Review of the use of remote sensing for biomass estimation to support renewable energy generation

Lalit Kumar; Priyakant Sinha; Subhashni Taylor; Abdullah F. Alqurashi

doi:10.1117/1.JRS.9.097696

16 June 2015 Review of the use of remote sensing for biomass estimation to support renewable energy generation

Lalit Kumar; Priyakant Sinha; Subhashni Taylor; Abdullah F. Alqurashi;

Author Affiliations +

J. of Applied Remote Sensing, 9(1), 097696 (2015). doi:10.1117/1.JRS.9.097696

Abstract

The quantification, mapping and monitoring of biomass are now central issues due to the importance of biomass as a renewable energy source in many countries of the world. The estimation of biomass is a challenging task, especially in areas with complex stands and varying environmental conditions, and requires accurate and consistent measurement methods. To efficiently and effectively use biomass as a renewable energy source, it is important to have detailed knowledge of its distribution, abundance, and quality. Remote sensing offers the technology to enable rapid assessment of biomass over large areas relatively quickly and at a low cost. This paper provides a comprehensive review of biomass assessment techniques using remote sensing in different environments and using different sensing techniques. It covers forests, savannah, and grasslands/rangelands, and for each of these environments, reviews key work that has been undertaken and compares the techniques that have been the most successful.

Kumar, Sinha, Taylor, and Alqurashi: Review of the use of remote sensing for biomass estimation to support renewable energy generation

1. Introduction

Lignocellulosic biomass or plant dry matter (biomass) is a highly abundant renewable energy resource that can be used to generate a continuous supply of heat and electricity as well as solid, liquid, and gaseous fuels.¹ Therefore, plant biomass plays an important role in the global quest for sustainable energy solutions since it is a renewable energy source that is easily available to humans. Although it is considered that all fossil fuels such as coal and oil originated from buried living material, they are usually excluded from the definition of biomass. Biomass has stored energy through the process of photosynthesis. It exists in one form as plants and may be transferred through the food chain to animal bodies and their wastes, all of which can be converted to energy through processes such as combustion. Biomass has been converted by partial pyrolysis to charcoal for thousands of years. Charcoal, in turn, has been used for forging metals and for light industry for hundreds of years. Both wood and charcoal formed part of the backbone of the early industrial revolution prior to the discovery of coal for energy. Wood is still used extensively for energy in both household situations and in industry, particularly in the timber, paper, pulp, and other forestry-related industries. The easiest and most efficient way to use biomass as energy is through burning. When it is burned, a part of the internal chemical energy converts to heat. Biomass can also be burned in special plants called waste-to-energy plants which use the heat energy to create steam, which is then used to either heat buildings or create electricity.

The main benefit of biomass is that it is a renewable fuel. Not only does this give us a renewable source of energy to heat our homes, power our vehicles, and produce electricity, but it also helps us to utilize discarded waste that is filling up large dump sites. Many Asian countries are looking to biomass power plants to increase domestic energy outputs and reduce reliance on foreign energy supplies. Asia is expected to construct about 1000 MW of biomass energy capacity annually by 2020—twice as much as is expected in Europe.² Thailand, Indonesia, Malaysia, and the Philippines all have introduced feed-in tariffs to encourage biomass energy production.³ Also, due to Asian climates, many countries can produce sufficient amounts of biomass.

Many countries of the world are now expanding resources toward quantifying, mapping and monitoring biomass due to its importance as a renewable energy source. However, biomass resources are distributed over wide geographical areas and their biochemical properties are highly variable over space. Furthermore, its suitability as a renewable resource is also site-specific. This makes biomass estimation a challenging task, especially in areas with complex forest stand structures and environmental conditions,³ and requires accurate and consistent measurement methods.⁴ Traditionally, two methods are available for the determination of biomass.⁵ The first method is destructive sampling, which involves the complete harvesting of plots and subsequent extrapolation to a unit area of hectare.⁶ The second method is based on allometry where allometric equations are used to extrapolate both in situ and remotely sampled data to a larger area to derive biomass and canopy volume from an easily measured attribute such as diameter at breast height (DBH), tree height, etc. Allometric relationships are used for estimating tree allometry which establishes quantitative relations between some key tree characteristic such as dimensions of trees (easy to measure) and other properties (which are difficult to assess). Both these traditional methods are accurate but are extremely time-consuming, costly, and generally limited to small areas and small tree sample sizes.⁷8.^–⁹ Moreover, extending this method to map forest biomass across a large area is extremely challenging when factors such as ecological differences, variations in inventory systems, and scattered sources of biomass data are considered. In addition, since the allometric coefficients are site and species specific and are based on a certain range of tree diameters, the use of standard allometric equations can lead to significant errors in vegetation biomass estimations if used outside the area where they were originally produced.¹⁰ There have been efforts in developing generalized regional and national tree biomass equations that could be applied to a larger geographic footprint than most existing allometric equations.¹¹^,¹²

