A Multi-Sensor Approach for High Resolution Airborne Soil Moisture Mapping

Airborne remote sensing provides a viable option for high resolution mapping of nearsurface soil moisture that allows larger areas to be covered in greater spatial and temporal detail than has hereto been possible from traditional ground based techniques. However, the current retrieval algorithms require information on near-surface soil temperature and vegetation water content in order to estimate the soil moisture from well calibrated horizontally polarised passive microwave data. This paper presents a methodology for retrieving this ancillary data, through relationships with other remote sensing data. Near-surface soil temperature is estimated from land cover specific relationships with thermal infrared data, and vegetation water content is estimated from land cover specific relationships with nearand shortwave-infrared data. The methodology is tested with data from the National Airborne Field Experiment (NAFE) conducted in the Goulburn River catchment in New South Wales Australia, during November 2005. This intensive month-long field campaign involved daily airborne flights with a Polarimetric L-band Multibeam Radiometer (PLMR) and thermal imager flown onboard a small environmental aircraft at altitudes ranging from 150m to 3000m AGL, yielding passive microwave data at resolutions from 62.5m across entire farms to 1km across entire regions. This study presents a preliminary analysis of the PLMR derived soil moisture product at 250m resolution across a 2000ha farm. This is the first airborne remote sensing study to both provide such high resolution soil moisture data and to take this multi-sensor approach to soil moisture retrieval. The remotely sensed soil moisture data is compared against ground near-surface soil moisture measurements taken at resolutions ranging from 250m to 500m across the same farm on the same days. These preliminary results indicate a good agreement of the retrieved and measured soil moisture spatial distribution, with an overall absolute retrieval error of around 6% v/v.