Citation: Balzter, H.

(2001): Forest mapping and monitoring with interferometric Synthetic

Aperture Radar (InSAR). Progress in Physical Geography 25, 159-177.

Forest mapping and monitoring with interferometric

Synthetic Aperture Radar (InSAR)

BALZTER, Heiko

Address for correspondence:

Heiko Balzter
Centre for Ecology and Hydrology
Institute of Terrestrial Ecology
Section for Earth Observation
Monks Wood
Abbots Ripton
Huntingdon
Cambridgeshire
PE28 2LS
UK
Tel +44 (0) 1487 77 2471
Fax +44 (0) 1487 77 3277
E-mail: [email protected]

1
 1 Keywords

4 biomass, stem volume, forest structure, fire scars, deforestation, freeze-thaw transition, tree

5 height, land cover mapping, SAR

8 Abstract

10

11 A Synthetic Aperture Radar (SAR) is an active sensor transmitting pulses of polarized

12 electromagnetic waves and receiving the backscattered radiation. SAR sensors at different

13 wavelengths and with different polarimetric capabilities are being used in remote sensing of

14 the Earth. The value of an analysis of backscattered energy alone is limited due to ambiguities

15 in the possible ecological factor configurations causing the signal. From two SAR images

16 taken from similar viewing positions with a short time-lag, interference between the two

17 waves can be observed. By subtracting the two phases of the signals, it is feasible to eliminate

18 the random contribution of the scatterers to the phase. The interferometric correlation and the

19 interferometric phase contain additional information on the three-dimensional structure of the

20 scattering elements in the imaged area.

21 A brief review of SAR sensors is given, followed by an outline of the physical foundations of

22 SAR interferometry and the practical data processing steps involved. An overview of

23 applications of InSAR to forest mapping and monitoring is given, covering tree bole volume

24 and biomass, forest types and land cover, fire scars, forest thermal state and forest canopy

2
25 height.

26

27

28 I Introduction

29

30

31 Remote sensing of forests has an important role in mapping large forest tracts that are

32 difficult to access on the ground, and in monitoring changes in these forests. The forest

33 canopy is characterised by different vegetation layers, like weeds, shrubs, undergrowth and

34 different tree canopy layers. Optical sensors can only detect the upper canopy, where the

35 absorption and reflection of parts of the spectrum of visible light and infrared is taking place.

36 Therefore, radar sensors have widely been used for large-scale forest mapping. A radar

37 operates in the microwave spectrum at wavelengths typically between 3 and 25 cm, much

38 longer than visible light. The sensor actively transmits pulses of electromagnetic energy and

39 receives the response from the imaged area. Because of the longer wavelength, the radiation

40 penetrates the top vegetation layer to a certain extent and is scattered by stems, branches,

41 twigs, leaves or needles. Microwaves also enable a weather- and illumination-independent

42 imaging process: They can penetrate clouds, dry snow and to some extent rain. Virtually no

43 part of the Earth's surface is permanently covered with rain of sufficient intensity to cause

44 major difficulties. Observations can be made at day and night, for instance throughout the

45 winter darkness of the polar regions.

46 Analysis of the data produced by Synthetic Aperture Radar (SAR) sensors can be used to

47 provide estimates of parameters such as forest area, forest thermal state (frozen / thawed),

48 forest biomass density and tree height. Such studies have been conducted over a wide range

3
49 of climate zones but more recently have been concentrated on the boreal forests of North

50 America and Eurasia. This article gives an introduction to the imaging process, available SAR

51 systems and the different areas of application of interferometric SAR to forest mapping.

52 Polarimetric SAR is only covered if it is also interferometric. A wide range of references

53 concerning the use of conventional SAR (i.e. backscatter at one wavelength and polarization)

54 is deliberately not covered here.

55

56

57 II Synthetic Aperture Radar

58

59

60 1 Sensor characteristics

61

62

63 A radar sensor transmits pulses of electromagnetic radiation in the microwave spectrum.

64 When the radiation hits an object, the electromagnetic wave is scattered and a fraction of it is

65 reflected in the direction of the sensor. The amount of radiation received by the sensor is

66 called radar backscatter. In contrast to Real Aperture Radar systems, Synthetic Aperture

67 Radar (SAR) makes use of the Doppler effect of the aircraft or satellite motion to increase the

68 resolution of the images. A SAR is a coherent imaging sensor measuring both real and

69 imaginary components of the backscattered signal. A sensor is mainly characterised by the

70 wavelength (or frequency), polarization, range and azimuth resolution. Operational radar

71 wavelengths used for forest mapping are X-band (3.1 or 3.5 cm wavelength), C-band (5.65

72 cm), L-band (24 cm) and P-band (30-60 cm). Important platform features are the available

73 swath widths and the repeat cycle. Longer wavelengths tend to penetrate deeper into the

4
74 vegetation canopy. The sensor transmits a polarized wave. If the electrical field of the

75 electromagnetic wave oscillates horizontally, the transmitted wave is horizontally (H)

76 polarized. Vertical (V) polarization is defined in the same way. Dual-polarization sensors can

77 transmit and receive in both polarizations, but only quad-polarized (fully polarimetric)

78 sensors record the whole polarization vector of the backscattered signal.

79 The backscattered intensity is higher for vegetated areas than for bare soil, because of the

80 multiple scattering in the vegetation layer. From calm water surfaces, the backscatter is very

81 low, as most of the radiation is reflected from the water surface away from the sensor.

82 The magnitude of microwave backscatter is a result of the geometric and dielectric properties

83 of the surfaces or volumes imaged. It is sensitive to the surface geometry (topography),

84 surface roughness (surface slope, variation of surface height, plant geometry) and water

85 content of surface materials (crop and soil moisture, snow wetness).

86 The received backscatter in one pixel is the spatial sum of radar echoes of all scatterers within

87 the imaged area on the ground. Adding up electromagnetic waves with different phases may

88 result in constructive and destructive interference. Constructive interference means that the

89 amplitude of the resulting radar echo is larger than that of the interacting waves, and

90 destructive interference means that the resulting amplitude is smaller. This phenomenon is

91 called “speckle” and causes SAR images to look like “salt and pepper”. To improve the

92 accuracy of the backscatter estimation, backscatter values of adjacent pixels in the single-look

93 image are averaged. This process of multi-looking improves the radiometric resolution at the

94 expense of spatial resolution. It changes the distribution function of the backscattered power.

95

96

97 2 Airborne and spaceborne SAR sensors

98

5
 99

100 Five spaceborne imaging radars systems were until recently or are currently in operation,

101 ERS-1, JERS-1 (terminated in October 1998), SIR-C (operated for two 10 day periods during

102 1994), ERS-2 and Radarsat. Missions planned for the near future include the European

103 ENVISAT and the Japanese ALOS satellites. Details of these systems are shown in Table 1.

104 The suitability of a SAR sensor for interferometric purposes requires a well calibrated phase

105 (Freeman 1992).

