Abstract:
Soils are under threat globally, with declining soil productivity and soil condition in many places. As a key indicator of soil functioning, soil organic carbon (SOC) is crucial for ensuring food, soil, water and energy security, together with biodiversity protection. While there are global efforts to map SOC stock and status, SOC is a dynamic soil property and can change rapidly as a function of land management and land use. Here, we introduce a semi-mechanistic model to monitor SOC stocks at a global scale, underpinned by one of the largest worldwide soil database to date. Our model generates a SOC stock baseline using machine learning methods, which is then propagated through time by keeping track of annual landcover changes obtained from remote sensing products with loss and gain dynamics dependent on temperature and precipitation, which finally define the magnitude, rate and direction of the SOC changes. We will share what this monitoring system enable us to do in terms of global SOC stock accounting, how it relates to soil productivity and its contribution in the context of green house gas emissions. We will also discuss the future improvements necessary to turn this project into the soil monitoring system needed to secure Earth's soils.
Bio:
José is a Soil Scientist currently working as a Postdoctoral Research Associate in the School of Life and Environmental Sciences and Sydney Institute of Agriculture (University of Sydney). José is particularly interested in spatio-temporal soil modelling and soil spectroscopy from the regional to the global scale, and on how to use machine learning methods to tackle the methodological challenges associated with them. He leads the application of deep learning in soil sciences, developing new modelling frameworks for digital soil mapping and soil spectroscopy. Besides the interest on improving the accuracy of soil models, he is also interested in the interpretability of these usually considered "black-boxes", privacy-preserving training methods and on how to improve the connectivity/transportability between global and local-scale models.
Summary:
Soils
Store water and solutes
Healthy spoils critical for agriculture and societal health
Soil security framework
Use soil indicators and utility graphs
Water capacity
Salinization
Key indicator: Soil organic carbon (SOC)
Many initiatives
GlobalSoilMap.net: bottom->up
ISRIC: top->bottom
FAO, 4Pour1000
Challenge: traditional approaches are static but soil properties change dynamically
E.g. SOC loss .04%-1.2%/year
Expansion of agriculture/forestry is threatening soils
Critical to integrate dynamic landcover changes in SOC assessment
Approach: SOC Monitoring at global scale
Baseline map
Landcover tracking: Where, when, change type, duration of change persistence
Baseline methodology: digital soil mapping
Based on covariates:
Soil properties
Climate
Organisms/landcover
Topography
Parent material/geology
Age
Geographic location
Trained ML model (Cubist) on dataset
Dynamics: Landcover tracking
MODIS MCF12Q1, 500m resolution
IGBP classification scheme
SOC amount: Depends on regional ecology and change in land cover type
Rate of change: depends on temperature and precipitation
Shape of change: gaining/losing SOC
Landcover tracking via Google Earth Engine
Key findings
Evolution maps of various regions of how changes in landcover types drive changes in SOC
Annual losses:
1.9Pg SOC / year (topsoil): 20% larger than annual production-based emission in US in 2018
Global total : 700-800 Pg
Tropic and sub-tropic : ~50% of global loss
Soil productivity:
Critical SOC limits:
1.1% tropical: 11 Mha/year
2% tropical: 6 Mha/year
Soil on these lands is very hard to recover (limited resource)
Towards a soil early warning system
500m-30m Landsat
Baseline for 1985
Future improvements:
Climate change effect, larger analytic window will make it possible to capture effect more clearly
Within-class changes
Different crops
Different management
Requires more soil data
Scenario analysis
Inform management and policy
Improve soil condition
Secure soils