Changes in Atmospheric Constituents and in Radiative Forcing

. Reactive gases and aerosols are produced by terrestrial ecosystems, processed within plant canopies, and can then be emitted into the above-canopy atmosphere. Estimates of the above-canopy fluxes are needed for quantitative earth system studies and assessments of past, present and future air quality and climate. The Model of Emissions of Gases and Aerosols from Nature (MEGAN) is described and used to quantify net terrestrial biosphere emission of isoprene into the atmosphere. MEGAN is designed for both global and regional emission modeling and has global coverage with ~1 km2 spatial resolution. Field and laboratory investigations of the processes controlling isoprene emission are described and data available for model development and evaluation are summarized. The factors controlling isoprene emissions include biological, physical and chemical driving variables. MEGAN driving variables are derived from models and satellite and ground observations. Tropical broadleaf trees contribute almost half of the estimated global annual isoprene emission due to their relatively high emission factors and because they are often exposed to conditions that are conducive for isoprene emission. The remaining flux is primarily from shrubs which have a widespread distribution. The annual global isoprene emission estimated with MEGAN ranges from about 500 to 750 Tg isoprene (440 to 660 Tg carbon) depending on the driving variables which include temperature, solar radiation, Leaf Area Index, and plant functional type. The global annual isoprene emission estimated using the standard driving variables is ~600 Tg isoprene. Differences in driving variables result in emission estimates that differ by more than a factor of three for specific times and locations. It is difficult to evaluate isoprene emission estimates using the concentration distributions simulated using chemistry and transport models, due to the substantial uncertainties in other model components, but at least some global models produce reasonable results when using isoprene emission distributions similar to MEGAN estimates. In addition, comparison with isoprene emissions estimated from satellite formaldehyde observations indicates reasonable agreement. The sensitivity of isoprene emissions to earth system changes (e.g., climate and land-use) demonstrates the potential for large future changes in emissions. Using temperature distributions simulated by global climate models for year 2100, MEGAN estimates that isoprene emissions increase by more than a factor of two. This is considerably greater than previous estimates and additional observations are needed to evaluate and improve the methods used to predict future isoprene emissions.

/pdf/estimates-of-global-terrestrial-isoprene-emissions-using-2inhk36wkp.pdf

Estimates of global terrestrial isoprene emissions using MEGAN (Model of Emissions of Gases and Aerosols from Nature)

Image classification is a complex process that may be affected by many factors. This paper examines current practices, problems, and prospects of image classification. The emphasis is placed on the summarization of major advanced classification approaches and the techniques used for improving classification accuracy. In addition, some important issues affecting classification performance are discussed. This literature review suggests that designing a suitable image-processing procedure is a prerequisite for a successful classification of remotely sensed data into a thematic map. Effective use of multiple features of remotely sensed data and the selection of a suitable classification method are especially significant for improving classification accuracy. Non-parametric classifiers such as neural network, decision tree classifier, and knowledge-based classification have increasingly become important approaches for multisource data classification. Integration of remote sensing, geographical information systems (GIS), and expert system emerges as a new research frontier. More research, however, is needed to identify and reduce uncertainties in the image-processing chain to improve classification accuracy.

/pdf/a-survey-of-image-classification-methods-and-techniques-for-5bc5ltr9dq.pdf

A survey of image classification methods and techniques for improving classification performance

MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets

. New burned area datasets and top-down constraints from atmospheric concentration measurements of pyrogenic gases have decreased the large uncertainty in fire emissions estimates. However, significant gaps remain in our understanding of the contribution of deforestation, savanna, forest, agricultural waste, and peat fires to total global fire emissions. Here we used a revised version of the Carnegie-Ames-Stanford-Approach (CASA) biogeochemical model and improved satellite-derived estimates of area burned, fire activity, and plant productivity to calculate fire emissions for the 1997–2009 period on a 0.5° spatial resolution with a monthly time step. For November 2000 onwards, estimates were based on burned area, active fire detections, and plant productivity from the MODerate resolution Imaging Spectroradiometer (MODIS) sensor. For the partitioning we focused on the MODIS era. We used maps of burned area derived from the Tropical Rainfall Measuring Mission (TRMM) Visible and Infrared Scanner (VIRS) and Along-Track Scanning Radiometer (ATSR) active fire data prior to MODIS (1997–2000) and estimates of plant productivity derived from Advanced Very High Resolution Radiometer (AVHRR) observations during the same period. Average global fire carbon emissions according to this version 3 of the Global Fire Emissions Database (GFED3) were 2.0 Pg C year−1 with significant interannual variability during 1997–2001 (2.8 Pg C year−1 in 1998 and 1.6 Pg C year−1 in 2001). Globally, emissions during 2002–2007 were relatively constant (around 2.1 Pg C year−1) before declining in 2008 (1.7 Pg C year−1) and 2009 (1.5 Pg C year−1) partly due to lower deforestation fire emissions in South America and tropical Asia. On a regional basis, emissions were highly variable during 2002–2007 (e.g., boreal Asia, South America, and Indonesia), but these regional differences canceled out at a global level. During the MODIS era (2001–2009), most carbon emissions were from fires in grasslands and savannas (44%) with smaller contributions from tropical deforestation and degradation fires (20%), woodland fires (mostly confined to the tropics, 16%), forest fires (mostly in the extratropics, 15%), agricultural waste burning (3%), and tropical peat fires (3%). The contribution from agricultural waste fires was likely a lower bound because our approach for measuring burned area could not detect all of these relatively small fires. Total carbon emissions were on average 13% lower than in our previous (GFED2) work. For reduced trace gases such as CO and CH4, deforestation, degradation, and peat fires were more important contributors because of higher emissions of reduced trace gases per unit carbon combusted compared to savanna fires. Carbon emissions from tropical deforestation, degradation, and peatland fires were on average 0.5 Pg C year−1. The carbon emissions from these fires may not be balanced by regrowth following fire. Our results provide the first global assessment of the contribution of different sources to total global fire emissions for the past decade, and supply the community with an improved 13-year fire emissions time series.

/pdf/global-fire-emissions-and-the-contribution-of-deforestation-r24zw2ke8k.pdf

Global fire emissions and the contribution of deforestation, savanna, forest, agricultural, and peat fires (1997-2009)

Land cover influences numerous biogeophysical processes at the interface of the land surface and atmosphere. Increasingly, models are being used to better understand the global climate system. Such models require accurate characterization of land cover at global and regional scales. To provide more accurate maps of land cover, classification of coarse resolution remote sensing data is widely used. Although classification techniques have been commonly applied to remotely sensed data in the past, classification problems currently being addressed by the global remote sensing community pose new challenges. To meet these challenges, nonparametric machine learning algorithms are being used. However, these algorithms are less well understood for remote sensing applications than conventional approaches. 
The main objective of this dissertation is to improve understanding of nonparametric classification algorithms in remote sensing applications, and by extension, to improve global land cover mapping techniques. The issues examined address identifying and reducing uncertainty in coarse resolution land cover maps produced using nonparametric classification algorithms. The research draws heavily on a new algorithm known as boosting, which is used in conjunction with supervised classification algorithms. This dissertation provides four main results. First, pixel-scale uncertainty measures derived using boosting are reliable indicators of classification error, as demonstrated using several validation data sets. As a result, spatially explicit estimates of the likelihood of misclassification can be provided with classified maps. Second, a Bayesian approach for incorporating ancillary information with nonparametric classification algorithms improves classification accuracy. This approach reduces the sensitivity of classification predictions to the class frequency distribution of the training sample. Third, estimates of sub-pixel cover from several machine learning algorithms are more accurate than those produced using linear spectral mixture analysis because the nonparametric algorithms are better able to accommodate within-class spectral variability. Finally, comparison of four currently available global land cover products, including one produced using techniques developed in this dissertation, against validation data from the United States Forest Service for California shows systematic overprediction of forest cover. Future global mapping efforts will benefit from identifying problems with accurate, high resolution validation data.

Adapting machine learning methods for coarse resolution land cover classification

This paper describes research to obtain local or pixel scale estimates of classification quality from classification of remote sensing data produced with nonparametric machine learning algorithms. The approach utilizes boosting, a method of improving classification accuracy using multiple iterations of a classification algorithm. Boosting has been shown to successfully improve classification accuracy, to be resistant to overfitting, and to be effective with remote sensing data. A recent statistical examination of boosting has shown that this algorithm is a form of additive logistic regression. As such, boosting can provide probabilities of class membership. These probabilities of class membership are evaluated as a measure of local classification quality. This approach is shown to be an effective predictor of classification errors and therefore, provides a means of assessing local classification quality.

Local estimation of land cover classification quality using machine learning methods

Global multitemporal composited AVHRR 1-km/sup 2/ spatial resolution data sets are being extensively used to prototype the MODIS Land Cover Classification algorithm. Results indicate that this method of prototyping prior to instrument launch is valid and useful even though the absolute results are significantly limited by the inherent data incompatibilities described in this paper.

D.K. McIver

Papers

Adapting machine learning methods for coarse resolution land cover classification

Local estimation of land cover classification quality using machine learning methods

Data simulation impacts on MODIS Land Cover Classification development