Modelling of water reserves in mineral soils with different retention properties

Abstract

This work focuses on modelling soil water reserves using an Artificial Neural Network (ANN). Four model variants were established based on 843 records (verified through 268 measurements) of soil water content (SWC) measured at full-scale field sites located in Southwest Poland. It is revealed that commonly recorded climatic data (precipitation and temperature) linked with SWC and field water capacity (FWC) are applicable in the ANN modelling. The basic model (utilising the meteorological data) was the most suitable for soil profiles with thicknesses of 0-25 cm, while in profiles with thicknesses of 0-50 cm and 0-100 cm the comprehensive ANN model (linking climatic data, FWC and SWC) was the most appropriate. Furthermore, comparative studies of the measured and modelled data indicated their statistical convergence, thus providing support for the practical implementation of the proposed ANN modelling.

Assessing the applicability of groundwater monitoring data in the modelling of soil water retention characteristics

“…The evaluation was completed based on summing up the points scored for each soil thickness, as presented in Table . The ranking analysis revealed that the best accuracy (top score with 7 ranking points) for each of the three soil thicknesses was defined by three different model variants, and two of them were the newly developed models E and G. Yet, the basic model D was the most accurate for the middle thickness of the soil profile (0–50 cm), which corresponds with findings of the previous studies (Orzepowski et al ). In the deepest layer (0–100 cm), the top rating was achieved for model E, which exchanges the variable SWC in model D with GWT position measured at the beginning of the growing season.…”

“…For this purpose, combinations of the results of the measured and estimated data applied in the most comprehensive model G and in the initial model A (relying only on the basic climatic data) were compared. This particular combination was justified as the most reliable for the data comparison for two reasons: 1 -the present study revealed that model G is the most accurate (7 ranking points) only in the most upper soil layer (0-25 cm), 2 -the former studies by Orzepowski et al (2017) defined variant A as highly suitable (6 ranking points) for modelling in the shallowest soil layer. Results of the comparison are exemplified by water contents in the uppermost layer of soil profile Sm through the last six growing seasons (GS1-GS6, Fig.…”

“…These sites, known as Ługowinka – Lu, Samotwór – Sa, Smolec – Sm, Rybnica – Ra, Bliż – Bl, Rybna – Ry and Wyszków – Wy, were established on agricultural fields cultivated mostly with cereals. Soils of these fields variate from light sandy to heavy soils, as follows: Lu – sandy and silt loam and loamy sand; Sa – sandy loam and sand; Sm, Ra and Bl – silt loam; Ry – silty clay and clay; and Wy – silty clay loam and silt loam (Orzepowski et al ).…”

“…temperature and precipitation), B (applying the climatic data and FWC), C (applying the climatic data and SWC) and D (linking the climatic data with FWC and SWC). The most accurate data model fit was achieved for variant D (Orzepowski et al ), thus it constituted the original platform for the present study assessing GWT applicability in determining SWRC through the next three new model variants: E (developed on variant D exchanging SWC with a new variable – GWT, both measured at the beginning of growing season), F (upgraded variant E by implementing a new variable describing fluctuations of GWT through the entire growing season) and G (improved variant D by applying the GWT fluctuations), as follows:…”

“…The principal idea of the present study was to develop new model variants that fit the current needs for determining soil water retention characteristics (SWRC). The main objective was to continue with the modelling options and to complete the estimations of soil water reserves that were recently reported by the authors herein (Orzepowski et al ). The modelling options constituted the initial basis for applying ANN methods with input data from standard meteorological and hydrological measures, such as precipitation, temperature, soil water content (SWC) and field water capacity (FWC).…”