A combination of visual interpretation and digital classification can many times be a good method to generate information for an area (Bickel et al. 2006). Remote sensing can be used to detect locations of change when using data from two different periods. The method contains two categories, which have been used in the 2006 IPCC Guidelines for National Greenhouse Gas Inventories (Bickel et al. 2006). Post-classification change detection approach The approach of detecting postclassification change refers to techniques where two or more predefined land-use classifications exist for different points in time and where the changes are detected, usually by subtraction of the data sets. The techniques are straightforward but are also sensitive to inconsistencies in interpretation and classification of land-use categories. Pre-classification change detection approach The approach of detecting pre-classification change refers to more sophisticated and biophysical approaches to detecting change. Difference between data from spectral responses from two or more periods are compared statistically and used to obtain information on land-use change. This approach is less sensitive to inconsistencies in interpretation and can detect much more subtle changes than is possible with the post-classification approaches, but is less straightforward and requires access to the original remotely sensed data. Simple visual interpretation can be used to assist these two approaches.
Areas of change are highlighted through display of different band combinations, band differences or derived indices, such as NDVI.
The following steps illustrate the approach to estimate changes in biomass stocks using land-use changes detected through remote sensing. The post-classification change detection is used here to estimate biomass and hence carbon stock:
Step 1: Collect land-use information for the area
° The information can be in the form of maps or data that can be used as a check against satellite-derived classification
Step 2: Collect remote sensing data for the area
° Several different remote sensing products can be used; Landsat is probably the most common because of its temporal coverage and availability (see Table 14.1 for web site information) ° Several bands can be used, where visible and near-infrared spectra have proven suitable for identifying vegetation
Step 3: Make sure the remote sensing data are calibrated to reduce satellite or technical noise or atmospheric issues
° Data distributors usually have information on how this can be achieved
Step 4: Make sure the data are georeferenced so the inventory plots can be identified in corresponding pixels Step 5: Produce land-use classes using the remote sensing data
° Perform unsupervised classification using image processing programmes or GIS
Step 6: Evaluate the classification with the help of land-use information
° This can be done by using the error matrix (see Table 14.3)
Step 7: If the relationship is low, try other types of land-use classification
° The accuracy should correspond to the certainty needed for the project
Step 8: Calculate biomass based on the inventory data and areas under different land-use classes
Saturation effect Saturation effect with remote sensing occurs when vegetation is dense and the added biomass, chlorophyll or leaf layers are not shown in the obtained data. This makes the assessment of heavy vegetation more unreliable. In adopting NDVI, which ranges from -1 to 1, it is suggested that the added chlorophyll activity after 0.7 is uncertain. When measuring biomass, it is suggested that saturation occurs in optical remote sensing at about 15 kg/m2 (Steininger 2000). For several space-borne SAR radar products, the saturation is slightly extended to ~20 kg/m2. This limits the use of data for routinely quantifying biomass, particularly as the majority of forest and woodlands globally supports an above-ground biomass of more than 100 t/ha or 10 kg/m2 (Rosenqvist et al. 2003).
Whenever a map of land use is used, it is necessary to know the reliability of the information presented in the map. When results are generated from classification of remote sensing data, it should be recognized that the reliability of the map is likely to vary for different land-use categories: some categories may be uniquely distinguished while some may be confounded with others. For example, coniferous forest is often more accurately classified than deciduous forest because reflectance characteristics of a coniferous forest are more distinct. Similarly, it is often difficult through remote sensing to ascertain changes, such as a change from intensive t illage to reduced tillage, in land management practices, in a specific land area (IPCC 2006). The level of detail can increase by subdividing, for example, dividing cropland by crop or a forest into different basal area classes, depending on what is being monitored and to what degree of accuracy. To evaluate the accuracy of mapping the following steps for basic interpretation could be adopted:
Step 1: Select a number of sample plots from each land-use type shown on the map based on the interpretation. Record the coordinates for the plots Step 2: Collect actual real-world data on land use from some type of ground-truthing data
Step 3: Create a matrix (see Table 14.3) of interpreted/modelled and actually measured land-use types Step 4: Calculate percentage of accuracy from the matrix.
