Evaluation of near-surface soil moisture data from an AAFC monitoring network in Manitoba, Canada: Implications for L-band satellite validation

Abstract

Summary This paper examines near-surface soil moisture data (top ∼6 cm) from an Agriculture and Agri-food Canada (AAFC) monitoring network in southern Manitoba, Canada, over the 2012 and 2013 growing seasons. This network is intended for use (in part) as a core validation site for the Soil Moisture Active Passive (SMAP) mission. The network currently consists of 9 stations sited within agricultural fields. Each station uses a triplicate configuration of horizontally installed Hydra Probe soil moisture sensors at near-surface (∼5 cm) to root-zone soil depths, in addition to permanently installed vertical surface sensors (0–5.7 cm). This domain was studied intensively over the 6.5 week SMAPVEX12 ground campaign in June/July 2012, providing coincident soil moisture datasets to assess network up-scaling over the domain. Results demonstrate that statistical differences exist between the network-level data from the horizontal ( θ 3.5 – 6.5cm ) and vertical ( θ 0 – 5.7cm ) surface Hydra Probes over 2012–13. Bootstrap resampling shows that a mean difference of 0.018 m 3 m −3 (at 99% confidence) is observed between θ 3.5 – 6.5cm and θ 0 – 5.7cm measurement depths ( θ 0 – 5.7cm sensors exhibit dry bias). Individual station analysis shows that the largest differences between θ 3.5 – 6.5cm and θ 0 – 5.7cm measurement depth occur at sites with higher clay content, whereas sensors located in sandy soils exhibit negligible differences. Up-scaling comparisons of the AAFC network to field-sampled data and a dense temporary network over SMAPVEX12 demonstrate that it is representative of the domain within 0.04 m 3 m −3 RMSE. This suggests that the AAFC network is suitable for coarse resolution L-band microwave soil moisture validation. Comparison of Soil Moisture Ocean Salinity (SMOS) Level 2 soil moisture product data to the up-scaled AAFC network data over 2012–13 shows that superior temporal correlation is found with the θ 0 – 5.7cm dataset. The difference in RMSEs between the SMOS comparisons to θ 3.5 – 6.5cm and θ 0 – 5.7cm network datasets is approximately 0.018 m 3 m −3 . These results suggest that differences of near-surface measurement depth and sensor configuration should be considered when conducting L-band satellite validation.