Relevance of GEMAS for soil property mapping Rainer Baritz1, Dietmar Zirlewagen2, Vibeke Ernstsen3 1) Federal Institute for Geosciences and Natural Resources (BGR) 2) INTERRA 3) Geological Survey of Denmark and Greenland (GEUS) 1 Introduction GEMAS samples were taken from agricultural surface-close soil layers (Ap 0-20 cm, Grazing land 0-10 cm); Parameters also include TOC, pH, P, CEC; “Standard results“ are provided as geostatistical maps; The main objective of GEMAS is to collect information about the spatial distribution pattern of trace elements in the rooted zone of soil; Soil organic matter and acidity are important to interpret the potential of soil to store and release heavy metals. 2 Questions Technical questions: What is the possible contribution of GEMAS to soil monitoring? How can GEMAS information be integrated into different soil inventories? How can the representativity from GEMAS be assessed? Criteria? Are there alternative upscaling methods? 3 Upscaling method 4 Upscaling method Spatial Regression Stepwise multiple linear regression combined with geostatistics/kriging; Covariates as possible impact factors on the target variables (TOC, pH, P); Stratification is important to optimise upscaling models. 5 Upscaling method Database Biogeographical regions, soil regions, N deposition data (EMEP); DEM 90m/Relief parameters (aspect, slope, curvature, topographic wetness index, potential direct radiation, etc.); Parent material (ESDB, 2004); Land cover CORINE 2000 and 2006, and at GEMAS points LUCAS 2003 crop types (CEC, 2003); Climate (WORLDCLIM). 6 Upscaling method Stratification Stratum 1: Sweden Norway and Finland (‘Boreal’) Stratum 2: United Kingdom and Ireland Stratum 3: Eco-Regions (DMEER) with CodeNumbers 56, 27, 10 (‘High Mountainous’) Stratum 4: the remaining European target area (‘European continent’) 7 Total organic carbon 8 TOC [%] agricultural soil (From Baritz et al., 2014, Fig. 6.4B, p.123) 9 Step Influence of predictors [All_TOC] Variable Entered Number Vars In Partial Model F Value R-Square R-Square Pr > F 1 TMAX5 1 0.2015 0.2015 462.11 <.0001 2 Ap 2 0.0892 0.2907 230.09 <.0001 3 PREC11 3 0.0139 0.3046 36.58 <.0001 4 Histosol 4 0.0128 0.3174 34.31 <.0001 5 TMAX4 5 0.0059 0.3234 16.05 <.0001 6 Corine_pasture 6 0.0066 0.3299 17.91 <.0001 7 BIOCLIM_3 7 0.0041 0.3340 11.12 0.0009 8 LForm_4 8 0.0040 0.3380 11.04 0.0009 9 Corine_scrub 9 0.0037 0.3417 10.25 0.0014 10 BIOCLIM_14 10 0.0030 0.3447 8.45 0.0037 11 BIOCLIM_17 11 0.0073 0.3520 12 TMIN7 12 0.0035 0.3555 9.78 0.0018 13 Luvisol 13 0.0033 0.3587 9.25 0.0024 14 Leptosol 14 0.0023 0.3610 6.55 0.0106 15 Podzol 15 0.0028 0.3638 7.98 0.0048 16 TEXTSRF1 16 0.0049 0.3687 13.98 0.0002 17 ECO_CODE_11 17 0.0026 0.3713 7.58 0.0060 18 climagroup4 18 0.0021 0.3734 6.09 0.0137 19 PREC8 19 0.0023 0.3758 6.82 20.44 <.0001 10 0.0091 TOC TOC and crop types Stratum 1: Sweden, Norway and Finland (‘Boreal’) Stratum 2: British Isles and Ireland Stratum 3: Eco-Regions (‘High Mountainous’) Stratum 4: remaining Europe ‘(European continent’) 11 (From Baritz et al., 2014, Fig. 6.5, p.124) LUCAS 2003 and soil texture (ESDB, 2004) coarse Rye medium medium fine fine very fine Potatoes Olive groves Shrubland Barley Maize Sunflower Grassland Durum Wheat Rape seeds Common wheat Dry pulses Sugar beet Other non permanent industrial crops Cotton, Other fibre and oleaginous Oranges, crops Vineyards, Nuts 12 (From Baritz et al., 2014, Table 6.3, p.127) Soil acidity pH (CaCl2) 13 pH (CaCl2) agricultural soil 14 Validation and uncertainties 15 Validation Method Despite low sampling density (1 sample site/2500 km2), the sample size was large enough to separate a training, and a validation set both representing well the predictive population; Split of the data; random split inside large-scale squares stratified biogeographical region; Regional models are derived from the training data; Prediction error is then compared to the results from running the training set-based models with the validation data. Germany: 357,104 km2 total, 187,291 km2 , agriculture (1 site/600 km2) Europe: 10.5 million km2; agriculture: 1 site/2333 km2 (parts of Eastern Europe and Balkans not covered) 16 Results Modellskala (lognormal) Response TOC TOC TOC TOC TOC TOC TOC TOC Variante ALL ALL_AP_STRATEN ALL_GR_STRATEN AP_STRATEN AP_STRATEN_KRIGING GR_STRATEN GR_STRATEN_KRIGING STRATEN R² 0,343 0,267 0,29 0,324 0,926 0,35 0,871 