Another vegetation type of great interest is the tropical savanna, not only for the large regions it covers but also for the high interannual biomass dynamics. Grasslands and rangelands also have considerable biomass and thus energy generation capacities, especially since they cover around 40% of the earth’s land surface. Remote sensing can be used to ascertain the potential availability of biomass over large regions and also to estimate biomass energy potential for different land-cover classes.¹³ However, the actual recovery of this biomass will depend on the availability of technology to collect and utilize this material in an economical fashion.¹⁴ Remote sensing techniques can be used in combination with geographical information systems (GIS) to evaluate the feasibility of such initiatives. These techniques can be used to evaluate the feasibility of and optimization of the locations of new biomass power plants¹³ to evaluate the cost effectiveness of energy production from biomass¹ and to devise a framework for estimating residual biomass using satellite imagery and forest inventory data.¹⁵

Additionally, remote sensing is the best approach to estimate biomass at a regional level where field data are scarce or difficult to collect. Almost two decades have passed since pioneers such as Refs. 16 and 17 related biomass to reflectance recorded at the sensor. Since then, many studies in different regions have found strong correlations between biomass and reflectance at different wavelengths. In this paper, we review various techniques and platforms for biomass estimation. We look at forests, savanna, and grasslands/rangelands separately as each has its own characteristics and problems when it comes to biomass estimation. There have been several review papers on biomass estimation in the past few years; however, most of them have described remote sensing based estimation for forest biomass.³^,¹⁸19.^–²⁰ This current review incorporates remote sensing-based biomass estimation for three major vegetation ecosystems: forest, grassland and rangelands, and tropical savanna, that cover $\sim 80 %$ of earth’s vegetative cover.²¹^,²² These vegetative surfaces on earth are more “natural” ecosystems without much human disturbance, unlike agricultural lands which are heavily dependent on cropping management, and thus provide an opportunity to the reader to assess the challenges and differences in remote sensing-based biomass estimations for these natural ecosystems.

2. Remote Sensing

One of the recent advances in biomass estimation approaches is the incorporation of inferences derived from remote sensing. Remotely sensed data have the provision of a synoptic view of the surface area of interest, thereby capturing the spatial variability in attributes of interest like tree height, crown closure, etc. The spatial coverage of large area biomass estimates that are constrained by the limited spatial extent of forest inventories may be expanded through the use of remotely sensed data. Biomass and carbon stock estimates derived from forest inventory data usually have some spatial, attributional, and temporal gaps. Remotely sensed data can be used to fill these gaps, thereby leading to estimates closer to the actual value. Remote sensing data are available at different scales, from local to global, from various sources including optical or microwave, and hence are expected to provide information which can be related directly, and in different ways, to biomass information.²³^,²⁴ Although remote sensing technology cannot effectively be used for underground biomass, it has the ability to provide important information for aboveground biomass (AGB).³^,²⁵ A large range of studies has been conducted for biomass estimation from remote sensing data.²⁴^,²⁶27.28.29.30.^–³¹ The advantages of remote sensing include the ability to obtain measurements from every location in the forest, the speed with which remotely sensed data can be collected and processed, the relatively low cost of many remote sensing data types, and the ability to collect data easily in areas which are difficult to access on the ground.³² There are many sensors available with different characteristics of spectral, spatial, and temporal resolutions used for biomass estimation based on availability, efficiency and cost. Optical remote sensing, radar and light detection and ranging (LiDAR) sensors provide the three main sources of remotely sensed data for biomass estimation.

