http://bolin.su.se/data/wagner-2024-soc-1 Julia Wagner, Victoria Martin, Niek J. Speetjens, Willeke A’Campo, Luca Durstewitz, Rachele Lodi, Michael Fritz, George Tanski, Jorien E. Vonk, Andreas Richter, Annett Bartsch, Hugues Lantuit, Gustaf Hugelius Bolin Centre Database 2024 Datafile Terrestrial Soil Random forest Machine learning Soil organic carbon Tundra Permafrost Earth science > Land surface > Soils > Carbon Julia Wagner 2024-03-12T12:30:42+00:00 English 1 Zip-file containing four directories. ##### Sample data Sample data is provided in form of the `wagner-2024-soc-1.xlsx` spreadsheet for the sampling locations from the field campaigns in 2018 at Ptarmigan Bay and 2019 at Komakuk Beach. The data covers the full depth up to 1 m. Depth increments without sample data available were estimated. - `Spreadsheets_soil_data/PB_KB_soil_data.xlsx` - `ID` site ID - `sample_ID` sample ID - `AL_1_PF_2` sample from the Active Layer = 1, sample from the Permafrost = 2 - `OL_3yes_0no` sample from the organic layer = 3, not organic = 0 - `sample_estimated` values basted on sample = 1, values estimated = 3 (interpolated/ estimated), only values for bulk density (BD) estimated = 3 - `sample_CN_1yes2no` sample for carbon : nitrogen ratio (CN) measurements was available - `sample_BD1yes2no` sample for bulk density was available - `comment_sample_material_CN` comment on the sample material - `comment_CN` commenton the value for CN when no sample was available - `comment_BD` commenton the value for BD when no sample was avaiable - `BD` dry bulk density at 65 °C - `diff_bottom_top` thickness of the sampling depth in centimetre - `top_cm` top depth of the sampling depth in centimetre - `bottom_cm` bottom depth of the sampling depth in centimetre - `N_content` nitrogen content [%] - `C_content` carbon content [%] - `CN_ratio` C:N ratio - `SOC_stocks` soil organic carbon stocks [Kg / m²] - `N_stocks` nitrogen stocks [Kg / m²] - `d_13C_12C` ratio of 13C and 12C - `d_15N_14N` ratio of 15C and 14N - `water_content` water content/ ice content for permafrost samples, simple ratio between weight of dry and fresh material (PB_2018_soil_data: water content given as fraction of %, KB_2019_soil_data water content given in %) ##### Sampling sites The sampling locations are provided as shapefiles for the two study sites Komakuk Beach (KB) and Ptarmigan Bay (PB): a total of 46 sampling locations at KB and 40 at PB. - `sampling_locations` - `KB_sampling_locations` - `PB_sampling_locations` - `KB_sampling_locations_additional.shp` for samples that were taken in the field, but were not used in the study Wagner et al. 2023. The attribute table contains the sampling Date in the format YYYY-MM-DD ##### Models Random Forest models for each depth increment were created in R based on environmental variables and the target variables SOC and N (for more information see Wagner et al. 2023). The final models for each depth increment were saved as rds-files. `models` contains the random forest models created in R to be used within the package randomForest that is implemented in the package caret (Kuhn 2022) for the two study areas - `area_Ptarmigan_Bay` Ptarmigan Bay (PB) - `N-stocks` Nitrogen stocks - `SOC-content` soil organic carbon content (SOC content) - `SOC-stocks` soil organic carbon stocks - `SOC-stocks_SOC-stocks_above_as_variable` contains the models for SOC stock maps where predicted SOC stocks (all layers accumulated) - `area_Komakuk_Beach` Komakuk Beach (KB) - `N-stocks` Nitrogen stocks - `SOC-content` soil organic carbon content (SOC content) - `SOC-stocks` soil organic carbon stocks - `SOC-stocks_SOC-stocks_above_as_variable` contains the models for SOC stock maps where predicted SOC stocks (all layers