Keywords

1 Introduction

1.1 Why Do We Need Maps of Deforestation and Forest Degradation?

Although humans have been modifying land to obtain food and other essentials for thousands of years, the concept of land use competition in its current understanding emerged in land use and land cover change (LUCC) research only relatively recently with the perceived looming scarcity of land due to booming populations and increasing demand for resources. In the 1960s, land use competition centered on urban frontiers to explore the competition of land resources between agriculture use and urban/industrial expansion (Wibberley 1959; Best 1968; Snyder 1966). Since 1970s, with increasing concerns on global climate change and the realization of terrestrial ecosystems as important sources and sinks of carbon (Sagan et al. 1979; Woodwell et al. 1983; Stone et al. 1983; Houghton et al. 1985; Woodwell et al. 1986), LUCC studies began to focus on changes of forest and woodland—the major formation of terrestrial ecosystems. Since early 1990s, theoretical and methodological guidelines of calculating carbon emission from change of forest sinks are provided by international conventions to monitor the impact of human activities on global climate change. Land Use, Land-Use Change and Forestry (LULUCF) sectors within the United Nations Framework Convention on Climate Change (UNFCCC), and further stipulated under the Kyoto Protocol in 1997 (UNFCCC 2014b), regulate the definition and accounting for carbon from certain activities (UNFCCC 2001).

With global forest lost at an alarming rate of 13 million hectares per year from 1990 to 2005 (FAO 2005), deforestation and degradation have become an important topic. In 2005, a special agenda focus on reducing carbon emissions from deforestation and degradation in developing countries was introduced into the Conference of the Parties (COP) agenda, entitled Reducing Emissions from Deforestation and Forest Degradation (REDD) (UNFCCC 2006), which was then adopted in the Warsaw Framework in 2013, as reducing emissions from deforestation and forest degradation and the role of conservation, sustainable management of forests, and enhancement of forest carbon stocks in developing countries (REDD-plus) (UNFCCC 2014a). Meanwhile, world deforestation still continues at an alarmingly high rate according to Global Forest Resources Assessment (FRA) in 2010 (FAO 2010b). Thus, there is urgent need for accurate forest change maps to improve carbon accounting from deforestation and degradation for international regulations and applications in ecological, social, political, and scientific fields.

1.2 Mapping Forests: A Remotely Sensed Approach

It is not surprising that the thriving of LUCC studies is closely linked with the development of remote sensing (RS) and Geographic Information Systems (GIS) technologies. RS and GIS can provide crucial information for LUCC studies, namely data about spatial patterns and temporal dynamics of land use and land cover change, mainly in terms of maps. RS is a general term describing the process of obtaining information on a certain target or area from a distance, typically from an airborne or satellite platform (Campbell 2002). The information that can be gleaned from remotely sensed data is constrained by the type of both target object and the instrument or sensor employed. Optical, synthetic aperture RADAR (SAR) and Light Detection and Ranging (LIDAR) are the most commonly used sensors in detecting forest change.

Applications of optical remote sensing imagery in classifying land cover types date back to the early 1970s (Anderson 1976), with Landsat imagery widely used in forest cover change detection in Global Forest Resources Assessment since 1990. The Landsat satellite series provides the scientific community with the world’s longest continuously acquired imagery at a high resolution (30 m), 16-day repeat cycle, and global coverage since 1992 (USGU 2014). Other frequently used optical imagery for earth observation includes The Moderate Resolution Imaging Spectroradiometer (MODIS), Satellite Pour l’Observation de la Terre (SPOT), RapidEye, etc. This ‘family’ is expanding steadily with newly launched Landsat 8 and scheduled Sentinel 2. Based on the Landsat dataset, the first global forest change map from 2000 to 2012 at a 30 m resolution has been published last year (Hansen et al. 2013) (Fig. 6.1a).

