Automatic environmental quality assessment for mixed-land zones using lidar and intelligent techniques
Research highlights
► LiDAR can provide excellent information to improve thematic maps in especially interesting areas. ► Intelligent techniques are a key factor to provide fast and accurate results when lidar is applied to the study of the natural environment. ► Our experimentation on real data from a riparian area in the south of Spain shows decision trees (C4.5) provide the best results with the highest level of clarity for the final model.
Introduction
The coast of Huelva, in the south of Spain, is a clear example of industrial progression and environmental protection coexisting in a common space. In proximity to the capital, a great number of industrial areas exist including the refineries that feed fuel to the southern area of Spain as well as coastal areas with high touristic development. The combination of both factors puts large natural areas at risk such as the Doñana National Park, within which can be found endangered species such as the Iberian Lynx (lynx pardinus), in great risk of extinction. The human impact on this and other areas of the territory concerns the Regional Ministry for the Environment of Andalusia which is designating large quantities of resources to control it.
The human influence on natural surroundings is not only a local fact. Many authors recognize that human beings and their activities are the principle factor that influences the evolution of the natural environment. In addition, human impact is a key factor when planning protection policies for natural spaces. Two key factors that are often studied to detect human activity are transport networks and urban nuclei (Goetz et al., 2009, Svancara et al., 2009). These parameters permit evaluating the risk of damage to natural areas nearby (Jones et al., 2009) and to draw up policies to correct the possible effects of human development. One of the most important tools to control these factors is the control and monitoring of land uses and land cover (LULC) using remote sensing.
Since its appearance, remote sensing has been used with different purposes in terms of natural resources. Recently, many authors have used remote sensing to monitor species (Stow, Hamada, Coulter, & Anguelova, 2008), or changes in cities (Gamanya, Maeyer, & Dapper, 2009), characterize the morphology of urban nuclei (Gill et al., 2008), study the richness of bird species in natural areas (Goetz, Steinberg, Dubayah, & Blair, 2007) or the severity of fires (Kokaly, Rockwell, Haire, & King, 2007), etc. In terms of monitoring changes, the authors mainly tend to use classical remote sensing techniques based on satellite images (Fraser et al., 2009, Pignatti et al., 2009, Townsend et al., 2009). It is important to bear in mind that studies of monitoring changes in natural environments are based on the interpretation of LULC maps (Svancara et al., 2009, Wang et al., 2009). In this way, an advance in the generation of these products will imply a possible direct improvement of the results obtained with these methodologies. Furthermore to the monitoring of changes, LULC has been studied profusely (Schubert et al., 2008, Schneider et al., 2009) with the general objective of treating areas of particular interest from an economical or environmental point of view. In these cases, planning and management play an important role at the time of exploiting the resources but are always subject to the quality of the products extracted from remote sensing (Dorigo et al., 2007, Kennedy et al., 2009).
The new technologies based on novel sensors, such as lidar, have become an excellent tool for improving the results of traditional remote sensing (Chen, 2007). Its capacity to register the height of objects overcomes the limitations that usually come with working with images. Due to this ability, it is relatively simple to distinguish between the ground and objects to develop digital terrain models (DTM), a primary product for a wide variety of applications. To produce DTM’s, a multitude of techniques have been proposed from the lidar point cloud (Evans and Hudak, 2007, Sithole and Vosselman, 2003). It is very important to point out that the laser is not affected by shadows and their associated problems nor does it need to be flown during the day due to its special characteristics. All these advantages, along with a progressive decrease in the related costs compared to other data sources such as satellite images, make lidar one of the leading technologies in environmental investigation.
In accordance with the proven usefulness of lidar, many investigators have chosen to use it as supporting technology for images. In this way, they tend to fuse sensors, with the objective of improving the results obtained separately (Arroyo et al., 2008, Bork and Su, 2007, Chust et al., 2008, Dalponte et al., 2008) whilst others focus their efforts on lidar as the only data source with excellent results (Chen et al., 2009, Pascual et al., 2008). Each strategy has its own advantages and disadvantages. The fusion provides a large quantity of data that produces extra information for any classification method. But also, it requires greater effort to adapt data from multiple sensors giving place to an increase in development and testing time. Furthermore, some studies show little improvement in classifications based on fusion among lidar and other sensors when they are used to carry out determined tasks (Jensen et al., 2008, Thessler et al., 2008). Other works (Townsend et al., 2009) advise being cautious when merging data in general, including if they are of the same typology which could be the case with satellite images.
