Combined Analysis of Optical and SAR Remote Sensing Data for Forest Mapping and Monitoring

E. Lehmann, Z.-S. Zhou, P. Caccetta, A. Held

CSIRO, Division of Mathematics, Informatics and Statistics, Australia

A. Mitchell, A. Milne, K. Lowell

Cooperative Research Centre for Spatial Information (CRC-SI), Australia

S. McNeill
Landcare Research, New Zealand

Presentation Outline
Background
forest/non-forest (F/NF) mapping and monitoring motivation

Data and Study Area

study area optical and synthetic aperture radar (SAR) datasets data pre-processing

Combined SAROptical Forest Classification

canonical variate analysis (CVA) maximum-likelihood classification (MLC) band information (variable selection) classification results

Conclusion
summary of main outcomes strategies for non-coincident data processing
E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

Background
Assess and take advantage of the complementarity and interoperability of synthetic aperture radar (SAR) and optical sensors for forest mapping and monitoring

Motivation
technological advances in synthetic aperture radar (not cloudaffected) complement the existing optical datasets GEO-FCT: Forest Carbon Tracking task of the Group on Earth Observations (in support of global forest carbon estimation) Australias response to GEO-FCT: International Forest Carbon Initiative (IFCI) to increase forest monitoring capacity further development of the National Carbon Accounting System (NCAS) developed by CSIRO & partners: continental Landsatbased forest monitoring system
E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

Data and Study Area

Pilot study area
north-eastern Tasmania calibration site defined as part of Australias GEO-FCT demonstrator project under IFCI 66km x 50km area main land covers:
dry & wet eucalypt forest non-eucalypt forest rainforest plantations / deforestation agriculture & urban areas

significant topographic variations (elevation: 80m to 1500m)

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

Datasets
Landsat TM
6 spectral bands (thermal band omitted), 25m pixel size from the NCAS archive of MSS/TM/ETM+ imagery acquired Jan. 2009 (~18% fill-in)
Landsat TM (bands 5,4,2) PALSAR (HH,HV,HH-HV) RADARSAT-2 (VV,VH,VV-VH)

ALOS-PALSAR
fine-beam dual polarisation (HH and HV), L-band (23.6cm) ascending orbit (34.3 offnadir) pre-processed to 25m pixel size acquired Sept./Oct. 2009

RADARSAT-2
wide-beam dual polarisation (VV and VH), C-band (5.6cm) ascending orbit (42.2 offnadir) pre-processed to 25m pixel size acquired in Sept. 2009

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

Data Pre-Processing
SAR terrain illumination correction
Correct for illumination differences on forward/backward facing slopes

[Zhou et al., Terrain slope correction and precise registration of SAR data for forest mapping and monitoring, ISRSE 2011, Sydney]
E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

Data Pre-Processing
Assessment of the dataset co-registration: using crosscorrelation of image features (sub-pixel results) PALSARLandsat: 0.56 pixel accuracy, 96% or residuals below 1.5 pixels (127 features) RADARSATLandsat: 1.05 pixel accuracy, 74% of residuals below 1.5 pixels (90 features)
SAR HV (greyscale-inverted)

Landsat band 5

(HH,HV,Landsat Band 5)
E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

SAROptical F/NF Classification

Step 1: definition of spectral classes using Canonical Variate Analysis (CVA)
268 training sites selected for the classification, representing a broad range of landcover types over the study area Analyses carried out for: 1. Landsat data (6 bands) 2. PALSAR data (2 bands) 3. RADARSAT data (2 bands) 4. combined SARoptical data (10 bands, concatenation)
agriculture (crops)
15
265 253 245 267 266 208 184 185 255

10

forest (dense)
90

160 210

209 244 259254

236 212 213 172 207 242 180 246 241

water
-5

243 148 15 211 150 98 157 40 174 239 96 214 43 41 125 87 225 247 97 109 105 93 132 258 124 123 65 39 38 107 104 34 158 63127 61 147 36 7172 67 256 35 33 146 122 92 64 62 31 55 151 141 37 32 8 6 19 131 110 10 26 91 106 52 101 9551 94 47 25 17 18 42 30 3 138 9 13 12 89 21 129 44 68 102 155 143 69 22 28 85133 45 4 6 86 7 103 59 29 16 130 79 257 234 162 145 237 60 117 56 144824 137 248 50164 83 11 159 156 58 154 203 57 5 112 120 111 53 76 169 2 80 128 7349 66 99 204 181 20 220 149 88 70 173 77 126 108 152 153 54 170 199 81 144 84 188 228 75 201 240 115 202 74 182 249 114 119 205 193 195 135 1 23 262 27 4200 183 78 113 191 196 118 198 197 221 230 268 136 116 251 222 223 166 190 121 217 235 231 232 189 100 206 134 192 178 260 165139 177 179 82 140 252 216 194 227 176 250 175 264 186 187 224 261 215 219 218 238 263 171 233 168 167 163 142 161 229 226