2.1.

Optical Remote Sensing

Due to its coverage, repetitiveness and cost-effectiveness, optical remote sensing provides a potential alternative to tedious hand sampling as a means of estimating biomass over large areas.³³^,³⁴ Optical remote sensing data can be acquired at a variety of spatial and temporal resolutions. High-spatial resolution data from sensors such as Quickbird, WorldView, GeoEye, IKONOS, and DigitalGlobe as well as aerial photographs come in spatial resolutions ranging from submeters to $< 5 m$ in both multispectral and panchromatic images. Images at high resolution offer a fundamental shift in vegetation assessment capability where a multispectral pixel can image a single tree crown, unlike sensors with medium resolution such as Landsat or Systeme Probatoire D’Observation De La Terre (SPOT) where a single pixel can encompass many tree crowns or significant noncrown features.³⁵^,³⁶ Satellite data covering 10 to 100 m of ground in 1 pixel are termed as medium-spatial resolution data and Landsat time series and SPOT sensors have been the two primary sources of medium-resolution data. Coarse-resolution data ( $> 100 m$ ) [e.g., MODIS, national oceanic and atmospheric administration (NOAA), advanced very high resolution radiometry (AVHRR), SPOT vegetation] can be useful for biomass estimation at regional to continental scales since their high temporal frequency increases the probability of acquiring cloud-free data for generating consistent datasets over large areas. AVHRR data have been the most widely used datasets for studies of vegetation dynamics on a continental scale. However, the MODIS sensor has improved spectral and spatial resolutions compared to the widely used AVHRR and provides a suite of biophysical products that are useful in biomass estimation, including vegetation indices, leaf area index (LAI), fraction of absorbed photosynthetically active radiation (FAPAR), gross primary production, net photosynthesis, and net primary productivity (NPP).³⁷^,³⁸ The mid-infrared (MIR) reflectance from optical remote sensing data is closely related to biomass and thus was found to be more useful in assessing alterations in vegetation characteristics compared to reflectance in visible (VIS) and near-infrared (NIR) bands.³⁹ Hyperspectral remote sensing is an another important source of optical satellite data for biomass estimation. Unlike multispectral satellite sensors, hyperspectral remote sensing allows the acquisition of many, very narrow, contiguous spectral bands throughout the VIS, NIR, MIR, and thermal infrared portions of the electromagnetic spectrum.⁴⁰ This ability to collect reflectance in many narrow bands makes hyperspectral remote sensing particularly useful for extracting vegetation parameters, such as LAI, chlorophyll content, and leaf nutrient concentration.⁴¹ Optical sensors collect data from only the aboveground vegetation and have been used mainly for aboveground biomass assessment.

A range of techniques are used with optical remote sensing data to estimate biomass.⁴² A commonly used technique involves the use of vegetation indices such as ratio vegetation index (RVI), normalized difference vegetation index (NDVI) and soil adjusted vegetation index (SAVI).⁴³ Alternatively, remote sensing data can be used to obtain indirect estimates of absorbed photosynthetically active radiation (APAR) from the red and infrared reflectance characteristics of the vegetation.⁴⁴ The APAR gives an indication of how efficiently absorbed energy is converted into dry biomass by a vegetation type.⁴⁵ Another technique involves the use of process-based models which estimate biomass production from remote sensing data by combining canopy functioning process-based models with physical radiative transfer models.⁴⁶^,⁴⁷

2.2.

Radar

Over recent years, there has been increasing interest in synthetic aperture radar (SAR) data for aboveground biomass analyses, particularly in the areas of frequent cloud conditions where obtaining high quality optical data is difficult. The capability of radar systems to collect data in all weather and light conditions overcomes this issue. Furthermore, the SAR sensor can penetrate vegetation to different degrees and provides information on the amount and three-dimensional (3-D) distribution of structures within the vegetation.⁴⁸ Airborne SAR has been operating for many years, but since the 2000s, space-borne SAR sensors such as TerraSAR-X, Advanced Land Observing Satellite (ALOS) and Phased Array L-band SAR (PALSAR) have become available.⁴⁹ Many studies based on SAR have focused on the development of algorithms for classification and biomass estimation in closed-canopy forests.⁴⁸^,⁵⁰ A commonly used approach to biomass retrieval from SAR has been to establish empirical relationships between field-based estimates and single channel data.⁴⁸