Age of vegetation developed as ascertained from field measurements in combination with information obtained from near-infrared remote sensing from a Landsat ETM + image could give estimates of above-ground biomass of hardwood forest stands in northern US with a coefficient of determination (r2) of 0.95 (Zheng et al. 2004).
Uncertainty in relation to heterogeneity of land-use types The accuracy of interpretation is related to the homogeneity of the surface of vegetation being investigated. Remote sensing data from a large area of boreal forest are less variable than those from a dry tropical forest with hundreds of different tree species. Similarly, estimates of production of wheat from large-scale cultivation are likely to be more reliable than those of production of rice from rice paddies scattered among blocks of agroforestry. Differences in soil moisture, topography and patches of atmospheric disturbances such as clouds of haze can also lower the certainty of interpretation from remote sensing data.
Uncertainty in relation to topography Topographic features in an image prevent light or radar energy from being reflected back to the remote sensing detector. In an image of mountainous areas, there are large dark areas such as valleys and shadowed ridges. When using data with small pixels size, the impact of shadows is larger than that with large-scale or coarse images. For example, 50 m pixels reported a 14% error rate, which dropped to 3% in 1.1 km pixels (Brown 1997).
14.5 Feasibility of Remote Sensing for Different Project Types
Methods to distinguish between forest and other land-cover types using remote sensing data are fairly accurate when the contrast between them is high. An accuracy of 80-95% is expected with high-resolution images. The problem arises when other land-cover types also have green vegetation, perhaps even trees. Forest parameters to be determined from the image, such as canopy cover and extent of degeneration that are supported without ground-truth data are not readily available. The extent and scale of ground-truthing is determined by the resources available. Afforestation and reforestation Application of remote sensing for afforestation and reforestation projects is feasible since the main feature of project implementation is conversion of different land-use systems to forests and plantations. Distinction between clearly distinct land-use systems, such as forest and non-forest is more reliable than interpreting the difference between young and mature forest stands or that of partly degraded forest and non-disturbed forest. CDM sink projects under the Kyoto Protocol are required to prove that the land was not forested in the past. Remote sensing can play an important role in this process, since data are available at least as far back as the early 1980s.
Avoided deforestation Annual emissions from land-use change, mainly through deforestation and degradation in tropical developing countries, account for -20-25°% of the total anthropogenic emissions of greenhouse gases. The uncertainty of the estimate is varied, given the range. The factors contributing to uncertainty are lack of resources, lack of standard methods and lack of data and capacity at the national level. Standardization is required for using remote sensing data, tools and analytical methods that suit the variety of national conditions as well as meet acceptable levels of accuracy (UNFCCC 2006).
The role of remote sensing will be central for projects aimed at avoiding deforestation or reducing emission from forest degradation in the tropics, since the area to be covered is likely to be large and inaccessible in places. Apart from conversion of forest to non-forest uses, the process of degradation, thinning or regeneration is also important since degradation of forest decreases carbon content. The idea of reducing the emissions from these areas automatically calls for a baseline and a base period. A historical baseline could be constructed on the basis of area under forest cover and extrapolated to the future based on remote sensing data (Skutsch et al. 2007).
Deforestation and forest degradation need to be differentiated since both contribute to loss of carbon stocks. Remote sensing technique has the potential to help in distinguishing the two processes at a much lower cost than other methods.
In addition to monitoring land use and changes in land cover, remote sensing techniques are particularly useful in integrating other factors such as population density, markets and ownership (Skutsch et al. 2007) that drive changes in land use and land cover. By adopting a GIS approach, issues related to remote sensing can be incorporated with the other spatially determined features to be used in analysis and monitoring of avoided deforestation (Castillo-Santiago et al. 2006). Several models have been developed to deal with avoided deforestation and related management issues.