106 • The European Research Satellite - 1 (ERS-1) is the first of two identical polar orbiting

107 Earth-viewing satellites launched by the European Space Agency (ESA). The imaging radar is

108 C-band (5.7 cm wavelength or a frequency of 5.25 GHz) with vertical transmit and vertical

109 receive (VV) polarization. It illuminates the Earth's surface at an incidence angle of 23º with a

110 transmitted power of 4.8 kW per pulse. ERS-1 images a swath 100 km in width at a

111 resolution of 30 m (for 4 looks). The satellite altitude ranges from 775 km, providing a 3 day

112 repeat coverage for the winter ice phases for the commissioning phase which comprised the

113 first 3 months of the mission, to 781 km, providing a 35 day repeat for global access, to 783

114 km, providing a 168 day repeat for the geodetic portion of the mission.

115 • ERS-2 was launched in spring 1995 into a 781 km, 35 day repeat orbit identical to ERS-1.

116 On a number of occasions ERS-1 and ERS-2 have been operated as a one-day repeat Tandem

117 Mission primarily for interferometric applications. In addition the ERS repeat cycle of 35

118 days means that interferometric measurements can also be made with a temporal repeat cycle

119 of n*35 days. Interferometric pairs can be browsed using a software package from the

120 European Space Agency (Delia and Biasutti 1999). Rufino et al. (1998) reported that only

121 tandem pairs allow an efficient interferometric processing because of their sufficiently short

122 time-lag of one day, whereas correlation adequate for differential interferometry could not be

6
123 achieved.

124 • ESA plans to launch a further SAR system aboard its ENVISAT satellite in 2000. As with

125 ERS-1 and ERS-2 this will be a C-band system (though at a slightly different frequency

126 which precludes ENVISAT/ERS-2 interferometry). The advanced SAR (ASAR) will have

127 many modes, unlike the single mode of the ERS SAR. These modes include an alternating

128 polarization mode enabling it to transmit in horizontal or vertical polarization and receive in

129 both. Although still not a fully quad-polarized system, it will measure the cross polarized

130 (HV) return which is important for forestry applications. ASAR also has a ScanSAR mode

131 which permits imaging over wider swathes at coarser resolutions. Like ERS the ENVISAT

132 orbital repeat cycle will be 35 days, but unlike ERS for ENVISAT interferometric

133 measurements will need to be specifically programmed into the satellite mission as a result of

134 the flexibility of ASAR. In fact for vegetation studies at C-band a 35 day repeat cycle is likely

135 to be of only limited use when contrasted with the 1 day Tandem Mission repeat cycle of

136 ERS-1 / ERS-2.

137 • The Japanese Earth Resources Satellite - 1 (JERS-1) was launched by the National Space

138 Development Agency of Japan (NASDA) in Summer 1992 on a 3 year mission. The

139 characteristics of its SAR are very similar to those of NASA’s Seasat satellite which laid the

140 foundation for all later satellite SARs during its a brief 3 month mission in 1978. The JERS-1

141 imaging radar is L-band (20 cm wavelength or a frequency of 1.28 GHz) with horizontal

142 transmit and horizontal receive (HH) polarization. The transmitted power of a pulse is 1.3

143 kW. It illuminates the Earth's surface at an incidence angle of 35º from nadir, and images a

144 swath 75 km in width at a resolution of 18m (for 3 looks). The satellite orbital repeat cycle

145 means that interferometry can be conducted with a repeat-cycle of n*44 days (Rossi et al.

146 1996). The satellite orbit was not maintained as exactly as that of ERS so that the spatial

7
147 distance between the two antenna positions was often larger, but for JERS-1 this was less

148 critical because of the longer wavelength and the greater incidence angle. Imaging positions 5

149 km apart would still produce fringes in the interferogram, whereas for the ERS satellites 1.1

150 km is the largest antenna separation for which interferometry is theoretically possible (Rossi

151 et al. 1996). In 1998 JERS-1 stopped transmitting data because of a failure of the solar panels.

152 JERS-1 has been used in the Global Rain Forest Monitoring Project to make radar mosaics of

153 the entire tropical forest belt, comprising the Amazon basin and Congo basin (both at high

154 and low water levels), West Africa, South-East Asia, Indonesia and Papua New Guinea. Data

155 has also been acquired in the Global Boreal Forest Monitoring Project to make similarly large

156 scale mosaics of the entire boreal forest belt, comprising the North American forests (Alaska

157 and Canada) and Eurasian forests (Scandinavia and Russia).

158 • NASDA plan to launch an Advanced Land Observation Satellite (ALOS) in 2002, which

159 will carry a Phased Array type L-band SAR (PALSAR) capable of polarimetry and repeat-

160 pass interferometry.

161 • Two radar missions are planned for the near future by NASA: The Shuttle Radar

162 Topography Mission (SRTM) will generate a global high resolution digital elevation model

163 from X-band single-pass interferometry in which the second antenna will be deployed at the

164 end of a 80m boom extended from the Space Shuttle, and LightSAR will be used for the

165 development of commercial applications. LightSAR is planned to carry an L- and an X-band

166 antenna (Hilland et al. 1998).

167 • The Canadian RADARSAT satellite can observe in a number of resolution and swathe

168 width modes including a Spotlight (high resolution) mode and a ScanSAR mode which

169 permits imaging over swathes up to 500 km wide at a resolution of 100 m. As with Envisat,

170 because of the flexibility of the SAR, interferometric measurements have to be specifically

8
171 programmed into the satellite mission.

172 • The only fully polarimetric spaceborne SAR so far was the L- and C-band system aboard

173 SIR-C which was deployed during two 10-day missions during 1994.

174 • Polarimetric data are available from a number of airborne systems including JPL’s

175 TOPSAR system (Madsen et al. 1995), JPL’s AIRSAR aboard a NASA DC-8, TUD-DCRS’s

176 EMISAR aboard a Danish Airforce Fanjet (Christensen et al. 1998), DLR’s E-SAR aboard a

177 Dornier 228, and more recently Dornier’s DoSAR also aboard a Dornier 228. Do-SAR was

178 the first airborne single-pass interferometric SAR in Europe (Faller and Meier 1995), and has

179 been used to estimate terrain height with an accuracy of 2-5 m. Gray and Farrismanning

180 (1993) studied repeat-pass interferometry with an airborne SAR. Coherence between separate

181 images requires very accurate flightline control and very close flightpaths with offsets less

182 than a few tens of meters. Repeat-pass interferometry with airborne SAR opens the possibility

183 for temporal coherence studies and differential interferometric SAR experiments with the

184 flexibility afforded by the airborne platform. As well as a direct use for programmes

185 observing relatively small areas of the Earth, these airborne radars act as development

186 systems for future more sophisticated satellite radars.

187

188

189 III SAR Interferometry

190

191

192 In this section, a brief introduction to the background of SAR interferometry is provided. For

193 a thorough and detailed introduction the interested reader is referred to Bamler and Hartl

194 (1998), and for a review of techniques and applications see Gens and Van Genderen (1996).

9
195

196

197 1 Physical background

198

199

200 Radar backscatter is measured as a complex number, containing information about the

201 intensity of the signal and the phase (Figure 1). The phase is determined by the two-way path

202 length from the sensor to the resolution cell on the ground and the interference between

203 individual scatterers (e.g. trees) in that cell. If two complex SAR images have been acquired

204 over the same area from very close antenna positions then the within-cell interference

205 contribution to the phase is almost identical for both images. The temporal or spatial

206 separation between the two antennas of the interferometric SAR signals is called the baseline.

