Influence of soil texture on the estimation of soil organic carbon from Sentinel-2 temporal mosaics at 34 European sites

Supplementary information: Influence of soil texture on the estimation of soil organic carbon using Sentinel-2 temporal mosaics at 34 European sites Wetterlind, J.1, Simmler, M.2, Castaldi, F.3, Borůvka, L.4, Gabriel, J.L.5, Gomes, L.C.6, Khosravi, V.4, Kıvrak, C.7, Koparan, M.H.7, Lázaro-López, A.5, Łopatka, A.8, Liebisch, F.9, Rodriguez, J.A.5, Savaş, A.Ö.7, Stenberg, B.1, Tunçay, T.7, Vinci, I.10, Volungevičius, J.11, Žydelis, R.11, Vaudour, E.12 1 Swedish University of Agricultural Science (SLU), Department of Soil and Environment, Skara, Sweden 2 Agroscope, Sustainability Assessment and Agricultural Management, Ettenhausen, Switzerland 3 National Research Council of Italy (CNR) Institute of BioEconomy, Firenze, Italy 4 Czech University of Life Sciences Prague (CZU), Faculty of Agrobiology, Food and Natural Resources, Department of Soil Science and Soil Protection, Prague, Czech Republic 5 Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria (INIA-CSIC), Madrid, Spain 6 Aarhus University (AU), Department of Agroecology, Tjele, Denmark 7 Republic of Turkey Ministry of Agriculture and Forestry, General Directorate of Agricultural Research and Policies (TAGEM), Ankara, Türkiye 8 Institute of Soil Science and Plant Cultivation – National Research Institute (IUNG), Puławy, Poland; 9 Agroscope, Agroecology and Environment, Zürich, Switzerland 10 Environmental Protection Agency of the Veneto Region (ARPAV), Italy 11 Lithuanian Research Centre for Agriculture and Forestry (LAMMC), Institute of Agriculture, Kėdainiai, Lithuania 12 Université Paris-Saclay, INRAE, AgroParisTech, UMR EcoSys, 91120 Palaiseau, France Correspondence: Johanna Wetterlind. E-mail: johanna.wet[email protected] Supplementary section Table S1-3 present an overview of the 34 sites included in the study, the soil sampling and soil analyses. Figure S1 shows temporal mosaics based on the median and the R90 method at three example sites. Table S4 presents the performance of site-specific random forest models for predicting SOC and clay as determined in leave-one-out cross-validation. Similar to Table 1 in the paper for the PLSR models. Figure S2 shows the performance of the site-specific random forest models. Similar to Figure 5 in the paper for the PLSR models. Figure S3 shows the relative importance of the satellite bands in the PLSR models for predicting SOC and clay content (same data as in Figure 8 in the paper). Table S1. Site information, dominant soil type according to the WRB systems with one principal qualifier. Site Lat Long Soil type SWE 1 58.4 13.43 No information SWE 2 58.38 13.48 No information SWE 3 58.38 13.3 No information SWE 4 58.25 13.13 No information SWE 5 58.26 13.13 No information SWE 6 58.23 13.13 No information DNK 1 56.93 8.65 Chernic Phaeozems DNK 2 56.37 9.57 Haplic Umbrisols, Haplic Phaeozems, Histosols, Haplic Luvisols DNK 3 55.49 9.07 Chernic Phaeozems DNK 4 55.32 11.34 Chernic Phaeozems DNK 5 54.89 9.12 Umbric Podzols LTU 1 55.73 21.46 Glossic Retisols, Gleyic Luvisols LTU 2 55.38 23.86 Gleyic Luvisols LTU 3 55.25 24.85 Gleyic Luvisols, Eutric Planosols, Mollic Gleysols LTU 4 