4. Metodología y Resultados
4.2 Análisis Imágenes Satelitales Sentinel 2A
The carbon stored within tropical forests can be estimated using remotely sensed data (those data collected from instruments mounted on satellites or aeroplanes). As stated above, remote-sensing can only be used in combination with plot-based data as it cannot be used to directly measure carbon stocks (Rosenqvist et al., 2003, Drake et al., 2003). However, remote sensing provides a tool by which ground-based measurements can be extrapolated across landscapes, on both national and global scales. It must be noted that, combining these techniques compounds the errors associated with plot-based methods (described above) with those associated with remote sensing (see Section 2.5). Here, I will briefly discuss the two main methods (biome-based methods and correlation-based methods) by which remote sensing can be used for forest and woodland carbon monitoring, expanding on limitations and uncertainties specific to carbon stock estimation that have not been previously discussed in Section 2.5.
The earliest compilations of biome averages were made decades ago, and have been continuously updated by the research community (Brown and Lugo, 1984, Whittaker and Likens, 1973). Biomes likely represent the most important source of variation in landscape carbon stocks, thus the application of plot-derived carbon estimates to biome area obtained via remote sensing (see Section 2.5) is perhaps the simplest way to estimate forest and woodland carbon storage. Biome averages are currently freely available and are currently the only source of globally consistent forest carbon information (e.g. IPCC Tier 1 values; Section 2.4.4). However, forest carbon stocks vary within each biome (e.g. according to temperature, precipitation, soil; Section 2.3) and so an average value cannot adequately represent this variation, leading to high uncertainties in carbon flux estimations associated with LCC, particularly if deforestation primarily occurs in forests and woodlands that systematically differ from biome averages (Houghton et al., 2001).
To better represent variation within biomes, regression models can be used to correlate remotely sensed data with plot-based carbon estimates. This approach is similar to the IPCC Tier 3 methods described in Section 2.4.4 and shows reduced uncertainty when compared to biome-based (Tier 1) estimates (GOFC-GOLD, 2010). Similar to monitoring forest area, several satellite sensors are available, broadly falling into four categories: medium and high resolution optical data; very high resolution optical data; microwave or radar data; and LiDAR data.
Present optical satellite sensors (e.g. Landsat, MODIS) cannot be used to estimate carbon stocks of tropical forests and woodlands with high certainty (Thenkabail et al., 2004). Correlations have been developed between plot- based carbon estimates and vegetation indices (e.g. NDVI) (Lu, 2005, Foody and Cutler, 2003). However, optical satellite sensors tend to saturate in high biomass regions (Sánchez-Azofeifa et al., 2009, Thenkabail et al., 2004, Waring et al., 1995) and may be of limited availability due to cloud cover (Sánchez-Azofeifa et al., 2009, Asner, 2001). Furthermore, the correlations developed are often regionally specific and so not transferable between studies or applicable across the globe (Waring et al., 1995). Very high-resolution images can be collected, typically from aeroplanes, and used to directly measure tree height and crown area. However, due to the high cost, it is often impractical to collect these data over vast areas, and so this technique is only particularly efficient for estimating biomass in small regions (Brown et al., 2005).
Table 2.6 Benefits and limitations of available methods to estimate national-level forest carbon stocks (reproduced from Gibbs et al.
(2007)).
Method Description Benefits Limitations Uncertainty
Biome averages
Estimates of average forest carbon stocks for broad forest categories based on a variety of input data sources
Immediately available at no cost
Data refinements could increase accuracy Globally consistent
Fairly generalised
Data sources not properly sampled to describe large areas
High
Forest inventory
Relates ground-based measurements of tree diameters or volume to forest carbon stocks using allometric relationships
Generic relationships readily available Low-tech method widely understood
Can be relatively inexpensive as field-labour is largest cost
Generic relationships not appropriate for all regions Can be expensive and slow
Challenging to produce globally consistent results Low
Optical remote sensors
Uses visible and infrared wavelengths to measure spectral indices and correlate to ground- based forest carbon measurements
e.g. Landsat, MODIS
Satellite data routinely collected and freely available at global scale
Globally consistent
Limited ability to develop good models for tropical forests
Spectral indices saturate at relatively low carbon stocks
Can be technically demanding
High Very high- resolution airborne optical remote sensors
Uses very high-resolution (10–20 cm) images to measure tree height and crown area and allometry to estimate carbon stocks
e.g. Aerial photos, 3D digital aerial imagery
Reduces time and cost of collecting forest inventory data
Reasonable accuracy
Excellent ground verification for deforestation baseline
Only covers small areas (10,000s ha) Can be expensive and technically demanding No allometric relations based on crown area are available Low to medium Radar remote sensors
Uses microwave or radar signal to measure forest vertical structure
e.g. ALOS PALSAR, ERS-1, JERS-1, Envisat
Satellite data are generally free
New systems launched in 2005 expected to provide improved data
Can be accurate for young or sparse forest
Less accurate in complex canopies of mature forests because signal saturates
Mountainous terrain also increases errors Can be expensive and technically demanding
Medium
Laser remote sensors
LiDAR uses laser light to estimates forest height/vertical structure
e.g. Carbon 3-D satellite system combines Vegetation canopy LiDAR (VCL) with horizontal imager
Accurately estimates full spatial variability of forest carbon stocks
Potential for satellite-based system to estimate global forest carbon stocks
Airplane-mounted sensors only option Satellite system not yet funded
Requires extensive field data for calibration Can be expensive and technically demanding
Low to medium
In contrast to the above optical techniques, microwaves, radar and LiDAR signals are able to detect the top of the canopy, whilst also penetrating down to the underlying terrain. Thus, data on canopy height are collected and used to estimate carbon storage. Recent studies (Baccini et al., 2012, Saatchi et al., 2011) have tended to use LiDAR data over microwave and radar techniques as they are less likely to saturate in high-biomass regions (Means et al., 1999, Lefsky et al., 1999, Drake et al., 2003). However, due to the scattering of reflectance beams, these techniques have higher uncertainties for taller canopies and in montane regions, where terrain is more rugged (Reutebuch et al., 2003, Means et al., 1999). Despite this drawback, large-footprint LiDAR remote sensing far exceeds the capabilities of radar and optical sensors to estimate forest and woodland carbon stocks (Means et al., 1999, Lefsky et al., 1999, Drake et al., 2003). However, currently aeroplane-mounted LiDAR instruments are too costly for use at large scales, and satellite based LiDAR systems are not yet widely available (Hese et al., 2005, Gibbs et al., 2007). In addition, techniques that use height as a proxy for AGB have high uncertainty in regions that obtain maximum height rapidly but continue to accumulate biomass for many years (Feldpausch et al., 2011, Banin et al., 2012).
In this thesis, I utilise both the biome-based and correlation-based techniques. Firstly, I apply plot-derived carbon estimates of differing land cover types to land cover maps; and secondly, I correlate the ground based data with remotely sensed candidate variables, providing proxy data for climatic, edaphic and anthropogenic variables. This second technique has an advantage over the detection methods described above in that it provides indications as to what influential variables likely effect carbon storage (Sánchez-Azofeifa et al., 2009). Furthermore, both these forms of remotely sensed data are freely available, increasing the accessibility of methods developed here to less economically developed countries (LEDC).