Fig. 6.1
figure 1

Map of forest cover in Africa, cropped from the global forest map generated from different sensors. a Forest cover change map from 2000 to 2013 based on optical imagery, data source Hansen et al. (2013); b forest cover map generated from L-band radar, data source and explanation of HH/HV nomenclature JAXA (2014); and c forest height map generated from Lidar and optical, data source Simard et al. (2011)

Full size image

Radar sensors, on the other hand, observe forests with a totally different algorithm. Rather than ‘seeing’ forest canopy, microwave can penetrate through clouds and forest canopy and interact with trunks, brunches, and undergrowth, depending on the wavelength in use. This enables radar backscatter images that closely represent woody biomass and forest structure (Woodhouse et al. 2012). Long wave length, such as L-band (Olander et al. 2008; Mitchard et al. 2011; Stedham 2012) and P-band (ESA 2012), has proved to be most effective in mapping forest extent, especially in tropical regions where constant cloud cover may constrain the use of optical imagery. Available L-band SAR imagery includes Japanese Earth Resources Satellite 1 (JERS-1), Phased Array type L-band Synthetic Aperture Radar on Advanced Land Observing Satellite (ALOS PALSAR), and the newly launched ALOS-2. In 2014, Japan Aerospace Exploration (JAXA) published the first 25 m/50 m global forest/non-forest map (2007–2010) based on the L-band ALOS PALSAR dataset (JAXA 2014) (Fig. 6.1b). With L-band Satellites for Observation and Communications (SAOCOM) and BIOMASS mission scheduled to launch in the coming 5 years, abundance of data specified to detect different aspects of forests will be increased greatly.

LiDAR was introduced to forest ecosystem studies in early 2000 (Lefsky et al. 2002). Because of its ability to measure the three-dimensional distribution of plant canopies, LiDAR images are widely used to accurately estimate vegetation structural attributes, such as tree height, tree density, leaf area index (LAI), and above-ground biomass (AGB) (Sun et al. 2011; Huang et al. 2013; Mitchard et al. 2012) (Fig. 6.1c). Constrained by a smaller footprint and relatively higher cost, airborne LiDAR is usually applied at a regional scale ranging from one to several thousand hectares. However, multisource remote sensing data fusion technology made the first Global Forest Height map possible in 2010. This dataset is produced by combining LiDAR (ICESat) with optical (MODIS) imagery with 1 km resolution (Simard et al. 2011). Also, the first spaceborne LiDAR, Global Ecosystem Dynamics Investigation (GEDI), will be launched by the National Aeronautics and Space Administration (NASA) in 2018 monitoring tree height and canopy density at a global scale (NASA 2014).

1.3 Linking Deforestation and Degradation to Its Drivers: The Role of GIS

While RS provides essential information of forest cover and forest biomass at various scales, GIS serves as an important tool for exploring their spatial patterns and change dynamics. ‘Spatial pattern’ refers to the geographic variation across space, which describes the perceptual structure, placement, or distribution of forest or forest change (Burrough and McDonnell 1998), such as the number, shape, clustering, or line arrangement of deforestation and degradation patches. These patterns, which capture combined effects of complex LUCC drivers, can be further manipulated with descriptive spatial analysis models, such as hotspot analysis or land use change matrices (which describes the proportion of forest taken up by other land cover and land use types) are generally used.

Another primary function of GIS is to determine the spatial relationships between features that occupy the same location (Burrough and McDonnell 1998). This can be achieved by overlaying ancillary maps (census data, soil maps, vegetation map, terrain maps, etc.) with land use or land cover maps derived from RS, such as to explore the relationships between forest change and environmental, ecological, economic, and social variables. Based on these spatial relationships, various predictive spatially explicit models have been built to explain past patterns of LUCC or predict future trends. These models can be location-based statistical and cellular models, which emphasize neighborhood effects (Committee on Needs and Research Requirements for Land Change Modeling 2013), or agent-based with a focus mainly on processes.

1.4 Who Else Has Participated in Changing the Forests?

Spatial patterns of deforestation and forest degradation, which are captured by GIS and RS, are the direct and indirect consequences of ecological, social, and economic drivers. These drivers have shifted local land use for centuries through different processes. Over the past decades, competition for forest resources has been intensified predominantly by changing population dynamics, agricultural expansion, changing lifestyles, and new distributional dynamics of globalization. Geist and Lambin (2002) proposed to categorize these drivers into proximate causes and underlying drivers. Recently, a process-based conceptual framework has been proposed to bring the focus of LUCC studies in urban areas away from geographical colocated phenomena and toward a better understanding of distal drivers, which are linked to local events through a complex set of dynamic processes (Seto et al. 2012, and Chaps. 1 and 2 for more detail).