207 Figure 2 illustrates the viewing geometry of SAR interferometry. The phases of the two

208 signals interfere in a characteristic pattern. The phase difference between the two images for

209 each resolution cell is directly related to the difference between the viewing distances of the

210 two sensors. In particular the average three-dimensional position of the scattering elements

211 may be inferred leading to the capability to derive topographic maps from the phase

212 difference images.

213 SAR backscatter intensity is strongly affected by terrain properties (slope and aspect). SAR

214 interferometry provides a method of removing topographic effects from the backscatter

215 without the need for additional external data sets and leaving only backscatter variations

216 arising from changes in target parameters, such as vegetation biomass or soil moisture. The

217 capability to derive Digital Elevation Models (DEM) also provides a vital input into mapping

218 out drainage networks and separating water catchments, particularly in poorly surveyed areas.

219 The two images from which an interferogram is generated can either be acquired using one

10
220 antenna for repeated passes over the same area at two different times (Repeat Pass

221 Interferometry) or can be acquired simultaneously using two spatially separated antennas on

222 the same platform (Single Pass Interferometry).

223 To compute an interferogram the two single look complex (SLC) SAR images are first co-

224 registered to an accuracy of less than 0.1 pixel. The complex vector product is then formed on

225 a pixel by pixel basis to derive a phase difference and a correlation at each position. The

226 interferometric correlation is a measure of the accuracy of the estimation of the

227 interferometric phase. A normalised interferogram is defined as the complex degree of

228 coherence of the two complex image values s1 and s2 given by:

229

s1 s 2 *
230 [1]
s1 s1 * s2 s2 *

231

232 where the < > brackets represent an ensemble average, formed by coherently averaging the

233 complex values of n single look pixels, and * represents the complex conjugate:

234

n
1
235 s1 s 2 * s1,i s 2,i * [2]
n i 1

236

237 Interference phenomena such as fringes will be observed so long as there is at least partial

238 coherence between the two images. The phase of is the interferometric phase and the

239 magnitude of is the degree of coherence between s1 and s2. represents the fringe visibility

240 and lies within the range zero to unity. can be shown to be the product of three terms:

241

242 noise temporal spatial

[3]

11
243

244 representing decorrelation arising from the system noise, from differences in the target over

245 the temporal and over the spatial baseline (Zebker and Villasenor 1992).

246

247 | |noise is related to the signal to noise ratio (SNR) of the sensor, and is only significant for

248 areas of very low backscatter:

249

1
250 [4]
noise 1
1
SNR

251

252 | |spatial , the baseline decorrelation, arises from surface and volume scattering and can be

253 separated as follows:

254

255 spatial slantrange volume

[5]

256

257 | |slantrange can be increased to 1 by applying common band (spectral shift) filtering. However

258 | |volume is only 1 in the case when the scatterers are confined to a plane. When they are

259 distributed in depth, as for example with multiple scattering from a forest canopy, then there

260 is appreciable volume decorrelation.

261

262 | |temporal , the decorrelation of the target arising from the time separation of the two

263 observations, is very low for stable, man-made structures, moderate for bare soil surfaces and

264 agricultural crops and substantial for tall vegetation such as forests. In the case of vegetation

265 targets, these temporal changes are caused by changes of the scatterers (growth or loss of

12
266 foliage, wind motion) and changes in their dielectric constant (surface moisture film, freezing,

267 thawing). For agricultural crops the mechanical cultivation associated with farming activities

268 such as harvesting, ploughing, mowing, and tillage causes complete decorrelation as almost

269 all scatterers are changed.

270 In addition to being influenced by temporal and spatial effects, the interferometric coherence

271 is, like the backscatter intensity, also influenced by the observing wavelength. Backscatter

272 arises predominantly from target components on the scale of the radar wavelength. Thus in

273 the case of a forest canopy JERS-1 L-band backscatter arises more from the trunk and

274 branches of the trees whereas ERS C-band backscatter arises more from their twigs and

275 leaves or needles. Since the longer wavelength scatterers are more stable, coherence tends to

276 be maintained over a longer temporal interval at longer wavelengths. For example for a forest

277 canopy coherence is maintained over the 44 day repeat cycle of JERS-1, but not over the 35

278 day repeat cycle of ERS. In the latter case image pairs from the Ice Phase with a three day

279 repeat cycle for ERS-1 or from the Tandem Mission with one day repeat cycle for ERS-1 /

280 ERS-2 are required to maintain coherence. In all cases the coherence or fringe visibility must

281 be sufficient to enable the fringe phase to be derived accurately, for an error in phase

282 translates directly into an error in height measurement.

283 Interferogram images derived from repeat-pass spaceborne SAR systems are known to exhibit

284 artefacts due to the time and space variations of atmospheric water vapour. Zebker et al.

285 (1997) therefore recommended to use the longest radar wavelengths possible and to maximise

286 the spatial baseline within decorrelation limits. To detect surface deformation they suggest to

287 use multiple observations and to average the InSAR derived products.

288 To ensure the highest coherence over vegetation targets, temporal decorrelation can be

289 avoided completely by making the interferometric measurements almost coincident in time as

290 well as almost coincident in space. Rather than a Repeat Pass Interferometric system, a Single

13
291 Pass Interferometric system is required - the two measurements from which the interferogram

292 is to be generated are made from two antennas across the track. The first such system in space

293 will be the Shuttle Radar Topographic Mission (SRTM).

294 Tough et al. (1995) analyse the statistical distributions of the amplitude and phase difference

295 in single look data. However, multi-look data have completely different statistical properties,

296 and Lee et al. (1994) examine the probability distribution functions of the multi-look phase

297 difference, magnitude of complex product, and intensity and amplitude ratios between two

298 components of the scattering matrix. They conclude that the distribution functions depend on

299 the complex correlation coefficient and the number of looks.

300 In broad terms, SAR interferometry provides information on the spatial distribution of the

301 scatterers which make up the target while SAR polarimetry provides information on the

302 scattering mechanisms predominating in each target, for example surface scattering, double-

303 bounce scattering and volume scattering. Polarimetric SAR interferometry provides

304 information on the spatial distributions of the scattering mechanisms making up the target. In

305 particular, the decorrelation components | |spatial and | |temporal in eq. 3 are known to be

306 polarization-dependent (Cloude and Papathanassiou 1998).

307 In the case of a forest observed at L-band, where the total backscatter is made up of

308 contributions from a number of different scattering mechanisms, the co-polarized return tends

309 to have a substantial component arising from returns from the ground while the cross-

310 polarized return tends to be dominated by returns from multiple scattering within the forest

311 canopy. The vertical spatial separation of the phase scattering centres of different scattering

312 mechanisms in the canopy provides the basis for an improved retrieval of vegetation height.

313

314

14
315 2 Interferometric processing chain

316

317

318 The first processing step is co-registration of the Single Look Complex (SLC) images (Lin et

319 al. 1992; Fornaro and Franceschetti 1995; Rufino et al. 1998). To achieve a high quality

320 interferogram, co-registration at sub-pixel accuracy is required, ideally better than 0.2 pixel,

321 otherwise the interferometric coherence is reduced considerably. In the following step

322 common band filtering is performed to improve coherence estimation by increasing | |slantrange

323 in eq. 5.

