Land cover change analysis of a Mediterranean area in Spain using different sources of data: Multi-seasonal Landsat images, land surface temperature, digital terrain models and texture
Highlights
► We apply a methodology for the mapping and analysis of land cover changes. ► We integrate different sources of data for the mapping process. ► We compute the land-cover transitions of Granada Province between 1998 and 2004. ► The applied methodology results in an increase in the mapping accuracy of 33 percentage points.
Introduction
Over the last few decades, an important transformation in the land cover/use of Spain has taken place. This transformation process is even more notorious in peri-urban areas. Urban expansion makes it difficult to make land use compatible in these regions, which are mainly agricultural but are strongly influenced by the bordering urban areas. For many years, this conflict has been the case for the province of Granada, especially the coastal regions and, also very notoriously, the area of influence of the city of Granada. This area is a geographic space of great socio-economic value in which, historically, agriculture has shaped its territorial structure. However, the swift urban development of the city of Granada—and its clusters of peri-urban and industrial area agglomerations—has caused a rapid transformation of the traditionally agricultural spaces (Menor-Toribio, 1997).
In general, the importance of accurate and updated cartographic information that describes the nature and extent of agricultural and natural resources is growing, especially in rapidly developing metropolitan areas. In these situations, it is crucial to have detailed geospatial information regarding the patterns and tendencies of land use. These data would constitute a base of information for decision-making regarding territorial management and regulation. The importance of making an inventory, quantifying changes and monitoring the changes of the physical characteristics of land-covers has been widely recognized by the international scientific community as a key element in the study of global change (Fearnside, 2000, Foley et al., 2005, IPCCC, 2000).
Although traditional samplings and inventories can be used to monitor land cover/use, remote sensing provides great amounts of information about the distribution of covers and their spatio-temporal changes. This method allows us to chart large areas using a small amount of time and economic resources. Change detection is the process of identifying differences in the state of an object or phenomenon on different dates (Singh, 1989). Three aspects are important when discussing changes: detecting changes that have occurred, identifying the nature of the change and measuring the total area of the change (Malila, 1980). Remote sensing has been a highly popular method to monitor changes because it has great potential for the study of these three aspects.
After the launch of the Landsat satellite in 1972, digital techniques of change detection and analysis improved significantly (Crapper and Hynson, 1983, Howarth and Boasson, 1983, Singh, 1989, Wickware and Howarth, 1981). Using satellite images obtained on different dates, change detection in a surface is conducted using numerical methods and algorithms that, in essence, produce an image (or multi-image) from which one can analyze the observed changes. Many geographers have used satellite data to address land-cover change-detection problems (Abd El-Kawy et al., 2011, Aguirre-Gutiérrez et al., 2012, Bakr et al., 2010, Dewan and Yamaguchi, 2009, Kennedy et al., 2009, Shalaby and Tateishi, 2007). Multi-temporal spectral change-detection techniques can be divided into two main categories. The first group includes categorical methods, also known as post-classification techniques, which consist of a comparative analysis of two different classifications from separate dates. The second group corresponds to continuous changes, also known as pre-classification enhancement techniques. The goal of such methods is to measure the magnitude of change produced in any of the attributes related to land cover that could be measured in a continuous way, for example, the quantity or concentration of the vegetation or urban cover. Many studies conduct a deep review of the different methodologies in the study of land-cover change (Collins and Woodcock, 1996, Coppin et al., 2004, Kennedy et al., 2009, Lu et al., 2004, Mas, 1999, Rogan and Chen, 2004, Singh, 1989).
