3
Measuring Fluxes from Land-Use Sources and Sinks

The agriculture, forestry, and other land-use (AFOLU) sector is the second-largest emitter of greenhouse gases, but the single greatest source of uncertainty. Uncertainty in carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O) emitted by AFOLU activities is typically 50-100 percent or more. Much of the uncertainty associated with total AFOLU emissions is caused by uncertainty in measurements of carbon stocks associated with deforestation and rates of tropical forest cover change. For example, the standing biomass of tropical forests is uncertain by a factor of 2 (Houghton et al., 2001; Saatchi et al., 2007), and estimates of the annual flux of CO2 released through forest clearing are uncertain by the same amount (Houghton, 2003; Achard et al., 2004; DeFries et al., 2007). For this reason, it is useful to focus on trends in AFOLU activity levels, rather than on the emissions themselves. For example, if the annually deforested area in a country decreases by a factor of 2, then, all else equal, CO2 emissions from deforestation have also decreased by a factor of 2 (van der Werf et al., 2009a).

This chapter first describes remote sensing methods that are able to estimate the activities responsible for the majority of AFOLU emissions (e.g., deforestation) and removals by sinks. It then identifies research that the United States could undertake to reduce uncertainty in these estimates. Such methodological improvements are periodically incorporated into United Nations Framework Convention on Climate Change (UNFCCC) inventory methods developed by the Intergovernmental Panel on Climate Change (IPCC). The chapter concludes with a discussion of research that could lead to long-term improvements in remote sensing capabilities.

REMOTE SENSING

Remote sensing provides a means to survey vegetation and land surface properties over large areas. It can be used to estimate the area within each country of recently harvested forest, mature forest, pasture, and various kinds of cropland, including rice paddies, which are a dominant source of CH4 emissions. It is also an effective means to measure fire and logging, which do not always lead immediately to detectable changes in forest cover. Time-series analysis of multiple scenes can be used to detect changes arising from deforestation, forest degradation, and afforestation. Deforestation refers to the conversion of forestland to agricultural cropland, grassland, and settlements. Degradation refers to a decrease in carbon stocks of ecosystems (e.g., through selective harvesting, draining of peatlands, or burning). Afforestation is the conversion of other land categories to forest.

In general, the transition from forest to cropland, urban areas, or pasture emits CO2 because of the decomposition of woody debris and short-lived wood products and because the remaining material is often burned to facilitate conversion. After clearing, forests can remain an annual net source of CO2 to the atmosphere for 5 to 20 years, with shorter time frames in warmer and wetter climates (Luyssaert et al., 2008) and in areas where the conversion process is rapid and



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 37
3 Measuring Fluxes from Land-Use Sources and Sinks T he agriculture, forestry, and other land-use (IPCC). The chapter concludes with a discussion of (AFOLU) sector is the second-largest emitter research that could lead to long-term improvements in of greenhouse gases, but the single greatest remote sensing capabilities. source of uncertainty. Uncertainty in carbon dioxide (CO 2), methane (CH 4), and nitrous oxide (N 2O) REMOTE SENSING e mitted by AFOLU activities is typically 50-100 percent or more. Much of the uncertainty associated Remote sensing provides a means to survey veg- with total AFOLU emissions is caused by uncertainty etation and land surface properties over large areas. It in measurements of carbon stocks associated with can be used to estimate the area within each country deforestation and rates of tropical forest cover change. of recently harvested forest, mature forest, pasture, and For example, the standing biomass of tropical forests various kinds of cropland, including rice paddies, which is uncertain by a factor of 2 (Houghton et al., 2001; are a dominant source of CH4 emissions. It is also an Saatchi et al., 2007), and estimates of the annual flux effective means to measure fire and logging, which do of CO2 released through forest clearing are uncertain not always lead immediately to detectable changes in by the same amount (Houghton, 2003; Achard et al., forest cover. Time-series analysis of multiple scenes can 2004; DeFries et al., 2007). For this reason, it is useful be used to detect changes arising from deforestation, to focus on trends in AFOLU activity levels, rather forest degradation, and afforestation. Deforestation than on the emissions themselves. For example, if the refers to the conversion of forestland to agricultural annually deforested area in a country decreases by a cropland, grassland, and settlements. Degradation factor of 2, then, all else equal, CO2 emissions from refers to a decrease in carbon stocks of ecosystems (e.g., deforestation have also decreased by a factor of 2 (van through selective harvesting, draining of peatlands, or der Werf et al., 2009a). burning). Afforestation is the conversion of other land This chapter first describes remote sensing meth- categories to forest. ods that are able to estimate the activities responsible In general, the transition from forest to cropland, for the majority of AFOLU emissions (e.g., deforesta- urban areas, or pasture emits CO2 because of the tion) and removals by sinks. It then identifies research decomposition of woody debris and short-lived wood that the United States could undertake to reduce products and because the remaining material is often uncertainty in these estimates. Such methodologi- burned to facilitate conversion. After clearing, forests cal improvements are periodically incorporated into can remain an annual net source of CO2 to the atmo- United Nations Framework Convention on Climate sphere for 5 to 20 years, with shorter time frames in Change (UNFCCC) inventory methods developed warmer and wetter climates (Luyssaert et al., 2008) by the Intergovernmental Panel on Climate Change and in areas where the conversion process is rapid and 

