Measuring Forest Degradation for REDD+: A Synthesis Study Across Five SilvaCarbon Countries
PIs: Prof. M. Herold, Dr. V. Avitabile, and K. Calders (Wageningen University), Dr. L. Verchot and Dr. C. Martius (CIFOR Indonesia)
Improving forest emission estimates requires better biomass measurements before and after the change events at local levels, and the
effective use and integration with remote sensing data to monitor impacts over larger areas. Novel technologies such as terrestrial
laser scanning that provide detailed 3D measurements of tree, canopy structure and allometry rapidly and non-destructively, and the
use of high-resolution remote sensing time series (i.e. from RapidEye) offer avenues to increase REDD+ measurement accuracy and
precision, and support improved monitoring capacities in developing countries. The research team aims to use both in combination to
systematically explore this potential by improving the underlying science, conduct a research synthesis across multiple tropical tests
site, and make a direct contribution to monitoring and training in REDD+ countries.
The research team aims to address four questions:
Better allometry: How can allometries for estimating tropical forest biomass be improved with terrestrial LiDAR
(higher accuracies, more samples) and without employing destructive harvesting?
Logging impacts: What are the impacts of selective logging (incl. collateral damage) on forest biomass and forest structure
and canopy in the short and long terms?
Link ground data to remote sensing: What is the capability and sensitivity of high-resolution satellite time series data to
detect changes and related emissions due to selective logging at sub-national scales across several tropical test sites?
Uptake by national monitoring: How can the novel sub-national monitoring activities of forest degradation be integrated into
a national REDD+ monitoring system and related capacity building through SilvaCarbon?
Biomass in Degraded Forests in Peru and Brazil: Evaluation Using Airborne Lidar Remote Sensing
PIs: Michael Keller (USDA Forest Service) and Ted Feldpausch (University of Exeter, INPA, and UNEMAT)
Degraded forests are poorly studied. Despite the rapidly accumulating number of lidar studies, degraded forests are rarely used
for calibration. With the exception of some long-term studies of logging, permanent tropical forest research plots have generally
avoided degraded forests. Lidar studies of degraded forest structure, particularly for forests that have suffered understory fires,
are rare although there are excellent counter-examples of specific studies in logged forests.
This study aims to resolve whether field calibration in degraded forest is necessary for accurate lidar biomass estimation in
degraded forests. The research team will test whether lidar biomass calibrations developed from old-growth and secondary forests
or “universal” approaches are sufficient for biomass estimation in degraded forests.
The study will respond to the following questions:
How accurate are lidar biomass calibrations using only old-growth and secondary forests or “universal” equations for
How much is calibration uncertainty reduced when degraded forest plots are included as part of the calibration data set
for lidar biomass estimation over degraded forests?
Do different degradation pathways (e.g. logging and fire) result in similar structures? And how do those differing
structures affect biomass calibration?
Can calibration curves derived in one region of the Amazon be used for another region?
Do varying degradation pathways have the same outcome for biomass?
A Prototype MRV System for a Sub-region in Colombia Compliant with IPCC Approach for Securing Activity Data
PIs: Dr. Pontus Olofsson (Boston University)
In this proposal, the research team proposes an alternative method for monitoring land change that makes use of all available
observations ever acquired by the Landsat satellite for a pixel. Studying a time series of observations rather than comparing
individual images or maps makes it possible to continuously monitor the land cover at pixel-level in time. While never implemented
in Colombia, the proposed methodology has proven capable of mapping stable and changing land cover with high levels of accuracy
and certainty. This research will evaluate the full utility of US satellite data for the development of MRV systems in
deforestation hotspots. It will provide a methodology compliant with IPCC Approach 3 for securing activity data for Colombia, and
when combined with emission factors provide estimates of carbon emissions and removals as a result of land transitions.
Specifically, this study has the following objectives:
Estimate the rates of change between the IPCC land categories from 2000 until 2014 for the Department of Caqueta in the
Colombian Amazon. This will be accomplished by production of annual maps of activity data with known uncertainty according
to IPCC Approach 3.
Estimate carbon emissions and removals in a gain/loss approach using three different sources of emission factors: field
measurements and data from two pan-tropical biomass maps.
