Title: The SoilSCAPE in-situ soil moisture network: innovations in network design and approaches to upscaling in support of SMAP
Presenting Author: Mahta Moghaddam
Organization: University of Southern California

Abstract:
The Soil Moisture Active Passive (SMAP) mission, launched in January 2015, will provide soil moisture at 3, 9, and 36 km scales through the use of radar and radiometer data. The soil moisture products need to be validated at all of these scales. Typically, validation plans consist of several sensors installed nearly uniformly in the scene. To upscale the soil moisture estimates to the scales of SMAP products requires a large number of sensors, distributed throughout the instrument footprint. Even for the higher resolution SMAP products (3 km) it is difficult and expensive to deploy uniformly distributed sensors with sufficient sampling density to capture sub-500 m heterogeneity of topography, landcover, soil texture, etc. The Soil moisture Sensing Controller and oPtimal Estimator (SoilSCAPE) project provides a new adaptive validation strategy, including upscaled estimates of soil moisture. By utilizing smarter network technology and optimized sensor placement, more representative measurements of soil moisture are obtained, at a range of spatial scales with lower costs than traditional networks. A large network was established around the Tonzi Ranch site in central California. The network design comprises multiple sites, each with a 10-30 node cluster taking measurements from up to 4 sensors installed at different depths. The nodes wirelessly communicate to a Local Coordinator, which collects data and transmits to a server (http://soilscape.usc.edu). Each node can communicate with the Coordinator up to a distance of 400m. Each station supports up to 60 nodes. The SoilSCAPE nonuniform placement of sensors requires novel upscaling methods. Previous studies have used regression, which works well when the measurement is well correlated with other variables. However, soil moisture dependence on various variables could be complex and nonlinear. To account for such complexities, we use the Random Forests statistical classification and regression algorithm, which is capable of modelling complex non-linear system and can handle continuous and categorical data. The algorithm has not previously been applied to upscale estimates of soil moisture. Using AirMOSS P-band radar data and other data layers, we describe a technique for upscaling in-situ soil moisture data from the SoilSCAPE network to resolutions of 3 km, 9 km, and 36 km for the SMAP mission. The method and results will be presented, including the latest advances and enabling data products from SoilSCAPE.