Another important decision to bear in mind is the work paradigm selection. In classical remote sensing, the smallest significant unit is the pixel which is characterized by having a standard size. The data sources are divided in one series of pixels from which information is extracted to be used later for general classification. Instead, recently, a new paradigm of work has started to be applied, object-oriented approaches (Gamanya, Maeyer, & Dapper, 2007). An object, contrary to a pixel, does not have a fixed size but depends on the type of the object in question. Thus, the data to classify is decomposed in diverse objects of variable size which have been extracted through some type of segmentation technique at a previous stage. Lately, the application of object-oriented techniques has commenced on lidar as a unique data source to resolve various tasks with good results (Antonarakis et al., 2008, Donoghue et al., 2007). In this case, the object-oriented techniques apply a segmentation from computer vision techniques using a set of features extracted from lidar. Afterwards, the classification method proceeds to learn from the segmented objects to classify future instances. Despite the results being very promising, unresolved problems still exist, essentially because segmentation of lidar data is not a process easily automated needing the interaction with the user to achieve good results. An additional problem is that the return intensity is one of the main parameters used in carrying out segmentation with lidar. This datum can be affected by other factors (impact angle, sensor distance (Hofle & Pfeifer, 2007) which can modify its value and falsify the final result. As opposed to object orientation, the traditional pixel-based approaches and work with models resulting of the application of advanced intelligent techniques (Witten & Frank, 2005) could be applied on lidar with good results and with much greater levels of automation. In this sense, numerous studies show that intelligent techniques could be applied to lidar data, such as vector-support machines (Koetz, Morsdorf, van der Linden, Curt, & Allgower, 2008), artificial neural networks (Brzank, Heipke, Goepfert, & Segel, 2008) or clustering (Pascual et al., 2008).
In this work we show a new application of intelligent techniques with the objective of extracting hidden knowledge in lidar data and use it to work on urban and natural areas with the objective of:
- •
Defining a general method based on intelligent techniques to classify high resolution LULC using lidar as a unique data source and to demonstrate its value to detect human activity in an automated manner.
- •
Using this method to evaluate the human impact on a riparian area located on the Atlantic coast of Huelva province (Andalusia, Spain) close to the Doñana National Park.
Section snippets
Data description
The data for this work was provided by REDIAM (Andalusia Environment Information Network) which pertains to the Regional Ministry for the Environment of Andalusia. The data was taken in coastal areas of the province of Huelva (see Fig. 1) and Cadiz between the 23rd and 25th of September 2007 and the flight was operated at an average height of 1200 m with low inclination angles (<11°) and nominal density of 2 pulses/m2. The pulses were geo-referenced and correctly validated by the distributor of
Method
The general form in which the authors evaluate the environmental impact is based on products generated by remote sensing techniques such as LULC maps. In our case, to develop these products, the classification of lidar data is required. With this objective, this study proposes a method based on the application of intelligent techniques on a collection of features previously generated. In Fig. 2, an overall view of the whole classification process can be found which will be described in detail
Comparison of methods
To select the technique which is best adapted to our data, a 10-fold cross validation was carried out also used to evaluate the classification methods in remote sensing (Tooke, Coops, Goodwin, & Voogt, 2009). The tests carried out show that the decision trees obtained the best results. In Table 2, Table 3, Table 4, the total and partial precisions and the kappa index of agreement (KIA) obtained for the three techniques on the 2168 training pixels are shown. The three techniques obtained high
Discussion
In terms of the results of the classification method (Table 6), it must be taken into account that although the riparian areas present great difficulties for classification, the results show a great overall precision. According to the latest studies (Shao & Wu, 2008), the general precisions of LULC maps developed for many organizations do not exceed 85% despite this being the standard for considering a LULC map a useful product. The proposed method demonstrates achieving this level generally
Conclusions and future work
Public investments destined for conservation of the environment can be seen to reduce in the context of economic crisis. The low level of automation to apply policies and environment protection methodologies causes implementation costs to be higher in many cases. It is therefore necessary, to apply new techniques that reduce such costs and improve the quality of the final product to increase productivity.
In this work, an approach based on lidar data and intelligent techniques to classify land
Acknowledgements
The authors thank the Regional Ministry of Andalusia for all the support received in the development of this work and especially, to thank Irene Carpintero, Juan José Vales, Daniel Laguna and Michael Grainge for all the time they invested that allowed this work to be completed.
References (46)
- et al.
Object-based land covers classification using airborne lidar
Remote Sensing of Environment
(2008) - et al.
Fire models and methods to map fuel types: The role of remote sensing
Forest Ecology and Management
(2008) - et al.
Integrating lidar data and multispectral imagery for enhanced classification of rangeland vegetation: A meta analysis
Remote Sensing of Environment
(2007) - et al.