Assess uncertainty in estimated carbon emissions and removals in a Monte Carlo based approach to provide insight to where
resources should be allocated to refine the MRV system.
Addressing Carbon Emissions and Removals from Selective Logging In Support of MRV System Capabilities in Gabon
PIs: Dr. Sassan Saatchi (UCLA), Dr. John Poulsen and Dr. Vincent Medjibe (Duke University)
In Central African countries, where deforestation has historically been low, but where logging occurs in over 70% of the forests
in some countries, forest degradation may be the most important source of carbon emissions. The uncertainty in quantifying the area
affected and the carbon loss through degradation, particularly from selective logging, is large because conventional methods of
remote sensing and surveying are not sensitive enough to precisely measure degradation. However, advances in high resolution remote
sensing techniques using Light Detection and Ranging (LiDAR) provide an opportunity to improve NFMS by accurately monitoring areas
affected by degradation, quantifying the emission factors from different types of logging and estimating carbon sequestration after
The research team will investigate the following research questions:
Can airborne LiDAR data accurately detect selective logging intensity and biomass loss?
To what extent can LiDAR observations distinguish biomass loss from tree felling and gap creation from the residual damage
created by skidtrails, landing yards, and logging roads?
What are the emission and gain factors from selective logging activities?
Integration of Remote Sensing Data with Ground Plot Information for MRV
PIs: Charles T. Scott (USDA Forest Service), Doug Muchoney (USGS), Andrew Lister (USDA Forest Service), and John Poulsen (Duke University)
There are two general approaches to carbon estimation that show promise for MRV-model-based and model-assisted estimation. Model-based
estimation in the context of forest attribute mapping relies on a set of modeled, pixel-based estimates, generally in the form of a
map derived from remotely sensed data. Precision estimates can come from analyzing the set of pixel values, their uncertainties, and
their spatial covariance. Model-assisted estimation is based on a probabilistic design in which ground plots along with auxiliary data
from maps derived from remote sensing are used to generate estimates of forest parameters and their variance.
In both approaches, there are two research questions that require additional study:
How best to use GIS and ground inventory data collection procedures to spatially integrate the two data types so as to:
Generate better maps of the forest attribute for both estimation types (i.e., maximize training data information
content for modeling)
Develop the best relationship between the plots and auxiliary information used in the model-assisted or -based
estimation (i.e. maximize the relationship between the dependent and auxiliary variables)
What combination of ground inventory plot type and remote sensing data type usage will lead to the most cost efficient
avenue for achieving desired precision of inventory estimates?
This study aims to identify existing ground data in tropical forests and to alter the design of some planned pilot studies to gather
variance and cost/time data in 3 or more countries with different ecological conditions. Sources will include the Gabon Phase I
inventory, and potentially existing or upcoming data from Peru, Ecuador, Colombia and Panama.
Integrating Earth Observation and Forest Inventory Data in Quantifying Biomass in Degraded Forests of the Republic of Congo
PIs: Matthew Hansen (University of Maryland), Peter Potapov (University of Maryland), Alexandra Tyukavina (University of Maryland),
and Ifo Averti Suspense (University of Marien Ngouabi, Republic of Congo)
The Republic of Congo is one of a subset of countries where the suspected dominant factor in greenhouse gas emissions from land use
change is forest degradation rather than deforestation. For a region such as central Africa, where few trees are harvested per
hectare, direct methods of mapping partial canopy cover are not feasible. Indirect methods have been implemented to delineate
degraded natural forests. Such approaches use indications of human activity to assign degradation to adjacent natural forests.
Quantifying biomass dynamics within degraded forests is a challenge.
This study combines remotely sensed-derived degradation time-series maps with field data collection to assess biomass change within
the logged forests of the Republic of Congo. By combining time-series of indirectly mapped degradation, the research team will in
effect swap space for time, targeting forests of varying intervals since disturbance. Additionally, directly observable forest
cover loss due to infrastructure development in support of logging will be employed to estimate aboveground biomass loss.