324 In the second step the normalised complex interferogram is computed. The two co-registered

325 images are multi-looked to improve estimation accuracy and then cross-correlated. The

326 resulting interferogram consists of complex values with the magnitude corresponding to the

327 multi-looked interferometric correlation and the phase to the interferometric phase.

328 The phase trends in azimuth and range direction resulting from the Earth’s curvature is then

329 removed from the interferogram (phase flattening).

330 To retrieve the effective height from the phase of the complex interferogram the correct

331 multiple of 2 has to be added in the phase unwrapping step. Phase unwrapping is

332 problematic due to fringe discontinuities caused by layover (Gelautz et al. 1996), areas of low

333 coherence, and phase noise. Filtering and multi-looking can be used to reduce the phase

334 noise. A review of phase unwrapping techniques is given by Griffiths and Wilkinson (1994).

335 Phase-unwrapping is often based on Goldstein's branch-cut approach (Goldstein and Werner

336 1998). However, holes that are isolated by branch-cuts often remain in interferograms with

337 high noise levels. Wang and Li (1999) present two algorithms to improve the unwrapped

338 phase image. Just and Bamler (1994) studied the dependence of the phase bias and variance

339 on processor parameters. Phase noise can be characterised by an additive noise model. Lee et

15
340 al. (1998) developed an adaptive filtering algorithm based on this noise model. Their

341 algorithm can be included in an iterative phase unwrapping step. Other approaches to phase

342 unwrapping are based on fringe detection (Lin et al. 1992), region growing (Fornaro and

343 Sansosti 1999, Xu and Cumming 1999), weighted least squares (Pritt 1996), the finite

344 element method (Fornaro et al. 1997a), the Green’s function and the Helmholtz equation

345 (Fornaro et al. 1996, Lyuboshenko and Maitre 1999), the fast Fourier transform (Costantini et

346 al. 1999), the minimum cost flow on a network (Costantini 1998) and local frequency

347 estimates (Trouve et al. 1998). Zebker and Lu (1998) present a synthesis of two frequently

348 used phase unwrapping algorithms, the residue-cut and the least-squares techniques. Their

349 synthesis offers greater spatial coverage with less distortion. Fornaro et al. (1997b) compared

350 global and local phase unwrapping techniques. Bamler and Hartl (1998) discuss approaches

351 to phase unwrapping in more detail.

352 After successful phase unwrapping, a height map can be derived (Madsen et al. 1993; Zebker

353 et al. 1994a; Zebker et al. 1994b). For this step a precise baseline estimate is crucial, and a

354 refined baseline estimation needs to be carried out with a number of ground control points of

355 known height. Accurate baseline estimation is crucial for derivation of a height map, and the

356 orbital state vectors may not always provide the required precision. A mathematical method

357 for the estimation of the two-dimensional orbital shift based on the fringe pattern in the

358 interferogram has been developed by Goyal and Verma (1996) to tackle the problem.

359 In the final processing step, the height values in SAR image geometry (slant range) are

360 transformed to orthonormal coordinates.

361

362

16
363 IV Applications of InSAR to forest mapping and monitoring

364

365

366 1 Tree biomass and bole volume

367

368

369 Within NASA’s Mission to Planet Earth Kasischke et al. (1997) describe the capability of

370 imaging radars to monitor variations in biomass in forest ecosystems. Radar backscatter is

371 increasing in a non-linear way with forest biomass. The shape of this function is known to be

372 dependent on wavelength, polarization, forest type and moisture conditions. As Baker et al.

373 (1994) showed for Corsican Pine stands, the cross-polarized term (horizontal transmit,

374 vertical receive, HV) is often most strongly correlated with forest biomass. This phenomenon

375 is caused by the depolarization of the electromagnetic waves by multiple scattering events in

376 the canopy. At a certain biomass level, the radar signal saturates. Dobson et al. (1992)

377 analysed radar responses at P-, L- and C-band to biomass of mono-species conifer plantations

378 at Les Landes, France, and Duke forest, North Carolina, and found an approximately linear

379 response of backscatter to increasing biomass with wavelength-dependent saturation levels

380 around 200 t/ha for P-band and 100 t/ha for L-band. In the study of Imhoff (1995) saturation

381 was reached at 100 t/ha for P-, 40 t/ha for L- and 20 t/ha for C-band in coniferous and

382 broadleaf evergreen forests. Luckman et al. (1998) found a saturation at 60 t/ha for L-band in

383 tropical forests. Ranson et al. (1995) used the ratio of the cross-polarized intensities at two

384 wavelengths (L-HV / C-HV), to estimate boreal forest biomass in Canada with a 95%

385 confidence interval of ±20 t/ha, and found that saturation was reached at 200 t/ha. The

386 accuracy with which biophysical parameters can be retrieved from SAR measurements of

387 forests depends considerably upon vegetation structure and ground conditions (Baker and

17
388 Luckman 1999). The upper levels of sensitivity for L-band and C-band systems such as SIR-

389 C range between <100 t/ha for complex tropical forest canopies to 250 t/ha for simpler forests

390 dominated by a single tree species. Best performance for biomass estimation is achieved

391 using lower frequency (P- and L-band) radar systems with a cross- polarized (HV or VH)

392 channel. The long wavelength is required to penetrate the upper canopy layer and interact

393 with branches and stems. Interferometric SAR can improve biomass retrieval from radar

394 backscatter through the interferometric information about the imaging geometry: The fringe

395 frequencies of the interferogram can be used to correct radar backscatter intensity for terrain

396 effects (Ulander 1996), and thus improve radiometric calibration.

397 The interferometric coherence is sometimes found to decrease with increasing forest biomass,

398 but this relationship is temporally unstable and may be affected by changing weather

399 conditions between the repeated image acquisitions, terrain effects, wind, rain, snow and

400 moisture, freezing and thawing or the spatial baseline for different image pairs.

401

402

403 2 Classification of forest types and land cover

404

405

406 Radar provides a means to classify land-cover patterns because of its sensitivity to variations

407 in vegetation structure and vegetation and ground-layer moisture (Kasischke et al. 1997).

408 Like-polarized imaging radars (HH or VV) are well suited for the detection of flooding under

409 vegetation canopies, as has been demonstrated with JERS-1 SAR for the Amazon basin.

410 Lower frequency radars (P- and L-band) are best suited for detecting flooding under forests,

411 whereas higher frequency radars (C- band) work best for wetlands dominated by herbaceous

412 vegetation (Kasischke et al. 1997).

18
413 During the Shuttle Imaging Radar C (SIR-C) campaigns of April and October 1994 the orbit

414 was tightly controlled to give baselines sufficiently short for interferometry to be conducted

415 using data from the missions. Rignot (1996) made repeat-pass interferometric radar

416 observations of tropical rain forest in the state of Rondonia, Brazil, and found that over the

417 forest no coherence between the two signals was found at C-band but that at L-band

418 coherence was maintained over the entire landscape. Similar observations were made by

419 Rosen et al. (1996), who made coherence measurements of Kilauea volcano, Hawaii. This is

420 caused by the different scattering mechanisms affecting the signal. Short wavelengths are

421 scattered mostly by leaves, twigs or needles in the tree crowns, whereas longer wavelengths

422 penetrate deeper into the canopy and are scattered by large branches and stems. Because the

423 orientation of the leaves in the tree crown changes with wind, but the upper branches remain

424 geometrically more or less unchanged, coherence is lost at C-band but preserved at L-band.