Distal drivers, also defined to address the increasing importance of distal connections and flows of driving current land use change, increasingly shape local land use disputes as global markets (Rueda and Lambin 2013) and flows of goods (Garrett et al. 2013), information, and knowledge interact with local constellations. Distal drivers need to receive more attention, because neither population increases nor resource scarcity alone constitute the sole and major underlying causes of land cover change (Lambin et al. 2001). In this chapter, I explore if looking through a lens of remote sensing and Geographic Information Systems will allow me to zoom in on a specific local situation of land use competition, but extend the scope of the analysis to discuss distal drivers of that competition.

2 Capture Distal Drivers in the Beira Corridor, Manica, Mozambique

2.1 Land Cover and Land Use History in Beira Corridor

The Beira corridor is historically an important road and railway network, which links Beira—the second largest city in Mozambique and one of the most important ports of southern Africa—with Zambia, Zimbabwe, and Malawi (Fig. 6.2). This area, which covers the districts of Manica and Sofala, suffered greatly from the civil war between 1977 and 1992 and is still largely underdeveloped. Since 1992, this area is undergoing rapid land use change, accompanied by large population movement mainly as resettlement in previously abandoned rural areas (Stedham 2012). For decades, small-scale agriculture, notably for maize production, has been the dominant form of income for rural people and the main driver of deforestation and degradation of Miombo woodlands (Bradley and Dewees 1993; Cavendish 2000). Recently, charcoal making has become an additional main driver, particularly in Mozambique (FAO 2010c). With economic and population growth (INE 2010), Mimobo woodland is disappearing at a rate of 219,000 ha (3.1 %) per year (FAO 2010c). This trend is also captured in the Beira corridor by the Global Forest Change map from 2000 to 2012.

Fig. 6.2
figure 2

Location of Beira corridor

Full size image

Although the deforestation and degradation rate of Mozambique is not as astonishing as Indonesia or Brazil, Manica has been identified as a pilot area of REDD+. This is because the original biomass stock of Miombo woodland is relatively low (characterized by an open tree canopy and a continuous grass layer) (Campbell 1996; Mayaux et al. 2004), and small biomass loss will make a large difference for the ecosystem services that Miombo woodland can provide, such as carbon sequestration and biodiversity preservation (Shackleton et al. 2002; Bourguignon 2006). Moreover, local people’s livelihood heavily depends on forest produce, such as food and fuelwood (FAO 2010a). 75 million people inhabit miombo woodland regions, with an additional 25 million urban dwellers relying on miombo wood or charcoal as a source of energy (Dewees et al. 2010).

To alleviate poverty, the government of Mozambique launched the Beira Agricultural Growth Corridor initiatives in 2010 (BAGC 2014), aiming at structural change among the smallholder farmers by boosting agricultural productivity and promoting commercial agriculture, which currently consists of less than 0.3 % of arable land in the Beira corridor region. These initiatives stimulated cooperation between government, the private sector, local farmers, and the international community in land use decision-making processes, increasing the competition primarily between woodland conservation interests and other land uses. Actors connected with Miombo woodland are shown in Fig. 6.3.

Fig. 6.3
figure 3

Actors related to deforestation and degradation of Miombo woodland in the Beira corridor

Full size image

2.2 Can Spatial Patterns Reveal Hidden Distal Drivers?

2.2.1 Patterns of Deforestation and Degradation in Beira Corridor Region

Human activities may alter the biophysical land surface with certain characteristics in spatial patterns in terms of shape, size, intensity, and proximity to other features such as roads or houses. For example, in central Mozambique, clearance for small-scale agriculture (Fig. 6.4b) in Miombo woodland (Fig. 6.4a) usually results in high-intensity above-ground woody biomass loss with plot sizes between ~1 and 2 ha. Clearance for commercial farming, with similar high intensity of biomass loss, usually results in plots over 4 ha (Ryan et al. 2014).

