Topic C 5 Remotely sensed assessment of tropical

Topic C 5. Remotely sensed assessment of tropical wetlands Erik Lilleskov, Belinda Margono, and Laura Bourgeau-Chavez

Topic C 5. Slide 2 of 26 Learning outcomes In this presentation you will be introduced to approaches for using remote sensing to map wetland extent and change

Topic C 5. Slide 3 of 26 Outline § Rationale § Background § Choice of sensors and resolutions § Airborne/spaceborne or ground-based sensors § Generating maps from sensor data • Wetlands • Special case: Peatlands § Ground truthing § Validation § Change detection

Topic C 5. Slide 4 of 26 Rationale § Deforestation and forest degradation have been reported to be the 2 nd leading cause of anthropogenic greenhouse gas emissions § Wetlands, especially peatlands, represent one of the largest terrestrial, biological carbon pools and are important wildlife habitats § Tropical peatlands and mangroves are being lost at high rates § Quantifying wetland type, extent, distribution and condition is vital for mitigation efforts, MRV for REDD+, IPCC and related efforts § Remote sensing is a major tool in wetland mapping

Topic C 5. Slide 5 of 26 Background: Wetland mapping and remote sensing § Remote sensing data is the main data source for monitoring and mapping wide areas, including • wetland extent and distribution • wetland type – Including extent of mangrove, freshwater peat swamps and non-forested peatlands • § Top: National wetlands map of Indonesia. Bottom: Peatland map for Central Kalimantan Province, Indonesia. (Margono et al. 2014) Land-use/land-cover change Remote sensing provides activity data, a critical component of estimating human impacts on wetlands • Field studies provide emissions factor (impact of human activity on greenhouse gas emissions) • Both activity data and emission factors are vital for estimating change in wetland carbon content • Baseline wetland extent maps (right) can be used to assess impacts of land use

Topic C 5. Slide 6 of 26 Approaches to wetland mapping § Selected remote sensing tools should detect some or all of the following: • water presence; • water temporal dynamics; • landforms likely to retain water; • vegetation type and floristic differences. § Fusion of multiple data sources often provides improved maps § Digital mapping suggests that water presence and dynamics, landform and vegetation type can be observed using multisource data sets

Topic C 5. Slide 7 of 26 Overall schematic of map development

Topic C 5. Slide 8 of 26 Possible data sources § http: //science. nasa. goe/missions/landsat 7 § http: //gliht. gsfc. nasa. gov § Spaceborne are most important for mapping large regions • Multispectral, e. g. Landsat TM, SPOT, MODIS • Hyperspectral – Hyperspectral Imager (HSI) on the Lewis satellite • Radar e. g. ALOS PALSAR, SRTM • Li. DAR e. g. ICESat/GLAS Airborne can provide higher resolution data for smaller regions • Hyperspectral, e. g. , AVIRIS, AHS, HYDICE, AISA • Li. DAR • Multispectral • Multiplatform, e. g. G-Li. HT (Li. DAR, hyperspectral, thermal) Ground-based sensors are used primarily at the site level or to validate remote methods • http: //en. wikipedia. org/wiki/Lidar#mediav iewer/File: Lidar_P 1270901. jpg Tripod-mounted Li. DAR

Topic C 5. Slide 9 of 26 Landsat § Landsat is a passive data source, i. e. it relies on incoming solar radiation. It does not see through clouds. § § Series of Landsat TM 5, Landsat 7 ETM+ and Landsat 8 § Landsat imagery captures floristic differences that can be associated with wetland status, as well as water extent and leaf moisture content § Available with 30 m spatial resolution, sufficient for mapping at scale 1 : 100, 000 or even 1 : 50, 000 § § Timely data acquisitions are limited by cloud cover Band 3, 4, 5 and 7 are commonly used and are: • suitable for soil-vegetation discrimination (B, G, R) • good for mapping biomass content (NIR) • very good at detecting and analyzing vegetation (NIR) • provides good contrast between different types of vegetation (SWIR) • useful for measuring the moisture content of soil and vegetation (SWIR) The image to the right shows a false color composite of bands 3, 4 and 5 from Landsat 7 of a region of the Peruvian Amazon basin near the Marañón River (lower right) that has previously been shown to contain a peat dome (black star) (Bourgeau-Chavez et al. 2009).

Topic C 5. Slide 10 of 26 PALSAR § Phased Array type L-band Synthetic Aperture (PALSAR) is an active source because it sends out a microwave energy pulse and collects the returns. § Uses L-band to achieve cloud-free and day-and-night land observation § 10– 20 m data are available, but for most national-level applications, 50 m spatial resolution is suitable § Data available in polarization mode, which enhances land-cover information § The different interactions of microwave data (PALSAR) with surface water compared to vegetation enable improved discrimination of wetlands § Comparing images from multiple dates (multi-temporal) improves understanding of hydrology and helps to distinguish wetlands and wetland types § The image to the right shows a false color composite of three different dates from ALOS PALSAR of a region of the Peruvian Amazon Basin near the Marañón River (lower right) that has previously been shown to contain a peat dome (black star). Color variation is mostly driven by differences in hydrologic condition. The areas in brighter colors are sloping portions of the peat dome (Bourgeau-Chavez et al. 2009).