425 Coltelli et al. (1996) used multifrequency repeat-pass interferometry over Mount Etna, Sicily,

426 and found that the coherence maps allowed vegetated and unvegetated areas to be separated.

427 Askne et al. (1997) used ERS InSAR coherence to separate forested and non-forested areas. A

428 hierarchical unsupervised segmentation algorithm for land cover classification from multi-

429 temporal InSAR images has been developed by Dammert et al. (1999). Wegmuller and

430 Werner (1995) studied the potential of SAR interferometry for forest mapping and

431 monitoring, using interferometric correlation and backscatter intensities from ERS-1 SAR

432 repeat-pass interferometric data. Forest could be clearly discriminated from other land

433 categories. Coherence increased from coniferous, mixed to deciduous forest. Because

434 Wegmuller and Werner (1995) used a November image pair, the deciduous trees had shed

435 their leaves. The branches of these deciduous winter forests act as scatterers which are more

436 stable over time than the needles of winter-green coniferous trees. Dependencies on the

437 spatial and temporal baselines and the seasons were also analysed. The results found for the

19
438 temperate forest site around Bern, Switzerland, were extended to a boreal forest site along the

439 Tanana River, Alaska. Repeat-pass interferometry was found to be particularly sensitive to

440 changes such as soil freezing, mechanical cultivation of agricultural fields, and vegetation

441 growth. In the case of soil moisture changes and freezing, the dielectric properties change

442 without a simultaneous geometric change so that the backscatter difference is high but the

443 coherence is the same. To visualise the information content of the interferometric signatures

444 Wegmuller and Werner (1995) proposed a colour-composite comprising the coherence in red,

445 mean backscatter in green and backscatter change in blue (see image 4). In this representation

446 different targets have readily identifyable colours, for example forests tend to appear in green

447 (low coherence, high backscatter and little backscatter change), water in blue (low coherence,

448 low backscatter but significant backscatter change) and bare soil areas in red (high coherence,

449 medium backscatter and little backscatter change). In a subsequent paper Wegmuller and

450 Werner (1997) studied the retrieval of vegetation parameters using SAR interferometry.

451 Based upon the interferometrically derived forest map generated above, a classification was

452 derived and then geocoded using the interferometrically derived height map generated from

453 the same ERS SAR data pair. From a digital forest map the remotely sensed classification

454 was validated and mapping accuracies of over 90% were achieved. Coniferous, deciduous and

455 mixed forest stands could be distinguished and separated from orchards, regrowth and clear-

456 cut areas. Another example for land cover classification is given by Dutra and Huber (1999),

457 who compared different classification algorithms for four land-cover classes of the Czech

458 Republic from ERS InSAR imagery with an overall classification accuracy exceeding 90%.

459

460

461 3 Fire scars

462

20
463

464 The large volume of ERS data collected at the Canadian and Alaskan receiving stations have

465 been used for a detailed study of high-latitude terrestrial ecosystems in North America. The

466 SAR imagery has been used to study carbon dioxide fluxes from boreal forests, particularly

467 the effect of boreal forest fires on the CO2 flux and the measurement of the length of the

468 growing season as an aid in determination of the seasonal CO2 flux, many of these studies

469 being parts of the BOREAS project (Moghaddam and Saatchi 1995; Ranson et al. 1995;

470 Chang et al. 1997; Ranson et al. 1997; Way et al. 1997).

471

472

473 4 Forest thermal state

474

475

476 In boreal forests the summer frost-free period bounds the growing season length for

477 coniferous forest species and the period of root and soil respiration and decomposition in the

478 broader landscape, while for both coniferous and deciduous forest species the growth

479 potential is further limited by their capability for mineral and water uptake. In studying the

480 seasonal dynamics of the boreal forest ecosystem the onset and duration of favourable soil

481 temperatures is therefore as important as the temperature regime of the forest canopy.

482 The length of the growing season can be determined by using imaging radar data to monitor

483 freeze/thaw transitions. At microwave frequencies freezing results in a large decrease of the

484 dielectric constant of soil and vegetation because the crystal structure in frozen water prevents

485 the rotation of the polar water molecules which they contain. Wegmuller and Werner (1995)

486 attributed a 3-4 dB drop in radar backscatter from bare soils during day-night freeze-thaw

487 cycles to this phenomenon. Backscatter change resulting from freezing and thawing was first

21
488 observed in image data in a series of L band aircraft radar data sets that were acquired over

489 the Bonanza Creek Experimental Forest site near Fairbanks, Alaska, in 1988 (Way et al.

490 1990; Way et al. 1994).

491 The most recent study by Way et al. (1997) looked at ERS-1 Ice Phase data (3 day repeat) and

492 used stem temperature to assess changes in backscatter. They found three distinct regimes:

493 i) below freezing, where backscatter is low and relatively constant;

494 ii) above freezing, where backscatter is higher and may reflect variations in the moisture

495 status of the forest; and

496 iii) a transition from lower to higher values as different components of the ecosystem thaw

497 (indicated by a range of stem temperatures which include values below zero, when canopy

498 components thaw, to values above zero, when the soil thaws).

499

500

501 5 Forest canopy height

502

503

504 Because of the deeper penetration into the forest canopy at longer radar wavelengths, the

505 interferometric effective height inferred from L-band and C-band SAR interferometry will

506 constitute different fractions of the overall canopy height. Mean tree height in a pixel can

507 therefore be inferred

508 i) from the height discontinuity at a boundary between a tree canopy and an adjacent cleared

509 area;

510 ii) from the difference between the phase scattering centres as a function of wavelength;

511 iii) from the difference between the interferometric effective height from InSAR and a

512 sufficiently accurate digital elevation model (DEM).

22
513

514 Sarabandi (1997) presented a study of the theoretical aspects of estimating vegetation

515 parameters from SAR interferometry. The phase of the interferogram is proportional to the

516 phase scattering centre of the target, and the coherence is inversely proportional to the

517 uncertainty with which the phase can be estimated. For distributed targets such as a forest

518 canopy the phase of the interferogram is a random variable which is a function of the system

519 parameters and target scattering mechanisms. However, despite the complications arising

520 from a three-dimensional array of scatterers, Sarabandi (1997) found that for a uniform closed

521 canopy the extinction and the physical height of the canopy top could be estimated very

522 accurately.

523 Hagberg et al. (1995) studied a dense boreal forest with InSAR. The interferometric height

524 discontinuity at the forest to non-forest boundary showed good agreement with in-situ tree

525 height measurements, although for a less dense forest the discontinuity was found to decrease,

526 suggesting the possibility of estimating bole volume from the interferometric tree height and a

527 ground DEM. Hagberg et al. (1995) also used the decrease of coherence over a dense forest

528 with increasing baseline to estimate the effective scattering layer thickness.

529 Treuhaft et al. (1996) modelled the interferometric radar response to vegetation and

530 topography. Four parameters were used to describe vegetation and topography: vegetation

531 layer depth, vegetation extinction coefficient, a parameter involving the product of the

532 average backscattering amplitude and scatterer number density, and the elevation of the

533 ground. Their analysis of airborne InSAR data from Bonanza Creek Experimental Forest in

534 Alaska showed approximately 5 m average ground truth agreement for vegetation layer

535 depths and ground-surface heights, with a dependence of error on stand height.