OCR for page 37
 VERIFYING GREENHOUSE GAS EMISSIONS more complete (Morton et al., 2008; van der Werf et action to reduce uncertainties. More information on al., 2009b). In contrast, afforestation causes a small trends in fertilizer consumption would provide insight long-term CO2 sink because woody biomass and soil about changing emissions when combined with infor- organic matter build up slowly on the site for decades mation on management practices, crop production, and to centuries. Forested land that is harvested (e.g., forest weather. degradation) and then allowed to regenerate produces a large and short-lived CO2 source, followed by a small The next five sections describe the methods and and long-lived CO2 sink. studies behind these conclusions. Uncertainty of AFOLU Activities, Emissions, and Available Satellites and Classification Methods Removals by Sinks Although a variety of satellite and aircraft sensors Table 3.1 summarizes the current levels of uncer- have been used to map land cover and land use (e.g., tainty for estimates of forest area, carbon stocks, and see Table 3.2), Landsat is widely used and offers several AFOLU emissions by remote sensing and the improve- advantages for regional- and country-level greenhouse ments in accuracy that could be accomplished using gas inventory applications. First, the pixel resolution remote sensing and ground monitoring within several (30 m) is high enough to distinguish most plant cover years. The most important messages in the table are characteristics, while providing a sufficiently large area the following: of coverage per scene to allow regular global coverage to be practical. Second, the 16-day repeat cycle allows 1. The area of land that is deforested within a the seasonal information needed for classification pur- country can currently be assessed with an uncertainty poses, such as greening and browning of grasslands, to of less than 50 percent over a period of 5 years (e.g., be obtained in areas with little or no cloud cover. Third, Hansen et al., 2008), but this could be improved to Landsat data provide a time series of observations 10-25 percent annually by expanding efforts to analyze beginning in 1984, although the older data have some satellite imagery. significant spatial gaps. A global map of land cover in 2. Uncertainty in remote sensing-based estimates 1990, when coverage is most complete, could serve as of annual carbon fluxes from deforestation, refores- a baseline from which to identify subsequent changes tation, and forest degradation is now high (25-100 (see Figure 3.1 for a Landsat map of the United States percent), but could be reduced to 10-25 percent by using data from the early to mid-1990s). integrating remote sensing observations with new Forests can be distinguished from nonforests at ground observations of forest biomass and peatland soil accuracies of 80-95 percent using Landsat-type imag- organic matter and biogeochemical models for spatial ery (Table 3.1; Lu et al., 2007; GOFC-GOLD, 2008). estimates. Accuracies can be validated with in situ observations 3. The area of flooded soils that emit CH4 (rice, or very high resolution aircraft or satellite data at a wetlands) can be determined each year with a relatively statistically defensible subsample of locations (e.g., low uncertainty (~10 percent), although uncertainty systematic or stratified sampling using Ikonos satel- in emissions is high (50-100 percent). Ground survey lite data). The accuracy of remote sensing-derived data on specific management practices would provide land cover maps can be assessed in a number of ways, the greatest reduction in uncertainty. including comparisons with independent data collected 4. Uncertainty in annual N2O emissions from in ground surveys. Errors are divided into those due to managed soils is high (about 50 percent) for the best omission (exclusion of an area from a category in which current inventory methods, and even higher (>100 it belongs) and commission (inclusion of an area in a percent) for developing countries. More flux measure- category to which it does not belong), and categorized ments for different nitrogen management practices by land cover class and region. All approaches are based and agronomic systems, coupled with improvements on statistical sampling. Land cover change also includes in process-based models, would be the most effective errors in geolocation and other factors.

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS TABLE 3.1 Reducing Uncertainties of Greenhouse Gas Emissions from Land Use Through Remote Sensing and Improved Monitoring Systems Current Attainable Annual Annual Measurement Uncertaintya Nature of Improvement Recommendationb Uncertainty Notes Forest Area Northern forest area 1c Systematic land cover mapping A 1+ across different countries using multiple satellite data streams Northern forest 2-3d Long-term continuity of ~30 m B, A 1-2 Loss of LDCM would increase deforestation- and ~1 m satellite observations; uncertainty for both northern afforestation new investment to map changes at and tropical deforestation rates an annual scale using new change to level 5 (greater than 100%) detection algorithms Tropical forest area 1-3e Same needs as for mapping B, A 1-2 northern forest area; improved access to international satellite observations Tropical forest 3-4f Same as above B, A, C 1-2 deforestation- afforestation Carbon Stocks Northern ecosystem 2-4g Improved access to existing E, D, C 2 Many Annex I forest carbon carbon stocks country inventories; a new inventories are not publicly initiative to improve the spatial available distribution of emission factors by combining P-band radar and lidar observations with inventory data Tropical ecosystem 3-4h Capacity building for tropical E, D, C 2 Forest biomass measurements carbon stocks forest and peatland inventories; the have not been systematically emission factor initiative described organized; peatland areas and above depths have not been accurately mapped Ecosystem Degradation Logging 3-4i Dedicated high-resolution (~1 m) B, A 2 observations for assessing logging rates in deforestation hot spots and for relating degradation patterns to Landsat observations; a wood products tracking system Fire emissions from 3-4j Improved atmospheric E, F 2 An OCO rebuild with existing tropical forests and emission ratios of CO/CO2 for MOPITT and TES CO peatlands deforestation and peatland fires observations would enable improved estimates of peat emissions CO2 Emissions Northern ecosystem 2-4k Improved inventories with A-E 2 Higher uncertainty levels for carbon fluxes belowground carbon monitoring countries without inventories in cropland, grassland, and forests; integration of improved observations with biogeochemical models continued

OCR for page 37
0 VERIFYING GREENHOUSE GAS EMISSIONS TABLE 3.1 Continued Current Attainable Annual Annual Measurement Uncertaintya Nature of Improvement Recommendationb Uncertainty Notes Tropical ecosystem 4-5l Improved estimates of forest A-F 2 carbon fluxes cover change, emission factors, measurement inventories including post-clearing land uses, and top- down constraints from column CO2 satellite measurements; integration of improved observations with biogeochemical models CH4 Emissions Rice production area 2-3m Multisensor mapping (e.g., A, C 1 Landsat, Ikonos, SAR) Flooded rice area 3n Multisensor mapping (e.g., SMAP, A 1 SSM/I, Landsat) Soil CH4 emissions 3-4o Ground surveys of management E 2 practices; flux measurements as a function of key management practices, including midseason drainage and fertilizer application Fire CH4 emissions 4-5p Continuous sampling instruments E, F 3 established on towers near fires and used to calibrate models and remote sensing data in relatively homogeneous areas N2O Emissions Soil N2O emissions 4-5q Flux measurement network E 3 targeting different hot spots and soil, fertilizer, and manure management practices NOTES: CO = carbon monoxide; LDCM = Landsat Data Continuity Mission; MOPITT = Measurements of Pollution in the Troposphere; OCO = Or- biting Carbon Observatory; SAR = synthetic aperture radar; SMAP = Soil Moisture Active-Passive; SSM/I = Special Sensor Microwave Imager; TES = Tropospheric Emission Spectrometer. aUncertaintylevels are 1 = 100%. bRecommended improvements: A. A new U.S. initiative to map global land cover and land-use change at an annual resolution using multiscale remote sensing data, including but not limited to the instruments listed in Table 3.2. A key part of this initiative is the design of multiscale approaches for ground truthing in rapidly changing areas, drawing on regional expertise from ground observers and high-resolution (~1 m) satellite and aircraft imagery. B. Successful launch and operation of the LDCM, together with immediate investment in follow-on missions for 30 m and 1 m resolution continuous data, which are crucial for monitoring logging and for improving emission factor estimates in hot-spot areas. C. Improved remote sensing data policies for sharing moderate- and high-resolution observations for forest cover and forest biomass monitoring (e.g., existing moderate-resolution visible and infrared reflectance measurements and radar observations from several countries cannot be downloaded for perusal or exploration by U.S. scientists). D. Improved data policies for sharing forest inventory observations among different countries. E. A new U.S. initiative to reduce uncertainties associated with emission factors related to land-use change and ecosystem degradation. This would include capacity building and investment in international scientific efforts to develop forest and peatland inventories in tropical countries, including carbon stocks associated with key post-clearing trajectories of land use. It would also involve a research program to find the most efficient ways to combine inventory observations with satellite observations (including existing P-band radar and lidar observations) to map spatial patterns of forest biomass and peatland carbon. For the fire component of forest and peatland degradation, new information of the emissions ratios of CO and CO2 from different types of tropical burning would enable more effective use of column CO2 and CO observations to constrain emissions. Emission factors for CH4 and N2O as a function of agronomic system and management practice (e.g., midseason drainage in rice agriculture) are not well characterized and represent a primary source of uncertainty in estimating regional scale emissions. F. Rebuilding and launching an Orbiting Carbon Observatory satellite to measure column CO2. Fire emissions at the deforestation frontier in South America and Southeast Asia would be detectable by an OCO-like instrument. An OCO rebuild would also help to constrain estimates of emissions from peatland degradation in Southeast Asia. cGOFC-GOLD (2008).