The SAR sensor can detect the horizontal (H) or the vertical (V) components of the backscattered radiation. Hence, there are four possible polarization configurations for an SAR system: horizontal transmit and horizontal receive (HH), vertical transmit and vertical receive (VV), horizontal transmit and vertical receive (HV), and vertical transmit and horizontal receive, depending on the polarization states of the transmitted and received radar signals. The SAR on the ERS satellite is VV polarized while the RADARSAT satellite is HH polarized. Radar backscatters (P and L bands) have been found to be positively correlated with major forest parameters, such as tree age, tree height, DBH, basal area, and total aboveground dry biomass.²⁸^,⁵¹52.53.^–⁵⁴ A detailed review on the use of radar data for biomass estimation can be found in the literature.⁵⁵^,⁵⁶ Various studies have utilized radar data in biomass analyses of a range of biomes.⁵³^,⁵⁴^,⁵⁷^,⁵⁸

There are a number of advantages to radar remote sensing compared to optical remote sensing in terms of its utility in biomass estimation in savannas. The ability of radar to penetrate cloud and haze makes it especially useful in the tropics. Furthermore, radar based sensors are active and have a controlled power outlet, which ensures consistent transmit and return rates. Thus, radar sensors can function independently of solar radiation variations, unlike optical sensors where spectral reflectance measurements are affected by variations in solar radiation.⁵⁹ On the other hand, radar use has limited applications in regional studies due to the small swath width, high costs of airborne acquisitions, lower sampling density of the large footprint waveform, and the limited extent of coverage.⁴⁸

2.3.

LiDAR

The two-dimensional (2-D) nature of optical remote sensing data limits its use in direct quantification of some vegetation characteristics like tree height, canopy height, volume, etc. LiDAR is a relatively new and sophisticated technology that helps to overcome this limitation due to its ability to extend the spatial analysis to a third dimension. LiDAR instruments have the ability to sample the vertical distribution of canopy and ground surfaces,⁶⁰^,⁶¹ and several studies have established a strong correlation between LiDAR metrics and aboveground biomass, thus allowing estimation of biomass in forested environments.⁶²63.^–⁶⁴ LiDAR technology has seen considerable advancement with the advent of full waveform digitizing sensors,⁶⁵ which has allowed this tool to be increasingly used in the study of forest structures in a variety of forest environments.⁶⁶67.^–⁶⁸ It has become the most efficient technology for structural assessment since it captures landscape structural data that are suitable for volume and biomass estimation.⁶⁹ Biomass can be estimated at the individual tree level with allometric equations using LiDAR data of sufficient post spacing (e.g., $> 1 return / m^{2}$ ).⁴⁸ A detailed review of LiDAR data application in forestry can be found in Lim et al.⁷⁰

The 3-D LiDAR points represent latitude, longitude, and ellipsoidal height based on the WGS84 reference ellipsoid. Ellipsoidal heights are converted to elevations. There are currently two types of LiDAR in operation: (1) discrete return LiDAR (small footprint) and (2) full waveform LiDAR (large footprint).⁷¹ Both are generally calibrated to operate in the 900- to 1064-nm wavelengths where vegetation reflectance is highest.⁶⁸ A combination of either small or large footprint LiDAR systems along with GPS and accurate time referencing allow the extraction of position in 3-D of the reflecting surface.⁶⁸ Discrete return airborne LiDAR systems are more suitable for fine-scale biomass mapping, while waveform space-borne LiDAR, e.g., The Geoscience Laser Altimeter System (GLAS) on board Ice, Cloud, and Land Elevation Satellite (ICESat) has the potential for broad-scale biomass mapping.⁷²^,⁷³

Although LiDAR data have some advantages over optical data, there are a few issues that restrict its use for field applications. For example, LiDAR data analyses are not simple and require more image processing knowledge and skill and specific software. The LiDAR data acquisition process is expensive and covers smaller areas, hence study areas are still limited to specific areas and have not been applied extensively to larger areas for biomass estimation.