Fig. 6.4
figure 4

Typical types of human disturbance to Miombo woodland

Full size image

Patterns of charcoal production, however, are more complex as our interviews in the area indicate. It can involve selective removal of certain tree species (such as Brachystegia spiciformis) or medium size stems (Fig. 6.4c) from an area of ~0.2 ha surrounding kilns (Ryan et al. 2014). Yet, it can also involve the clearance of a patch of woodland. In this case, patterns of charcoal making are featured with ~2 ha intensive biomass loss in the center, with low biomass loss in the surrounding areas. Interviews also suggested a tight relationship between charcoal making and clearance for farming. The majority of the kilns are built temporarily at the first stage of setting up a new farm, when woodland is cleared and trees are burnt and sold as charcoal. In this case, charcoal making is more like a by-product of farming, such that the income can be used for buying necessities for agriculture. At the same time, however, such charcoal-making activities may degrade the surrounding woodland, which may stimulate the transition into framing even further.

Both optical and radar remote sensing imagery are capable of detecting change in Miombo woodland, with advantages and limitations to both data types. Hansen et al. (2013) forest change maps in the study area from 2007 to 2010 (based on optical Landsat imagery, 30 m resolution, Fig. 6.5a) and radar ALOS PALSAR biomass change maps at 25 m resolution (Fig. 6.5b) show that both of these two sensors capture the high-intensity biomass loss (deforestation) from clearance for agriculture or charcoal. These woodland disturbance activities are mostly carried out at a small scale with plot size for all detected forest loss events less than 1 ha (Table 6.1). From the data, obvious signs of commercial farmland cannot be discerned with less than 1 % detected forest loss events exceeding 4 ha. This result is supported by the Beira agricultural initiative report 2013 (BAGC 2014), as well as field interviews in 2013. Large-scale commercial agriculture is still rare in this area.

Fig. 6.5
figure 5

Woodland change maps from optical and radar sensor for a subsection of the Beira corridor in Central Mozambique

Full size image
Table 6.1 Statistic and spatial patterns of woodland loss events captured by optical and radar sensor
Full size table

Statistcs (Table 6.1) show that ALOS PALSAR radar imagery captures six times as many change events as Hansen optical imagery, especially for woodland loss at a very small plot size (less than 0.04 ha). This result is likely due to the fact that optical sensors provide information at the top-of-canopy level, while radar waves can penetrate woodland canopy and detect subtle changes. Yet, the discrepancy may also result from differences in method, e.g., different definitions of forest and deforestation classifications, use of different minimum mapping units, or different levels of uncertainties. To exclude the effects from different forest definitions, we masked out change events from ALOS PALSAR imagery that are located in areas with a biomass value of less than 20 tC/ha to match the forest definition used by the Hansen map. The obvious difference persists.

2.2.2 Effects of Maize and Charcoal Markets

Woody biomass loss is related to small-scale agriculture and charcoal making, which are identified as main reasons for woodland deforestation and degradation in Mozambique in various international reports and studies. To explore this relationship, annual biomass loss from 2007 to 2010 has been plotted against the change of charcoal price at the local market and the change of maize price at both local and global markets. Constrained by the availability of reliable data on charcoal price, an estimation of 37 % annual increase is used (EUEI 2012; Herd 2007). Results are inconclusive. Changes in price of charcoal and maize do not readily correlate with the trend of biomass loss captured by radar.Footnote 1 Limited by the frequency, period, and accuracy of data available for the Beira Corridor region, it is difficult to conclude about dynamics from these data. This demonstrates, first and foremost, that small-scale agricultural activity and local markets operate according to dynamics that are not easily captured with data that are spatially detailed but suffer from too coarse a temporal and social or institutional resolution (see also Chap. 7). With more frequent RS imagery covering a longer time span, it might be possible to capture the connection between markets and woodland deforestation and degradation, or the interconnections between charcoal making and opening a new farm (farmland arising as a collateral land cover from charcoal making, or charcoal making as a by-product of clearance for farmland) (Fig. 6.6).