accumulated) for the depth intervals 0-5, 5-15, 15-30, 30-60 and 60-100 cm. Structure of the files e.g.: `PB_model_0_05_N_stocks.rds` with area-code (PB, KB), model depth interval and soil property (N_stocks, SOC_stocks, SOC_content) ##### Predictions Predictions were made using random forest models based on the target variable and environmental variables (for more information see Wagner et al. 2023). The predictions are 2 m raster resolution saved in tiff format and in WGS 84 / UTM zone 7N ([EPSG: 32607](https://epsg.io/32607)). `outputs/Predictions` contains the random forest maps created in R with the package randomForest implemented in the caret package for the two study areas - `area_Ptarmigan_Bay` Ptarmigan Bay (PB) - `N-stocks` Nitrogen stocks in kg / m² - `cubicm` Nitrogen stocks in kg / m³ - `SOC-content` soil organic carbon content (SOC content) in % - `SOC-stocks` soil organic carbon stocks in kg / m² - `cubicm` soil organic carbon stocks in kg / m³ - `SOC-stocks_SOC-stocks_above_as_variable` contains the models for SOC stock maps where predicted SOC stocks in % (all layers accumulated) - `area_Komakuk_Beach` Komakuk Beach (KB) - `N-stocks` Nitrogen stocks in kg / m² - `cubicm` Nitrogen stocks in kg / m³ - `SOC-content` soil organic carbon content (SOC content) in % - `SOC-stocks` soil organic carbon stocks in kg / m² - `cubicm` soil organic carbon stocks in kg / m³ - `SOC-stocks_SOC-stocks_above_as_variable` contains the models for SOC stock maps where predicted SOC stocks in % (all layers accumulated) at depth intervals 0 – 5, 5 – 15, 15 – 30, 30 – 60 and 60 – 100 cm. Structure of the files e.g.: `KB_prediction_model_N_stocks_1_0_05.tif` with the area-code (PB, KB), model soil property (N_stocks, SOC_stocks, SOC_content) and deph interval (e.g. 0 – 05,5 – 15, etc.). The folders containing the `SOC stocks` contain a subfolder called "cubicm" where kg / m² were converted into kg / m³ for better comparison (as in Wagner et al. 2023) . ##### Uncertainty assessment Uncertainty estimations were created based on quantile regression forest (Yigini et al. 2018). The files are 2 m raster resolution saved in tiff format and in WGS 84 / UTM zone 7N ([EPSG: 32607](https://epsg.io/32607)). `outputs/Uncertainty_maps` contains the uncertainty maps created in R using quantile regression forest (workflow of Yigini et al 2018, "Soil Organic Carbon Mapping Cookbook") for the two study areas - `area_Ptarmigan_Bay` Ptarmigan Bay (PB) - `N-stocks` Nitrogen stocks - `SOC-content` soil organic carbon content - `SOC-stocks` soil organic carbon stocks - `area_Komakuk_Beach` Komakuk Beach (KB) - `N-stocks` Nitrogen stocks - `SOC-content` soil organic carbon content - `SOC-stocks` soil organic carbon stocks at depth intervals 0 – 5, 5 – 15, 15 – 30, 30 – 60 and 60 – 100 cm. Structure of the files e.g.: `KB_uncertainty_N_stocks_1_0_05.tif` for the area-code (PB, KB), model uncertainty (N_stocks, SOC_stocks, SOC_content) and deph interval (e.g. 0 – 05,5 – 15, etc.) All tif-files are in a WGS 84 UTM Zone 7N projection. This study is part of the [Nunataryuk project](https://www.nunataryuk.org). The collected soil data contributes towards a better understanding of the sparsely sampled permafrost region in general, may contribute to global databases of soil properties and may enhance future digital soil mapping efforts. Maps of predicted SOC and N stocks contribute to a better understanding of small scale distribution of SOC and nitrogen stocks and may be used for further studies that estimate potential release of SOC and N from permafrost affected soils. The sites were sampled by horizon from soil pits of the active layer. The permafrost was cored in 10 cm depth increments by using a steel pipe.