Estimating the Hydraulic Properties of Two Soil Profiles Under Pasture and Forest Land Use in Southeastern Amazonia.

Simulating water flow in soils usually requires information on soil hydraulic properties which are difficult and costly to measure directly. Inverse modeling techniques have been used as a convenient way to estimate soil hydraulic properties, particularly when sufficient field data is available. However, data required by many inverse modeling techniques is very limited in remote regions including the Amazon Biome. One way to overcome this problem is to use sparse data efficiently to obtain distributions of soil hydraulic parameters. Estimating soil hydraulic parameters with a Markov chain Monte Carlo (MCMC) approach has several advantages compared to other inversion techniques. The MCMC approach provides posterior distributions of parameters that are independent of the initial values and allows for uncertainty analyses. The objectives of this study are: 1) apply the Differential Evolution Adaptive Metropolis (DREAM) algorithm to determine distributions of soil hydraulic parameters from sparse in-situ data at two sites (pasture and forest) located in the southeastern Amazon region, and 2) generate the hydraulic parameters from the distributions for other sites with similar soils but lacking field data. The DREAM algorithm was used to fit volumetric water content (VWC) and tension observations (h) obtained from several soil depths at each of the two sites. The VWC and h data were collected after rainfall events during a 207-day period. DREAM executed HYDRUS-1D 10,000 times and the posterior distributions of each hydraulic parameter along with the 95% confidence intervals (CI) of the VWC and h time series were calculated from the last 25% of the runs. These posterior distributions were also used to generate hydraulic parameters for two nearby toposequences with similar soils, totaling 6 soil profiles under forest (3) and under pasture (3). HYDRUS-2D models were built for both forest and pasture toposequences and were run 5,000 times using randomly selected parameters from the posterior distributions. The new set of parameters for the toposequences were accepted when the HYDRUS-2D model converged and the predicted tension time series felt within the 95% CI of the tension time series from the calibration sites. The effects pasture and forest degradation scenarios will be investigated using the estimated parameters of each profile.