54.75 24.78 Calcaric Luvisols POL 1 51.47 22.05 Podzols CZE 1 50.52 15.42 Eutric Cambisols CZE 2 50.39 15.26 Haplic Luvisols, Albic Luvisols, Luvic Chernozems CZE 3 50.08 14.9 Calcic Chernozems, Luvic Chernozems, Eutric Regosols CZE 4 49.86 14.74 Haplic Stagnosols, Stagnic Cambisols, Eutric Leptosols FRA 1 48.9 1.85 Haplic Luvisols, Calcaric Cambisols, Solimovic Cambisols CHE 1 47.67 9.06 No information CHE 2 47.58 8.79 No information CHE 3 47.56 9.08 Dystric Gleysols, Haplic Alisols CHE 4 47.45 8.68 No information CHE 5 47.43 8.52 No information ITA 1 45.86 12.5 Haplic Luvisols, Hypereutric Cambisols ITA 2 45.6 12.49 Calcaric Gleysol, Gleyic Calcisol ITA 3 45.4 12.17 Gleyic Calcisols, Calcic Gleysol ITA 4 45.25 11.08 Gleyic Calcisols ITA 5 45.17 11.49 Gleyic Calcisols TUR 1 38.44 33.87 Calcaric Fluvisols, Haplic Vertisols TUR 2 36.73 28.78 Calcaric Fluvisols ESP 1 37.66 -0.86 Haplic Calcisols Table S2. Site information, soil sampling. Area is calculated using a convex hull round the sampling points. Nr = number of sampling points. Site area (ha) Nr Density (samples /ha) sampling design based on Sampling depth (cm) Sample area (m2) Soil sampling year SWE 1 259 83 1* targeted sampling satellite data 0-20 30 2015 SWE 2 56 51 1 targeted sampling satellite data 0-20 30 2015 SWE 3 44 38 1 targeted sampling satellite data 0-20 30 2010 SWE 4 1136 175 0.2 targeted sampling soil property maps 0-20 30 2005 SWE 5 25 122 5 regular grid - 0-20 30 2001 SWE 6 37 159 4 regular grid - 0-20 30 2001 DNK 1 0.9 53 60 regular grid - 0-20 20 2010 DNK 2 11 280 26 regular grid - 0-20 20 1992 DNK 3 0.6 37 59 regular grid - 0-20 20 2011 DNK 4 1.7 81 47 regular grid - 0-20 20 2011 DNK 5 1.6 88 54 regular grid - 0-20 20 2012 LTU 1 1.8 53 29 regular grid - 0-20 1 2021 LTU 2 1.5 47 31 regular grid - 0-20 1 2021 LTU 3 4.0 109 27 regular grid - 0-20 1 2021 LTU 4 1.9 56 30 regular grid - 0-20 1 2021 POL 1 101 32 0.3 transects - CZE 1 300 78 0.6* targeted sampling soil map and satellite data 0-20 6.25 2021 CZE 2 565 76 0.5* targeted sampling soil map and satellite data 0-20 6.25 2021 CZE 3 631 80 0.5 targeted sampling soil map and satellite data 0-20 6.25 2021 CZE 4 82 78 1 targeted sampling soil map and satellite data 0-20 cm 6.25 2021 FRA 1 13 28 2 targeted sampling soil map and satellite data 0-10 cm 2013 & 2015 CHE 1 0.7 22 33 regular grid - 0-20 cm 0.25 2022 CHE 2 1.4 16 11 regular grid - 0-20 cm 0.25 2022 CHE 3 2.8 35 13 regular grid - 0-20 cm 0.25 2021 CHE 4 0.6 20 31 regular grid - 0-20 cm 0.25 2022 CHE 5 0.8 20 25 regular grid - 0-20 cm 0.25 2021 ITA 1 12 18 2 targeted sampling soil map 0-30 cm 79 2018 ITA 2 12 18 2 targeted sampling soil map 0-30 cm 79 2018 ITA 3 20 54 3 targeted sampling soil map 0-30 cm 79 2018 ITA 4 20 54 3 targeted sampling soil map 0-30 cm 79 2018 ITA 5 27 53 2 targeted sampling soil map 0-30 cm 79 2018 TUR 1 477 120 0.3 evenly distributed - 0-20 cm 4 2021 TUR 2 104 82 1 evenly distributed - 0-20 cm 4 2021 ESP 1 114 50 1* regular grid - 0-10 cm 40 2022 * Sites with more than one field where the fields are not always next to one another. Sample density refers here to the density at the sampled fields. Table S3. Site information, SOC and soil texture analyses methods Site SOC analysis method soil