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS TABLE 3.1 Continued dExpert opinion based on Landsat change detection approaches taken by Masek et al. (2008) and Kennedy et al. (2007). Masek et al. (2008) report for the conterminous United States that the rate of disturbance between 1990 and 2000 was 0.9 ± 0.2% per year, primarily as a consequence of harvesting and fire. eFor global humid tropical forest area, the difference between the Hansen et al. (2008) estimate for the year 2000 (1,152 Mha) and the Achard et al. (2002) estimate for the year 1997 (1,076 Mha) is relatively small (less than 10%). For tropical forest area as a whole (including forest area within 90 tropical countries), the 2005 Food and Agriculture Organization (FAO) Forest Resource Assessment estimate for the year 2000 is 1,828 Mha (Grainger, 2008). FAO assessments for this larger domain have varied considerably in different reports over the last three decades, particularly at the scale of individual countries (see Table S5 in Grainger, 2008). Estimates of year 2000 forest cover, for example, differed by more than 10% for 6 of the 12 countries with the largest amounts of tropical forest area (Grainger, 2008). fAssessments of tropical forest cover loss by Hansen et al. (2008) and the 2005 FAO Forest Resource Assessment differ by about a factor of 2 (5.4 Mha yr–1 vs. 12 Mha yr–1 for 2000-2005). Although part of this difference is associated with study domain; at a country level differences between the two ap- proaches remain substantial, with a ~16% difference for Brazil between the two approaches and more than a factor of 2 difference for Indonesia (Hansen et al., 2008). gFor some northern regions, such as the coterminous United States, forest biomass mapping approaches—which combine remote sensing and detailed inventory information—appear to be converging (see Table 11.2 of CCSP, 2007, and Table 5 of Blackard et al., 2008), implying that the uncertainty range is currently within ±25%. In other temperate and boreal regions, where inventory information is not as complete, the uncertainties are higher (Houghton et al., 2009; Goodale et al., 2002). hRegional-scale inventories in tropical regions depend on both the quality of the inventory observations and the remote sensing techniques used to extend these observations in space and time. Based on differences between extrapolation approaches, uncertainties for aboveground live biomass within the Amazon basin are within ±50% (Houghton et al., 2001; Saatchi et al., 2007). Peatlands within the Indonesian Archipelago are extensive (Page et al., 2004) but have not been systematically mapped in terms of carbon content or depth. iFor the Brazilian Amazon, Asner et al. (2005) estimate wood extraction to range between 27 million and 50 million cubic meters of wood using Landsat observations to identify degraded areas. Detailed estimates using high-resolution satellite observations are not available for many other tropical countries. jVan der Werf et al. (2008) estimate emissions from Indonesia, Malaysia, and Papua New Guinea during 2000-2006 to be 128 ± 51 Tg C yr –1, with a large component of the uncertainty attributable to an incomplete characterization of emission factors (CO/CO2 ratios) for peat fires. kExpert opinion based on syntheses by Goodale et al. (2002) and CCSP (2007). lGlobal land-use change was estimated to be 1.2 ± 0.7 Pg C yr–1 for 2008 in Le Quéré et al. (2009), with uncertainties dominated by tropical forest areas. These uncertainty levels remain similar to those reported in Working Group 1 of the IPCC Fourth Assessment (Denman et al., 2007). mExpert opinion based on analysis of data on rough rice area and yield; available from the International Rice Research Institute at . nExpert opinion. Key factors influencing flooded area include water demand for other sectors and changing rice management practices (Frolking et al., 2004). Flooded areas can be detected using passive microwave satellite observations (Prigent et al., 2007). oExpert opinion based on studies by Wassmann et al. (1996), van der Gon (1999), Jain et al. (2000), and Li et al. (2002). pCampbell et al. (2007). qThe committee assumed that the –40% to 70% uncertainty range for global annual agricultural N O emissions reported by Bouwman et al. (2002) was 2 conservative and that uncertainties are likely to be considerably higher in many regions and individual countries where activities are not consistently measured. Key agricultural components that need to be sampled more effectively include manure generated from livestock management and different types of fertilizer application (e.g., Davidson, 2009). Longer sampling intervals (e.g., 5 to 10 years) lead satellite sensors that provide frequent global coverage to higher rates of omissions of land cover change because (>250 m resolution), although significant deforesta- tion also occurs in smaller patches <100 m2 in size or of forest regrowth after harvest. The uncertainty can be reduced in many cases by using scenes collected more in fine-scale linear patterns (Skole and Tucker, 1993). frequently (e.g., <1 year to 2 years) in the analysis. A Landsat-type remote sensing (30 m resolution) can global analysis using the Global Land Survey Landsat detect deforestation at fine scales (Huang et al., 2009). data record has been produced at a decadal interval, and Nearly complete coverage of the globe from Landsat efforts are under way to reduce this to a 5-year interval satellites is available for the early 1990s and 2000s. at least back to 1990 (Townshend et al., 2008). The Lower resolution (1 km) Advanced Very High Resolu- reduction from 10 to 5 years will reduce omissions of tion Radiometer (AVHRR) data on global land cover characteristics are available from 1992 to 1993.1 land cover change, although 2-year intervals are likely to be necessary in regions where vegetation regrows Deforestation by complete clearcut harvest or quickly (e.g., wet tropical forests). stand-replacing fire is currently being assessed with annual Landsat data in some regions (e.g., Oregon, California, Washington; Law et al., 2006; Kennedy et Remote Sensing of Deforestation, Forest al., 2007). An accuracy assessment in Law et al. (2006) Degradation, and Afforestation Deforestation and Afforestation. Large-scale contigu- 1 See the Global Land Cover Characteristics database at .

