Three dimensional prediction of soil moisture at any support using multiple datasets with different spatial supports

Abstract:

There are numerous approaches for estimating soil moisture. Examples include direct estimates with soil moisture probes and radar estimates of soil moisture, or indirect estimates based on modelling the water balance equation. Each has their advantages and disadvantages and for the most part these are related to the spatial support over which the measurements are made. In this work we define the spatial support as the vertical and horizontal area over which the estimate is made. In the case of soil moisture probes we effectively have a point horizontal support as in most cases the measurement volume is quite small, and in most cases sensors are arrayed vertically giving discrete point measurements throughout the soil profile. In terms of mapping soil moisture, probes are generally sparse in density. Radar estimates of soil moisture offer full spatial coverage and have a horizontal supports up to 20-40 km, however their vertical support is the first few cms of the soil profile. One major issue is how to combine observation from a range of different supports into a single prediction model. We address this here by presenting an approach which allows estimates of model parameters for datasets of different supports in both vertical and horizontal dimensions, and subsequent prediction with this data at any support vertically and horizontally. The approach is based on area-to-point kriging with residual maximum likelihood estimation of model parameters. We use a case study from the Murrumbidgee catchment with two types of soil moisture estimates; point measurements from soil moisture probes which measure at 5 depths in the soil profile to 1.0 m, and a whole-profile estimate of soil moisture from a water balance model driven by remotely sensed components of ET from the MODIS 16 product and rainfall from TRMM.