Tanzania is one of the countries that has embarked on a national programme under the United Nations collaborative initiative on Reducing Emissions from Deforestation and forest Degradation (REDD). ...Tanzania is currently developing the capacity to enter into a carbon monitoring REDD+ regime. In this context spatially representative soil carbon datasets and accurate predictive maps are important for determining the soil organic carbon pool. The main objective of this study was to model and map the SOC stock for the 0–30-cm soil layer to provide baseline information for REDD+ purposes. Topsoil data of over 1400 locations spread throughout Tanzania from the National Forest Monitoring and Assessment (NAFORMA), were used, supplemented by two legacy datasets, to calibrate simple kriging with varying local means models. Maps of SOC concentrations (g kg−1) were generated for the 0–10-cm, 10–20-cm, 20–30-cm, 0–30-cm layers, and maps of bulk density and SOC stock (kg m−2) for the 0–30-cm layer. Two approaches for modelling SOC stocks were considered here: the calculate-then-model (CTM) approach and the model-then-calculate approach (MTC). The spatial predictions were validated by means of 10-fold cross-validation. Uncertainty associated to the estimated SOC stocks was quantified through conditional Gaussian simulation. Estimates of SOC stocks for the main land cover classes are provided. Environmental covariates related to soil and terrain proved to be the strongest predictors for all properties modelled. The mean predicted SOC stock for the 0–30-cm layer was 4.1 kg m−2 (CTM approach) translating to a total national stock of 3.6 Pg. The MTC approach gave similar results. The largest stocks are found in forest and grassland ecosystems, while woodlands and bushlands contain two thirds of the total SOC stock. The root mean squared error for the 0–30-cm layer was 1.8 kg m−2, and the R2-value was 0.51. The R2-value of SOC concentration for the 0–30-cm layer was 0.60 and that of bulk density 0.56. The R2-values of the predicted SOC concentrations for the 10-cm layers vary between 0.46 and 0.54. The 95% confidence interval of the predicted average SOC stock is 4.01–4.15 kg m−2, and that of the national total SOC stock 3.54–3.65 Pg. Uncertainty associated with SOC concentration had the largest contribution to SOC stock uncertainty. These findings have relevance for the ongoing REDD+ readiness process in Tanzania by supplementing the previous knowledge of significant carbon pools. The soil organic carbon pool makes up a relatively large proportion of carbon in Tanzania and is therefore an important carbon pool to consider alongside the ones related to the woody biomass. Going forward, the soil organic carbon data can potentially be used in the determination of reference emission levels and the future monitoring, reporting and verification of organic carbon pools.
•SOC stock was mapped and uncertainty assessed using a recent, nationwide dataset.•Two mapping methods were compared: calculate-then-model and model-then-calculate.•Average SOC stock was is 4.1 kg m−2, translating to a total stock of 3.6 Pg.•Uncertainty about SOC concentration contributed most to SOC stock uncertainty.•The SOC maps provide baseline information to prepare Tanzania for REDD+.
Inadequacy of spatial soil information is one of the limiting factors to making evidence-based decisions to improve food security and land management in the developing countries. Various digital soil ...mapping (DSM) techniques have been applied in many parts of the world to improve availability and usability of soil data, but less has been done in Africa, particularly in Tanzania and at the scale necessary to make farm management decisions. The Kilombero Valley has been identified for intensified rice production. However the valley lacks detailed and up-to-date soil information for decision-making. The overall objective of this study was to develop a predictive soil map of a portion of Kilombero Valley using DSM techniques. Two widely used decision tree algorithms and three sources of Digital Elevation Models (DEMs) were evaluated for their predictive ability. Firstly, a numerical classification was performed on the collected soil profile data to arrive at soil taxa. Secondly, the derived taxa were spatially predicted and mapped following SCORPAN framework using Random Forest (RF) and J48 machine learning algorithms. Datasets to train the model were derived from legacy soil map, RapidEye satellite image and three DEMs: 1 arc SRTM, 30m ASTER, and 12m WorldDEM. Separate predictive models were built using each DEM source. Mapping showed that RF was less sensitive to the training set sampling intensity. Results also showed that predictions of soil taxa using 1 arc SRTM and 12m WordDEM were identical. We suggest the use of RF algorithm and the freely available SRTM DEM combination for mapping the soils for the whole Kilombero Valley. This combination can be tested and applied in other areas which have relatively flat terrain like the Kilombero Valley.
•RF is less sensitive to training set sampling intensity than J48 algorithms.•Soil taxa predictions from 1 arc SRTM and 12m WordDEM are identical.•RF and SRTM combination is suggested to predict soil taxa in Kilombero valley.