The consortium focuses its work on the WP and the tasks defined below:
WP0 Management of the project
WP leader: D. Arrouays
Consortium management group: D. Arrouays, B. Minasny, L. Poggio, Z. Libohova, T. Mulder, P. Roudier
Secretariat: Z. Libohova, A. Richer-de-Forges.
WP1 Legacy and ancillary data for Digital Soil Mapping
WP leader: Titia Mulder
Task 1.1 Test the potential of new ancillary data for Digital Soil Mapping
Task Leader: P. Roudier
The number of freely accessible sources of soil covariates (Digital Elevation Models, Remote Sensing products, Climate data, Lithology data, etc) is increasing exponentially. There is an urgent need to explore and assess the potentials and limitations of these new data (e.g., Poggio & Gimona, 2017b). Moreover, increasing the number of co-variates in predictive modelling, in some cases, may lead to spurious relationships and over-fitting. One temptation with the exponential increase of data is to incorporate all the available data and then let algorithms or statistical indicators decide. This uncontrolled use of co-variates may lead to strong biases (e.g., Fourcade et al., 2018) and overlook of accumulated tacit knowledge from decades of field surveying. Therefore, we should use all the information with caution and rigorously test their advantages and limitations and their biophysical meaning. Among the products we want to test are new remote sensing products at high spatial and spectral resolutions (e.g., Lagacherie et al., 2012), airborne sensing products (e.g., Martelet et al., 2013), high-resolution digital elevation models, time series of remote sensing data, historical maps of land use, etc.
Task 1.2 Explore methodologies to merge and/or harmonize different products
Task Leader: T. Mulder
GlobalSoilMap and SoilGrids are now producing gridded prediction of soil attributes both at local and global scales. As a consequence, several spatial data sources (maps, grids, or point data) referring to the same soil properties are commonly available for a particular zone. Indeed, each model and data used to create maps has its own strengths and weaknesses. However, the information provided by different sources may be complementary and merging them may be a way to create the most accurate map possible. We will deal with these issues by applying model ensemble (or model averaging) methods (Malone et al., 2014; Roman Dobarco et al., 2017) and assessing their performance. We will also test the feasibility of down-scaling global products such as SoilGrids, by adding more precise co-variates or by adding soil data available only at local scale.
Task 1.3 Propose methods for harmonizing products to a common date
Task Leader: B. Minasny
Monitoring changes in soil properties is essential to assess the effects of global change and policy implementation on their quality. One way forward is to go from Digital Soil Mapping to Digital Soil Monitoring. We will develop methods for harmonizing data to common dates. On the one hand, these methods will allow assessing changes a posteriori. On the other hand, a very practical objective of harmonization to a common date is that it will constitute a baseline for designing future soil monitoring sampling scheme. This task will be done in close collaboration with task 3.3.
WP2 Methods for sampling, modelling and mapping soils in space and time
WP Leader: B. Minasny
Task 2.1 Testing and developing new methods/models for prediction
Task Leader: L. Poggio
Statistical, geospatial and machine learning methods are constantly being improved. In order to stimulate and develop innovation, the consortium will provide a scientific and technical watch and monitor new emerging methods at an international level (e.g., Poggio & Gimona, 2017a&b, Poggio et al., 2016; Roudier et al., 2017). Promising methods will be tested, by comparing their performance with results previously obtained in several parts of the world, using the same calibration and validation data. A benchmarking approach will be defined for estimating unbiased and significant performance indicators for the tested models, in link with task 3.1. This task will also address sampling optimization, especially for new sampling campaigns, and assess the added value of model averaging approaches. One major challenge is also to develop 3-D modelling and models able to predict a set of soil properties that are inherently correlated.
Task 2.2 Testing methods for estimating complete probability distribution functions
Task Leader: B. Minasny
When predicted soil properties are used as an input data for further modelling, a major challenge is not only the prediction of the uncertainty (e.g., a 90% confidence interval), but the prediction of a complete probability distribution function (pdf). Indeed, knowing this pdf is crucial for studying the propagation of errors or for assessing model sensibility to input parameters. However, observed soil data are rarely numerous enough to derive a pdf. A way to estimate pdf could be to reconstruct them using values derived from expert knowledge, such as modal, low and high estimated values provided by soil surveyors when characterizing soil mapping units of legacy soil maps, and using these values to fit simple models of probability distribution.
