Victoria Naipal Max-Planck Institute for Meteorology

1. From Local to Global Scale Soil Erosion Modelling A Sensitivity Analysis of the RUSLE Model Victoria Naipal Max-Planck Institute forMeteorology Land Department; Vegetation Modelling Group Supervisor: Ch.Reick CO-Supervisor: J.Pongratz EGU, 29.04.2014 Photo: NRCS

2. Background MethodsResults & Validation Analysis Conclusions Outlook Soilerosionandfeedbacks Land cover/ land use and soil erosion in Germany since the Early Middle Ages (data from Bork et al., 1989) • Estimations of present global soil erosion range between 20.5 – 201 Pg/yr • Current Dynamic Global Vegetation Models (DGVMs) do not accountforsoilerosionand do not treatsoilas a dynamicalsytem • This limitsourcapabilitiestoinvestigatethe global carbonandnutrientcycles 2

3. Why soil erosion modelling on global scale? Background MethodsResults & Validation Analysis Conclusions Outlook Cumulative carbon emissions as a result of anthropogenic land-cover change in the non-alpine Rhine Basin (Hoffmann et al., 2013) • Sediment storage data indicates that hillslopes and floodplains could have sequestered an amount of OC similar to cumulative carbon emissions from anthropogenic land-cover change Objective: Investigate the effects of soil erosion on the global carbon cycle for different climate and land-use scenarios • How to represent complex soil erosion processes, which occur on local scales, in a simplified but realistic way on global scale? 3

4. Tool – RUSLE Model(Renard, 1997) Background MethodsResults & Validation Analysis Conclusions Outlook • Semi empirical/process-based model • Based on the largest set of experiments on soil erosion in the US • Parameterized for agriculture areas and the agriculture plot scale • Does not simulate deposition and sediment transport • Originally does not have a “limit to soil erosion” 4

5. Background MethodsResults & Validation Analysis Conclusions Outlook Input datasets • Topography: 1 km global GTOPO DEM • Climate: 0.25 degree monthly GPCC precipitation 1951 – 2000 • Soil: 1km Harmonized World Soil Database • Land cover: 0.05 degree MODIS NDVI • Support practice: MODIS landcoverdata + Yang et al (2003) • Additional changesmadetothe RUSLE model: • Implementation ofgravel (fromthesoildatabase) as a limittosoilerosion • Estimatethe land-use/land-cover parameterbased on NDVI 5

6. Current global soil erosion rates Background MethodsResults & Validation Analysis Conclusions Outlook Mean global soil erosion rate: 7 tons ha-1 yr-1 6

7. Validation Background MethodsResults & Validation Analysis Conclusions Outlook • Overestimation of soil erosion rates in mountainous areas – Cerdan et al. (2010) • Underestimation of soil erosion rates on agricultural land - Van Oost et al.(2007), Doetterl et al.(2011) 7

8. Sensitivity analysis and scaling 8

9. Scaling slope – fractal method Background MethodsResults & Validation AnalysisConclusions Outlook Mean global slope versus DEM resolution • The topographical factor of RUSLE is a function of the slope, and the slope is highly dependent on the spatial resolution of the DEM • Topography generally obeys fractal statistics (Huang and Turcotte, 1989) Meanslope (degrees) DEM resolution (arc-minutes) • Statistical variation of elevation between samples varies with the distance between them ( ex.: Klinkenberg and Goodchild, 1992) S = percentage slope α = fractal coefficient d = specified finer grid resolution D = fractal dimension 9

10. Implementing the fractal method BackgroundMethodsResults & Validation AnalysisConclusions Outlook Applying the fractal method for the scaling of slope from a 1km global DEM shows a significant increase in detail and improves the global mean Slope derived from 1km DEM Slope estimated by the fractal method for 30m grid resolution 10

11. Sensitivity of the erosivity Background MethodsResults & Validation AnalysisConclusions Outlook • Large erosionamountsasmodelled in mountainousareascanberelatedtobiases in theerosivityfactor • Original R formula (Renard, 1997): • j, • where: • E = kinetic energy • I30 = 30-minute rainfall intensity • Formula Renard and Freimund (1994): • , for P <= 850mm • , • where: • P = Total yearly precipitation USLE Erosivity, NRI data Erosivity in currentstudy 11

12. Validation of the erosivity Background MethodsResults & Validation AnalysisConclusions Outlook • In thecurrentstudyerosivityislargelyoverestimatedforregionswithsmallerosivityvalues in the NRI data • High resolution PRISM precipitationdata (800m) forthe US was usedtovalidatethe global GPCC precipitation (0.25 degree) • The goodcorrelationbetweentheprecipitationdataindicatesthatthespatialscalecannotbe a reasonforthe large biases in theerosivity 12

13. Erosivity dependence on climate conditions Background MethodsResults & Validation AnalysisConclusions Outlook • The biases in erosivityascomputedwithprecipitationtotals, couldbeattributedto large variability in precipitationintensitiesfor different climaticconditions Temperateandtundrazonesshowthelargestbiases 13

14. Background MethodsResults & Validation Analysis Conclusions Outlook Conclusions • Implementing RUSLE on a global scaleresults in overestimatingerosionrates on mountainousareaswhileunderestimatingerosion in agricultural/flatterareas • Soil erosion is highly dependent on the spatial resolution – use of scaling methods is essential • The fractalmethodcanbeusedforimprovingtheeffectoftopography on erosion in RUSLE • Improvementoftheerosivityfactorofthe RUSLE modelcanremovebiases in mountainousareas, on global scale 14

15. Background MethodsResults & Validation Analysis ConclusionsOutlook Outlook: Sediment mass-balance model • Toestimateeffects on thecarboncycle, otherprocesses such asdepositionandsedimenttransporthavetobeincluded • A global approachneedstherefore a simplifiedmethodthatcancapturetherealisticbehaviouroftheseprocesses from Hoffmann et al., 2013 (GBC) Residence-time Residence-time 15