3. Biomass Estimation in Forests

The remote sensing methods, data types, and some examples for forest biomass estimation are shown in Table 1.

Table 1

Summary of the remote sensing methods, data types, and some examples for forest biomass estimation.

Category	Methods	Data used	Characteristics	Examples
Remote sensing-based methods	Methods based on fine spatial resolution data (	Aerial photographs, IKONOS, Quick Bird, GeoEye, WorldView	Per-pixel level	Refs.
	Methods based on medium-spatial resolution data (10–100 m) (linear, exponential and multiple regression analysis, neural network, k-nearest neighbor method, productivity model)	Landsat 4 5 7	Per-pixel level	Refs.
	Methods based on coarse-spatial resolution data (	IRS-1C WiFS, AVHRR, MODIS, SPOT vegetation	Per-pixel level	Refs.
	Methods based on radar data (regression models, canopy height model, multiplicative models)	SIR-C, SAR-L JERS-1 SAR-L, AeS-1 SAR-P, InSAR, airborne laser, large and small footprint LiDAR	Per-pixel level	Refs.
	Method based on image fusion techniques (intensity hue and saturation (HIS), Brovey, PCA	Multispectral and PAN		Refs.
	Vegetation index-based method (NDVI, ratio)			Refs.
	Object based (segmentation and classification, ANNs, k-nearest neighbor, statistical models; random forest)		Object-level	Refs.
	Advanced classifier spectral mixture analysis (SVM), random forest, support vector machine (SVM)	Multispectral	Per-pixel level	Refs.

3.1.

Use of Optical Remote Sensing

Optical remote sensing data, with a variety of spatial and temporal resolutions, have been widely used for forest biomass estimation using different types of image processing techniques.⁴^,⁷^,²⁴^,²⁹^,³⁰^,⁸⁴^,⁸⁷^,¹¹⁷118.119.120.^–¹²¹ For biomass estimation from optical data, the commonly used approaches are multiple regression analysis, $k$ -nearest neighbor, and neural network.²⁴^,²⁹^,³⁰^,¹²²^,¹²³ Optical data can be used to carry out spatial stratification of vegetation from which estimates of biomass distribution can be generated. For indirect biomass estimation, remote sensing data are used to determine tree canopy parameters, such as crown diameter using multiple regression analysis or canopy reflectance models.¹²⁴^,¹²⁵ Different types of vegetation indices and band ratios derived from optical data are also used to extract biomass by correlating vegetation index values or band ratio values with field estimations.⁸⁷

The ready availability of high-resolution data from a range of sensors has permitted the modeling of tree parameters or forest canopy structures. For example, Song et al.³⁶ estimated tree crown size from IKONOS and Quickbird images and concluded that this approach could provide estimates of average tree crown size for hardwood stands. Greenberg et al.⁷⁷ have effectively used IKONOS data (spatial resolution 4 m) in estimating crown projected area, DBH and stem density. There are numerous methods applied for the extraction of biophysical parameters using high-spatial resolution data.¹²⁶ Large scale photographs and photomensuration methods have been used to measure various forest characteristics, such as tree height, crown diameter, crown closure, and stand area.⁷⁵^,¹²⁷ De Jong et al.⁷⁶ used digital airborne data to estimate biomass in southern France using linear regression analysis. In another study, Thenkabail et al.⁴ used IKONOS data to estimate biomass of oil palm plantations in Africa. Although high-spatial resolution and associated multispectral characteristics may become an important data source for forest biomass estimation and have attained great success, the shadows and intracrown spectral variance and the low spectral separability between tree crowns and other vegetated surfaces in the understory¹²⁸129.^–¹³⁰ create difficulty in developing biomass estimation models. High-resolution data need large data storage and processing time and are much more expensive to cover a given area. These factors influence the application of high-spatial resolution images for biomass estimation over broad areas. The absence of shortwave-infrared images, an important parameter for biomass estimation, also limits its application in biomass assessment. The problem is greater when traditional pixel-based spectral classifiers are used for vegetation classification. However, the incorporation of contextual information and object-based methods into the classification process has overcome this problem to an extent.¹⁰⁹^,¹¹¹ Object-based methods consider both spectral and context information during the classification process by segmenting the image into meaningful objects.¹¹⁰^,¹¹² The size of the image objects is determined by a scale parameter.¹³¹ The selections of segmentation parameters are subjective and determined through a combination of trial and error steps. Statistics on spectral bands (mean, standard deviation, etc.) along with other contextual information, such as geometric features (area, length, compactness, shape, etc.), and texture features-gray-level co-occurrence matrix (GLCM) (homogeneity, contrast, entropy, dissimilarity, correlation, etc.), and gray-level difference vector (entropy, contrast, etc.) of spectral bands are used to statistically derive features for each object that best separate the vegetation classes. Numerous studies have extracted GLCM textures from remote sensing images.¹¹¹^,¹¹³^,¹³² In Rondônia State, Brazil, Lu and Batistella¹¹³ used the GLCM texture (mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment, and correlation) with different moving window sizes and Landsat thematic mapper (TM) spectral bands 2 to 5 and 7 to examine the relationships between biomass and textural images for secondary and mature forest. They found a stronger relationship between textural images and biomass for mature forest with complex stand structure than original spectral bands. However, for secondary forest with a simple stand structure, biomass was closely related to spectral bands.

