Chapter 37 Regional Organic Carbon Storage Maps of the Western Brazilian Amazon Based on Prior Soil Maps and Geostatistical Interpolation

Abstract The study area was defined as 6°-square portion of the Brazilian Amazon, from longitudes 60–66° West and latitudes 8–14° South. This region corresponds to ∼334,000 km 2 of the western Brazilian Amazon, that is 6.7% of the Brazilian Amazon basin. A database was constructed from soil profile information reported in published and unpublished soil surveys. Bulk density (BD) values were derived from multiple regressions with other available parameters. Two digitised soil maps were considered: a 1:5,000,000 soil map and 1:1,000,000 soil map of the region that was digitised for the purpose of this study. Individual C stock for each profile was calculated by two procedures. The first and classical procedure consists in summing C densities by horizon, determined as a product of BD, C concentration, and horizon thickness. The second calculation used a double exponential model of organic C vertical distribution in soils: (C(X) − C2)/(C1 – C2)=(e −b X − e −b X2 )/(e −b X1 − e −b X2 ), where X and C(X) are the depth and the C content, X1 and C1 the depth and the C content of a fixed upper position, and X2 and C2 for a fixed deeper position. Regional C estimates were derived using different map approaches or geostatistical interpolations. Geostatistical interpolations were run not only on the individual C stock, but also on the parameters of the vertical model separately. Two types of validation were conducted, an “internal” validation known as “jackknifing” and an external validation with preserved data (about 10%). The regional stock was obtained multiplying the carbon content of each taxonomic unit by its area. This calculation leads to similar results using the soil map either at 1:5,000,000 scale or at 1:1,000,000. Regional stocks (0–100 cm) ranged between 2100 Tg using median values, and 2300 Tg, using means, and exhibit an associated error (based on SD) of 900 Tg. Variographic analysis showed that carbon stocks exhibit a spatial structure at the regional scale. However the high nugget effect reveals that about half of the spatial variability may appear within distances less than a few tens of kilometres. Results of the fitting of the experimental semivariogram for the top metre showed the same pattern with a nugget effect of 4.70 (kg C m −2 ) 2 and a sill value of 4.26 (kg C m −2 ) 2 , but the range is reduced to 513 km. Geostatistical interpolations run either on the C stocks values or independently on each parameters of the vertical model led to similar results.