Forest Monitoring over Hilly Terrain Using ERS INSAR Data
Recent studies using ERS SAR data showed that its configuration is somewhat limited for forest applications, due to ERS small incidence angle (23deg.) and low penetration depth in forest covers at C-band. In addition to low-level cover discrimination such as forest/non-forest mapping, possible applications are mainly related to the monitoring of natural or man-made forest perturbations such as clear-cutting (tropical deforestation), regrowth (low biomass levels) and environmental changes (thaw/freeze events). Fortunately, applications can be broadened significantly when interferometric phase and correlation derived from repeat-pass ERS INSAR data are considered, in addition to the usual backscatter information.
On one hand, the interferometric correlation, which is an indicator of the temporal stability of the target in terms of geometric and dielectric properties, proved to be a good discrimination in cultivated areas (Wegmuller and Werner 1995a) and forested landscapes (Wegmuller and Werner 1995b, Herland, 1995 ); multi-pairs correlation appeared to be particulary powerful (Wegmuller et al. 1995). On the other hand, it has been showed that the interforemetric phase, usually used to derive the terrain altitude, can also be linked to the height of the forest canopy (Askne et al. 1995, Hagberg et al. 1995, Ulander et al. 1995).
However, these are very recent results and the potential of such new data type is far from being fully explored. The interferometric, topographic and environmental conditions, for which INSAR data can provide useful thematic applications, have to be assessed furthermore. In this paper, we address the potential use of INSAR data over hilly terrain, towards the generalization of SAR-retrieved forest information in various conditions. The approach consists in using a differential interferometric technique developed at CNES (French Space Agency), to remove the topographic effects and keep the residual correlation and fringes. Then, differential interfrometric data are analyzed on a well-documented forest test-site featuring large plantations of coniferous trees found over hilly terrain.
The test-site is situated in the central part of the Département de la Lozère, in Southeastern Massif Central (France). It is centered approximately at 44.5deg. N and 3.5deg. E. The area is characterized by large and gently rolling limestone plateaus culminating around 1200m, which are intersected by gorges with 300-500m depth and steep slopes up to 50deg. (see DEM in Fig.1a). Main land-uses are natural short grasslands, cultivated areas (small pot holes on the plateaus and valleys) and coniferous plantations. These latter are found on gorges slopes and partially on plateaus tops for erosion prevention and pulpwood production.
The planted species adapted to the difficult site conditions (soil thinness and summer drought) is almost exclusively Austrian pine (Pinus nigra). They are found in state-owned forest plantations made of even-aged and relatively homogeneous stands. The two main forest test-sites cover respectively 5 400 and 1200 ha with more than 500 stands (average area 5-15 ha), offering a large range of growth stages (0 to 140 year-old) as well as topographic situations.
Different ground data concerning these stands have been collected and entered in a GIS: 1) stand limits, 2) age classes, 3) detailed measurements of forest parameters and 4) a Digital Elevation Model (DEM) with a 50 m grid size in Lambert III projection (IGN (c)). Main stand limits and the DEM are included in Fig.1a. More than 300 homogeneous stands of different age classes and with area greater than 3ha were kept for the SAR data analysis. In addition, 103 stands were sampled for stem density, diameter at breast height (dbh) and stand height. Then, bole volume were estimated from simple allometric equations, with an estimation error around 10 to 20% (see Beaudoin et al. 1995 for more details).
INSAR acquisition and processing
Different ERS INSAR pairs, including one tandem acquisition, were acquired on this test-site. The various acquisition parameters that must be accounted for in the analysis are presented in Table 1. Only pairs with relatively small baselines (< 300m) were kept to prevent from excessive spatial decorrelation. Ignoring the tandem pair, it can be seen that the time acquisition interval is large, providing contrasted seasons and thus, variability in ground, forest and weather conditions. In particular, near-freezing conditions with a shallow snow layer on the ground were present for one winter scene included in pairs 1 & 3. However, low wind conditions (10 to 30 km/h) prevailed for all pairs.
Several differential interferometric techniques, involving either SAR data pairs or triplets, have already been considered, mainly for the detection of ground displacements related to various types of hazards such as earthquakes, landslides or volcanic activity (Carnec et al. 1995, Massonnet et al. 1993, Massonnet et al. 1995). The concept of differential interferometry can also be applied to detect relative height differences within the interferogram : in that case, fringes patterns corresponding to these variations are superposed to those related to topography.
In our case, INSAR data were processed using a DEM differential technique (Massonnet 1994) : the site DEM is used in combination with the orbit geometry to simulate the topographic component and to remove it directly from the interferogram. Another approach would consist in unwrapping first the raw interferogram - including thus the terrain altitude and the forest height - and then substracting the terrain altitude provided by the DEM. The first method is more interesting in the way that there is no need for phase unwrapping if the average height of the forest cover is small compared to the altitude of ambiguity, that is to say the altitude variation corresponding to one interferometric fringe. However, the results may be corrupted if atmospheric perturbations occurred over the site area.
