Applications: Mapping

Can the height of a rice plant be measured from space? TanDEM-X, a mission designed to generate high accuracy global elevation models, has been shown to be capable of doing just that [1]. The growth of a rice plant can be observed starting from its vegetative phase, through its maturation phase until harvest. The measured height is very accurate, with a root mean square error ranging from a few up to 20 centimeters, depending on the phenological phase. This result is of particular interest considering that the plant grows to only about 1.5 meters and the TanDEM-X system’s relative and absolute elevation accuracy requirements are on the order of meters.

In this paper, the processing strategies for the interferometric height derivation are first addressed. The data stack is composed of 32 dual-pol strip-map X-band images over an agricultural area surrounding Lake Gala (Turkey). Figure 1 shows the exact test site location, as well as backscattering in HH and VV polarizations, coherence and phase difference. The processor configuration and strategies to handle the stack are firstly presented.  Since the evaluation is performed on a field-by-field basis, crop segmentation is an essential step. The segmentation is performed in a fully automated way, by inspecting the early vegetative stage of the plants. In this stage, plants are fully submerged, so that the segmentation operation turns into a water detection problem. The combined use of amplitude and coherence data helps this step. The employed approach is the same used for the water body layer generation of the TanDEM-X DEM.  The final product to validate is a stack of digital elevation models, generated with HH and VV polarizations, for different times during the plant growth. Each detected crop can be then temporally followed in its development.

For the result validation and interpretation, an on-site campaign has been performed. Several parameters have been collected for 8 reference fields. Among them, plant height over ground has been used for the InSAR height validation. To assess the plant development, terrain height is also needed, in order to subtract it from the InSAR measurement and making it then proportional to the plant height. At this purpose, a DEM generated with a post-harvesting take has been used as terrain model. Before validating the result, expected height can be predicted in the general electromagnetic framework. It is well known that the interferometric phase is proportional to the height of the scattering phase center. The center location depends on the complex interaction between the SAR signal and the imaged surface. For canopies, it is usually below the vegetation top layer as the signal usually penetrates into it. The key parameter is the extinction coefficient. The smaller it is, the lower the scattering center. Parameters as wavelength and local moisture impact in its determination. A purpose of the paper is a deep study of the extinction for different growing stages and different polarizations. For instance, it is shown that milky grains produced in the maturation stage well reflect the signal at X-band, so that signal penetration is a rather small issue, with height underestimation in a centimetric level. Contrariwise, before this stage, signal interaction with lower portions of plants brings underestimations in the decimetric level. Differences between HH and VV DEMs are also addressed. It is shown that wave polarization is not particular relevant for absolute ranging studies over rice paddies. Only small elevation discrepancies around 8 cm were measured during dry season. Nevertheless, wave polarization is considerable in backscatter studies aimed to classify the phonological stages. This study is also briefly introduced and applied to the dataset.

Finally, canopy height influences yield estimation in precise farming. The paper is concluded with the current research about the combined use of texture analysis and InSAR height towards the yield assessment on a field-by-field basis.

(see pdf for figure)

Figure 1. The agricultural study area in Ipsala, Turkey. The backscattering images acquired on 21^st of June, 2014, in HH and VV polarizations, and their coherence and phase difference measurements are placed over the optical image extracted from Google Earth.

[1] Rossi, C., and Erten, E., 2015. Paddy rice monitoring using TanDEM-X, IEEE Transaction on Geoscience and Remote Sensing, Vol. 53 (2), pp. 900-910.

Remote sensing at radar wavelengths has the potential to be a beneficial tool to detect subsurface features in desert areas. Penetration depth varies from different wavelengths of different sensors with the greater the penetration the longer the wavelength and conversely the shorter the frequency. It has been shown that microwave imaging at low frequency (e.g. 1.25 GHz/24 cm, L-band) can penetrate down to about 2 m depth under sand cover in Libyan Desert with extremely low surface and bulk humidity. Thus, subsurface features, such as paleo-channels and buried craters, have been unveiled. On the other hand, global Digital Elevation Models (DEMs) are now being produced from spaceborne earth observation sensors, such as at microwave frequencies from SRTM, TanDEM-X and ALOS-PALSAR as well as stereo-optical, such ASTER or PRISM or echo waveform lidar such as ICESat, which can be useful data resources to detect subsurface features over desert regions. By using such DEMs, surface geomorphological features can be detected, while subsurface feature detection in desert regions can sometimes have surface expression but mostly not. 