Aspects of generating precise digital terrain models in the wadden sea form lidar-water classification and structure line extraction
ISPRS Journal of Photogrammetry and Remote Sensing
(2008) - et al.
Hierarchical object oriented classification using very high resolution imagery and lidar data over urban areas
Advances in Space Research
(2009) - et al.
Coastal and estuarine habitat mapping, using lidar height and intensity and multi-spectral imagery
Estuarine, Coastal and Shelf Science
(2008) - et al.
A review on reflective remote sensing and data assimilation techniques for enhanced agroecosystem modeling
International Journal of Applied Earth Observation and Geoinformation
(2007) - et al.
Monitoring land cover change and ecological integrity in Canada’s National Parks
Remote Sensing of Environment
(2009) - et al.
Decision tree classification of land cover from remotely sensed data
Remote Sensing of Environment
(1997) - et al.
An automated satellite image classification design using object-oriented segmentation algorithms: A move towards standardization
Expert Systems with Applications
(2007)
Object-oriented change detection for the city of Harare, Zimbabwe
Expert Systems with Applications
Characterising the urban environment of UK cities and towns: A template for landscape planning
Landscape and Urban Planning
Connectivity of core habitat in the Northeastern United States: Parks and protected areas in a landscape context
Remote Sensing of Environment
Laser remote sensing of canopy habitat heterogeneity as a predictor of bird species richness in an eastern temperature forest, USA
Remote Sensing of Environment
Correction of laser scanning intensity data: Data and model-driven approaches
ISPRS Journal of Photogrammetry and Remote Sensing
Nearest neighbor imputation of species-level, plot-scale forest structure attributes from lidar data
Remote Sensing of Environment
Monitoring land use and cover around parks: A conceptual approach
Remote Sensing of Environment
Remote sensing change detection tools for natural resource managers: Understanding concepts and tradeoffs in the design of landscape monitoring projects
Remote Sensing of Environment
Multi-source land cover classification for forest fire management based on imaging spectrometry and lidar data
Forest Ecology and Management
Characterization of post-fire surface cover, soils, and burn severity at the cerro grande fire, new mexico, using hyperspectral and multispectral remote sensing
Remote Sensing of Environment
Object-based semi-automatic approach for forest structure characterization using lidar data in heterogeneus pinus sylvestris stands
Forest Ecology and Management
Evaluating hyperion capability for land cover mapping in a fragmented ecosystem: Pollino National Park, Italy
Remote Sensing of Environment
Land cover classification of tundra environments in the arctic lena delta based on landsat 7 etm+ data and its application for upscaling of methane emissions
Remote Sensing of Environment
Cited by (7)
An evolutionary voting for k-nearest neighbours
2016, Expert Systems with ApplicationsCitation Excerpt :In this context, we have experience with remote sensing data (García-Gutierrez, Gonçalves-Seco, & Riquelme-Santos, 2011) and this approach could compliment the cited study in a new work.
An evolutionary-weighted majority voting and support vector machines applied to contextual classification of LiDAR and imagery data fusion
2015, NeurocomputingCitation Excerpt :Eight labels were selected for the final classification: water, marshland, roads and railways, low vegetation (pastures), middle vegetation (bushes), tall vegetation (mainly eucalyptus), buildings (industrial areas, port facilities) and landfills. LiDAR data and aerial orthophotography (with resolution 1 m for site A and 0.5 m for site B and only visible spectrum bands: red, green and blue) were provided for both data sites which had already been presented and used in previous works after preprocessing (see Garcia-Gutierrez et al. [9] for a deeper explanation about the preprocessing techniques applied). The last study site (hereafter site C) was located in the Natural Park of the Alto Tajo in Guadalajara (central Spain, UTM30; 563385E 4533140N).
Urban land cover classification using airborne LiDAR data: A review
2015, Remote Sensing of EnvironmentCitation Excerpt :Through airborne LiDAR data classification and object recognition, the results can contribute different perspectives to the control of urban planning and monitoring. Examples are not limited to urban impervious surface extraction (Germaine & Hung, 2011; Hodgson et al., 2003), urban environmental quality assessment (Garcia-Gutierrez, Gonçalves-Seco, & Riquelme-Santos, 2011) and urban change detection (Stal, Tack, De Maeyer, De Wulf, & Goossens, 2013; Teo & Shih, 2013). In view of these studies, the majority of contribution from airborne LiDAR toward urban environment can be categorized in terms of urban morphology and green analysis.
Design and development of a cost-effective LiDAR system for transportation
2019, ASME International Mechanical Engineering Congress and Exposition, Proceedings (IMECE)