Evaluation of vertical monitoring strategies to predict soil hydraulic characteristics and water contents by inverse modeling

Having a numerical model able to predict soil water content correctly is a very useful tool for many different objectives. However, it depends on the correct election of the soil hydraulic properties (SHP) on which the simulations are based. Measuring SHP in laboratory is time and economic-consuming and their inference by soil water monitoring and inverse modelling is a smart alternative. 

However, the resources needed to obtain copious data are sometimes unavailable and questions arise regarding what is the best monitoring strategy that let to obtain the best SHP with the fewest number of sensors.  When null or scarce data is present for some soil layers several solutions of the same problem are encountered. SHP estimations by inverse modeling could vary according to the data available and the vertical distribution of the measurement points.  The aim of this work is to evaluate different monitoring strategies to obtain an accurate hydraulic model with a limited number of observations depths. For this purpose, data monitored in an experimental plot in Bahía Blanca (Argentina) was used to run several inverse numerical simulations with the HYDRUS software.  Several scenarios of available data were considered: (1) six monitoring depths (6-MD) (30 cm, 60 cm, 90 cm, 120 cm, 150 cm, and 180 cm); (2) five monitoring depths (5-MD) subtracting the information from one soil depth at a time; (3) four monitoring depths (4-MD) subtracting the information from two soil depths, simultaneously. Each depth included soil water content, ϴ, and soil pressure head, h, measurements.

The best fit was achieved with the 6-MD strategy. The Nash-Sutcliffe coefficient of efficiency (EF) were 0.784 and 0.665 for the ϴ and h, respectively. Besides, the relative root mean square error (rRMSE) was 0.134 for ϴ and 0.127 for h. For the 5-MD strategy the best performance was achieved by removing the information from depths of 90 cm, 120 cm, or 150 cm. In those cases, EF was between 0.715-0.717 and rRMSE ranged from 0.132-0.133. Statistics reported a worse fit when removing data from the upper and the lower layers. For the 4-MD strategy, the best performance was accomplished by suppressing data from 90 cm and 120 cm (EF=0.707; rRMSE=0.135).

The observation points that had less weight in parameter prediction corresponded to the intermedium vadose zone. If data from the upper and lower boundaries of the soil profile are available, ϴ and h from the middle section could be predicted reasonably well, anyway. The inversely model SHP from the 5-MD and 4-MD strategies correctly represent field retention data points θ (h). If the optimal monitoring depths are recognized, the time, cost, and labor needed to a correctly soil manage practice will be greatly reduced.