In order to detect subsurface features under dry sand cover, we investigate the penetration ability of different DEMs acquired by SRTM (X- and C-band), ALOS-PALSAR (L-band), ASTER (stereo-optical) and ICESat (1064nm lidar) over three study sites in Africa/Middle East, which includes Gilf Kebir Plateau, Kufrah River and Kuwait. Gilf Kebir Plateau is located in southwest Egypt, where 1,300 craterlike features over 4,000 km² were found by Paillou et al. This study site covers a 23,000 km² region between latitude 23°N-24°N and longitude 26°E-28°E. Kufrah River is located in southeast Libya, the study site covers an 11,000 km² region between latitude 23°N-24°N and longitude 23°E-24°E. There is a paleodrainage system which passes through this region, which has been discussed in former pulications. The last study site is located in Kuwait, covering Kuwait city between 28°N-30°N and longitude 46°E-48°E. Since elevation data obtained from different sources have different vertical datums, they were brought into the same reference frame. SRTM-X DEM and ICESat/GLA14 data were transformed into the EGM96 datum, thus, the elevation used in this study is orthometric height referenced to EGM96. Horizontal offsets between ASTER and SRTM DEM were checked, which appear to be within one pixel. Although the ICESat/GLA 14 data are referenced to a different ellipsoid cf. SRTM-C DEM and ASTER DEM, the differences in geodetic latitude and longitude only produce a horizontal displacement of less than a metre, which can be ignored. 

Our preliminary analysis shows that, in terms of subsurface features detection, SAR (SRTM at C-band and PALSAR at L-band) and lidar (ICESat) are superior to stereo-photogrammetric (ASTER), which is affected by severe noise due to the lack of contrast in the stereo image pairs and possible atmospheric sand particles obscuring the surface. This is consistent with previous studies [3] that indicate that electromagnetic waves of lower frequency can penetrate deeper in sandy layers. Surprisingly some subsurface features are obvious on the SRTM-C DEM, which also conforms well with elevations measured by ICESat. Besides that, The elevations extracted from SRTM-X DEM are also consistent with elevations measured by ICESat with a slight deviation at the edge of SRTM-X DEM’s strip. This appears to be due to sand filling-up previous water channels. However, these are absent in the ASTER DEMs probably because of the high noise content of these DEMs. By analysing topographic features of the DEMs such as those detected using bi-quadric modelling of the DEM height points using a surface patch model, a preliminary automated detection of subsurface Kufrah river has been realised. But the extraction of Kufrah river is affected by refilled sand and dunes, which change the slope of riverbed. Further work will be shown on how to deal with this problem and the adopted method will be applied to the other two study sites. Results will be shown from different areas using this method to auto-detect paleochannels and buried craters from recently released SRTM DEM.

In this study, we assess the potential of using seasonal effects demonstrated by TanDEM-X InSAR data in land cover and forest mapping over two study sites in Finland.

Forest tree height and above ground biomass (or forest stem volume) represent key parameters suitable for characterization of forests from SAR backscatter and interferometric SAR coherence.

Tree heights retrieved by single baseline polarimetric interferometric SAR (PolInSAR), or multibaseline PolInSAR, increase the accuracy of biomass estimation. If an external digital elevation model is available, even single-pol interferometric coherence can also be used to retrieve the forest heights.

While X-band is considered to be sub-optimal for forest characterization, it was shown to provide promising results in tree height retrieval of relatively sparse boreal forest with low forest stem volume. Also, identified variability of the X-band radar signal penetration depth to the forest due to seasonal changes should be accounted for. Its potential can be increasingly tested using data from spaceborne interferometric SAR mission TanDEM-X, making it possible to collect relatively long time series of bistatic SAR interferometric acquisitions.