The selection of the methodology of changes to apply will be determined by the necessities of the user. When the objective of the user is the study of the transitions that are produced between the different land covers/uses, post-classification analysis techniques will be used to compare previously classified images or classify images from different dates together. The analyst can produce change maps and their corresponding confusion matrixes. This process facilitates interpretation by environmental managers (Crabtree et al., 2009, Nagler et al., 2009, Nemani et al., 2009, Svancara et al., 2009, Townsend et al., 2009, Wang et al., 2009). These techniques provide information about the nature of the change by comparing two maps obtained under a classification process (Hepcan et al., 2011, Potapov et al., 2011, Yuan et al., 2005b). The accuracy of this technique depends on the precision of the original classifications. The resulting change map will be as accurate as the previous classifications because its accuracy will be the product of the accuracies of each individual classification (Mas, 1999, Rogan and Chen, 2004).
Mediterranean covers have a similar spectral behavior (low inter-class separability) and a complex spatial structure of the landscape, which presents a great variability of highly fragmented spatial patterns. The low inter-class separability of Mediterranean areas is a direct consequence of the climate and the characteristics of the covers present in these regions. Hydric resource scarcity in these regions gives rise to the presence of bare soils, which are usually calcareous with very light shades and, hence, a high reflectance similar to that of urban areas. This high reflectance of the soils may overwhelm the relatively small component reflected from sparse vegetation. Thus, the accuracy with which urban areas, soil and non-dense vegetation covers (e.g., olive groves) can be separated spectrally is low (Berberoglu et al., 2007, Berberoglu et al., 2000). Different approaches could be used to increase the separability between spectrally similar categories. Information about variation in the phenological state of vegetal covers can be added by incorporating multi-seasonal images. Additionally, auxiliary variables that describe environmental gradients that improve the characterization of the covers, such as temperature, digital terrain models, or humidity, could be included. Characterization of the spatial variability in these images provides important information about the disposition of the objects and their spatial relations within the image through textural measures. Lastly, a great number of auxiliary variables could be used in the classification of land covers/uses. The majority of land-cover/use mapping applications only use satellite images as input variables to the classification (Dixon and Candade, 2008, Oetter et al., 2001, Yuan et al., 2005a). In the last few years, the use of auxiliary variables to improve the classification process has increased (Franklin, 1998, Rogan et al., 2008, Watanachaturaporn et al., 2008). The results of land-cover/use mapping studies indicate that including non-spectral variables in the classification process helps to improve the discrimination of the land-cover categories, resulting in more accurate maps (e.g., 5–10% greater accuracy) (Franklin, 1995, Watanachaturaporn et al., 2008, Wright and Gallant, 2007). Several studies show that the combination of multi-stationery images allows a better distinction of certain covers (Lunetta and Balogh, 1999, Oetter et al., 2001, Wolter et al., 1995, Yuan et al., 2005a). Additionally, other land-cover/use mapping studies have proven that including textural variables provides additional information on the classification process to improve the accuracy of the maps (Agüera et al., 2008, Asner et al., 2002, Chan et al., 2003, Chica-Olmo and Abarca-Hernández, 2000, Franklin et al., 2000, Johansen et al., 2007).
This study focuses on the application of a land-cover mapping method and change analysis of complex and heterogeneous areas. The study area chosen for this research is the province of Granada. However, the suggested methodologies are general and could be applied to other areas with similar characteristics. The method integrates the use of a machine learning classifier, which is able to classify complex data spaces (random forest) with the inclusion of new spectral and auxiliary variables that improve the accuracy of the classifications, including multi-seasonal satellite images, digital terrain models, land-surface temperature and textural images.