OCR for page 37
 VERIFYING GREENHOUSE GAS EMISSIONS TABLE 3.2 Current Land Remote Sensing Instruments in the Public Domain Resolution and Instrument Measurement Coverage Data Availability Land Remote Sensing Satellite Provides the longest continuous record of the Earth’s 15-60 m, global Landsat 7: 1999-present (Landsat) continental surfaces Landsat 5: 1984-present Advanced Spaceborne Thermal Emission Provides high-resolution images of the land surface, 15-90 m, global 1999-present and Reflection Radiometer (ASTER) water, ice, and clouds Moderate Resolution Imaging Measures biological and physical processes occurring 250 m-1 km, global 1999-present Spectrometer (MODIS) on the surface of the Earth, in the oceans, and in the lower atmosphere Airborne Visible/Infrared Imaging Measures constituents of the Earth’s surface and 5-20 m, aircraft is tasked 1998-present Spectrometer (AVIRIS) atmosphere showed an uncertainty of 20 percent when automated Forest Degradation. Remote sensing techniques can techniques were compared with air photos (Table 3.1). be used to identify partial biomass removals over large These methods may be suitable for countries with areas, particularly biofuels harvest (Asner et al., 2005) appropriate technical expertise and software capabili- and selective harvest of high-grade trees in the tropics. ties for automation. In northern forests, degradation from logging can be FIGURE 3.1  Major land cover features for the coterminous United States, based on early to mid-1990s Landsat Thematic Mapper  satellite  data.  SOURCE:  Vogelmann  et  al.  (2001).  Reprinted  with  permission  from  the  American  Society  for  Photogrammetry  and  Remote Sensing.

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS distinguished by inferring changes from the differences based on techniques of spectral mixture analysis and between two satellite images or, more recently, by exam- N ormalized Difference Fraction Index (GOFC- ining as many as 25 images over a time series to identify GOLD, 2008). The system was tested in Brazil and segments that are changing over time. The former is Bolivia using 30 m Landsat data, and it failed to detect typically done using annual to subannual Landsat data selective harvest. The system is now being tested using and research-level algorithms. The latter (trajectory- Satellite Pour l’Observation de la Terre (SPOT) 5 (10 based image analysis), which is currently operational m resolution) and ASTER (15 m resolution) imagery, across the Pacific Northwest region of the United which have spatial resolutions more appropriate for the States, has advantages over traditional approaches in size of individual tree canopies. Even if this approach that it can detect forest thinning and trends such as is successful, it has three limitations: (1) it requires progressive change from one land cover type to another, frequent (at least annual) mapping; (2) natural (e.g., spreading mortality, and slow regrowth of forests over windthrow) and human-caused degradation can be time (Kennedy et al., 2007). It can also detect a wide confused, possibly requiring additional ground or air range of disturbance and recovery phenomena that were photo interpretation; and (3) it requires a higher level previously too ambiguous to label, capturing types of of expertise and software for automated techniques. degradation with accuracies two to five times higher At the national scale, the most effective method for than previous change detection methods. Combining detecting areas of selective harvest is to apply high spa- trajectory-based image analysis and high-resolution tial and temporal resolution remote sensing approaches data improves the accuracy of regional estimates of ter- to areas suspected of thinning, such as those deter- restrial carbon fluxes and enables identification of the mined by detection of landings along roads. Asner et al. type of forest degradation (e.g., thinning versus mor- (2005) applied an automated image analysis approach tality from insects or diseases; Kennedy et al., 2007). to annual Landsat data and pattern recognition tech- Such subtle disturbances may have potentially large niques for detecting selective logging in the Brazilian cumulative impacts on carbon cycling at the regional Amazon. The analysis required initial ground-based scale (e.g., large-scale mortality of boreal forests from spectroscopic characterization of surface features and insect attack; Kurz et al., 2009). tree species canopy spectra from a spaceborne hyper- Landsat data are also being used in time-series spectral sensor (Hyperion). The authors found an over- analysis across North America to identify forest areas all uncertainty of up to 14 percent in total logged area, subject to harvest and wildfire with a repeat inter- based on seasonal Landsat data, atmospheric modeling, val of 2 years (Goward et al., 2008). An assessment detection of forest canopy openings, surface debris, and over southeastern and northern U.S. national forests bare soil exposed by forest disturbances. Alternatively, a indicated overall accuracy values of 80 percent when combination of seasonal Landsat-type remote sensing comparing automated to human-identified disturbance and lidar or P-band radar may be required to reduce mapping (Table 3.1). Most of the omissions were par- uncertainty (Treuhaft et al., 2004). tial disturbances, such as thinning and storm damage, Anthropogenic fires in tropical peatlands and at although some clearing harvests may not be detectable the deforestation frontier contribute substantially to with temporal intervals of 2 years or more in areas of interannual variation in the growth rate of atmospheric rapid forest regrowth (Huang et al., 2009). This type of CO2 and CH4, so fire monitoring is crucial to separate approach has the potential to be applied globally. natural trends in atmospheric concentrations from Degradation from selective logging is more difficult the effects of mitigation. In addition, fire is used in to detect. In tropical forests, selective logging may leave some parts of the world to clear forest for pasture or a forest canopy that fills in within a year or that does agriculture, and fire is an important source of atmo- spheric CH4 (14-88 Tg CH4 yr–1; Mikaloff Fletcher not appear to have been thinned. The trajectory-based change detection approach has not been tested in the et al., 2004; van der Werf et al., 2006; Denman et al., tropics. A new forest degradation monitoring system 2007). A variety of remote sensing methods are being (Real-time Detection of Deforestation, DETER) has used to identify the location, area, and intensity of fire. been developed in Brazil to detect selective logging, The Global Observations of Forest and Land Cover