536 The methods described above require precise estimates of the interferometric effective height.

537 This precision crucially depends on sufficiently high coherence between the images.

23
538 Particularly at shorter wavelengths, decorrelation over forested areas can be high. At C- and

539 X-band even light to moderate wind can cause a loss of coherence (Gray and Farrismanning

540 1993).

541 Askne et al. (1997) estimated the interferometric effective height of the forest by comparison

542 of the InSAR height map with a digital elevation model. They also developed a model to

543 relate basic forest properties to interferometric SAR observations, showing that the coherence

544 and interferometric effective height change between image pairs. The model demonstrated

545 how these properties are related to temporal decorrelation and to scattering from the

546 vegetation canopy and the ground surface, and showed the important effect of gaps in the

547 vegetation canopy. They inferred that the information content of the SAR backscatter

548 intensity alone was limited, but that considerably more information about forest parameters

549 could be derived if coherence and interferometric effective height could also be included.

550 Most published work on SAR interferometry uses a single-wavelength, single-polarization

551 sensor. However, fully polarimetric SAR sensors, with both horizontal and vertical transmit

552 and receive polarization, have been employed in the past and are being used at present, like

553 the SIR-C mission or airborne sensors like JPL AIRSAR or the Danish EMISAR. If the

554 baseline between repeated overflights is small enough, techniques of polarimetric SAR

555 interferometry can be used to improve mapping capabilities. Cloude and Papathanassiou

556 (1998) have published a pioneering paper on the theoretical background of polarimetry in

557 SAR interferometry. They proposed a general formulation for vector wave interferometry,

558 which includes conventional scalar interferometry as a special case. They show how

559 interferograms between all possible linear combinations of polarization states can be formed,

560 from any one of the linear polarization states (HH, HV, VH, VV) measured at time 1, and any

561 one of these four polarization states at time 2 - e.g. HH1VV2. This approach revealed the

562 strong polarization dependence of the interferometric coherence. Cloude and Papathanassiou

24
563 (1998) describe an algorithm for coherence maximisation and formulate a new coherent

564 decomposition for polarimetric SAR interferometry that allows separation of the effective

565 phase scattering centres of different scattering mechanisms. This analysis gives the height of

566 the phase scattering centre for volume scattering in the tree crowns and for double-bounce

567 scattering at the tree-trunks. Cloude and Papathanassiou (1998) introduced a scattering model

568 for an elevated forest canopy to demonstrate the effectiveness of the algorithms and the

569 importance of wave polarization for the physical interpretation of SAR interferograms. The

570 potential of polarimetric SAR interferometry was investigated using results from fully

571 polarimetric interferometric SIR-C data collected over the Selenga delta region at Lake

572 Baikal, Russia. The scattering mechanisms in the forest, arising from different types of

573 interactions at the ground, branches and canopy top, can be separated polarimetrically and

574 located at different heights so that the phase differences at different polarizations can be

575 interpreted as canopy height differences. Over the forested area near Lake Baikal these height

576 differences were found to be around 20-30m, and the authors suggest a direct relationship to

577 forest canopy height. A prerequisite for this analysis is significant canopy penetration so that

578 a fully polarimetric SAR operating at L-band or at a longer wavelength is required.

579 An alternative approach to polarimetric interferometric SAR image analysis is the inversion

580 of a microwave scattering model. Because the number of forest structural parameters is higher

581 than the number of SAR parameters measured, such a model inversion is not straightforward,

582 and multiple solutions may exist for a given set of observations. Lin and Sarabandi (1999)

583 presented a fractal-based coherent scattering model of the polarimetric and interferometric

584 response to forest as a function of incidence angle, tree density, tree height, trunk diameter,

585 branching angle, wood moisture, soil moisture, and finer structural features. A genetic

586 algorithm has been used to estimate the input parameters of a forest stand from a set of

587 measured polarimetric and interferometric backscatter responses of the stand.

25
588

589

590 V Outlook

591

592

593 SAR interferometry, and particularly polarimetric SAR interferometry has the potential to

594 operationally deliver three-dimensional structural information on the Earth’s surface and its

595 vegetation cover. The Japanese satellite ALOS will be the first spaceborne SAR mission

596 carrying an L-band SAR capable of repeat-pass polarimetric interferometry. Other future

597 missions are in earlier planning stages, like the German-British collaborative project

598 InfoTerra / TerraSAR or the American LightSAR.

599 As a conclusion from this review of application of InSAR to forest mapping and monitoring,

600 SAR interferometry can estimate biophysical variables of forest ecosystems on continental

601 scales which could hardly be retrieved with other methods. These variables may have future

602 applications in forest ecosystem models, models of the global carbon cycle and the impacts of

603 global climate change, and will certainly prove useful in the efforts to monitor sustainable

604 forest management with respect to the international commitments to the UN Biodiversity

605 Convention.

606

607

608 VI References

609

610

611 Askne, J. I. H., Dammert, P. B. G., Ulander, L. M. H. and Smith, G. 1997: C-band repeat-pass

26
612 interferometric SAR observations of the forest. IEEE Transactions On Geoscience and

613 Remote Sensing 35, 25-35.

614 Baker, J. R., Mitchell, P. L., Cordey, R. A., Groom, G. B., Settle, J. J. and Stileman, M. R.

615 1994: Relationships between physical characteristics and polarimetric radar backscatter for

616 Corsican Pine stands in Thetford Forest, UK. International Journal of Remote Sensing 15,

617 2827-2849.

618 Baker, J. R. and Luckman, A. J. (1999, in press): Microwave observations of boreal forests in

619 the NOPEX area of Sweden and a comparison with observations of a temperate plantation in

620 the UK. Journal of Agricultural and Forest Meteorology

621 Bamler, R. and Hartl, P. 1998: Synthetic Aperture Radar interferometry. Inverse Problems 14,

622 R1-R54.

623 Chang, A. T. C., Foster, J. L., Hall, D. K., Goodison, B. E., Walker, A. E., Metcalfe, J. R. and

624 Harby, A. 1997: Snow parameters derived from microwave measurements during the

625 BOREAS winter field campaign. Journal of Geophysical Research-Atmospheres 102, 29663-

626 29671.

627 Christensen, E. L., Skou, N., Dall, J., Woelders, K. W., Jorgensen, J. H., Granholm, J. and

628 Madsen, S. N. 1998: EMISAR: An absolutely calibrated polarimetric L- and C-band SAR.

629 IEEE Transactions On Geoscience and Remote Sensing 36, 1852-1865.

630 Cloude, S. R. and Papathanassiou, K. P. 1998: Polarimetric SAR interferometry. IEEE

631 Transactions On Geoscience and Remote Sensing 36, 1551-1565.

632 Coltelli, M., Fornaro, G., Franceschetti, G. et al. 1996: SIR-C/X-SAR multifrequency

633 multipass interferometry: A new tool for geological interpretation. Journal of Geophysical

634 Research-Planets 101, 23127-23148.

635 Costantini, M. 1998: A novel phase unwrapping method based on network programming.

636 IEEE Transactions On Geoscience and Remote Sensing 36, 813-821.

27
637 Costantini, M., Farina, A. and Zirilli, F. 1999: A fast phase unwrapping algorithm for SAR

638 interferometry. IEEE Transactions On Geoscience and Remote Sensing 37, 452-460.

