dsm for soil erosion risk in scotland with uncertainty propagation · dsm for soil erosion risk in...

DSM for soil erosion risk in Scotland with uncertainty propagation

Laura Poggio, Alessandro Gimona, Jim McLeod, Marie Castellazzi, Andrea Baggio Compagnucci, Justin Irvine

Brief outline

Erosion modelling

Digital Soil Mapping approach for input data :

soil, climate, land management

Correction for organic soils

Uncertainty propagation

R factor K factorC factor P factor

Land useVegetation MorphologyClimate Data Soil Data

K factorpeatsLS factor

RUSLEK * R * C * P * LS

modified from Panagos et al, 2015+

expert knowledge rules for peats (Lilly et al, 2002)

Soil OM

Input point data Points input data interpolated with a scorpan-kriging flavour: Soil properties: e.g. Organic soils Soil C stocks Input for K factor

Soil C stocks (1 m depth, kg/m2)

K Factor

Input point dataPoints input data interpolated with a scorpan-kriging flavour: Soil properties: e.g. Organic soils Soil C stocks Input for K factor

Soil C stocks (1 m depth, kg/m2)

K Factor

R Factor

Climate station data Inputs for R factor

Tested covariates

Snow

NDWI

EVI

Elevation

• MODIS• LANDSAT• Sentinel 1 and 2

Validation and GAM plotsK factor Organic soils R factor

xy *** *** ***

Elevation --- ** *

Slope * * ---

TWI --- --- ---

Soil Organic matter *** NA ---

EVI *** *** ---

NDWI *** * ---

Snow --- --- ---

Productivity * * ---

LST *** ** ---

Seasonality vegetation ** ** ---

Radar (Sentinel 1) *** *** *

Bioclimatic variables --- --- ***

Explained deviance 26.6% 60.5% 35.1%

RMSE 0.01 0.1 0.05

Input spatial data

C & P & S Factors

Spatial layers derived from basic GIS analysis:

C and P factorsRS, land use integration

S(heep) factor Downscale of district data for sheep and cowsDeer grazing(McLeod et al,2014,Unpublished)

Erodibility for Organic soilsExpert knowledge rules (Lilly et al, 2002)

Difference between

RUSLE without

organic soils

correction and with

organic soils

correction

Uncertainty propagation

1. Simulations from each of input factor interpolations with

summary statistics : median, quantiles a lot of

combinations (10^3) ^ 5

This can be simplified with e.g. Latin Hypercube Sampling

Uncertainty propagation

1. Simulations from each of input factor interpolations with

summary statistics : median, quantiles a lot of

combinations (10^3) ^ 5

This can be simplified with e.g. Latin Hypercube Sampling

1. Bayesian approach: e.g. Bayesian Belief Networks for risk

modelling output is a probability distribution of risk

High erosion risk Low erosion risk

Probability distribution of erosion risk

Thank you for your attention

This work was funded by the ScottishGovernment Environment, Land use andRural Stewardship research programme.