Section snippets
Classification method (random forest)
A type of machine learning technique that uses ensembles of classifications has received increased interest during the last decade (Friedl et al., 1999, Ghimire et al., 2010, Gislason et al., 2006, Hansen and Salamon, 1990, Krogh and Vedelsby, 1995, Sesnie et al., 2008, Steele, 2000). Ensemble learning algorithms use the same base classifier to produce repeat classifications of the same data, which are combined using a rule-based approach (such as maximum voting) (Breiman, 2001, Friedl et al.,
Study area
The province of Granada (GP) is the study area chosen for this project. This province is located in the south of Spain on the Mediterranean coast, surrounded by the Penibetica mountain range (Fig. 1). This area occupies 12,635 km2 and the elevation ranges from sea level to the Mulhacen Peak (3482 m) in Sierra Nevada National Park. The climate of Granada is Mediterranean with a continental influence, characterized by hot and dry summers and wet and cold winters. The average annual temperatures
Methodology
Basically, there are two methodologies to detect and analyze the changes that have occurred in an area from images taken on different dates: a comparative analysis of two independent classifications and a simultaneous analysis of multi-spectral data or change-enhancement techniques. Both techniques have their limitations. Techniques based on classification do not detect subtle differences in vegetal classes and the final change map will accumulate the errors from each individual classification
Evaluation of the classifier and different data sources
The selection of the classifier and the subset of variables to be used had an important influence on the accuracy of the generated land-cover maps. Table 2, Table 3 show improvements in the obtained accuracies as a consequence of using RF and adding different subsets of additional information to the spectral variables from the satellite images. To confirm if the results of the classifications were linked to the selection of the classification algorithm, a maximum-likelihood (MLH) classifier was
Conclusions
The province of Granada is a highly complex Mediterranean area that is extremely heterogeneous and compounded by numerous land-covers that are difficult to map due to the spectral similarities between categories or land-covers. Therefore, it is necessary to apply robust classification methodologies to map and analyze changes. The method of analysis used in the present study was post-classification by comparing two supervised classifications obtained from the application of the RF classifier to
Acknowledgments
We are grateful for the financial support given by the Spanish MICINN (Project CGL2010-17629) and Junta de Andalucía (Group RNM122). We thank the reviewers for their constructive criticism.
References (75)
- et al.
Land use and land cover change detection in the western Nile delta of Egypt using remote sensing data
Applied Geography
(2011) - et al.
Using texture analysis to improve per-pixel classification of very high resolution images for mapping plastic greenhouses
ISPRS Journal of Photogrammetry and Remote Sensing
(2008) - et al.
Optimizing land cover classification accuracy for change detection, a combined pixel-based and object-based approach in a mountainous area in Mexico
Applied Geography
(2012) - et al.
Remote sensing of selective logging in Amazonia: assessing limitations based on detailed field observations, Landsat ETM+, and textural analysis
Remote Sensing of Environment
(2002) - et al.
Monitoring land cover changes in a newly reclaimed area of Egypt using multi-temporal Landsat data
Applied Geography
(2010) - et al.
Texture classification of Mediterranean land cover
International Journal of Applied Earth Observation and Geoinformation
(2007) - et al.
The integration of spectral and textural information using neural networks for land cover mapping in the Mediterranean
Computers & Geosciences
(2000) - et al.
An assessment of several linear change detection techniques for mapping forest mortality using multitemporal landsat TM data
Remote Sensing of Environment
(1996) - et al.
A modeling and spatio-temporal analysis framework for monitoring environmental change using NPP as an ecosystem indicator
Remote Sensing of Environment
(2009) - et al.
Change detection using Landsat photographic imagery
Remote Sensing of Environment
(1983)
Evaluation of random forest and adaboost tree-based ensemble classification and spectral band selection for ecotope mapping using airborne hyperspectral imagery
Remote Sensing of Environment
Computing geostatistical image texture for remotely sensed data classification
Computers & Geosciences
Land use and land cover change in Greater Dhaka, Bangladesh: using remote sensing to promote sustainable urbanization
Applied Geography
Random forests for land cover classification
Pattern Recognition Letters
Relevance of airborne lidar and multispectral image data for urban scene classification using Random Forests
ISPRS Journal of Photogrammetry and Remote Sensing
Landsat digital enhancements for chage detection in urban environments
Remote Sensing of Environment
Application of high spatial resolution satellite imagery for riparian and forest ecosystem classification
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
Synthesis of ground and remote sensing data for monitoring ecosystem functions in the Colorado River Delta, Mexico
Remote Sensing of Environment
Monitoring and forecasting ecosystem dynamics using the terrestrial observation and prediction system (TOPS)
Remote Sensing of Environment
Land cover mapping in an agricultural setting using multiseasonal Thematic Mapper data
Remote Sensing of Environment
Monitoring agricultural lands in Egypt with multitemporal Landsat TM imagery: how many images are needed?
