Main content area

Can next-generation soil data products improve soil moisture modelling at the continental scale? An assessment using a new microclimate package for the R programming environment

Kearney, Michael R., Maino, James L.
Journal of hydrology 2018 v.561 pp. 662-673
correlation, data collection, databases, landscapes, meteorological data, meteorology, microclimate, models, monitoring, neutrons, prediction, remote sensing, rhizosphere, satellites, soil profiles, soil properties, soil water, Australia
Accurate models of soil moisture are vital for solving core problems in meteorology, hydrology, agriculture and ecology. The capacity for soil moisture modelling is growing rapidly with the development of high-resolution, continent-scale gridded weather and soil data together with advances in modelling methods. In particular, the initiative represents next-generation, depth-specific gridded soil products that may substantially increase soil moisture modelling capacity. Here we present an implementation of Campbell’s infiltration and redistribution model within the NicheMapR microclimate modelling package for the R environment, and use it to assess the predictive power provided by the product Soil and Landscape Grid of Australia (SLGA, ∼100 m) as well as the coarser resolution global product SoilGrids (SG, ∼250 m). Predictions were tested in detail against 3 years of root-zone (3–75 cm) soil moisture observation data from 35 monitoring sites within the OzNet project in Australia, with additional tests of the finalised modelling approach against cosmic-ray neutron (CosmOz, 0–50 cm, 9 sites from 2011 to 2017) and satellite (ASCAT, 0–2 cm, continent-wide from 2007 to 2009) observations. The model was forced by daily 0.05° (∼5 km) gridded meteorological data. The NicheMapR system predicted soil moisture to within experimental error for all data sets. Using the SLGA or the SG soil database, the OzNet soil moisture could be predicted with a root mean square error (rmse) of ∼0.075 m3 m−3 and a correlation coefficient (r) of 0.65 consistently through the soil profile without any parameter tuning. Soil moisture predictions based on the SLGA and SG datasets were ≈ 17% closer to the observations than when using a chloropleth-derived soil data set (Digital Atlas of Australian Soils), with the greatest improvements occurring for deeper layers. The CosmOz observations were predicted with similar accuracy (r = 0.76 and rmse of ∼0.085 m3 m−3). Comparisons at the continental scale to 0–2 cm satellite data (ASCAT) showed that the SLGA/SG datasets increased model fit over simulations using the DAAS soil properties (r ∼ 0.63 &rmse 15% vs. r 0.48 &rmse 18%, respectively). Overall, our results demonstrate the advantages of using products in combination with gridded weather data for modelling soil moisture at fine spatial and temporal resolution at the continental scale.