CV2

bare ground
0 5 10 15 CV1

20

25

30

35

Plot of training sites in CV1-CV2 space for Landsat data (4 out of 7 sub-classes shown). Colour legend: forest sites, nonforest sites, cleared/immature plantations.

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

SAROptical F/NF Classification

Definition of spectral classes using CVA
The number of selected sub-classes reflects the ability of each dataset to discriminate between different land covers

(*) mix of forest and non-forest sites.

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

SAROptical F/NF Classification

Step 2: Maximum-likelihood classification (using CVA-based classes)
F/NF accuracy: 89.7% F/NF accuracy: 93.6%

F/NF accuracy: 84.8%

F/NF accuracy: 90.7%

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

SAROptical F/NF Classification

Band information
variable selection: percentage of total F/NF information provided by the optical and SAR bands (and their combinations) contrasts between various subclasses (clusters) of forest and non-forest sites
141 129 137

35

161 227

219 217 204

159

209

216 196 229

210

193 201

205 228

215

152 199 131 146 144 140 128 151 150 200 145 139 153 132 157 163 143 155 148 147 223 206 164 127 214 121 82 1 99 84 119 130 100 81 218156 41 118 27 63 62 177 176 194 116 23 40 87 98 165 207 154 115 83 30 134 135 61 97 15149 96 123 31 54 133 184 49 224 32 113 7876 733 28 2 198136 36 17 67 122 80 24 138 75 21 48 166 174 94 16 55 74 120 4 747 12 37 26 90 91 124 142 8 182 179 88 20 114 50 69 53 107 160 29 643 95 158 14 18 72 92 169 86 93 65 180 39 42 34 22 79 104 183 77 167 64 38 33 101 35 125 5106 213 66 19 71 168 185 10 52 25 111 45 175 225 46 112 13 44 68 70 89 105 181 230 57 51 222 172 211 117 103 109 188 9 58 173 178 59102 85 110 60 56 11 190 108 191 189 126 195 171 192 220

208

CV2

25

30

221

212

10 CV1

15

20

25

Plot of selected training sites in CV1-CV2 space for SARoptical data. Colour legend: forest sites, nonforest sites, cleared/growing plantations.

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

Multi-Temporal SAROptical Processing

assume that the datasets are not coincident temporally consider independent forest probability maps from each dataset refinement of the single-date forest classifications using a Bayesian Conditional Probability Network (CPN): spatial-temporal model
Landsat time series (NCAS)
Landsat prob. image Landsat Landsat Landsat prob. image prob. image

SARLandsat time series

Landsat SAR prob. prob. image image 2008 Landsat Landsat prob. image

1972

1972

2010

2010

CPN

CPN

forest map forest map

1972

forest map forest map

1972

2010
E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

2010

Conclusion
Summary
SAR and optical sensors provide complementary information for forest mapping and monitoring jointly considering PALSAR, RADARSAT-2 and Landsat data improves the forest/non-forest classification significantly:

overall, L-band SAR data allows more separation than C-band with SAR, the cross-polarisation (HV or VH) provides most of the
discrimination information significant variation exists in the respective contribution of the SAR and optical bands towards the separation of specific sub-classes of forest and non-forest sites

strategies for dealing with non-coincident datasets:

use of a multi-temporal approach (e.g. conditional probability network) check for atypical spectral signatures in the maximum-likelihood
classification

E. Lehmann et al.: Combined analysis of SAR and optical data for forest mapping and monitoring

CSIRO Mathematics, Informatics & Statistics

Eric A. Lehmann, Research Scientist Phone: +61 (0)8 9333 6123 Email: eric.lehmann@csiro.au Web: http://www.csiro.au/org/CMIS.html

Thank you...
Contact Us Phone: 1300 363 400 or +61 3 9545 2176 Email: enquiries@csiro.au Web: www.csiro.au