Soil databases and the problem of establishing regional biogeochemical trends

Regional and global environmental modeling depend on soil data for input layers or parameterization. However, randomly located observations, such as provided by agricultural databases, are not always representative of trends identified in field studies conducted under carefully controlled conditions. Many researchers lament the paucity of soil profile data in Amazônia, and suggest that given more data, regional studies would more closely approximate field research results. We assess the ability of a well‐populated regional database collected in the southwestern Brazilian Amazon to reproduce expected biogeochemical trends associated with forest clearing and pasture establishment, and explore the ramifications of relying on independently collected soil data for regional modeling. The Soteron database includes analyses of approximately 3000 soil cores collected for zoning purposes in the state of Rondônia. Pasture ages were determined from a time series of Landsat TM images classified using spectral mixture analysis. Although regional averages showed some of the temporal trends expected based on field study results (e.g. increase in pH following forest clearing), the trends were not statistically significant. Stratification by precipitation and other variables showed pasture age to be important but difficult to separate from other potential controls on soil conditions, mainly because of the reduced number of observations in each stratum. Using multiple regression, which permitted the inclusion of all potential explanatory factors and interactions, pasture age was shown to be a statistically significant predictor of soil conditions. However, the expected temporal sequence of changes documented by field chronosequence studies could not be reproduced. Properties dominated by large‐scale environmental gradients – pH, sum of base cations, aluminum saturation, and exchangeable calcium – were moderately well modeled, while those more strongly linked to dynamic spatially heterogeneous processes such as biological cycling and land management, particularly organic carbon and nitrogen, could not be modeled. Management‐induced soil changes occur at too fine a scale to be captured by most maps, and the relative changes are small compared with spatial heterogeneity caused by controls on soil development over large regions. Therefore, regardless of whether chronosequence‐derived models of biogeochemical response to land‐cover change are correct, the results of these models will not lead to spatially explicit maps that can be validated by regional reconnaissance, nor will they facilitate realistic predictions of the regional biogeochemical consequences of land‐cover change. The change from local to regional scale entails a change in the relative importance of processes controlling soil property behavior.