639 Dammert, P. B. G., Askne, J. I. H. and Kuhlmann, S. 1999: Unsupervised segmentation of

640 multitemporal interferometric SAR images. IEEE Transactions On Geoscience and Remote

641 Sensing 37, 2259-2271.

642 Delia, S. and Biasutti, R. 1999: DESCW: PC software supporting remote sensing data. ESA

643 Bulletin - European Space Agency 84-88.

644 Dobson, M. C., Ulaby, F. T., Letoan, T., Beaudoin, A., Kasischke, E. S. and Christensen, N.

645 1992: Dependence of radar backscatter on coniferous forest biomass. IEEE Transactions On

646 Geoscience and Remote Sensing 30, 412-415.

647 Dutra, L. V. and Huber, R. 1999: Feature extraction and selection for ERS-1/2 InSAR

648 classification. International Journal of Remote Sensing 20, 993-1016.

649 Faller, N. P. and Meier, E. H. 1995: First results with the airborne single-pass Do-SAR

650 interferometer. IEEE Transactions On Geoscience and Remote Sensing 33, 1230-1237.

651 Fornaro, G. and Franceschetti, G. 1995: Image registration in interferometric SAR processing.

652 Iee Proceedings-Radar Sonar and Navigation 142, 313-320.

653 Fornaro, G., Franceschetti, G. and Lanari, R. 1996: Interferometric SAR phase unwrapping

654 using Green's formulation. IEEE Transactions On Geoscience and Remote Sensing 34, 720-

655 727.

656 Fornaro, G., Franceschetti, G., Lanari, R., Rossi, D. and Tesauro, M. 1997a: Interferometric

657 SAR phase unwrapping using the finite element method. IEE Proceedings-Radar Sonar and

658 Navigation 144, 266-274.

659 Fornaro, G., Franceschetti, G., Lanari, R., Sansosti, E. and Tesauro, M. 1997b: Global and

660 local phase-unwrapping techniques: a comparison. Journal of the Optical Society of America

661 A - Optics Image Science and Vision 14, 2702-2708.

28
662 Fornaro, G. and Sansosti, E. 1999: A two-dimensional region growing least squares phase

663 unwrapping algorithm for interferometric SAR processing. IEEE Transactions On Geoscience

664 and Remote Sensing 37, 2215-2226.

665 Freeman, A. 1992: SAR calibration - an overview. IEEE Transactions On Geoscience and

666 Remote Sensing 30, 1107-1121.

667 Gelautz, M., Leberl, F. and Kellerer-Pirklbauer, W. 1996: Image enhancement and

668 evaluation: SAR layover and shadows. AEU-Archiv für Elektronik und Übertragungstechnik-

669 International Journal of Electronics and Communications 50, 100-105.

670 Gens, R. and VanGenderen, J. L. 1996: SAR interferometry - Issues, techniques, applications.

671 International Journal of Remote Sensing 17, 1803-1835.

672 Goldstein, R. M. and Werner, C. L. 1998: Radar interferogram filtering for geophysical

673 applications. Geophysical Research Letters 25, 4035-4038.

674 Goyal, R. K. and Verma, A. K. 1996: Mathematical formulation for estimation of baseline in

675 synthetic aperture radar interferometry. Sadhana-Academy Proceedings in Engineering

676 Sciences 21, 511-522.

677 Gray, A. L. and Farrismanning, P. J. 1993: Repeat-pass interferometry with airborne

678 Synthetic Aperture Radar. IEEE Transactions On Geoscience and Remote Sensing 31, 180-

679 191.

680 Griffiths, H. D. and Wilkinson, A. J. 1994: Improvements in phase unwrapping algorithms

681 for interferometric SAR. Onde Electrique 74, 46-52.

682 Hagberg, J. O., Ulander, L. M. H. and Askne, J. 1995: Repeat-pass SAR interferometry over

683 forested terrain. IEEE Transactions On Geoscience and Remote Sensing 33, 331-340.

684 Hilland, J. E., Stuhr, F. V., Freeman, A., Imel, D., Shen, Y., Jordan, R. L. and Caro, E. R.

685 1998: Future NASA spaceborne SAR missions. IEEE Aerospace and Electronic Systems

686 Magazine 13, 9-16.

29
687 Imhoff, M. L. 1995: Radar backscatter and biomass saturation - Ramifications for global

688 biomass inventory. IEEE Transactions On Geoscience and Remote Sensing 33, 511-518.

689 Just, D. and Bamler, R. 1994: Phase statistics of interferograms with applications to Synthetic

690 Aperture Radar. Applied Optics 33, 4361-4368.

691 Kasischke, E. S., Melack, J. M. and Dobson, M. C. 1997: The use of imaging radars for

692 ecological applications - A review. Remote Sensing of Environment 59, 141-156.

693 Lee, J. S., Hoppel, K. W., Mango, S. A. and Miller, A. R. 1994: Intensity and phase statistics

694 of multilook polarimetric and interferometric SAR imagery. IEEE Transactions On

695 Geoscience and Remote Sensing 32, 1017-1028.

696 Lee, J. S., Papathanassiou, K. P., Ainsworth, T. L., Grunes, M. R. and Reigber, A. 1998: A

697 new technique for noise filtering of SAR interferometric phase images. IEEE Transactions On

698 Geoscience and Remote Sensing 36, 1456-1465.

699 Lin, Q., Vesecky, J. F. and Zebker, H. A. 1992: New approaches in interferometric SAR data

700 processing. IEEE Transactions On Geoscience and Remote Sensing 30, 560-567.

701 Lin, Y. C. and Sarabandi, K. 1999: Retrieval of forest parameters using a fractal-based

702 coherent scattering model and a genetic algorithm. IEEE Transactions On Geoscience and

703 Remote Sensing 37, 1415-1424.

704 Luckman, A., Baker, J., Honzak, M. and Lucas, R. 1998: Tropical forest biomass density

705 estimation using JERS-1 SAR: Seasonal variation, confidence limits, and application to

706 image mosaics. Remote Sensing of Environment 63, 126-139.

707 Lyuboshenko, I. and Maitre, H. 1999: Phase unwrapping for interferometric synthetic

708 aperture radar by use of Helmholtz equation eigenfunctions and the first Green's identity.

709 Journal of the Optical Society of America A - Optics Image Science and Vision 16, 378-395.

710 Madsen, S. N., Martin, J. M. and Zebker, H. A. 1995: Analysis and evaluation of the

711 NASA/JPL TOPSAR across-track interferometric SAR system. IEEE Transactions On

30
712 Geoscience and Remote Sensing 33, 383-391.