Medium-spatial resolution data have also been widely used in forest biomass estimation. For example, Lefsky et al.⁸⁰ estimated stand tree structure attributes such as basal area, biomass and DBH using remote sensing data. Linear or nonlinear regression models, $k$ -nearest neighbor, neural network, and vegetation canopy models are the main methods applied in this case. In a Bornean tropical rain forest, Foody et al.⁸² used neural networks for biomass estimation using Landsat TM. Ghasemi et al.¹³³ used SPOT 5 data to estimate aboveground forest biomass from canopy reflectance model inversion in the mountainous terrain of Kananaskis, Alberta. Landsat TM data were used to estimate tree volume and biomass using the $k$ -nearest neighbor estimation method.⁷⁸^,⁷⁹^,⁸¹ The task of estimating biomass from optical data for humid tropical forests is challenging because of its complex multilayered closed canopy structure combined with high levels of biomass.³^,²⁴^,²⁹^,⁸²^,¹¹⁸^,¹²³^,¹³⁴ In such cases, spectral reflectance and vegetation indices were found not to be reliable indicators of biomass²⁴ and were not sensitive to biomass change.²⁹ However, with the inclusion of some other factors, a few studies have shown positive results in estimating tropical forest biomass. For example, Nelson et al.¹²³ included the age of the forest into Landsat TM image analysis to estimate tropical forest biomass, while with the use of texture information into the image analysis process, Lu¹¹⁸ and Sarker and Nichol¹³⁵ improved biomass estimation results in tropical forests. Lu¹¹⁸ concluded image texture features to be more important than spectral reflectance for biomass estimation for forests with more complex stand structure. However, it is critical to identify suitable image textures that are strongly correlated with biomass but are weakly correlated with each other and this requires a great deal of effort.¹³⁶ In addition, image textures vary with the landscape and images used, therefore, not all texture measures can effectively extract biomass information and guidelines on how to select an appropriate texture needs more research. Several vegetation indices have been developed, mostly from VIS and infrared bands and applied to biomass estimation and biophysical parameter studies.¹⁰⁵^,¹⁰⁶ Vegetation indices have been found useful in minimizing spectral variability caused by canopy geometry, soil background, sun view angles, and atmospheric conditions when measuring biophysical properties.¹⁰⁷^,¹⁰⁸ Although not all vegetation indices were found to be directly correlated with biomass,²⁴ by minimizing the impact of environmental conditions and shadow on spectral reflectance, there was improved correlation between biomass and vegetation indices, especially in complex vegetation stand structures.¹⁰⁵ Therefore, a combination of image textures and spectral responses can be considered useful in determining forest stand parameters and to establish more accurate biomass estimation models.¹¹⁸ In addition to pixel-based spectral responses and textural images, subpixel-based variables such as green vegetation, shade, and soil can also be used as input variables for biomass estimation.²⁰^,¹³⁷ Spectral mixture analysis (SMA) has been found useful in developing these fractional images from multispectral images such as Landsat TM.¹¹³^,¹¹⁵ Lu and Batistella¹¹³ used SMA to extract fractional images from a Landsat TM image to examine the relationship between biomass and the subpixel variables for secondary and mature forests in Rondônia State. They found fractional images to be more useful for biomass estimation as compared to individual spectral bands. A detailed description of the SMA approach and its applications can be found in the literature.¹¹⁴115.^–¹¹⁶