This research will integrate large area forest monitoring data from earth observations and in situ inventory data. Forest
degradation maps will help target the allocation of biomass plots in assessing carbon stock dynamics within logged areas. The
research will advance national-scale RoC monitoring by developing a new method for integrating remotely sensed-derived forest
cover loss and degradation maps with inventory data collection. The proposed activity will quantify carbon loss and gain through
the life cycle of RoC logging concessions by sampling various aged concessions from 1990 to present. In doing so, a targeted
method for quantifying carbon stock changes due to logging activities will be realized.
Mapping Deforestation and Degradation in Mexico, Colombia and Peru Using Time Series of SENTINEL-1 Radar Data
PIs: Dr. Kellndorfer and Dr. Cartus (Woods Hole Research Center)
While change detection from optical time series has progressed well in recent years, the use of radar data for forest cover change
detection, which due to its ability to penetrate clouds could be a valuable additional source of information on forest cover change
in areas where cloud cover tends to be persistent (i.e., the tropics), is largely underutilized. On April 3rd 2014, the European
Space Agency (ESA) successfully launched the first in a new series of earth observation satellites, SENTINEL-1, which will acquire,
for the first time, dense time series of C-band (~5 cm wavelength) radar data at medium (~25 m) spatial resolution consistently every
three days and at a global scale. According to the European Delegated Act on Copernicus data, ESA will provide free, full and open
access to Sentinel-1 data.
SENTINEL-1 hence opens up new possibilities for mining time series of spaceborne optical and radar data for improved operational
forest monitoring, in particular in tropical countries. This study aims to support the development of national MRV systems by
investigating, in collaboration with governmental agencies in Mexico, Colombia, and Peru, the potential of SENTINEL-1 data for
mapping forest cover change.
A synthesis of tropical forest degradation scenarios and carbon emissions trajectories for REDD+
PIs: Dr. Jennifer K. Balch (Penn State University)
Human-caused disturbance to tropical forests, such as through intentional use and resource extraction or through unintentional
wildfires, cause substantial losses of carbon stocks. But does tropical forest degradation lead to permanent carbon losses? This
is a critical question to address in the context of policy discussions to implement REDD+ (Reduced Emissions from Deforestation
and Forest Degradation Plus enhancement of forest carbon stocks through conservation and sustainable forest management). We propose
to review the current scientific knowledge about the temporal and spatial dynamics of degradation--‐induced carbon emissions to
build a coherent picture of the pattern of emissions from different types of degradation across tropical forest regions. Using best
available information, we will: i) develop emissions factors (per area) for different types and scenarios of degradation; ii)
describe the temporal pattern of degradation emissions and recovery trajectory post--‐disturbance; and iii) assess the evidence that
demonstrates how tropical forest degradation leads to a lower carbon state, either through arrested succession, a switch to an
alternate vegetation state, or facilitation of future deforestation. The overarching goal of this research is to synthesize existing
knowledge on the range of initial gross and longer-term net carbon emissions from different types of degradation activities across
Theoretical underpinnings: what’s known about the trajectory of tropical forest recovery after human-induced degradation?
What is the spatial and temporal pattern of emissions from different types and scenarios of forest degradation across
When and where is degradation just a harbinger of deforestation?
Through what pathways does forest degradation lead to substantial and permanent carbon losses?
Alternatively, under what degradation scenarios Do tropical forests tend to recover carbon stocks quickly?