0,397 Spatial distribution of the inaccuracy (standard error) RMSE 0,43818 0,38471 0,48785 0,36742 0,12247 0,46583 0,21213 0,41833 Training MSE 0,192 0,148 0,238 0,135 0,015 0,217 0,045 0,175 STD 0,553 0,463 0,59 0,457 0,457 0,587 0,587 0,549 OBS 1865 951 911 949 949 904 904 1853 R² 0,366 0,296 0,316 0,338 0,33 0,352 0,35 0,399 Validierung RMSE MSE 0,43932 0,193 0,40988 0,168 0,46904 0,22 0,40866 0,167 0,41833 0,175 0,45935 0,211 0,46797 0,219 0,43359 0,188 STD 0,566 0,502 0,58 0,511 0,511 0,581 0,581 0,57 Spatial distribution of the residuals 17 OBS 1862 952 910 948 948 907 907 1855 Outlook 18 Outlook Include N and CEC, include soil texture data; Re-upscale with the new parent material map; Condense regionally, then also improve stratification; Interpret covariates; Include integrated evaluations (e.g., potential heavy metal release relative to SOM and acidity). 19 New BGR parent material map alluvium/colluvium calcearous rocks clayey materials crystalline rocks detrital formations European Soil Database glaciofluvial materials loamy/silty marl other/organic sandstone/flysch/molasse sandy materials schists volcanic rocks (From Günther et al., 2013, Fig. 2, p.299 & Baritz et al., 2014, Fig. 6.2, p.120) 201 classes (aggregated from 671 initial classes) Regional studies BGR GEMAS: N=310 (completely sampled and analysed soil profiles) + BGR soil profiles: N=1567 (agricultural land) = Representative data set for higher resolution evaluations, 2.5 D 21 Conclusions The quality of the GEMAS inventory (analysis, georeferencing) is high so that satisfactory regression models can be built (950 plots for the ‘learning’ data set; stratification is important. Option: Integration into a larger soil monitoring and soil quality assessment scheme (country-level/Europe). Added value to facilitate a closer exchange between geoscientists and soil scientists. Thank you for your attention! d.zirlewagen@interra.biz ve@geus.dk rainer.baritz@bgr.de References 23 References SLIDES 7, 9, 11, 12, 20: Baritz, R., Ernstsen, V. & Zirlewagen, D., 2014. Carbon concentrations in European agricultural and grazing land soil. Chapter 6 In: C. Reimann, M. Birke, A. Demetriades, P. Filzmoser & P. O’Connor (Editors), Chemistry of Europe's agricultural soils – Part B: General background information and further analysis of the GEMAS data set. Geologisches Jahrbuch (Reihe B 103), Schweizerbarth, 117-129. SLIDE 6: CEC (Commission of the European Communities), 2003. The LUCAS survey. European statisticians monitor territory. Theme 5: Agriculture and fisheries, Series Office for Official Publications of the European Communities, Luxembourg, 24 pp. Corine land cover 2000 (CLC2000) seamless vector database. http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2000clc2000-seamless-vector-database Corine Land Cover 2006 (CLC2006)s eamless vector data. http://www.eea.europa.eu/data-and-maps/data/clc-2006-vector-data-version ESDB, 2004. The European Soil Database distribution version 2.0. European Commission and the European Soil Bureau Network, CD-ROM, EUR 19945, http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDB_Data_Distribution/ESDB_data.html . SLIDE 9: Baritz, R., Ernstsen, V. & Zirlewagen, D., 2014. Carbon concentrations in European agricultural and grazing land soil. Chapter 6 In: C. Reimann, M. Birke, A. Demetriades, P. Filzmoser & P. O’Connor (Editors), Chemistry of Europe's agricultural soils – Part B: General background information and further analysis of the GEMAS data set. Geologisches Jahrbuch (Reihe B 103), Schweizerbarth, 117-129. SLIDE 17: Baritz, R., D. Zirlewagen and E. Van Ranst (2006). Methodical standards to detect forest soil carbon stocks and stock changes related to land use change and forestry – landscape scale effects. Final report Deliverable 3.5-II. Multi-source inventory methods for quantifying carbon stocks and stock changes in European forests (CarboInvent) EU EVK2-2001-00287. SLIDE 20: Günther, A., Van Den Eeckhaut, M., Reichenbach, P., Hervás, J., Malet, J.-P., Foster, C. & Guzzetti, F., 2013. New developments in harmonized landslide susceptibility mapping over Europe in the framework of the European Soil Thematic Strategy. Proceedings Second World Landslide Forum, 3-7 October 2011, Rome. In: C. Margottini, P. Canuti, K. Sassa (Editors), Landslide Science and Practice. Springer-Verlag, Berlin, Vol. 1, 297-301. doi: 10.1007/978-3-642-31325-7_39.