texture analysis method SWE 1 Loss on ignition corrected for structural water in clay minerals, multiplied by a factor 0.58 wetsieving and sedimentation (SS ISO 11277 2001) SWE 2 Loss on ignition corrected for structural water in clay minerals, multiplied by a factor 0.58 wetsieving and sedimentation (SS ISO 11277 2001) SWE 3 Loss on ignition corrected for structural water in clay minerals, multiplied by a factor 0.58 wetsieving and sedimentation (SS ISO 11277 2001) SWE 4 Loss on ignition corrected for structural water in clay minerals, multiplied by a factor 0.58 wetsieving and sedimentation (SS ISO 11277 2001) SWE 5 Loss on ignition corrected for structural water in clay minerals, multiplied by a factor 0.58 wetsieving and sedimentation (SS ISO 11277 2001) SWE 6 Loss on ignition corrected for structural water in clay minerals, multiplied by a factor 0.58 wetsieving and sedimentation (SS ISO 11277 2001) DNK 1 combustion in a LECO induction furnace Pipette method (ISO 11277:2009) DNK 2 combustion in a LECO induction furnace Pipette method (ISO 11277:2009) DNK 3 combustion in a LECO induction furnace Pipette method (ISO 11277:2009) DNK 4 combustion in a LECO induction furnace Pipette method (ISO 11277:2009) DNK 5 combustion in a LECO induction furnace Pipette method (ISO 11277:2009) LTU 1 SOC content was determined byphotometric procedure at the wavelength of 590 nm using UV– VISspectrophotometer Cary (Varian) and using glucose as a standard. Pipette method (ISO 11277:2009) LTU 2 SOC content was determined byphotometric procedure at the wavelength of 590 nm using UV– VISspectrophotometer Cary (Varian) and using glucose as a standard. Pipette method (ISO 11277:2009) LTU 3 SOC content was determined byphotometric procedure at the wavelength of 590 nm using UV– VISspectrophotometer Cary (Varian) and using glucose as a standard. Pipette method (ISO 11277:2009) LTU 4 SOC content was determined byphotometric procedure at the wavelength of 590 nm using UV– VISspectrophotometer Cary (Varian) and using glucose as a standard. Pipette method (ISO 11277:2009) POL 1 Modified Tiurin's method Sedimentation, areometric Casagrande method (PN-ISO 11277) CZE 1 Walkley–Black Pipette method (ISO 11277:2009) CZE 2 Walkley–Black Pipette method (ISO 11277:2009) CZE 3 Walkley–Black Pipette method (ISO 11277:2009) CZE 4 Walkley–Black Pipette method (ISO 11277:2009) FRA 1 dry combustion (NF ISO 10694) pipette method (NF X31-107) CHE 1 dry combustion (NF ISO 13 878) Pipette method (NFX 31-107) CHE 2 dry combustion (NF ISO 13 878) Pipette method (NFX 31-107) CHE 3 dry combustion (NF ISO 13 878) Pipette method (NFX 31-107) CHE 4 dry combustion (NF ISO 13 878) Pipette method (NFX 31-107) CHE 5 dry combustion (NF ISO 13 878) Pipette method (NFX 31-107) ITA 1 EN 17505:2023 Dry combustion by thermal gradient (Temperature dependent differentiation of total carbon) ISO 11277:2020 ITA 2 EN 17505:2023 Dry combustion by thermal gradient (Temperature dependent differentiation of total carbon) ISO 11277:2020 ITA 3 EN 17505:2023 Dry combustion by thermal gradient (Temperature dependent differentiation of total carbon) ISO 11277:2020 ITA 4 EN 17505:2023 Dry combustion by thermal gradient (Temperature dependent differentiation of total carbon) ISO 