High-resolution digital mapping of soil organic carbon in permafrost terrain using machine-learning: A case study in a sub- Arctic peatland environment Matthias B. Siewert12 1Department of Physical Geography, Stockholm University, Stockholm, SE 106 91, Sweden 2New Affiliation: Department of Ecology and Environmental Science, Ume( University, Ume(, SE 901 87, Sweden Correspondence to: Matthias B. Siewert (matthias.siewert,umu.se. Supplementary material TableA. 1: Description of the individual land cover classes adapted from Andersson (2016) /and cover class Description Bare ground E0posed 1edrock, 1lockfields, rock fragments covered ground or 1arren unvegetated soil with no or very little vegetation cover2 vegetation includes mosses in patches of few centimeters in diameter and lichens. Alpine heath tundra Areas a1ove the tree line 34500 m a.s.l., covered 1y low growing vegetation from few centimeters to few decimeter height, often alternating with e0posed 1are soil or rocks2 low growing deciduous and evergreen shru1s, including Betula nana, mpetrum sp., mosses and lichens. Can 1e associated with cryotur1ation landforms such as solifluction lo1es and mud 1oils. Alpine willow Alpine areas a1ove the Birch tree line covered 1y low growing willow (Salix sp. shru1s mostly confined to wind protected depressions along river pathways with wet soil conditions and micro climate. Birch forest 7orest dominated 1y mountain 1irch (Betula pubescens ssp. c$erepanovii.2 tree height is 2815 m2 trees can have single stems (monocormic. or grow with many stems divided near the ground (polycormic.2 1irch forest is restricted to areas 1elow 4600 m altitude. Sparse 1irch forest /ower elevation areas with spread out clusters of 1irch trees (Betula pubescens ssp. C$erepanovii. over 1oulder fields with disrupted soil cover. Dwarf shru1 Areas covered 1y small shru1s up to 450 cm high. Mainly in su1alpine terrain at the transition from 1irch forest to alpine heath tundra, 1ut also ad9acent to lowland wetland areas over 1oulder terrain. 7orested wetland 7orested areas with trees higher than 2 m (Betula pubescens ssp. C$erepanovii and Salix sp.. and seasonally waterlogged ground conditions. The ground has significant peat accumulation. On alluvial fans fern can cover the ground. /owland shru1 7en areas with su1stantial growth of dwarf shru1s, mostly low growing willow (Salix sp.)% ground wetland cover is tall graminoids and Sphagnum sp. 7ound close to streams and rivers. Sphagnum wetland Waterlogged fen areas with continuous Sphagnum sp. ground cover and some graminoids. No distinction 1etween om1rotrophic and minerotrophic areas was made. Sedge wetland Waterlogged wetland areas with standing water and sedges ( riophorum angustifolium and Carex spp). Often transitions into lakes and ponds. Mostly minerotrophic areas surrounding the peat plateaus or small wetlands surrounded 1y 1irch forest. Peat Bog Dry elevated om1rotrophic wetland areas, mostly in form of palsa plateaus elevated 1y the occurrence of permafrost2 vegetation includes growing evergreen and deciduous shru1s such as mpetrum hermaphroditum and Betula nana, Rubus chamaemorus, mosses and lichens. Water Water 1odies with a minimum area of 130 m> including rivers, ponds and lakes. Including Tornetr@sk lake2 masked layer. Artificial surfaces Street, railway and 1uildings including a 1uffer margin2 masked layer. TableA. 2: Confusion matrix of the land cover classification based on 108 ground control points. Alpine Sparse /owland Samples Bare heath Dwarf Alpine Birch Birch 7orested Sedge shru1 Sphagnum Peat (N: rock tundra shru1 willow 7orest 7orest wetland wetland wetland wetland 1og 108. Bare rock 5 3 1 0 0 0 0 0 0 0 0 9 Alpine heath tundra 0 6 0 0 0 0 0 0 0 0 0 6 Dwarf shru1 1 1 A 1 0 0 0 0 0 0 0 7 Alpine willow 0 1 1 3 1 1 0 0 0 0 0 7 Sparse Birch 7orest 0 0 1 0 11 0 0 0 0 0 0 12 Birch 7orest 0 0 1 1 A 21 0 0 0 0 0 27 7orested wetland 0 0 0 0 0 0 6 0 0 0 0 6 Sedge wetland 0 0 1 0 0 0 0 7 0 1 3 12 /owland shru1 wetland 0 0 0 0 0 0 1 0 A 0 0 5 Sphagnum wetland 0 0 0 0 0 0 0 0 0 2 2 A Peat 1og 1 0 0 0 0 0 0 0 0 1 11 13 Overall Accuracy 7AB Cappa 0.71 Fig.A. 1: Variable importance plots for individual random forests prediction models of SOC,30, SOC 0,100, SOC, O. and O.,Depth measured in mean decrease in accuracy (0 1nc2S ). .