Coarse-spatial resolution AVHRR NDVI data have been used to estimate biomass in Africa⁸⁶ and boreal and temperate forest woody biomass in Canada, Finland, Norway, Russia, Sweden, and the USA.⁸⁷ The advantages of a large number of spectral bands of MODIS data and their availability have improved biomass estimation accuracy at the continental or global scale. Recent studies have achieved promising results using tree-based models and metrics derived from MODIS data, in combination with radar data and ancillary information (climate, topography, and vegetation maps), to map the biomass distribution for the Amazon basin,⁸⁹ the United States,¹³⁸ and tropical Africa.⁸⁵ Baccini et al.⁸⁴ used MODIS data in combination with precipitation, temperature, and elevation for mapping biomass in national forest lands in California, USA. Overall, the application of forest biomass estimation using coarse-spatial resolution data is limited due to the occurrence of mixed pixels, saturation of spectral data at high biomass density and by the mismatch between the size of field plots and pixel size. A few studies have used coarse-resolution data along with medium-resolution data in combination with different modeling approaches to get more accurate biomass estimates for large areas. For example, Hame et al.⁸⁸ used Landsat TM and AVHRR data to estimate coniferous forest biomass. In another study, Tomppo et al.⁸¹ used TM as an intermediate step between field data and IRS-1C wide field sensors data to estimate tree stem volume and biomass in Finland and Sweden.

Overall, optical sensor data are found suitable for extracting horizontal vegetation structures such as vegetation types and canopy cover; however, the 2-D data have limitations in estimating vertical vegetation structures such as canopy height, which is one of the critical parameters for biomass estimation. Recently, optical data such as ALOS, panchromatic remote-sensing instrument for stereo mapping (PRISM), IKONOS stereo satellite images, and SPOT provide a stereo viewing capability that can be used to develop vegetation canopy height, thus can improve biomass estimation performance.¹³⁹^,¹⁴⁰ For example, St‐Onge et al.¹³⁹ assessed the accuracy of the forest height and biomass estimates derived from an IKONOS stereo pair and an LiDAR digital terrain model. Reinartz et al.¹⁴¹ used SPOT 5 HRS for forest height estimations in Bavaria and Spain, while Wallerman et al.¹⁴² investigated 3-D information derived from SPOT 5 stereo imagery to map forest variables such as tree height, stem diameter and volume. These studies show that high-resolution stereo data can be used as a valuable alternative to derive vegetation height information; however, more studies are needed to support this.

3.2.