After this processing, interferometric coherence (or correlation) and residual phase are estimated from spatial averages within forest stands (more than 150 pixels) and are compared with parameters of interest. At this stage, we will focus only on the information content of the interferometric coherence, which must be analysed before investigating the phase, as the relevance of the latter directly depends on the value of the former.
2) EXPERIMENTAL RESULTS
2.1) General correlation behavior
Fig. 1b presents the geocoded correlation image including the forest stand limits, while Fig. 2 presents mean correlation values for the 3 main land-use types, plotted against time interval. Globally, brighter areas correspond to grasslands (Causses area in Fig. 1b ), which always exhibit a good correlation in Fig. 2 (>0.85), even for the pairs including the winter scene. Forested areas are darker and correspond very well to the stand limits. However, bright areas are also found in these areas (see under Mende in Fig. 1b). In Fig. 2, young stands have a higher correlation than mature ones; in addition, the correlation decreases for both stand ages with acquisition interval. Finally, cultivated areas also present in the scene can be found as small dark patches on the Causses (cultivated pot holes) or dark stripes in the main valley (Florac area). In most cases, agricultural areas are confused with mature forests. Beside this, the potential of interferometric correlation can be foreseen for general land-use discrimination, which was not possible using ERS backscatter intensity alone.
2.2)Forest correlation behavior
Looking at Fig. 1b , large correlation variations are found within forested areas. Correlation was estimated on 103 stands presenting different biomass levels and topographic situations. Fig. 3a presents correlation estimates for the tandem acquisition, as a function of biomass level. A decrease with biomass is observed with r2=0.52 from linear regression, but the trend is too weak to estimate correctly the biomass. Lower r2 were observed for other pairs, but the same trend appeared with a more or less pronounced negative slope. Such decrease was partially observed also by Herland (1995) between some forest stands.
To interpret qualitatively above observations, simple theoretical considerations are addressed in the following. Considering that most of the spatial decorrelation due to terrain slope was remove during the DEM-differential interferometric processing, the total decorrelation rho can be decomposed in its different sources (Askne et al. 1995, Zebker and Villasenor, 1992):
rho total = rho thermal * rho spatial * rho temporal (1)
For interpretation purposes, we first consider the backscatter from a discontinuous forest layer over a ground surface. At C-band, the backscatter mainly originates from needles and twigs volume scattering sigma forest, in addition to a direct ground scattering sigma ground which contributes more or less with soil moisture and roughness, along with transmissivity T and forest fraction cover F, which in turn are linked to the growth stage. This can be approximated by:
sigma = F (sigma forest + T2 sigma ground) + (1- F)
sigma ground (2)
When adding a forest layer (here, F within [0.6-1.0] and crown layer depth from 3 to 8 meters), the ground contribution will decrease with growth stage. In addition, temporal decorrelation due to volume scattering at C-band will add and should be considered at 2 different time spans:
· On a monthly basis during the growing season, the crown geometry changes due to annual new shoots, which is probably a source of significant decorrelation.
· On a very short time basis (maybe down to few minutes), decorrelation occurs due to random displacements of the scatterers (needles and twigs) under the wind influence.
Zebker and Villasenor (1992) linked this decorrelation to the RMS displacement of the scatterers. For ERS configuration, only few RMS cm are enough to decorrelate the signal, which we qualitatively observed even for weak winds. However, the link between the displacement and the wind and forest parameters has been merely investigated experimentally (Askne et al. 1995). It should depend on:
· local wind speed over the canopy, in addition to the topographic exposure to wind which will be described in the following;
· tree stiffness, expected to be related to tree species and structure (Hagberg et al. 1993).
No model exists to relate the scatterers displacement to above parameters. Askne et al.( 1995) use a parameter k in a simple exponential model to describe the temporal decorrelation within the crown layer, k being related in an unknown way to above parameters. This will be investigated in a near future on our data set.
At this stage, we will focus on the topographic effect on the correlation, knowing the hourly wind speed |W| and the topography through the DEM. We consider that the topography attenuates the speed of the wind vector W, giving a local wind vector Wloc through:
Wloc = W*T(thetaw) (3)
where T is an unknown attenuation function assumed to depend on a wind exposure angle, that we defined as:
with alpha and beta being the terrain slope and aspect angles, and betaw the wind direction. thetaw is simply the angle between the wind vector W and the normal vector to the terrain N, as sketched in Fig. 4.
First, from all above considerations, we can interpret Fig. 2 as follows :
· as usually observed, forests have lower correlation due to volume decorrelation;
· higher correlation for young pines compared to mature ones can be explained by higher ground contribution and/or higher tree stiffness, as observed in the field;
· lower correlation for larger time spans, for both pine
ages, could be explained by an additional volume decorrelation
due to new annual shoots (about 30 cm increase).