In this study we investigate seasonal dynamics of TanDEM-X InSAR data acquired during several seasons over several forest types (coniferous, deciduous, mixed), in order to assess potential of differentiating different land cover and forest classes based on observed multitemporal signatures.

InSAR data was acquired over two test sites: in the vicinity of Kirkonummi in southern Finland and over Hyytiälä forestry station in central Finland. Kirkkonummi InSAR dataset represents five sets of dual-pol (HH,VV) Coregistered Single look Slant Range Complex (CoSSC) datatakes covering September to November of 2011. A set of 12 dual-polarization (HH and VV) interferometric CoSSC pairs acquired over Hyytiälä during 2013 and 2014 were investigated as well. Also, a set of three single-polarization (HH) interferometric CoSSC pairs acquired during 2012-2013 in standard DEM production mode were studied separately.

Reference data include airborne laser scanning measurements and forest management plans, as well as CORINE land cover data of Finnish National Land Survey.

Quantitative results on multitemporal dynamics of interferometric phase, as well as its suitability for delineating primary forest types, are reported in detail at the workshop.

In the semi-arid northeast of Brazil, numerous reservoirs of various sizes have been constructed to ensure water supply in the dry season. Among them, more than 30,000 are small reservoirs with average storage capacities of 0.5 hm³. To support local water management on catchment to regional scale, an accurate estimation of the water storage capacities of reservoirs is required. However, the currently bathymetric surveys in the study area and elsewhere are conducted manually either with echo sounders or with depth meters at different times, which involves enormous efforts and point-wise measurements with varying accuracies. SAR interferometry based on TanDEM-X in the bistatic mode has been proven to be very effective to generate DEM of fine resolution. However, water bodies including the reservoirs in the test area of this study have been masked out in the global TanDEM-X DEM mission.

This study, therefore, aims at deriving high resolution DEMs of the reservoir areas with TanDEM-X data acquired in the dry season, i.e. when the water level is lowest and most reservoirs fall nearly or completely empty. Subsequently height-area-volume curves are generated with respect to each reservoir of interest based on their topography derived from InSAR DEM. These curves assist the subsequent estimation of available water volume from water surface changes derived from satellite time series data. Field data from DGPS measurements and existing bathymetric surveys are employed for validation purpose and comparison of DEM generation approaches.

Svein Solberg, Norwegian Forest and Landscape Institute

Storm damage in forests can be detected in several ways with optical satellite remote sensing, including increased brightness in particular in the SWIR band and reduced vegetation index, and Sentinel-2 will be the first choice for detection in most cases. However, in Nordic countries the stormy season in mid-winter is typically having low sun angles, short days and cloudy weather making this difficult. SAR sensors can overcome these problems, however; with SAR data it is in general difficult to detect such damage based on backscatter intensity and based on coherence. Wind-thrown trees still retain branches on. An alternative method is to use InSAR and detect the damage as a decrease in height. We have tested the method on damage after the storm Dagmar in south Norway 26^th December 2011, where 1.5 million m³ were blown over. We made a ground truth data set based on the track-log data from the harvester machine that did the cleaning up after the storm. One Tandem-X stripmap pair was acquired on 11^th January 2011 and served as the reference data before the storm. Another Tandem-X data set was acquired on 12^th April 2012 and served as post-storm data. We identified a threshold of -2 m for classifying a given stand as damaged, while this was distinguished from regular clear-cuts that were identified by having more than -9 m decrease. The 3 classes undisturbed, storm damage and clear-cut could be classified at the stand level with an overall accuracy of 84% and a Kappa value of 0.48. Two addition benefits were obtained by using Tandem-X data. First, the volume of wind-thrown timber could be estimated for each forest stand, as well as the entire area, by aggregating the height decreases and multiplying them with a volume density factor of 24 m³/ha per m change in InSAR height. Secondly, we used the Tandem-X DSM and the InSAR height to map the storm risk over the area for the actual wind direction of Dagmar (SW), which turned out to fit reasonably well to the observed height decreases, i.e. storm damage.

During the round table, seed questions proposed by the chairs will be discussed with the audience.