Pedotransfer functions to estimate water retention parameters of soils in northeastern Brazil

Abstract

SUMMARYPedotransfer functions (PTF) were developed to estimate the parameters (α α α α α, n, θ θ θ θ θr and θ θ θ θ θs) of the van Genuchten model (1980) to describe soil water retention curves. The data came from various sources, mainly from studies conducted by universities in Northeast Brazil, by the Brazilian Agricultural Research Corporation (Embrapa) and by a corporation for the development of the São Francisco and Parnaíba river basins (Codevasf), totaling 786 retention curves, which were divided into two data sets: 85 % for the development of PTFs, and 15 % for testing and validation, considered independent data. Aside from the development of general PTFs for all soils together, specific PTFs were developed for the soil classes Ultisols, Oxisols, Entisols, and Alfisols by multiple regression techniques, using a stepwise procedure (forward and backward) to select the best predictors. Two types of PTFs were developed: the first included all predictors (soil density, proportions of sand, silt, clay, and organic matter), and the second only the proportions of sand, silt and clay. The evaluation of adequacy of the PTFs was based on the correlation coefficient (R) and Willmott index (d). To evaluate the PTF for the moisture content at specific pressure heads, we used the root mean square error (RMSE). The PTFpredicted retention curve is relatively poor, except for the residual water content. The inclusion of organic matter as a PTF predictor improved the prediction of parameter α α α α α of van Genuchten. The performance of soil-class-specific PTFs was not better than of the general PTF. Except for the water content of saturated soil estimated by particle size distribution, the tested models for water content prediction at specific pressure heads proved satisfactory. Predictions of water content at pressure heads more negative than -0.6 m, using a PTF considering

Development and Uncertainty Assessment of Pedotransfer Functions for Predicting Water Contents at Specific Pressure Heads

“…The GP‐PTF was not able to predict the wilting point under covariate shift with the same accuracy as the other methods. The obtained RMSE of 0.020 m 3 m −3 is similar to the results of van den Berg et al (1997) and Barros et al (2013), who developed PTFs for Brazilian soils. The important feature of regression‐based methods is the similarity of performance prediction between training and testing stages, a feature not observed in ENS and explained by overtraining.…”

“…The PTFs developed by Tomasella et al (2003) for Brazilian soils predicted θ 0.6 with an RMSE of 0.046 m 3 m −3 but require detailed information about the particle size distribution (fine and coarse sand fractions), which is not always available. The RMSE values of 0.046 and 0.050 m 3 m −3 for the prediction of this water content were also obtained by Barros et al (2013) using soil samples from the northeast of Brazil. For sandy and clayey soils from different parts of Brazil, da Silva et al (2017) reported RMSE values of about 0.030 and 0.050 m 3 m −3 obtained by predictions from a semi‐deterministic PTF.…”

“…Data from samples collected in the northeastern region of Brazil (Brazil‐NE, Fig. 1a) were retrieved from Barros et al (2013) and a dataset of Embrapa, the Brazilian Agricultural Research Corporation. The database comprised 838 samples with water contents determined at −0.6, −1, −3, −5, −10, −20, and −150 m pressure head (θ 0.6 , θ 1 , θ 3 , θ 5 , θ 10 , θ 20 , θ 150 , respectively), particle size fractions (sand, silt, and clay contents), OM, and BD.…”

Pedotransfer functions to estimate retention and availability of water in soils of the state of Santa Catarina, Brazil

“…In Brazil, the generation of point PTFs has already been studied for soils in the Amazon region (Tomasella & Hodnett, 1998) and of the States of São Paulo (Arruda et al, 1987), Pernambuco (Oliveira et al, 2002) and Rio Grande do Sul (Reichert et al, 2009), as well as the generation of parametric PTFs for estimation of water retention using samples from different soils in northeastern Brazil (Barros et al, 2013) and the generation of point and parametric PTFs for estimation of water retention from samples from several Brazilian States (Tomasella et al, 2003). Nevertheless, the PTFs for Brazilian soils were generated using a database composed predominantly of soil samples from the Southeast and North of Brazil.…”

Evaluating pedotransfer functions of the Splintex model

“…However, the precision and accuracy of PTF1 estimates evaluated by Barros et al . () were larger for θ r with r = 0.60 and RMSE = 0.0671 m 3 m −3 compared with r = 0.47 and RMSE = 0.0731 m 3 m −3 for PTF2 (Figure c). This could be explained by the fact that the measured θ (1.0 m) was used in PTF2.…”