memory (LSTM) network. ten-year phenological derived from moderate-resolution imaging spectroradiometer (MODIS) observations predictors historical changes addition commonly variables. random forest (RF) reference comparison. Our results indicate adding produce more accurate map, tested five-fold cross-validation, demonstrate potentially effective predicting at regional scale long-term extra input. We highlight great hybrid deep learning models, which simultaneously extract different types variables, future applications mapping.

Язык: Английский

---

Digital mapping of soil pH and carbonates at the European scale using environmental variables and machine learning

The Science of The Total Environment, Год журнала: 2022, Номер 856, С. 159171 - 159171

Опубликована: Сен. 30, 2022

Язык: Английский

---

Spatial statistics and soil mapping: A blossoming partnership under pressure

Spatial Statistics, Год журнала: 2022, Номер 50, С. 100639 - 100639

Опубликована: Фев. 15, 2022

For the better part of 20th century pedologists mapped soil by drawing boundaries between different classes which they identified from survey on foot or vehicle, supplemented air-photo interpretation, and backed an understanding landscape processes is formed. Its limitations for representing gradual spatial variation predicting conditions at unvisited sites became evident, in 1980s introduction geostatistics specifically ordinary kriging revolutionized thinking to a large extent practice. Ordinary based solely sample data variable interest—it takes no account related covariates. The latter were incorporated 1990s onward as fixed effects regression predictors, giving rise with external drift kriging. Simultaneous estimation coefficients variogram parameters best done residual maximum likelihood estimation. In recent years machine learning has become feasible huge sets environmental obtained sensors aboard satellites other sources produce digital maps. techniques are classification regression, but take correlations. Further, effectively 'black boxes'; lack transparency, their output needs be validated if trusted. They undoubtedly have merit; here stay. too, however, shortcomings when applied data, statisticians can help overcome. Spatial pedometricians still much do incorporate uncertainty into predictions, averages totals over regions, errors measurement positions data. must also communicate these uncertainties end users maps, whatever means made.

Язык: Английский

---

Assessing Machine Learning-Based Prediction under Different Agricultural Practices for Digital Mapping of Soil Organic Carbon and Available Phosphorus

Agriculture, Год журнала: 2022, Номер 12(7), С. 1062 - 1062

Опубликована: Июль 20, 2022

Predicting soil chemical properties such as organic carbon (SOC) and available phosphorus (Ava-P) content is critical in areas where different land uses exist. The distribution of SOC Ava-P influenced by both natural anthropogenic factors. This study aimed at (1) predicting a piedmont plain Northeast Iran using the Random Forests (RF) Cubist mathematical models hybrid (Regression Kriging), (2) comparing models’ results, (3) identifying key variables that influence spatial dynamics under agricultural practices. machine learning were trained with 201 composite surface samples 24 ancillary data, including climate (C), organism (O), topography- relief (R), parent material (P) features (S) according to SCORPAN digital mapping framework, which can predictively represent formation factors spatially. Clay, one most well-known relationship SOC, was important predictor followed open-access multispectral satellite images-based vegetation indices. had similar set effective variables. Hybrid approaches did not improve model accuracy significantly, but they reduce map uncertainty. In validation set, calculated RF algorithm normalized root mean square (NRMSE) 96.8, while an NRMSE 94.2. These values change when technique for Ava-P; however, changed just 1% SOC. management supply activities be guided maps. Produced maps scientist plays active role used identify concentrations are high need protected, uncertainty sampling required further monitoring.

Язык: Английский

---

Beyond prediction: methods for interpreting complex models of soil variation

Geoderma, Год журнала: 2022, Номер 422, С. 115953 - 115953

Опубликована: Май 27, 2022

Understanding the spatial variation of soil properties is central to many sub-disciplines science. Commonly in mapping studies, a map constructed through prediction by statistical or non-statistical model calibrated with measured values property and environmental covariates which maps are available. In recent years, field has gradually shifted attention towards more complex algorithmic tools from machine learning. These models particularly useful for their predictive capabilities often accurate than classical models, but they lack interpretability functioning cannot be readily visualized. There need understand how these can used purposes other making whether it possible extract information on relationships among variables found models. this paper we describe evaluate set methods interpretation variation. An overview presented model-independent serve purpose interpreting visualizing different aspects model. We illustrate two case study topsoil organic carbon France. reveal importance each driver variation, interaction, as well functional form association between covariate property. Interpretation also conducted locally an area locations distinct land use climate. show that all cases important insights obtained, both into overall decision made at location. This underpins going beyond studies. predictions help us formulating hypotheses underlying processes mechanisms driving

Язык: Английский

---