Visual interpretation of images is often used in identifying sampling sites for ground inventories. The method is simple and reliable. However, it is labour intensive and therefore restricted to limited areas and may be affected by subjective interpretations by different interpreters. There has been a revolution in the way information about environment is acquired, processed and stored, which has been attributed to advances in computers for data collection and manipulation. Geographic information systems, or GIS, have played a key role in the development (Rosenqvist et al. 2003). The power of GIS comes from a database management system that is designed to store and manipulate data (Lillesand et al. 2004).
Application of remote sensing requires integration of the extensive remote sensing data with ground-truth measurements or data to characterize areas associated with multiple features. This is generally achieved most cost effectively using a GIS (IPCC 2006). Apart from its application for remote sensing data, GIS also offers the possibility of integrating for further analysis other types of information including data on soil types, population of a certain area and infrastructure or management practices, as presented in Fig. 14.4 (Ostwald 2002).
Technological advances in remote sensing techniques will enhance their use in many different land-use monitoring applications. The data are getting more reliable, available and affordable. Avoided deforestation is one area where remote sensing is being targeted as the key approach to holistic and acceptable methods of monitoring changes. These changes extend beyond the straightforward "forest to deforestation" process to also include degradation of forest resources (DeFries et al. 2006). Modelling could be employed to increase accuracy and reduce the cost of estimating carbon stocks. Over the last two decades, many biophysical models of forest growth dynamics have been developed, many of them with the specific objective of using data from satellite images as input to drive the models (Porté and Bartelink 2002; Skutsch et al. 2007).
Recent developments in remote sensing technology have advanced its application in estimating carbon stocks in land-use systems. Here, radar and laser systems are the most promising, with radar giving high accuracy in estimating biomass under cloudy conditions and laser soundings giving a three-dimensional picture of the forest (Skutsch et al. 2007). Several radar products using SAR such as CAOSMO-SkyMEd, Radarsat-2, TerraSAR-X and L, TanDEM-X and RiSat-1 are being developed. The limitation of these techniques is the expertise required for the analysis and the cost of the data, but the techniques hold great promise for the future.
Remote sensing and GIS can be useful to most types of land-use project but certain characteristics should be taken in to account when using it. Remote sensing and GIS could be used to
° Define land area and boundary, since GIS allows for good data storages and handling that can be handy throughout the different phases of the project. ° Detect distinct land-use categories and its boundary, but land-use categories with similar vegetation features or blurry boundaries are harder to assess accurately.
° Estimate land cover and land-use change since data can span effectively over time and remote sensing can compute consistent classification for change detections. However, certain land-use practices such as selective felling in forests is hard to detect with remote sensing. ° Estimate homogeneous vegetation such as single species plantations with high certainty, but, heterogeneous vegetation such as tropical natural forest have larger uncertainties.
° Estimate carbon stock if the remote sensed data is accompanied with biomass data to establish relationships. ° Assess land-use over large areas at low cost. Small areas can also be assessed depending on resolution of data. However, its interpretation should always be used together with ground-truthing.
Remote sensing techniques are already used extensively for monitoring changes in land use and land cover in most countries, largely at macro-levels such as global, national and regional, but their application in monitoring changes in land use, biomass production and carbon stocks at the project level is still evolving. In principle, remote sensing and GIS can be used for estimating changes in area and carbon stocks, even at micro-level. However, large-scale application requires lower costs and more reliable estimates. Remote sensing in combination with GIS will be increasingly used in the years to come for national greenhouse gas inventory programmes as well as land-based carbon mitigation and roundwood production projects.
Was this article helpful?
What you need to know about… Project Management Made Easy! Project management consists of more than just a large building project and can encompass small projects as well. No matter what the size of your project, you need to have some sort of project management. How you manage your project has everything to do with its outcome.