OCR for page 37
 VERIFYING GREENHOUSE GAS EMISSIONS Dynamics (GOFC-GOLD) project is working to may be able to detect large homogeneous fires (27-34 develop a global system of geostationary monitoring percent uncertainty), but it is less effective in areas of active fires. Multiyear burned area products derived with high tree cover or heterogeneous burned areas from moderate resolution satellite imagery are avail- (45 percent uncertainty; Hawbaker et al., 2008; Giglio able from several sources including SPOT (Tansey et al., 2009). With both the regional and the global et al., 2008) and the Moderate Resolution Imaging a pproaches described above, parameterizations of Spectrometer (MODIS; Giglio et al., 2006; Roy et al., combustion completeness are now the most uncertain 2008). Landsat provides more accurate mapping of fire component of these models. area and severity in forests and shrublands, and it has Post-fire emissions of CO2 can be substantial and been used for this purpose in the United States since may persist for several years after fire as a consequence 1984.2 Tests using field data indicate that change detec- of decomposition of remaining organic matter, includ- tion using Landsat identifies high-severity fires with ing vegetation killed but not consumed by the fire 10-30 percent uncertainty, moderate-severity fires with (e.g., McMillan et al., 2008). Key factors that regulate 40-50 percent uncertainty, and low-severity fires with the time it takes for an ecosystem to transition from 30 percent uncertainty in the western United States a source to a sink after fire include the severity of (Miller et al., 2009). In four fires in the western United the fire, recruitment and growth of new plant species States, low- and moderate-severity fire released 58 and within burned areas, and fire-induced changes in the 82 percent as much carbon emissions, respectively, as microclimate, which influence rates of decomposition high-severity fire (Meigs et al., 2009), so determining and levels of soil moisture. the severity of all fires would reduce uncertainty in emissions estimates. National reporting is inconsistent, Soil Carbon limiting our understanding of fire effects on forest and ecosystem degradation. Remote sensing of soil carbon stocks is not possible Approaches to estimating fire emissions vary because the soil immediately below the surface is opaque widely and some result in overestimates because of to the portions of the electromagnetic spectrum used to faulty assumptions about the amount of fuel that is detect properties of organic matter. Further, variations combusted (Wiedinmyer et al., 2006; Campbell et in soil moisture, soil mineralogy, plant residue, and veg- al., 2007). A common approach is to use Landsat (30 etation cover make even soil surface carbon estimates m pixels) estimates of burn area (change detection) problematic, although low- versus high-organic-matter and fire severity, and combustion completeness fac- soils can be broadly differentiated (Sullivan et al., 2005; tors derived from field observations of live and dead Yadav and Malanson, 2007). However, remote sensing biomass and surface organic matter before and after provides information on aboveground vegetation (e.g., fire. Global-scale estimates use similar methods, but plant phenology, leaf area, photosynthetically active draw on moderate-resolution burned area observations radiation) and residue cover. In conjunction with mod- (e.g., 500 m MODIS data) that do not differentiate fire els, these data can be useful in estimating soil carbon severity. They require the use of models to estimate bio- stocks and stock changes and greenhouse gas emissions. mass and combustion completeness (e.g., van der Werf Nevertheless, a much more extensive set of ground- et al., 2006). An alternate method for global estimates is based soil carbon measurements will be necessary to to use fire radiative power estimated at a subpixel level improve the soil carbon models before they are reliable (e.g., using MODIS) and calibrated based on relation- enough for treaty purposes (see below). ships between radiated energy and field estimates of combustion (Wooster et al., 2005). The results are at Cropland and Pastures a coarser resolution than those yielded by the Landsat approach because active fire observations are currently Most comprehensive maps of agricultural land available only at 1 km spatial resolution. The approach cover derived from satellite imagery provide informa- tion only on aggregate classes (e.g., row crops) and often at relatively coarse (e.g., 1 km) resolution (Friedl 2 See .

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS et al., 2002). However, accurate mapping (>90 percent duration and timing of flooding. Studies combining fidelity) of major field crop types (e.g., corn, soybean, remote sensing with ground-based surveys and model wheat) has been demonstrated using Landsat (Daugh- simulations have produced estimates of country-scale try et al., 2006), and higher-resolution (30-50 m) CH4 emissions from rice in China (Li et al., 2005; Yao crop maps are now being generated for U.S. croplands et al., 2006) and India (Manjunath et al., 2006). Similar ( West et al., 2008). Recently, progress has been made techniques to map natural wetlands using remote sens- in assessing crop residue coverage, which is closely ing (Kaheil and Creed, 2009) and climate information correlated with tillage management, using space-borne could help in estimating CH4 emissions from wetlands hyperspectral instruments. For example, Daughtry et and other flooded soils that are not included in con- al. (2006) were able to accurately differentiate mini- ventional inventories. Accuracies of 90-95 percent have mum (conservation) tillage fields from more intensive been achieved in mapping and classifying different rice (reduced plus intensive) tillage practices 80 percent of production systems using multiscale, multispectral sat- the time in corn and soybean fields in central Iowa, ellite data (Biradar et al., 2008). However, actual CH4 based on comparisons with ground surveys. They used emissions have greater uncertainties (50-100 percent) a cellulose absorption index based on reflectance in the at regional to national scales because they are influenced upper shortwave infrared wavelength region from the by variable water and crop residue and manure manage- EOS-1 Hyperion sensor.3 Differentiation of three till- ment and plant varieties, and because plant varieties age classes (conservation, reduced, intensive) was only must be quantified using ground survey information. 60 percent accurate. Further development to correct Virtually all soils emit nitrous oxide, but the main for interference from certain types of soil minerals and driver for increased N2O emissions is external input to screen out pixels with more green vegetation could of nitrogen, particularly from synthetic fertilizers but further improve accuracy (Serbin et al., 2009). also from greater use of legume crops, manure, and deposition of the nitrate generated by combustion of fossil fuel. Emissions occur both at the site of nitrogen Methane and Nitrous Oxide application (direct emissions) and in adjacent or down- Flooded soils, including rice fields, natural wet- stream ecosystems (indirect emissions) that receive lands, and reservoirs, are significant sources of CH4 nitrogen that was lost from where it was originally emissions (Denman et al., 2007). Currently, only applied. Remote sensing of vegetation characteristics flooded rice is included as an anthropogenic source in (e.g., species, leaf area, leaf chlorophyll) that are related national inventories, although a provisional method- to plant nitrogen status can help constrain model-based ology for calculating CH4 emissions from reservoirs estimates of N2O (and nitrogen oxide) emissions (Mar- and other artificial water impoundments is included tin and Asner, 2005; Vuichard et al., 2007). However, in the IPCC guidelines. For rice methane, one of the emission estimates remain highly uncertain because of most significant management effects is the timing and the high spatial and temporal variability of fluxes and duration of flooding—a CH4 abatement option is to because of uncertainty in management practices related reduce the period of flooding and encourage midseason to nitrogen use and the amount and form of nitrogen in drainage of paddies. Landsat (e.g., Thimsuwan et al., the system. Direct monitoring of nitrogen management 2000; Biradar et al., 2008) and MODIS (e.g., Saka- practices via remote sensing is unfeasible, and ground moto et al., 2009) have been used to determine rice surveys would be required to verify the effects of any areas and phenology, and seasonal synthetic aperture changes in practices directed at reducing N2O emis- radar (SAR) images (e.g., Diuk-Wasser et al., 2006; sions (e.g., reduction in fertilizer use; change in tim- Salas et al., 2007) have been used to determine the ing, method of application, or type of fertilizer; use of nitrification inhibitors). Direct measurements of N2O using micrometeorological techniques (e.g., Phillips 3The EOS-1 Hyperion sensor was flown as a test bed for sci- entific applications of high spectral resolution (unlike the Landsat et al., 2007; Fowler et al., 2009) with aircraft or tower fixed spectral bands). It has limited spatial coverage and is near observations, in addition to more conventional chamber the end of its lifetime, so this satellite cannot be used for future methods, offer the potential for estimating emissions applications.