11277:2020 ITA 5 EN 17505:2023 Dry combustion by thermal gradient (Temperature dependent differentiation of total carbon) ISO 11277:2020 TUR 1 Walkley–Black hydrometer method TUR 2 Walkley–Black hydrometer method ESP 1 Walkley–Black wetsieving and sedimentation (SS ISO 11277 2001) Figure S1. Temporal mosaics shown as RGB images (B4, B3 and B2) of the median and the R90 approach at A) the Polish site POL 1, B) the French site FRA 1, and C) the Spanish site ESP 1. Table S4. Performance of site-specific random forest models for predicting SOC and clay as determined in leave-one-out cross-validation. Predictors in the models for SOC include either satellite data only, satellite data and texture information, or texture information only. The best models (smallest RMSE) using either the median or R90 satellite data, as well as using either clay only or information on all three particle size fractions (clay, silt, sand; ‘allTex’) are listed. SOC predictions Clay predictions Satellite Satellite + texture Texture Satellite Site RMSE RPD RPIQ ccc RMSE RPD RPIQ ccc RMSE RPD RPIQ ccc RMSE RPD RPIQ ccc SWE 1 R90 0.58 1.82 1.23 0.81 R90 clay 0.60 1.78 1.20 0.80 clay 1.17 0.91 0.62 0.05 med 4.04 1.31 1.98 0.60 SWE 2 med 0.46 1.00 1.40 0.23 med allTex 0.42 1.09 1.53 0.38 allTex 0.45 1.01 1.41 0.39 R90 2.63 1.37 1.90 0.65 SWE 3 med 0.29 0.90 1.00 -0.03 med allTex 0.29 0.92 1.02 -0.02 allTex 0.27 0.97 1.07 0.10 med 5.07 2.72 5.03 0.93 SWE 4 med 0.36 1.10 1.46 0.39 med allTex 0.31 1.25 1.67 0.53 clay 0.40 0.98 1.31 0.18 med 6.14 1.75 2.60 0.80 SWE 5 med 0.48 1.16 1.30 0.50 med clay 0.36 1.52 1.71 0.73 clay 0.42 1.33 1.50 0.64 med 2.88 2.42 1.30 0.91 SWE 6 med 0.35 1.52 1.82 0.72 med clay 0.31 1.74 2.09 0.80 clay 0.42 1.26 1.51 0.59 med 4.70 1.40 1.38 0.67 DNK 1 med 0.07 1.54 1.94 0.73 med clay 0.07 1.55 1.96 0.74 clay 0.11 0.98 1.23 0.32 med 0.85 1.60 2.77 0.76 DNK 2 R90 1.09 1.58 0.22 0.75 med clay 0.91 1.89 0.26 0.84 allTex 1.69 1.02 0.14 0.24 med 1.23 1.41 2.18 0.67 DNK 3 med 0.57 2.87 2.64 0.93 med clay 0.56 2.91 2.68 0.93 allTex 1.64 0.99 0.91 0.28 R90 1.52 1.12 1.24 0.37 DNK 4 R90 0.11 1.19 1.64 0.48 R90 allTex 0.11 1.18 1.63 0.46 clay 0.12 1.06 1.47 0.43 med 1.47 0.94 1.00 0.00 DNK 5 R90 0.18 1.20 1.58 0.52 R90 clay 0.16 1.29 1.71 0.60 clay 0.21 1.02 1.34 0.40 R90 0.32 1.25 1.62 0.56 LTU 1 med 0.19 1.19 1.59 0.52 med allTex 0.17 1.33 1.77 0.61 allTex 0.19 1.21 1.61 0.54 R90 2.01 1.09 1.64 0.33 LTU 2 med 0.21 0.99 1.21 0.17 med allTex 0.21 0.95 1.17 0.17 allTex 0.23 0.87 1.07 0.00 med 2.31 0.90 0.97 -0.03 LTU 3 med 0.43 1.13 1.00 0.41 med allTex 0.41 1.19 1.06 0.54 allTex 0.42 1.17 1.04 0.54 med 2.85 1.01 1.51 0.16 LTU 4 med 0.25 0.97 1.23 0.11 R90 clay 0.19 1.27 1.61 0.57 clay 0.20 1.17 1.48 0.60 R90 3.23 1.06 1.76 0.34 POL 1 med 0.24 1.26 1.17 0.57 med clay 0.22 1.41 1.30 0.64 allTex 0.25 1.21 1.12 0.53 med 1.95 1.06 0.95 0.25 CZE 1 R90 0.46 0.91 0.83 0.22 R90 allTex 0.45 0.91 0.84 0.23 clay 0.47 0.88 0.81 0.13 R90 4.46 0.90 1.26 -0.12 CZE 2 R90 0.19 1.20 1.52 0.56 med allTex 0.18 1.30 1.64 0.63 allTex 0.22 1.08 1.37 0.42 med 2.58 1.83 2.20 0.82 CZE 3 R90 0.16 1.20 1.81 0.52 R90 clay 0.17 1.15 1.75 0.48 allTex 0.20 1.00 1.51 0.22 R90 3.65 1.14 0.88 0.45 CZE 4 R90 0.29 1.01 1.11 0.21 R90 allTex 0.29 1.02 