713 Madsen, S. N., Zebker, H. A. and Martin, J. 1993: Topographic mapping using radar

714 interferometry - Processing techniques. IEEE Transactions On Geoscience and Remote

715 Sensing 31, 246-256.

716 Moghaddam, M. and Saatchi, S. 1995: Analysis of scattering mechanisms in SAR imagery

717 over boreal forest - Results from BOREAS-93. IEEE Transactions On Geoscience and

718 Remote Sensing 33, 1290-1296.

719 Pritt, M. D. 1996: Phase unwrapping by means of multigrid techniques for interferometric

720 SAR. IEEE Transactions On Geoscience and Remote Sensing 34, 728-738.

721 Ranson, K. J., Saatchi, S. and Sun, G. Q. 1995: Boreal forest ecosystem characterization with

722 SIR-C/X-SAR. IEEE Transactions On Geoscience and Remote Sensing 33, 867-876.

723 Ranson, K. J., Sun, G. Q., Lang, R. H., Chauhan, N. S., Cacciola, R. J. and Kilic, O. 1997:

724 Mapping of boreal forest biomass from spaceborne Synthetic Aperture Radar. Journal of

725 Geophysical Research-Atmospheres 102, 29599-29610.

726 Rignot, E. 1996: Dual-frequency interferometric SAR observations of a tropical rain- forest.

727 Geophysical Research Letters 23, 993-996.

728 Rosen, P. A., Hensley, S., Zebker, H. A., Webb, F. H. and Fielding, E. J. 1996: Surface

729 deformation and coherence measurements of Kilauea volcano, Hawaii, from SIR-C radar

730 interferometry. Journal of Geophysical Research-Planets 101, 23109-23125.

731 Rossi, M., Rogron, B. and Massonnet, D. 1996: JERS-1 SAR image quality and

732 interferometric potential. IEEE Transactions On Geoscience and Remote Sensing 34, 824-

733 827.

734 Rufino, G., Moccia, A. and Esposito, S. 1998: DEM generation by means of ERS tandem

735 data. IEEE Transactions On Geoscience and Remote Sensing 36, 1905-1912.

736 Sarabandi, K. 1997: Delta k-radar equivalent of interferometric SAR's: A theoretical study for

31
737 determination of vegetation height. IEEE Transactions On Geoscience and Remote Sensing

738 35, 1267-1276.

739 Tough, R. J. A., Blacknell, D. and Quegan, S. 1995: A statistical description of polarimetric

740 and interferometric Synthetic-Aperture Radar data. Proceedings of the Royal Society of

741 London Series A - Mathematical and Physical Sciences 449, 567-589.

742 Treuhaft, R. N., Madsen, S. N., Moghaddam, M. and van Zyl, J. J. 1996: Vegetation

743 characteristics and underlying topography from interferometric radar. Radio Science 31,

744 1449-1485.

745 Trouve, E., Nicolas, J. M. and Maitre, H. 1998: Improving phase unwrapping techniques by

746 the use of local frequency estimates. IEEE Transactions On Geoscience and Remote Sensing

747 36, 1963-1972.

748 Ulander, L. M. H. 1996: Radiometric slope correction of Synthetic Aperture Radar images.

749 IEEE Transactions On Geoscience and Remote Sensing 34, 1115-1122.

750 Wang, Z. J. and Li, S. S. 1999: Phase unwrapping through a branch-cut-based cut-bridging

751 and window- patching method. Applied Optics 38, 805-814.

752 Way, J., Paris, J., Kasischke, E. et al. 1990: The effect of changing environmental conditions

753 on microwave signatures of forest ecosystems - Preliminary results of the March 1988

754 Alaskan aircraft SAR experiment. International Journal of Remote Sensing 11, 1119-1144.

755 Way, J., Rignot, E. J. M., McDonald, K. C., Oren, R., Kwok, R., Bonan, G., Dobson, M. C.,

756 Viereck, L. A. and Roth, J. E. 1994: Evaluating the type and state of Alaska taiga forests with

757 imaging radar for use in ecosystem models. IEEE Transactions On Geoscience and Remote

758 Sensing 32, 353-370.

759 Way, J., Zimmermann, R., Rignot, E., McDonald, K. and Oren, R. 1997: Winter and spring

760 thaw as observed with imaging radar at BOREAS. Journal of Geophysical Research-

761 Atmospheres 102, 29673-29684.

32
762 Wegmuller, U. and Werner, C.L. 1995: SAR interferometric signatures of forest. IEEE

763 Transactions On Geoscience and Remote Sensing 33, 1153-1161.

764 Wegmuller, U. and Werner, C. 1997: Retrieval of vegetation parameters with SAR

765 interferometry. IEEE Transactions On Geoscience and Remote Sensing 35, 18-24.

766 Xu, W. and Cumming, I. 1999: A region-growing algorithm for InSAR phase unwrapping.

767 IEEE Transactions On Geoscience and Remote Sensing 37, 124-134.

768 Zebker, H. A., Farr, T. G., Salazar, R. P. and Dixon, T. H. 1994a: Mapping the worlds

769 topography using radar interferometry - the Topsat Mission. Proceedings of the IEEE 82,

770 1774-1786.

771 Zebker, H. A., Werner, C. L., Rosen, P. A. and Hensley, S. 1994b: Accuracy of topographic

772 maps derived from ERS-1 interferometric radar. IEEE Transactions On Geoscience and

773 Remote Sensing 32, 823-836

774 Zebker, H. A. and Lu, Y. P. 1998: Phase unwrapping algorithms for radar interferometry:

775 Residue-cut, least-squares, and synthesis algorithms. Journal of the Optical Society of

776 America A - Optics Image Science and Vision 15, 586-598.

777 Zebker, H. A., Rosen, P. A. and Hensley, S. 1997: Atmospheric effects in interferometric

778 Synthetic Aperture Radar surface deformation and topographic maps. Journal of Geophysical

779 Research-Solid Earth 102, 7547-7563.

780 Zebker, H. A. and Villasenor, J. 1992: Decorrelation in interferometric radar echoes. IEEE

781 Transactions On Geoscience and Remote Sensing 30, 950-959.

782

33
Tables, figures and images

Table 1: Past, present and planned orbital SAR missions.

Satellite launch freq band polari- resolu- swath look inciden- repeat
date zations tion width angle ce angle cycle
(GHz) (m) (km) (deg) (deg) (days)
ERS-1 6/1991 5.25 C VV 30 100 20 23-35 3, 35,
168

JERS-1 summer 1.28 L HH 18 75 35 38 44

1992

SIR-C 4/1994 1.25, L, C HH, VV, 10-40 15-90 15-55 (1)

& 10/1/ 5.3 HV (quad.)
1994

ERS-2 spring 5.25 C VV 30 100 20 23-35 (1), 35

1995

Radarsat 10/ 5.3 C HH 10-100 45-500 20-59 (2-3)

1995

Envisat 2000 5.25 C HH, VV, 35

HV (not
quad.)

ALOS 2002 1.28 L HH, VV, 45

HV
(experi-
mental
quad-pol
mode)

34
Figures and images

1.5
phase
1
amplitude

0.5

0
0 5 10 15 20 25

-0.5

-1

wavelength
-1.5

Figure 1: Amplitude, phase and wavelength of a radar signal. The intensity of the signal is the

squared amplitude.

35
 SAR2
B
SAR1

Figure 2: Interference of iso-phase lines of two SAR sensors separated by a spatial baseline B.

The distance between two iso-phase lines is half the wavelength. The absolute phase

difference fluctuates between 0 and , as indicated by the two orthogonal lines.

36