OCR for page 37
 VERIFYING GREENHOUSE GAS EMISSIONS for land areas with a high concentration of agricultural unprecedentedly accurate estimates of CO2 emissions activities. Observations of NO (nitric oxide) emissions from U.S. AFOLU activities (i.e., 10 percent error from agricultural soils following application of fertilizer would be a realistic target given the experience of the and precipitation have also been detected with satellite FIA inventory). It would demonstrate new inventory observations (Bertram et al., 2005). Because the release methods that could be brought into the UNFCCC of N2O and NO is often proportional, depending on process, including better quantitative estimates of type of fertilizer (e.g., Akiyama and Tsuruta, 2003), important emission factors. Finally, it would provide an this approach may help quantify agricultural N2O accurate measure of the total CO2 flux from non-fossil- emissions. fuel sources in the United States that could be used to develop better atmospheric methods for estimating carbon emissions (see Chapter 4). IMPROVING UNFCCC INVENTORIES OF Several European countries, Japan, Canada, and LAND-USE EMISSIONS Mexico now have high-quality forest inventories, but Improvements in monitoring and verification most other nations rely on methods that are much less of AFOLU emissions will depend on reducing the accurate than the FIA. Global monitoring of forest uncertainties of UNFCCC inventory estimates. The carbon stocks with ground-based inventories would next three sections describe research that could lever- provide the most accurate estimates of any method age existing infrastructure to reduce uncertainties in but would require an FIA-level effort in all forested AFOLU emission estimates. countries, with spatially representative sampling, repeat visits to permanent plots, and measurements of above- ground live and dead biomass, forest floor carbon, Improved Carbon Inventories belowground live and dead biomass, and soil carbon. Changes in forest biomass carbon stocks can be Detailed methods for establishing a forest inventory measured directly from inventories of aboveground are laid out in Global Terrestrial Observing System- biomass (live and dead) at two points in time, supple- Terrestrial Carbon Observations (GTOS-TCO) pro- mented with plot data on coarse root carbon and tocols (Law et al., 2008). stumps. The U.S. Forest Service Inventory and Analy- Global inventories of carbon fluxes from all types sis (FIA) program measures every tree on more than of ecosystems would require the implementation of 100,000 plots every 5 years, and the cause of death is similar protocols with repeated measurements of eco- investigated if it died. Plots are selected using stratified system carbon stocks over time in croplands, pastures, random sampling of remote sensing imagery. Data on and nonforested natural ecosystems in all countries. soil carbon, woody debris, and quantities such as soil This is likely to be beyond the capacity of many nutrients, light levels, and tree health are collected on a developing nations. For example, the cost of the FIA subset of the plots. These data are generally converted program in the United States is roughly $65 million to carbon using wood density data from tree cores per year, although FIA was designed for other purposes (roughly 50 percent of stemwood biomass is carbon). and the cost reflects many additional objectives. The The Forest Service estimates that uncertainty is 20 U.S. inventory is heavily based on field data collection percent for the nation’s forest carbon uptake and 10 designed for state-level consistency in sampling inten- percent for the aboveground timber volume change that sity. An efficient design for greenhouse gas monitoring is actually measured (EPA, 2008). However, estimates would rely more heavily on remote sensing and would that include historically unmeasured pools and lands have a variable sampling intensity based on ecosystem put the uncertainty at 50 percent (CCSP, 2007). characteristics. To reduce uncertainty, measurements The United States does not conduct a similarly should include annual growth from tree cores (except rigorous carbon inventory for nonforested ecosystems, some tropical species), changes in dead material (tree such as croplands, pastures and natural grasslands, or stems, branches, bark, stumps), and changes in soil shrublands. However, doing so would provide a number carbon between the two measurement periods. In the of advantages. An expanded inventory would provide United States, an expanded carbon stock measurement

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS network for nonforested ecosystems could build on the carbon stocks and fluxes from the land. The measure- U.S. Department of Agriculture’s (USDA’s) existing ments can also be used to evaluate flux estimates from National Resource Inventory, which already collects ecosystem inventories. For example, records from the an extensive set of land-use and management data, site at Harvard Forest show that the annual net carbon for a cost of about $5 million per year ( Jeffery Goe- uptake over 15 years has averaged ~2.5 tons of carbon bel, USDA, personal communication, 2009). A good per hectare per year and has increased at an average example of an efficient greenhouse gas inventory design rate of ~0.2 tons of carbon per hectare per year, which is the Australian National Carbon Accounting System, is consistent with a comprehensive carbon inventory which was designed explicitly for carbon accounting at the same site (Barford et al., 2001). Where inven- in the absence of an existing forest inventory and was tory measurements are made infrequently or turnover budgeted for $35 million over 10 years.4 The Mexican rates of pools are low (e.g., slow decomposition rates), forest inventory, which is almost identical to the U.S. it will take 5 to 10 years to compile enough data to design but has ~17 percent of the forest area, costs ~$2 compare with eddy covariance measurements (Curtis million per year (18 percent of the U.S. cost per unit et al., 2002). forest area), largely due to significantly lower labor costs Eddy covariance measurements have several advan- (Richard Birdsey, USDA, personal communication, tages over ecosystem inventory methods for computing August 19, 2009). Given the Australian and Mexican the land-based net annual carbon uptake. In particular, examples, the cost of a comprehensive national inven- they measure contributions to the CO2 flux from all tory of all non-fossil-fuel carbon fluxes would likely carbon pools, including some that inventories may miss. be a few (i.e., of order 10) million dollars per year in In addition, eddy flux sites provide information on the large and populous countries and significantly less in rapid fluctuation in carbon exchange over 24 hours and most countries, because the number of measurement between days that is vital to constrain models of ter- sites scales with land area and the cost of labor. An restrial ecosystem carbon stocks and fluxes (Urbanski assessment of monitoring costs for the Reducing Emis- et al., 2007; Medvigy et al., 2009; Wang et al., 2009) sions from Deforestation and Forest Degradation in and estimates of carbon fluxes from atmospheric data Developing Countries (REDD) program suggested (see Chapter 4). that, depending on the policy framework and the preci- The distribution of flux sites is determined by sion needed to detect carbon stock and area changed, national scientific research programs, with a relatively monitoring costs may reach $550 per square kilometer large number in many developed countries, but few or (Böttcher et al., 2009). none in developing countries. China and India recently started their own networks. Over the past 10 years, the number of sites in the global network has increased Eddy Flux Networks fivefold to 500 sites worldwide and 103 in the Ameri- An eddy covariance tower is a device that under Flux network in the Americas. The regional networks most meteorological conditions measures the instan- operate independently, but protocols exist or are being taneous exchange of CO2 and other gases between the developed to coordinate or standardize measurements atmosphere and land surface for areas ranging from a across networks for various purposes. For example, the hectare to a few square kilometers, depending on the Integrated Carbon Observation System (ICOS) would height of the tower. Fluxes are computed half-hourly standardize measurements at a number of European and summed over an entire year to provide an estimate sites and combine them with other kinds of measure- of the annual net amount of CO2 absorbed or released ments to provide improved regional estimates of carbon fluxes.5 by the land ecosystem. Eddy covariance tower measure- ments are used to calibrate models that map annual Eddy flux towers are too expensive (i.e., $100,000 per year) to be used to verify emissions by themselves, given the heterogeneity of ecosystems. Recall that the 4 See and < http://www.globalcarbonproject.org/global/presentations/2_ 5 See Terrestrial/Richards.pdf>. .