1.12 0.21 clay 0.33 0.90 0.98 0.11 R90 4.01 0.92 1.00 0.04 FRA 1 med 0.26 1.26 1.53 0.60 med clay 0.26 1.27 1.54 0.60 clay 0.29 1.16 1.41 0.57 med 1.70 2.41 4.63 0.90 CHE 1 med 0.12 1.27 2.13 0.60 R90 allTex 0.09 1.63 2.72 0.77 allTex 0.10 1.46 2.44 0.73 R90 2.24 1.43 2.36 0.71 CHE 2 med 0.10 1.57 2.13 0.70 med clay 0.10 1.60 2.17 0.73 allTex 0.12 1.37 1.86 0.65 med 1.99 1.67 2.21 0.76 CHE 3 R90 0.26 1.04 1.49 0.33 R90 allTex 0.23 1.15 1.65 0.43 allTex 0.23 1.16 1.66 0.49 med 2.81 1.03 1.21 0.27 CHE 4 med 0.33 1.80 1.17 0.81 R90 clay 0.24 2.44 1.60 0.90 clay 0.27 2.19 1.43 0.89 med 2.66 1.89 2.40 0.83 CHE 5 R90 0.27 1.49 0.89 0.66 R90 clay 0.24 1.67 1.00 0.73 clay 0.22 1.88 1.12 0.80 R90 1.86 1.56 1.11 0.71 ITA 1 R90 0.13 0.96 1.43 0.22 med clay 0.13 1.00 1.49 0.19 clay 0.14 0.93 1.39 0.30 R90 3.39 1.29 1.44 0.63 ITA 2 R90 0.09 1.17 1.48 0.44 R90 clay 0.09 1.18 1.49 0.43 clay 0.10 1.12 1.41 0.41 med 3.41 1.17 1.23 0.45 ITA 3 med 0.18 1.32 1.86 0.64 med clay 0.16 1.48 2.08 0.70 allTex 0.20 1.18 1.66 0.56 R90 4.37 1.60 2.17 0.76 ITA 4 R90 0.06 1.04 1.53 0.29 R90 clay 0.06 1.06 1.55 0.29 allTex 0.08 0.87 1.27 -0.05 R90 1.64 1.06 1.51 0.24 ITA 5 med 0.18 1.09 1.42 0.40 med clay 0.18 1.11 1.44 0.41 allTex 0.21 0.95 1.24 0.19 med 3.47 1.74 2.04 0.81 TUR 1 R90 0.24 0.94 1.25 0.05 R90 allTex 0.21 1.05 1.39 0.28 allTex 0.24 0.93 1.24 0.19 med 6.15 1.74 2.55 0.81 TUR 2 R90 0.14 1.33 1.90 0.66 med clay 0.11 1.70 2.43 0.79 allTex 0.13 1.42 2.04 0.69 R90 6.15 1.33 1.53 0.65 ESP 1 med 0.12 1.40 1.52 0.67 med allTex 0.11 1.45 1.57 0.66 allTex 0.17 0.97 1.05 0.23 R90 4.48 1.44 1.73 0.68 Median 0.24 1.19 1.47 0.52 0.21 1.28 1.58 0.60 0.22 1.04 1.36 0.41 2.83 1.35 1.63 0.65 Mean 0.28 1.28 1.45 0.48 0.25 1.40 1.58 0.55 0.35 1.14 1.32 0.39 3.07 1.43 1.86 0.54 Figure S2. A) Performance of site-specific random forest models for predicting SOC from satellite data. B) Performance of site-specific random forest models for predicting clay from satellite data. RPD and ccc were determined in leave-one-out cross-validation. The results are shown for the best models (smallest RMSE) using either the median or R90 satellite data (corresponding to Table S4). The dotted lines indicate the performance of the dummy models, which are simple baseline models, always predicting the mean of the measured SOC or clay content in the training data (ignoring the satellite information). The solid lines are only eye-guides. The order of sites (x-axis) corresponds to Figigures 3 A, 3B, 4 A, and 5 in the paper. Figure S3. Relative importance of predictors in the PLSR models for predicting SOC and clay from satellite data (median or R90, whichever resulted in the smallest RMSE for each site). A) For models predicting SOC. B) For models predicting clay content. C) The difference between the relative importances for models predicting SOC and for models predicting clay content (difference of data shown in panel A and B). Sites are sorted from top to bottom according to decreasing prediction performance (ccc) of the SOC models. The dendrogram shows the redundancy structure of the set of predictors. It is based on a (complete-linkage) hierarchical clustering of the median satellite dataset with distance between the bands defines as 1-|r| (Pearson).