Topic C 5. Slide 11 of 26 PALSAR Principal Component Analysis § Principal Component Analysis (PCA) is a multivariate statistical technique that is used to identify the dominant spatial and temporal backscatter signatures of a landscape § PCA generates a set of new images, reducing most of the information to the first few new PC images § Several advantages including the ability to filter out temporal autocorrelation and reduce speckle § Helpful in understanding moisture patterns § The image to the right is a single PCA derived image that extracts the major axes of variation in the previous PALSAR image.

Topic C 5. Slide 12 of 26 DEM from SRTM or Li. DAR § Global DEM (topography map) derived from single-pass interferometric synthetic aperture radar (In. SAR) of SRTM § Available globally at 90 m spatial resolution, and 30 m resolution for some places § Spaceborne Li. DAR coverage e. g. ICESat/GLAS is limited to long transects § Airborne Li. DAR coverage varies by country § Using DEMs, a set of topographical indices capture landforms more likely to retain water. § Example to right: Topographic indices derived from SRTM for peatlands in Central Kalimantan, Indonesia. The top figure depicts a flatness index which has clear hydrologic predictive value; whereas the bottom index depicts relative elevation of catchments of 121. 5 km 2 and is indicative of slope (Margono et al. 2014). Both have been found to be useful predictors in wetland mapping.

Topic C 5. Slide 13 of 26 Data integration/fusion § Data integration (data fusion): Combining data from different sources § Geospatial data integration e. g. • vegetation type, generated from Landsat • landform derived from DEM • water presence, using topographical indices generated from DEM – First derivatives of elevation (e. g. slope) – Second-order derivatives of elevation (e. g. various curvatures) • vegetation and soil wetness, generated from ALOS-PALSAR

Topic C 5. Slide 14 of 26 Example of data integration using Landsat, ALOSPALSAR and SRTM (a) Landsat image with 5– 4– 3 spectral combination; (b) terrain flatness; (c) relative elevation of 121. 5 km 2 (medium) catchments; (d) Landsat band 5 represent soil/vegetation moisture; (e) false-color r-g-b of (b), (c), and (d); and (f) the initial resulting wetland map as a probability layer where blue is high wetland cover probability and white low wetland cover probability. Single date PALSAR (data not shown) contributed a small percentage to the final wetland model.

Topic C 5. Slide 15 of 26 Peatlands as a special case Peatlands are wetlands that accumulate peat (partially decomposed organic matter) and so contain large reserves of carbon vulnerable to anthropogenic disturbance, e. g. decomposition or fire triggered by drainage or climate change

Topic C 5. Slide 16 of 26 Mapping tropical peatlands § Unique vegetation • Known peat-forming plant associations – Peat swamp forests – Mountain fens • § Unique hydrology • Seasonal hydrologic dynamics of peatlands differ from other wetland classes • Multi-temporal PALSAR can be used to characterize hydrologic dynamics http: //onlinelibrary. wiley. com/10. 1002/agc 834/pdf § http: //www. fao. org/docrep/003/y 1899 e 04. htm Landsat can detect unique vegetation signals Unique geomorphology • Many peatlands have convex geomorphology (dome formation) • SRTM or Li. DAR-derived DEMs can be used to characterize and identify domes

Topic C 5. Slide 17 of 26 Peatland hydrology & SAR Hoekman (2007) § Peatland hydrology is driven by exogenous and endogenous factors. Doming, which is common in Indonesian peat swamp forests (and is being quantified elsewhere) regulates water flux patterns. § This SAR multi-temporal image reveals divergent hydrology across the width of a peat dome, with the flat top of this peat dome (light blue areas, A) showing a different time course of flooding than the edges and stream channels (redder areas, B)

Topic C 5. Slide 18 of 26 Peatland doming Ballhorn et al. 2011 § Peat accumulates over thousands of years where production outpaces decomposition § In some places, peat rises above the local water table, creating domes § Doming can be observed as regular, rounded topographic features sometimes many km across. § These features can be recognized when analyzing topographic relief, especially in conjunction with wetland mapping § Quantifying dome morphology can improve estimation of peatland carbon storage § The example at the right (Ballhorn et al. 2011) illustrates use of satellite-based Li. DAR (ICESat/GLAS) to determine dome morphology and forest structure on a peatland in Indonesia. In B the blue points delineate the dome height in meters over a horizontal distance of about 100 km. The green points represent canopy height. The method was validated using airborne Li. DAR and ground sampling.