OCR for page 37
 VERIFYING GREENHOUSE GAS EMISSIONS Reducing Uncertainties in AFOLU Emissions of FIA relies on 100,000 sites to characterize stemwood CH4 and N2O dimensions of the different forest types, climate, and soils in U.S. forests. Also, the eddy covariance method The greatest impediment to reducing uncertain- is vulnerable to systematic bias errors in nonideal ter- ties in soil N2O fluxes is the limited number of flux rain (Finnigan et al., 2003). Nonetheless, eddy flux measurements for different climate regions, soil types, networks would be an essential component of an inte- and management systems. An expanded network of grated system of measurements for monitoring land- flux measurement sites, using conventional chamber- use greenhouse gas fluxes in countries with sufficient based methods, at well-characterized field experiments capacity and funding. could provide data to calibrate process-based models Using integrated observation and modeling frame- that integrate variable climate, soil, and management works, annual carbon stocks and fluxes—including car- conditions. Although chamber methods are commonly bon sources and sinks from deforestation, degradation, used and are well suited for experimental plots compar- and afforestation—can be estimated for some countries ing differences between management systems, they are with uncertainties of ~30 percent (Luyssaert et al., subject to high spatial variability due to their small size. 2009). Uncertainty can be reduced by using improved Combining conventional chamber-based approaches observations (e.g., remote sensing of disturbance his- with micrometeorological techniques that are capable tory, soil carbon) and data assimilation methods in the of estimating integrated fluxes for larger areas could modeling framework. further improve emission estimates (Fowler et al., 2009). Additional basic research to improve under- Anthropogenic Sinks at Small Scales standing of biotic and abiotic controls on N2O fluxes and improved predictive flux models are also needed. Under the UNFCCC, deliberately enhancing car- Better quantification of indirect N2O emissions, which bon uptake (e.g., by planting a forest) can be counted are driven by the transport and subsequent emis - as an anthropogenic sink. The inventory and satel- sion of excess nitrogen to nonagricultural landscapes, lite methods described in this chapter can be used to will require flux measurements and nitrogen balance monitor emissions and removals (sources and sinks) measurements at watershed scales (Deay et al., 2003). from forests with similar accuracy. However, verifica- Improved survey data on soil nitrogen additions and tion of an emissions-trading and/or offset program management practices at local to national scales, which would require monitoring at scales as small as a forest are needed to drive predictive models, can further plantation or farm. Because ecosystem carbon uptake reduce uncertainties. and release fluctuate from year to year with changes in Soil emissions of CH4 are a minor component of the weather and other factors, carbon gains caused by U.S. land-use-related emissions but are a major green- deliberate management will be best measured against house gas source in rice growing regions, particularly in the baseline carbon flux on similar lands without the Asia. Additional flux studies, along with better survey management. data on rice cropping practices (e.g., water manage- Both high-resolution satellite imagery of forests ment, residue management, manuring) are needed to and the small inventory plots used by the FIA have reduce uncertainties in annual emissions. Although the spatial resolution required to assess forestry and CH4 emissions from native wetlands, reservoirs, and land use at small scales. High-resolution satellites are other flooded land are not currently required in national capable of monitoring the sizes of individual trees and greenhouse gas inventories, they are highly uncertain thus are able to monitor claimed increases in above- due to the paucity of field studies conducted. ground carbon stored by planted trees. Thus, conduct- Ground-based eddy flux sites (described above) ing regular inventories of all ecosystems and satellite- could be augmented with tunable diode laser instru- based assessments of land use would also facilitate mentation to measure CH4 and N2O fluxes (Pattey monitoring of forestry offset projects and identification et al., 2006). The CH4 and N2O flux instrumentation of emission leakage. is more expensive and difficult to manage than that

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS required for CO2, so deployment of such systems is scale using lidar and High-fidelity Imaging Spec- currently limited, but it is likely to grow in the future trometer (HiFIS) data in a tropical region with closed (Desjardins et al., 2007; Sutton et al., 2007; Denmead, canopies and complex terrain (Asner et al., 2009). The 2008). method explained ~80 percent of the variation in field observations of aboveground forest biomass. The great- est source of uncertainty was in the field measurements FUTURE (>5 YEARS) OPPORTUNITIES AND used to develop species-specific equations. The inten- THREATS sive measurements, which are currently made from aircraft, are experimental and require field observations Remote Sensing Methods for Estimating for developing the lidar equations. However, they show Aboveground Carbon Stocks promise for landscape applications that could be scaled to larger areas using models or other scaling approaches Improvements in spatial estimates of terrestrial (e.g., 10-year time frame). biomass (not including soils) are possible with planned It may be possible to calibrate the National Aero- satellite sensors. Currently, aboveground biomass is nautics and Space Administration’s (NASA’s) planned estimated using trajectory change detection maps to Deformation, Ecosystem Structure and Dynamics of first identify which areas are changing and then ascribe Ice (DESDynI) mission and, to a lesser extent, the Ice, biomass before change, biomass gained or lost, and Cloud, and Land Elevation Satellite-II (ICESat-II) thus biomass after the signal stabilizes. It is difficult mission,6 with aboveground biomass observations to to model biomass with Landsat alone, but Landsat produce estimates of aboveground carbon storage in time series could be combined with suitable ecosystem vegetation. These could be used to monitor changes inventory data in a modeling framework to estimate in aboveground biomass caused by major disturbances biomass and biomass change (Samuel Goward, Uni- (harvest, fire, storms). versity of Maryland, personal communication, April 2009; Powell et al., 2010), before an orbiting vegetation lidar (light detection and ranging) sensor is flown. For Threats to Continuity of Terrestrial Observations example, intensive plot data can be used along with Long-term trends are critical for detecting changes inventory tree dimension data to produce algorithms in vegetation caused by management or by changes for estimating carbon stocks in vegetation and soils. in climate, atmospheric CO2, or nitrogen deposi- The plot data are scaled spatially by developing mod- tion. However, the future continuity of observations els relating field-measured response variables to plot from remote sensing and flux networks is threatened. attributes (e.g., those related to stand age, soil fertility, Records from ground flux networks are now 7 to 15 climate), which are then used with satellite land cover years long, and they are beginning to show trends in and disturbance data and other spatial data (e.g., mete- ecosystem responses to management and climate, but orology) to map carbon stocks (Blackard et al., 2008; there is a high risk that flux sites in the Canadian Car- Hudiburg et al., 2009). If there is a sufficient density bon Program and CarboAfrica will be closed beyond of inventory plots, the data can be used to calibrate 2010 because of budget shortfalls (Hank Margolis, the remote sensing data (Landsat with lidar or P-band Canadian Carbon Program coordinator, and Ric - radar) to estimate biomass. cardo Valentini, CarboAfrica coordinator, personal Other promising improvements may arise from communication, August 2009). In the United States, research that combines lidar, radar, and airborne multi- AmeriFlux sites are supported by the Department of spectral technologies (Treuhaft et al., 2004). Small (<5 Energy (DOE), National Oceanic and Atmospheric m) and large (>25 m) footprint lidar have been most Administration, USDA, NASA, and National Science widely used to estimate forest carbon (e.g., Drake et Foundation, with DOE supporting more than half of al., 2002; Lefsky et al., 2002) over landscapes, but these the current sites. Because these sites were established methods also require field observations to develop the remote sensing algorithms and to assess accuracy. The multisensor approach has been applied at the landscape 6 Neither mission has been scheduled for launch.