WP3 Methods for estimating model and map uncertainty
WP leader: L. Poggio
Task 3.1 Develop methods of uncertainty spatial assessment
Task Leader: T. Mulder
The evaluation of uncertainty of soil predictions is in our view the scientific aspect for which most progress are needed. There is solid theory on uncertainty that is largely developed by pedometricians, and this has been incorporated in the GlobalSoilMap specifications (Heuvelink et al., 2014). The practical application to the sparse dataset often encountered in digital soil mapping remains challenging, and several approaches need to be tested (e.g., Malone et al., 2015; Odgers et al., 2015; Poggio et al., 2016; Vaysse & Lagacherie, 2017). Moreover, many countries may not have independent and unbiased validation datasets that allows for assessing prediction performance. Therefore, there is a need to develop straightforward, robust and costless ways to evaluate the performance of the predictions and to derive spatial estimates of uncertainty. The uncertainty also derives from many sources (e.g. sampling, measurement errors, co-variates accuracy, vertical splining of properties, spatial modelling) One aim of this task will be to develop methods to assess the importance of these various sources for the final predictions. Another aim is of course to contribute to the updates of the specifications described in Task 4.4.
Task 3.2 Develop methods do deal with censored data/soft data
Task Leader: P. Roudier
Many observed point data are not the exact value of the property of interest. They may be ‘censored data’, that is, information that the target value is higher or lower than a given observation (e.g. soil depth data for which the highest value is limited by the tool used for digging). Some other data may be imprecise or be considered as more or less good ‘proxies’ of the property of interest (e.g., soil colour, visible or near-infrared spectral information, gamma-ray data). Methods should be developed to deal with this kind of data, which may be available or acquired in huge quantities but are not real measurements of the target value.
Task 3.3 Solve the question of influence on the age of the rescued soil data on predictions
Task Leader: L. Poggio
As some soil properties evolve quite rapidly and as legacy soil data may date from the present to more than 50 yrs. ago, there is a need for the development of innovative methods that explicitly take into account space and time in modeling. New methods (e.g. new machine learning techniques) will be implemented and tested to solve this issue and make a better use of legacy soil data. This task is strongly related to the practical objective defined in task 1.3.
WP4 Scientific outreach and capacity building
WP leader: P. Roudier
Task 4.1 Produce an exhaustive review of GlobalSoilMap initiatives and results all over the world
Task Leader: D. Arrouays
This review is an important step in addressing the main issues to be dealt with by the consortium. It will also be highly beneficial to the scientific community. It will be accomplished using standard methods of reviews and meta-analysis, as well as by contacting directly our network of partners all over the world, which implies the contribution of all consortium partners. The consortium partners are well aware of GlobalSoilMap progress in many parts of the world; moreover, they are very familiar with producing literature reviews (Van Looy et al., 2017; Arrouays et al., 2017; Horta et al., 2015; Minasny & McBratney, 2015; Stockmann et al., 2015; Minasny et al., 2013; Mulder et al., 2011). The consortium will write a high-impact paper on the state-of-the-art GlobalSoilMap initiatives. This task will prvide major inputs to tasks 4.2 and 4.3.
Task 4.2 Revise and update the GlobalSoilMap specifications by keeping them at the state-of-the-art level
Task Leader: Z. libohova
The GSM specifications will be continuously revised and updated by the consortium, and revision proposals will be submitted to the scientific community through the IUSS, and to the FAO Global Soil Partnership, in order to sustain the consensus principle underlying their establishment. This will be one of the major outcomes of the project. These specifications are being used as benchmark in many international initiatives and thus it is of the upmost importance to deliver proper and up to date documents. In order to achieve this we need to do the other tasks detailed in this proposal. Indeed, due to changes in user needs and scientific advances, this document needs to be updated The revisions may include the proposal for new parameters or new soil attributes to be mapped, the proposal and the description of new methods for spatial modeling, and the proposal of new pedo-transfer functions to derive some unknown soil properties from already existing predicted soil properties.
Task 4.3 Show relevance of gridded, Global, Digital Soil Map by use cases and communication to end users
Task Leader: D. Arrouays
All efforts and investments in soil data and soil mapping, and in producing high-resolution gridded maps for the globe, only make sense if the soil information and the new soil information products are relevant for purposes such as food security, climate change, sustainability, field decisions, etc. The consortium will show the usefulness of delivering original baseline soil properties and derived functions contributing to global and local issues (e.g. carbon sequestration, erosion control, salinity, land-use planning). Showing the relevance by reporting on case studies will certainly lead to more and better use. It will also encourage users and stakeholders (e.g., countries) to buy into the approach and ultimately contribute to bottom-up initiatives and improvement in the current products. This task will include a review of grid utilizations and a prospective study of end-users’ needs, and work related to the ways to communicate (or to show on maps) the uncertainty and how this uncertainty could be incorporated into the decision-making process.
