Conference Agenda

EMPAC addresses different aspects related to the air quality (AQ) over China: aerosols, trace gases and their interaction through different processes, including effects of radiation and meteorological, geographical and topographical influences. Satellite and ground-based remote sensing together with detailed in situ measurements provide complimentary information on the contributions from different sources and processes affecting AQ, with scales varying from the whole of China to local studies and from the surface to the top of the boundary layer and above. Different species contributing to air quality are studied, i.e. aerosols, in AQ studies often represented as PM2.5, trace gases such as NO2, NH3, Volatile Organic Compounds (VOCs) and O3. The primary source of information in these studies is the use of a variety of satellite-based instruments providing data on atmospheric composition using different techniques. However, satellite observations provide column-integrated quantities, rather than near-surface concentrations. The relation between column-integrated and near-surface quantities depends on various processes. This relationship and the implications for the application of satellite observations in AQ studies are the focus of the EMPAC project. Initial results of detailed process studies using ground/based in situ measurements, instrumented towers, as well as remote sensing using lidar and Max-DOAS will be presented. A unique source of information on the vertical variation of NO2, O3, PM2.5 and BC is obtained from the use of an instrumented drone.
The results of the third year will be presented, including trend studies of air pollutants, algorithm development for aerosol retrieval, and analyses of NOx emissions from Sentinel 5p over the YRD.

This project is focused on multi-orbits applications of SAR imaging, and is intended to support use of multi-pass data stacks from: the upcoming P-Band mission BIOMASS; future L-Band missions, such as the SAOCOM constellation, the upcoming Chinese L-Band bistatic Mission Lu-Tan1, and potentially Tandem-L and Rose-L; the C-Band Sentinel Missions.

The main results until now are summarized into 4 contributions:

1. A detailed experimental analysis was carried out to compare two techniques for estimating forest height and vertical structure using airborne synthetic aperture radar (SAR) data, namely SAR tomography (TomoSAR) and the phase histogram (PH) technique. Using multiple SAR images, TomoSAR allows for a direct imaging of the three-dimensional (3D) electromagnetic structure of the vegetation layer, from which biophysical parameters such as forest height and underlying terrain topography can be extracted. The PH technique assigns each pixel in a SAR interferogram to a specific height bin based on the value of the interferometric phase, allowing for a local estimation of the vertical profile of forest scattering by accumulation of pixels fall within a given spatial window.

Results indicate that the PH technique can only loosely approximate the vertical structure produced by SAR tomography, but it can be used to produce a fairly good estimate of forest height. In particular, using the datasets collected by the TomoSense campaign, TomoSAR and PH techniques are observed to produce an average root mean square error (RMSE) of 2.63 m and 4.72 m in NW flight data, and 1.86 m and 5.26m in SE flight data, respectively. The observed results are interpreted in light of a simple physical model to predict phase variations in the two cases where forest scattering is determined by the presence of a dominant scatterer at each resolution cell or by a multitude of elementary scatterers, leading to the conclusion that the PH technique is best fit for the case of high- or very high-resolution data at higher frequency bands. Overall, the analysis in this paper demonstrates, both theoretically and experimentally, that the PH technique cannot achieve the same performance as multi-baseline tomography when applied to lower frequency data at a resolution of few meters. Yet, even in these conditions we remark that the PH technique allows for the retrieval of forest height based on a single interferogram at a single polarization. This makes the PH technique extremely interesting in the context of spaceborne missions.

2. A solution is proposed to the problem of atmospheric estimation using SAR data and a network of GNSS stations on the ground. The raw data from each station is processed to extract a GNSS-derived APS. Then, the SAR-derived APS is extracted on the spatial location of the GNSS stations. Such measurements are the sum of the true APS and the orbital error. We use the GNSS-derived APS as a ground truth, removing them from the SAR-derived estimates leading to a set of measurements of the pure orbital error. An inverse problem is solved, leading to two parameters characterizing the orbital error. The benefit of this inversion is double. First of all, the two estimated parameters can be used to provide a quality proxy for the trajectory. Second, the two parameters can be used to compute the forward model on the whole grid of the APS map (and not just on the set of GNSS stations as done before), leading to a calibration phase screen.

The procedure is tested using a dataset of more than 30 Sentinel-1 images and a network of GNSS stations in Sweden. The algorithm shows excellent performance. The validation process compares a set of independent GNSS stations with the SAR-derived APS before and after the calibration procedure. A second validation is carried out using a separate NWPM showing, once again, very good performances.

3) An alternative approach is proposed for NESZ estimation that exploits an interferometric pair of images over land. The method is based on the relation between coherence and noise. we use the stack to generate a set of interferograms with short temporal baselines. Each interferogram is fitted with a coherence model, and all measures are averaged to improve robustness. By repeating for each incidence angle, the NESZ profile of the new satellite can be characterized. The procedure was validated and tested using a stack of Sentinel-1A (S1A) and Sentinel-1B (S1B) images. First, the noise level of S1B was obtained separately according to Equation (1). Then, S1B’s NESZ was estimated by the procedure described above, using the stack of S1A data. The comparison between the two results confirmed that the noise level of a new satellite could be characterized over land, with as little as one available product.

4) InSAR has been widely recognized as an effective tool for landslide investigation. However, its measurement accuracy is largely limited by the complex atmospheric delay distortion in alpine valley regions, resulting in poor performance of landslide detection and monitoring. Particularly, the spatial atmospheric heterogeneity over wide areas cannot be accurately reflected by conventional empirical phase-elevation models or external data-based methods. Here we proposed a multi-temporal moving-window linear model (MMLM) to correct the tropospheric delay for wide-area landslide investigation. This is a linear regression model based on the elevation-phase relationship for modeling multi-temporal phases within a sliding local window. It mitigates the influence of local turbulent phase, local landslide deformation, and phase unwrapping error on parameter estimation, providing precise heterogeneous InSAR atmospheric corrections for wide-area landslide identification and deformation monitoring. Experimental results with descending and ascending Sentinel-1 data over the reservoir area of the Lianghekou hydropower station clearly demonstrated that the proposed MMLM model outperforms modern APS correction approaches including the ERA5, GACOS, spatial-temporal filtering, and traditional linear model.