Case Studies across South America: Country-level Experiences and comparisons of forest degradation
Inventory and remote sensing-based assessments of forest degradation
PIs: Ronald E. McRoberts (USFS), Michael Keller (USFS), Douglas C. Morton (NASA) and Erik Næsset
(Norwegian University of Life Sciences)
Deforestation and forest degradation account for nearly 20% of global greenhouse gas emissions, more than any sector other than the
energy sector (UN REDD, 2009). REDD (Reducing Emissions from Deforestation and Forest Degradation in Developing Countries) is a
mechanism designed under the United National Framework Convention on Climate Change to financially support developing countries that
are willing and able to reduce emissions from deforestation and invest in low carbon paths to sustainable development. The term
deforestation refers to the permanent removal of forests and withdrawal of land from forest use, whereas the term forest degradation
refers to detrimental changes that limit a forest’s production capacity. A relevant question pertains to the persistence component
of degradation, i.e., is a forest degraded if it recovers from detrimental change that only temporarily limits its productivity? The
overall objective of the proposal is to elaborate the definition of degradation by clarifying the persistence component. The
research questions are primarily methodological in nature with anticipated outcomes relating to the utility of remotely sensed data
for assessing forest degradation. To address these questions, three kinds of will be used, ground inventory, multi-spectral and
This study focuses on four specific objectives
to estimate temporal trends in carbon stocks using ground data and to use these estimates as a standard for comparison
to construct confidence intervals for estimates of areas of undisturbed and degraded forest land obtained from a combination
of ground and multi-spectral data
to construct confidence intervals for lidar-assisted estimates of carbon stock change for undisturbed and degraded forest land
to estimate the number of years post-logging after which change in carbon stocks can and/or cannot be detected using lidar data
Detecting and Monitoring Tropical Forest Degradation in Vietnam using Landsat Time Series Analysis
PIs: James E. Vogelmann, (USGS), Michael Wimberly (South Dakota State University)
The purpose of this proposed work is to explore the use of Landsat time series data for mapping and monitoring forest degradation
in Vietnam. The degradation caused by tree harvest and slash and burn agriculture is of serious concern in Vietnam
(Manley et al., 2013). The proposed work fits under the SilvaCarbon “third stream of work on degradation,” whereby alternative
approaches are being solicited for detecting, measuring and monitoring tropical forest degradation. In general, the spatial,
spectral and radiometric qualities of Landsat data are particularly well suited for providing landscape characterization, and
monitoring degradation in tropical environments (Hansen et al., 2008; Lambin, 1999).
In this proposed project, the overall question is: How can we best use Landsat time series data to provide accurate and meaningful
forest degradation information in Vietnam? Some specific questions of that we will address include:
How do patterns and rates of forest degradation vary among different tropical regions?
What are the underlying causes behind these differences?
How do the spatial patterns of forest degradation vary across various selected tropical regions during the previous
28 years (i.e., using the historical Landsat TM 5, 7 and 8 archive)?
How do the differences in patterns of degradation vary among different types of tropical forests (e.g. dry deciduous versus
What are the relationships between patterns of deforestation and degradation?
What are the patterns of degradation near or within select natural reserves?
Investigating the influence of airborne lidar data density on the ability to detect low-intensity forest degradation in the western
PIs: Dr. Hans-Erik Andersen (USDA Forest Service)
Previous studies conducted at Antimary State Forest (western Brazilian Amazon) have indicated that low-intensity selective logging
activities can be detected using three-dimensional canopy structure information derived from measurements from airborne laser scanning
data (d’Oliveira et al., 2012). Using very-high-density lidar data (> 24 pulses/sq.m.), a relative density model (RDM) can be developed
that represents the density of lidar returns (and vegetation) within a layer in the forest canopy between 1 and 5 meters height above
ground. Variability in the density within this layer was found to be highly sensitive to forest impacts associated with selective
logging, such as the development of skid trails and logging roads. A more recent study utilized multi-temporal airborne lidar data sets
to quantify the reduction in biomass/aboveground carbon due to selective logging activities (Andersen et al., in review). This study
showed that even relatively low levels of biomass change (10-20 Mg/ha) could be detected and quantified using changes observed in
airborne lidar structural metrics. While the results of these studies are highly encouraging and indicate the potential utility of
airborne lidar as a tool in detecting and characterizing forest degradation in tropical areas, there are several remaining critical
research questions that will determine practical value of lidar for this application. For example, it is unclear how much the lidar
density can be reduced and still maintain an adequate level of accuracy. The lidar data used in these studies was very high density
(24 pulses/sq.m. and 10 pulses/sq.m. for 2010 and 2011 data respectively). While these densities were appropriate for research
studies, they are not economically feasible for large-area acquisitions, where we would expect densities closer to 1-4 pulses/sq.m.
We propose to investigate whether the quality of 1)the lidar-derived terrain model, 2) the information provided by the RDM and
3) the lidar-based measurement of changes in biomass/carbon via a model-based approach is sensitive to the density of the airborne
lidar data, and provide recommendations as to the lowest lidar density (i.e. most economical) that still yields acceptable results
for these three applications.