Fig. 6.6
figure 6

Woodland loss detected from radar imagery plotted with change in price of maize and charcoal. Retail price of maize in Beira taken from FAO (2013); price in charcoal is estimated from Herd (2007) and EUEI (2012). The trend of biomass loss, rather than the value is comparable, because the huge difference in biomass loss area detected by radar and optical is due to the characteristics of the sensor and various other factors

Full size image

2.3 RS and GIS: An Unbiased Microscope or Distorting Mirror?

2.3.1 Linking Patterns to Processes

Recently, developments in spatial modeling, remote sensing, and computing performance have provided unprecedented amounts of data to capture, monitor, and model the spatial and temporal patterns of LUCC. This makes it more feasible to relate discernible patterns of land cover to the processes of land use and their drivers. For example, deep time series of optical imagery has been proved to be effective to reveal ecological processes (Kennedy et al. 2014). Woodland deforestation and degradation-related processes, meanwhile, are vastly complicated by the influences of global markets, exchange of knowledge and information, and international cooperation between actors in structures of governance.

The framework of distal drivers expands the focus of LUCC studies from a piece of land and its geographically neighboring areas to include new dimensions, both spatially and temporally. The process of telecoupling (Friis et al. 2015) alerts to land use drivers that cross spatial scales and urge analysts to take into account land use competition at various levels of institutional hierarchy and diverse spatial and temporal scales (see also Chap. 2). The action of distal drives on local contexts is commonly processed by selected local mediators (see Chap. 4). Distal drivers may create unique LUCC patterns as they amplify or attenuate local institutional factors, such as the distinctive spatial patterns shown for distally driven commercial farming and significantly more local, traditional small-scale agriculture for the Beira corridor. In other cases, distal driver may introduce a process altogether different from existing local factors, such as REDD+. In a less developed region as the Beira corridor region, the primary concern is to foster economic growth and market integration, alleviate poverty, and improve people’s livelihood, which usually heavily relies on resources that woodland provides (including land). These needs conflict with the aims of REDD+ with its focus on decelerating the deforestation and degradation rate in developing countries. Understanding and managing REDD+’s ‘distal’ needs and mediate cooperation with local actors by means of compensation, conserving, and sustainable management is still a major challenge. Depending on its implementation, REDD+, as a distal driver, may affect land use decision-making process at government level, improve agricultural efficiency, and exert a subtle awareness of forest preservation for local people.

2.3.2 Uncertainties of Deforestation and Degradation Estimation

Information captured in a remotely sensed image is not only constrained by the characteristic of the sensor itself (type and resolution), but also largely affected by the bias and uncertainties imposed during image processing and modeling processes. Concern has been raised over the accuracy of baseline maps used in REDD+ content for years (IPCC 2006), and recently research from Mitchard et al. (2013) showed that large uncertainties resulted from the comparison of two global biomass maps, especially when integrated only with few field data. Figure 6.7 summaries how backscatter signal from ALOS PALSAR imagery is interpreted to woodland deforestation and degradation maps. Because radar backscatter is a direct measurement neither of woodland nor of woody biomass, a ground plot is needed to build empirical (or semiempirical) relationships between woody biomass and radar backscatter. Three types of uncertainties may be introduced in the woodland deforestation and degradation maps produced in this process: plot-level uncertainty, which includes errors from ground measurement and algometric models; landscape uncertainty together with errors from remote sensing images; and the errors from unrepresentativeness of ground plots compared to the heterogeneity of actual landscape. Landscape uncertainty is perhaps most closely associated with biomass maps produced from radar backscatter. To further interpret biomass to woodland, woodland deforestation, and degradation, or other thematic maps, classification thresholds for biomass are needed to define these terms. Here, threshold uncertainty is introduced through the potential discrepancies between general thresholds applicable in all contexts and local thresholds defined for a specific local system. Integrating profound knowledge of the local system will help to reduce threshold uncertainty.

Fig. 6.7
figure 7

Processing chain of building deforestation and degradation maps from ALOS PALSAR radar imagery, with uncertainty estimation

Full size image

2.3.3 Can a Map Itself Become a Distal Driver?

As remote sensing-based global forest maps are largely available at low cost, forest maps have been widely used in ecologic, social, and political fields, where the map users are generally not involved in the process of map building. The process of producing remote sensing-based maps is not a neutral intermediary but a mediator that shapes the representation of land surfaces in particular, systematically biased ways (cf. Latour 2005). Problems may arise when such mediators are treated as if they were intermediaries, i.e., free of biases. It is therefore important for map users to be (made) aware of the enclosed potential biases so that decisions are made in full awareness of the choices, which needed to be made to produce the map and which may not be visible or legible in the final map.