Topic C 5. Slide 19 of 26 Ground truthing Plot selection Plot-level field data Image interpretation § Field surveys and image interpretation • Plot selection: Sampling should be statistically valid, stratified over putative wetland classes from initial unsupervised classification • Logistical constraints on plot selection should be included in sampling design • Plot characteristics: Plots should be sized and oriented to stay within a single map class. • Image interpretation can derive data from aerial imagery, e. g. urban areas, lakes, other distinct features

Topic C 5. Slide 20 of 26 Supervised classification § Supervised classification (e. g. Random Forests) § Supervised classification • Based on field or other independent data, a supervised classification can be run using a portion of the data • This divides the data into specific classes of similar properties that can be more or less resolved depending on goals of classification. Validation • Using plots not included in supervised classification, the quality of the classification can be evaluated. • Results can be presented as an accuracy assessment matrix – example below. Accuracy assessment matrix

Topic C 5. Slide 21 of 26 Change detection Pre-analysis steps • Image registration • Calibration or normalization • Selection for same spatial/spectral resolution • Mosaicking § Remote sensing can be used to quantify change in land use/land cover of wetlands § This can be accomplished by performing a change detection analysis using remote sensing data (e. g. Landsat) collected over time, known as a multitemporal data set § Involves change from one class to another (e. g. conversion to agriculture) or change within a class (e. g. thinning of forest) § There are many possible change detection approaches • Algebra-based • Transformation-based • Classification-based Choose change • Advanced models detection • GIS-based, Other method • Steps specific to method, involving direct comparison Perform change analysis of spectral data, or some sort of image processing (transformation, classification, etc. ) followed by comparison. • Requires field-based reference data, e. g. , forest inventory Perform accuracty assessment • Accuracy assessment matrix/ error matrix, as for other RS data.

Topic C 5. Slide 22 of 26 Example of change detection work flow using probability filters Klemas (2011).

Topic C 5. Slide 23 of 26 The future of change detection using remote sensing § The Landsat archive is available with free access to terrain-corrected data for many regions. § Automated image preprocessing and landcover characterization methods will soon be standard practice. § The images on the right show change detection results for the expansion of bare ground on a national scale from the US (top) and a close-up of a localized region, from the Web-Enabled Landsat Data (WELD) project. Blue areas are newly bare ground (Hansen and Loveland 2012). § These large-scale automated methods should greatly accelerate change analysis in wetlands. Hansen and Loveland (2012).

Topic C 5. Slide 24 of 26 References Adam E, Mutanga O and Rugege D. 2010. Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: A review. Wetlands Ecology and Management 18(3): 281– 96. Ballhorn U, Jubanski J and Siegert F. 2011. ICESat/GLAS data as a measurement tool for peatland topography and peat swamp forest biomass in Kalimantan, Indonesia. Remote Sensing 3(9): 1957– 82. Bourgeau-Chavez LL, Riordan K, Powell RB, Miller N and Nowels M. 2009. Improving wetland characterization with multi-sensor, multi-temporal SAR and optical/infrared data fusion. In Jedlovec G (ed). Advances in Geoscience and Remote Sensing. Vukovar, Croatia: In. Tech. 679– 708. Bwangoy JRB, Hansen MC, Roy DP, Grandi GD and Justice CO. 2010. Wetland mapping in the Congo Basin using optical and radar remotely sensed data and derived topographical indices. Remote Sensing of Environment 114(1): 73– 86. Hansen MC and Loveland TR. 2012. A review of large area monitoring of land cover change using Landsat data. Remote Sensing of Environment 122: 66– 74.

Topic C 5. Slide 25 of 26 References Hoekman DH. 2007. Satellite radar observation of tropical peat swamp forest as a tool for hydrological modelling and environmental protection. Aquatic Conservation: Marine and Freshwater Ecosystems 17(3): 265– 75. Klemas V. 2011. Remote sensing of wetlands: Case studies comparing practical techniques. Journal of Coastal Research 27(3): 418– 27. Margono BA, Bwangoy JRB, Potapov PV and Hansen MC. 2014. Mapping wetlands in Indonesia using Landsat and PALSAR data-sets and derived topographical indices. Geo-spatial Information Science 17(1): 60– 71. Ozesmi SL and Bauer ME. 2002. Satellite remote sensing of wetlands. Wetlands Ecology and Management 10(5): 381– 402.

Thank you The Sustainable Wetlands Adaptation and Mitigation Program (SWAMP) is a collaborative effort by CIFOR, the USDA Forest Service, and the Oregon State University with support from USAID. How to cite this file Liilleskov E, Margono B and Bourgeau-Chavez L. 2015. Remotely sensed assessment of tropical wetlands [Power. Point presentation]. In: SWAMP toolbox: Theme C section C 5 Retrieved from <www. cifor. org/swamp-toolbox> Photo credit Adam Gynch, Belinda Margono/Ministry of Environment and Forestry, Daniel Murdiyarso/CIFOR, Erik Lilleskov/USFS, Laura Bourgeau-Chavez, Michelle Cisz, Yayan Indriatmoko/CIFOR.