Remote Sensing of Environment
Regional-scale boreal forest cover and change mapping using Landsat data composites for European Russia
Remote Sensing of Environment
Random forest classification of Mediterranean land cover using multi-seasonal imagery and multi-seasonal texture
Remote Sensing of Environment
An assessment of the effectiveness of a random forest classifier for land-cover classification
ISPRS Journal of Photogrammetry and Remote Sensing
Downscaling Landsat 7 ETM+ thermal imagery using land surface temperature and NDVI images
International Journal of Applied Earth Observation and Geoinformation
Remote sensing technology for mapping and monitoring land-cover and land-use change
Progress in Planning
Mapping land-cover modifications over large areas: a comparison of machine learning algorithms
Remote Sensing of Environment
Integrating Landsat TM and SRTM-DEM derived variables with decision trees for habitat classification and change detection in complex neotropical environments
Remote Sensing of Environment
Remote sensing and GIS for mapping and monitoring land cover and land-use changes in the Northwestern coastal zone of Egypt
Applied Geography
Combining multiple classifiers: an application using spatial and remotely sensed information for land cover type mapping
Remote Sensing of Environment
Assessing the landscape context and conversion risk of protected areas using satellite data products
Remote Sensing of Environment
Spatial pattern analysis for monitoring protected areas
Remote Sensing of Environment
Remote sensing of land-cover change and landscape context of the National Parks: a case study of the northeast temperate network
Remote Sensing of Environment
Change detection in the Peace-Athabasca delta using digital Landsat data
Remote Sensing of Environment
Improved wetland remote sensing in Yellowstone National Park using classification trees to combine TM imagery and ancillary environmental data
Remote Sensing of Environment
NALC land cover change detection pilot study: Washington D.C. area experiments
Remote Sensing of Environment
Cited by (63)
Particulate matter concentration from open-cut coal mines: A hybrid machine learning estimation
2020, Environmental PollutionAuxiliary datasets improve accuracy of object-based land use/land cover classification in heterogeneous savanna landscapes
2019, Remote Sensing of EnvironmentCitation Excerpt :Our schema contains six land use classes of cultivated and managed vegetation, 10 land cover classes of natural and semi-natural vegetation, and one land use class of artificial surfaces and associated areas (Table 3). We chose the RF classification algorithm, because it is resilient to overfitting, training data reduction and outliers (Rodriguez-Galiano and Chica-Olmo, 2012; Zhu et al., 2016), provides measurements of feature importance, is relatively easy to parameterize, and has the ability to characterize high-dimensional data with many collinear features (Breiman, 2001; Pal, 2005; Foody, 2004; Belgiu and Drăgut, 2016; Fox et al., 2017). Throughout the RF model building steps, we applied R packages caret (Kuhn et al., 2017) and randomForest (Liaw and Wiener, 2002, 2018), which follow the original code of Breiman (2001).
Multi-temporal analysis for land use and land cover changes in an agricultural region using open source tools
2017, Remote Sensing Applications: Society and EnvironmentCitation Excerpt :These aspects required the application of techniques to monitor and properly assess the surface area over time. Spatiotemporal analysis using a combination of remote sensing techniques, geographical information systems (GIS) and statistical analysis has shown its potential in characterization studies of land cover, which integrates the resulting information to support the decision making process (Araya and Cabral, 2010; Rodriguez-Galiano and Chica-Olmo, 2012; Ruiz et al., 2013; Kiptala et al., 2013). Satellite images are useful in many applications related to environmental monitoring.