Use of Radar

Studies that utilized radar data in forest biomass estimations found SAR L-band data to be more useful⁵³ than SAR C-band data.⁹⁰ Beaudoin et al.¹⁴³ found that VV and HV radar backscatter at high frequencies (C-bands and X-bands) were linked to crown biomass while radar backscatter HH at lower frequencies (P-bands and L-bands) were related to both trunk and crown biomass. Harrell et al.¹⁴⁴ used SIR C- and L-band multipolarization radar data for pine forest biomass estimation in the southeastern USA and found L-band HH data to be critical in biomass estimation. They noted that the inclusion of C-band HV or HH significantly improved biomass estimation performance. For biomass estimation of regenerating forests, Kuplich et al.⁹¹ found JERS-1/SAR data to be useful when forests are regenerating after block logging and not after selective logging. For mountainous area forest biomass estimation, multipolarization L-band SAR data were found to be useful.⁵³ Santos et al.⁹² found that JERS-1/SAR double bounce scattering and forest structural-physiognomic characteristics are the two important factors for biomass estimation of forest and savanna. For biomass estimation, most of the previous studies used the radar system from JERS-1, ERS-1/2 of single polarization, single incident angle, and low resolution SAR sensor. However, with the establishment of PALSAR and RADARSAT-2 (C-band), data are now available in different polarizations, different resolutions, and varying incident angles, which offer more opportunities to the scientific community to re-examine the potential of SAR data in forest biomass estimation. PALSAR data results have shown its ability to map forest in the Amazon and Siberia; however, the retrieval of forest biomass is still typically limited to values less than $50 t {ha}^{- 1}$ , which excludes most temperate and tropical forests.¹⁴⁵ Sarker et al.⁵⁷ investigated the capability of RADARSAT-2 fine-beam dual-polarization (C-HV and C-HH) data for forest biomass estimation in complex subtropical forest and found encouraging results. Radar data saturation problem is greater in complex forest stand structure when backscattering values are used for biomass estimation.¹⁴⁶^,¹⁴⁷ Interferometry SAR (InSAR) has been found useful in reducing this problem by increasing the saturation range to a certain degree by coherently collecting data over a short time increment with two identical instruments.⁹³^,⁹⁴^,¹³³ This improves the height-based biomass and volume estimation when the L-band saturation point increases to $200 t {ha}^{- 1}$ .⁷³ Balzter⁹³ reviewed InSAR for forest mapping and monitoring covering tree volume and biomass, forest types and land cover, fire scars, forest thermal state, and forest canopy height. The high correlation between vegetation canopy height and biomass of InSAR makes it a promising tool for broad-scale biomass estimation, especially for tropical and subtropical regions where frequent cloud cover is a problem.⁹⁴^,⁹⁵ However, other weather conditions, such as wind speed, moisture, and temperature, affect the InSAR estimation accuracy.¹⁴⁸ Recently, the polarimetric SAR interferometry (Pol-InSAR), a combined polarization and interferometry, has been found useful in estimating forest height using coherence information¹⁴⁹ and then correlating it to biomass.¹⁵⁰

3.3.

Use of LiDAR

The structural forest measurements from LiDAR data permit the accurate estimation of height, crown size, basal area, stem volume, LAI, NPP, and aboveground biomass, even in high biomass forests, a difficult task with passive sensors.⁶⁶ Biomass mapping from airborne discrete return LiDAR is based on two approaches: (1) area-based and (2) individual tree-based methods.⁷² Area-based methods develop statistical models to relate biomass with metrics derived from a LiDAR point cloud at the plot or stand level and apply the models over the whole study area.¹⁵¹96.^–⁹⁷ The development of statistical models requires field data for calibration and validation. The most widely used area-based LiDAR metrics for biomass prediction are various height metrics⁷⁰^,¹⁵²^,¹⁵³ calculated based on first, last, or all returns. Height metrics can also be calculated from grids of the canopy height model.⁹⁶^,¹³⁹^,¹⁵² Individual tree-based methods identify individual tree crowns and extract individual tree information from LiDAR point cloud, such as tree height and crown size, which can be related to biomass and other canopy structure variables through allometric equations. ¹⁵⁴155.^–¹⁵⁶ In this case, the amount of fieldwork required is much smaller than that for area-based methods because field data are needed only for a sample tree and not for sample plots or stands. Discrete return systems have been used to estimate biomass at the individual tree level up to the stand level.¹⁵⁴^,¹⁵⁵^,¹⁵⁷^,¹⁵⁸ The DEMs generated from airborne LiDAR data are very accurate and widely used in forest mapping and tree parameter estimations. It captures elevation information from the forest canopy as well as the ground beneath and can be used to assess the complex 3-D patterns of canopy and forest stand structure such as tree density, stand height, basal area, LAI, and forest biomass and volume.⁶⁸^,¹⁵⁹ In densely vegetated areas when passive sensors saturate at high biomass levels (higher than $100 mg {ha}^{- 1}$ ),¹⁶⁰ LiDAR has been found to accurately estimate LAI and biomass in such high biomass ecosystems.⁶⁸ In British Colombia, Canada, Loos et al.¹⁶¹ identified understory canopies between the dominant canopies of Douglas-Fir and Western Hemlock tree species by creating bare earth DEM and DSMs (digital surface models). The estimation of biomass is generally based on regression equations relating vegetation biomass to LiDAR derived variables. Studies are being conducted using LiDAR to determine the most appropriate laser-based predictors in regression models for estimation of forest structural variables. For example, García et al.¹⁶² have explored several biomass estimation models based on LiDAR height or intensity, separately, or height-intensity combined. They found height-related