Advancing Inversion Techniques for Improved Retrievals of Soil Moisture and Vegetation Optical Depth using Spaceborne L-band Radiometry

Abstract

Satellite remote sensing techniques provide us an unprecedented opportunity to monitor global land surface water status, the outstanding progress of which has been witnessed in the past three decades. In comparison with traditional optical remote sensing with visible, near-infrared, and/or short-wave infrared sensors, microwave bands could provide an all-sky capability for accurate remote sensing of hydro-meteorological states. To measure global surface soil moisture (SM) and vegetation optical depth (VOD), NASA's Soil Moisture Active Passive (SMAP) satellite mission has been launched on 31 January 2015 using L-band (1-2 GHz) radiometry, which could provide corresponding measurements with a global revisiting time of 2-3 days. The retrieval of SM and VOD at L-band often relies on the inversion of the zeroth-order radiative transfer model that can well formulate the emission interaction between soil and canopy layer. However, due to the limited number of observations at vertical and horizontal polarization channels, inverting the model is typically an ill-posed problem. For improved retrievals of SM and VOD, this thesis strives to advance existing inversion algorithms in terms of the following aspects: (1) narrowing down the feasible ranges of SM and VOD retrievals using the information of soil texture and VOD climatology; (2) accounting for the slow temporal changes of VOD through a Sobolev-norm regularization; (3) incorporating the information of nearby spatial observations considering the theorem of spatial autocorrelation that states neighboring geographical elements tend to be similar on a map; (4) combining the merits of multiple SM products through a deep neural network. The results suggest the proposed strategies could effectively reduce the uncertainties in the retrievals and SM and/or VOD beyond existing SMAP official products.