Second, we can interpret in more details the correlation decrease with bole volume in Fig. 3 :
· Correlation decrease with bole volume is mainly due to decreased ground contribution and increased displacement with stand age due to decreased stiffness; this can be understood as for mature trees, needles are mainly holded by branches of higher order (3 or 4) which are long and thin, compared to young trees where needles are holded by low order branches (1 or 2) which are short and relatively big, thus stiffer;
· The high dispersion around the trend in Fig. 3a can be explained by the topographic exposure to wind. If we consider in Fig. 3b&c two data subsets, that is young and mature pines respectively, and plot the correlation against the wind exposure angle (eq. 4), we see that:
· for mature trees in Fig. 3c, the correlation is highly correlated to thetaw, with a quasi-linear decrease (r2=0.78); this explains also the correlation variation visually observed just below the Mende town in Fig. 1b, which seems to be linked to the topography (2 small parallel valleys ( NE-SW) with old pine stands);
· for young pines in Fig. 3b (see
also Fig. 1b on the Causses Méjean),
no relationship with thetaw is observed; probably due to the correlation
mainly driven by the ground contribution, in addition to the higher
stability of young trees under the local wind influence. Therefore,
dispersion in Fig. 3b could be explained
in the variability of the transmissivity and the cover fraction,
in addition to soil moisture and roughness that all modulate the
In this paper, ERS DEM-differential interferometric correlation was investigated over a hilly forested test-site. This correlation was found to be a good land-use discriminator for major land-use types, especially using shorter time intervals. On forest covers, the correlation is partially linked to the bole volume (growth stage of the forest in fact), due to decreased ground contribution and increased crown volume scattering, originating mainly from needles and twigs. With ERS characteristics (C-band 23deg.), such volume scattering rapidly decorrelates due to centimetric random displacements easily generated by low winds (10km/h). As is it rare that there is no wind at all, and as the wind speed can change rapidly, we can expect that very high correlation on forests can not be achieved using repeat-pass interferometry, towards a performant DEM generation over forested areas.
Although the wind conditions were quite similar for the different ERS pairs investigated, the correlation behavior with the wind speed was highlighted using the tandem scene, through the variation of the local wind speed due to the topographic exposure to wind. Therefore, in a sense, for similar mature pine forest, we can say in our case that the correlation provides us with an indicator of the local wind speed at the canopy level.
Globally, this study confirmed that the correlation is linked to many parameters, including those related to the interferometric configuration (baseline and incidence angle), the weather conditions (wind speed and direction), the topography and the ground and forest covers. Therefore, for generalization purposes, to which point this correlation can be linked robustly to characteristics of interest is a new and exciting challenge. Carying on such work is probably worthwhile, considering the potential of interferometric SAR data in addition to the usual backscatter information, especially at L-band.
To this aim, future works will include the use of semi-empirical models as proposed by Askne et al. (1995) for a better understanding of the data set. Also, other tandem couples will be processed. Finally, ERS differential interferometric pairs with sufficient correlation will be investigated for the phase information content, which should be linked to the forest height.
This work was performed within a joint LCT-SCOT project funded by the CNES (contract # CNES/94/0239), that we would like to warmly thank. Particularly, we thank Didier Massonnet for the access to the CNES interferometric processor. Thanks also go to ESA for providing the SAR data within the project-pilot PP2-F132, and P. Gigord from IGN who provided us with the DEM (BD CARTO IGN (c)) within a collaboration on the correction of SAR bacskcatter data acquired over hilly terrain. Final thanks are for Luce Castagnas, from LCT, for the html editing.
Askne, J., Dammert, P., Fransson, J., Israelsson,
H. and Ulander, L.M.H, 1995:
Beaudoin, A., Castel, T., Deshayes, M., Stachs, N., Stussi, N.
and Le Toan, T., 1995:
Carnec C., Massonnet D. and King C., 1995 :
Hagberg, J.O., Ulander, L.M.H. and Askne, J., 1995:
Herland, E.A., 1995:
Massonnet D., Rossi M.,Carmona C.,Adragna F.,Peltzer
G., Feigl K., Rabaute T.,1993:
Massonnet D., 1994 :
Massonnet D., Briole P. and Arnaud A., 1995 :
Ulander, L., Dammert, P.B.G. and Hagberg, J.O., 1995:
Wegmuller, U. and Werner, C.L., 1995a:
Wegmuller, U. and Werner, C.L., 1995b:
Wegmuller, U., Werner, C.L. and Nuesch, D., 1995:
Zebker, H.A. and Villasenor, J., 1992:
Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry
|Copyright 2000 - European Space Agency. All rights reserved.|