OCR for page 37
0 VERIFYING GREENHOUSE GAS EMISSIONS for research purposes, rather than for agency operations, is the Group on Earth Observations (GEO), which there is no guarantee that they will be maintained. is working to establish a Global Earth Observation Likewise, operational measurements of the land System of Systems to improve public access to obser- vations, datasets, tools, and expertise.8 The group’s surface from space are not within the mission of any U.S. agency, jeopardizing future continuity. Key obser- draft carbon strategy describes using inventories, eddy vations of land cover and land-use change have been covariance flux networks, atmospheric greenhouse made for more than 30 years by the Landsat series of gas observations, and ocean observations along with satellites, but the two satellites in orbit are deteriorating remote sensing observations of land cover and land-use and the Landsat Data Continuity Mission (LDCM) is change (Landsat, SPOT, IKONOS, MODIS, SAR) in not scheduled for launch until 2012. There are no firm a modeling framework to support climate agreements arising from UNFCCC conferences.9 Open data will plans to continue the series of observations after the nominal mission lifetime ends in 2018 (NRC, 2007).7 be required for this purpose. These and additional threats to carbon cycle observa- tions are detailed in Birdsey et al. (2009). RECOMMENDATIONS The following recommendations are aimed at inte- Access to Satellite and Inventory Observations grating remotely sensed land cover data with invento- Improved access to satellite remote sensing and ries and other data in a model-data framework. Landsat forest inventory observations is important for reducing is emphasized because global data are freely available uncertainties in carbon emissions associated with land- now and its 30-year record offers a historic baseline use change and for estimating emissions and removals from which to work for treaty purposes. Lidar and the from natural sources and sinks. Currently, different planned sensors discussed above are likely to remain in nations have different policies regarding access to mod- research mode for some time but may eventually be use- erate- and coarse-resolution satellite imagery. As noted ful for supplementing Landsat-type data for validating above, Landsat and MODIS imagery is free, publicly estimates of biomass at subregional to regional scales. available, and easily accessible. As a result, these data- sets have been widely used by the international scientific 1. Establish a long-term working group to produce community for both the development of UNFCCC publicly available global maps of land-use and land AFOLU inventories and independent land-use change cover change from Landsat and high-resolution satel- assessments. Both access and cost remain substan- lite imagery at least every 2 years. This will provide tial barriers to the widespread use of satellite data an independent check on the activities responsible for from other countries, such as France (SPOT), India the majority of AFOLU emissions in self-reported (India Remote Sensing Satellite), Japan (Advanced UNFCCC inventories. In particular, it will enable mea- Land Observing Satellite Phased Array type L-band surement of each nation’s level of deforestation, which Synthetic Aperture Radar), and China-Brazil (Earth are cumulatively responsible for approximately 12-22 Resources Satellite; see Achard et al., 2007; DeFries percent of global anthropogenic CO2 emissions. The et al., 2007; GOFC-GOLD, 2008). Moreover, data maps will also provide some independent constraints from a number of planned satellite missions (e.g., on estimates of the agricultural practices (e.g., rice cul- Japan’s Global Change Observation Mission Second tivation) responsible for a substantial fraction of CH4 Generation Global Imager, European Space Agency’s and N2O emissions. For countries with limited capac- BIOMASS mission) have great potential to reduce ity, the data provided under this initiative could also uncertainties in carbon emissions, but are expected to 8 As of September 2009, members of the Group on Earth Obser- carry restrictions. vations included 80 governments and the European Commission. An international avenue for discussing data access See . 9 See the open review version of the GEO carbon strategy at .

OCR for page 37
 MEASURING FLUXES FROM LAND-USE SOURCES AND SINKS lay the foundation for better self-reported inventories. 3. An interagency group, with broad participation Global maps of land-use and land cover change could from the research community, should undertake a com- be produced by the U.S. Geological Survey, which has prehensive review of existing information and design been disseminating land remote sensing imagery and a research program to improve and, where appropri- creating Landsat data products for several decades, ate, implement U.S. estimates of AFOLU emissions and/or by NASA or university scientists. of CO2, N2O, and CH4. Key elements are likely to 2. Increase the availability of moderate- and high- include continued research on the biogeochemical resolution satellite observations for mapping land cover cycles of these gases, supported by observations from change. This means that a successor to the LDCM eddy covariance towers, other flux measurements for should be added to the mission queue of NASA or N2O and CH4, and ecosystem inventories of all of another federal agency within the next year. The 30 m the major carbon pools and their trends in the United resolution of the Landsat instrument will have to be States. These observation systems will be necessary in a supplemented with 1 m imagery in a statistical sub- modeling framework (e.g., ecosystem biogeochemistry sampling of locations to detect and measure selective process modeling) to provide the accuracy needed for logging and to improve estimates of tree density. The annual, spatially explicit assessments within countries. high-resolution data could be obtained either by add- ing another instrument to the Landsat platform or by For recommendation 3, a realistic goal is to deploy acquiring commercial or national data (with the proviso the observing systems within 5 years, which would that they be made freely available). The current plan return data needed to reduce uncertainties in AFOLU to launch a single LDCM carries with it considerable CO2 fluxes in the United States to less than 10 percent risk. If the launch fails, it would be virtually impos- within the following 5 years. The improved estimates sible for the United States to monitor land-use change could be used to provide early warning of changes in using public domain information and may significantly the carbon cycle that could inform the design of a undermine the REDD component of a future global climate treaty, to facilitate improvements in models of c limate treaty by limiting the capability of tropical the carbon cycle, to provide data necessary to improve countries to produce realistic national inventories. The atmospheric methods for estimating emissions (see implementation issues associated with maintaining a Chapter 4), and to demonstrate new inventory methods U.S. capability for collecting moderate-resolution land that could become part of the UNFCCC reporting imaging data are discussed in FLIIWG (2007). process.

OCR for page 37