Firstly, despite the limitation of sensors and effects of noise, remote sensing images provide a mediated but nevertheless ‘direct’ visual representation of the world, compared to land use data from, e.g., interviews, where much longer chains of translation, perception, and interpretation are involved in connecting land with map. RS images do not contain biases toward any particular subset of land use processes. Instead, they deliver a ‘balanced’ result in as far as they show the actual outcome of negotiations between various LUCC processes. Interpreting these RS images to a thematic or land use classification map, however, is always a process of selection such that only some of the available information is focused and extracted. Secondly, uncertainty of LULC classification should be considered in any application of the maps. Because maps, especially global scale maps, are normally based on understanding from a limited number of areas. A universal definition of categories across the entire study region may not be suitable for one particular area. A good illustration of this effect is the previous discussion on how different forest definitions used in Hansen’s deforestation map and a radar map result in very different areas being represented in the map as deforestation and degradation patches (Fig. 6.6). It is worth mentioning in this context that over 1000 forest definitions are currently in use in various regions and study fields (Lund 2012). Thirdly, it appears that the vulnerability of analyses to bias increases with the various kinds of ‘distalness’ (Chap. 2) and thus the complexity of LUCC drivers. A simple example: In the Beira corridor, the tight and complex relationship between charcoal making and clearing forest for a new farm makes classifying land use into any of these two categories questionable; particularly when these attributions are treated as mutually exclusive from then on. Fourthly, the focus and thus the labeling of a map depend to a significant degree on the interests and perceptions of the mapmaker. Any particular land use at a certain location (pixel) always has physical, economic, and political dimensions that can be made more or less visible not least through the map labels. In the case presented here, the forest degradation map could also be labeled as a charcoal burning events map or smallholder farmers’ livelihood map, depending on the perspective of actors and agencies. What appears as environmental problems in one map shows as local economic performance in another based on the same set of data. Labeling processes embed one particular land use interest in land use maps, which may obstruct other understandings of land use competition in that particular location. This is not to say that maps are simply social constructions in any trivial sense of the term. It is merely pointing out the looping effects between maps, the world they try to represent and the map users developing interventions into the world based on an understanding of land use competition dynamics that is influenced by maps among other information. In this sense, the maps available for any particular area will have an impact on the understanding of the dynamics of land use competition in that area. In many cases, they are not a major driver such as markets for land-based products and they certainly should not be understood in any deterministic sense. Yet, they introduce expertise into political decision-making processes and as such mediate land use competition.

3 Conclusion

Remote sensing provides us with the unprecedented ability to sense various aspects of land surface change at global scale and high temporal resolution. While the advantages of remote sensing have been widely recognized in land use competition studies, revealing the underlying processes that shape patterns of change is still challenging. This is exacerbated as geographically, socially, and institutionally distal factors play an increasingly important role in shaping local landscapes. These factors, such as global markets, international organizations, and NGOs, alter local land use through mediating and shaping flows of capital, material, and information, often across spatial and temporal scales. This creates complex LUCC patterns on the land surface. To capture them, ‘sensing’ needs to operate on different spatial and temporal scales; i.e., a profound understanding of local LUCC processes needs to be integrated with various ancillary datasets including GIS, statistics, and time series information.

What can we ‘see’ from RS and GIS?

What can we not ‘see’?

Aperture of what we can observe

(1) RS imagery captures spatial patterns of LCLU from local to global scale, which exhibit the ‘balanced’ results of previous land use competition

(2) Stacking time series of RS imageries may reveal distal driver that creates certain spatial patterns.

(3) Fusing optical and radar RS might help reveal hidden land use change patterns

(4) Spatial analysis combined with ancillary data on a GIS platform may help reveal LULC and change patterns

(1) Agencies and drivers are always standing behind the maps, though we might distinguish certain drivers by unique spatial change patterns they create

(2) Process of land use change, such as information flow, decision-making process, capital flow, etc

(3) Pathways that distal driver mediated with local factors

(4) Information we can interpolate from RS map also constrained by its resolution and type of remote sensing sensors

(1) Land use classification categories can be subjective, which only reflects part of the information captured by RS in that pixel, or the most important information that interests the map builder

(2) The same map can be labeled differently according to the perspective of different actors and agencies