Agriculture and Food Availability – Cultivating the City: Mapping and Characterizing Urban Agriculture with Satellite Imagery
Large cities of the modern world mostly conjure images of tall buildings, paved surfaces, heavy traffic, and tertiary workers dressing in suits and in a hurry. But that is not the whole picture. Present in cities of both rich and poor countries, urban agriculture refers to farming and food production in a mostly urbanized setting. This activity performs valuable social, cultural, and environmental roles: it provides food and income, offers recreational opportunities, and benefits the urban ecological system (e.g., biodiversity, green corridors, CO2 sinks, and urban climate) (Netzband et al. 2007). It also contributes to sustainable local development, and in the present context of worldwide crisis and instability, it can also increase food security (Veenhuizen 2006).
The city of Lisbon, Portugal, has historically expanded towards areas occupied by farms, orchards, and olive groves, thus integrating some rural character. This process was complemented by the influx of immigrants from the rural countryside or from abroad who had farming habits. While until recently this land use activity was perceived as marginal and simply tolerated by public officials, there are now municipal plans to expand it, organize it, and integrate it in the city planning process.
Within the city, agriculture typically takes place in small farms or clusters of plots in residual or vacant public land, occupying valleys, hillsides, and roadsides. After peaking in 1987, it is estimated that currently this land use occupies close to 84 hectares. However, since this is mostly an informal activity with limited official supervision, there is a lack of available accurate and updated records or geospatial data for assessment and analysis.
Mapping Urban Agriculture from Space
For almost forty years, images acquired by artificial satellites orbiting the Earth have proven useful for mapping, environmental monitoring, and planning, offering the benefits of frequent repetition, wide geographical coverage, and low cost, in addition to the ability to capture reflected and emitted radiation of surfaces in different spectral bands.
Remote sensing with satellite imagery and spatial analysis of agricultural land use within the urban fringe is generally an under-researched topic, especially in cities of the ‘developed world’. The few case studies for cities of the developing world generally utilize medium resolution satellite imagery (e.g., Landsat, SPOT), which lack the spatial detail to capture intra-urban agriculture taking place in small plots (e.g., Thapa et al. 2004; Luedeling et al. 2007). Kemeling et al. (2002; 2003) have innovated by using high spatial resolution IKONOS imagery for surveying agricultural activities in the city of Ouagadougou, Burkina Faso.
The current and future generation of very high spatial resolution (VHR) satellite imagery provides an advantageous alternative to detect and closely monitor urban agriculture. However, their effective use requires the development of feature extraction software that enables a timely and consistent discrimination, classification and delineation of these specific land uses. Such semi-automated approaches that aim at extracting real-world features from imagery have been grouped under the new acronym GEOBIA – Geographic Object-Based Image Analysis (Hay and Castilla 2008).
The GeoSat Project
The fast-changing urban landscape makes keeping the municipal spatial databases up-to-date, a challenging task, especially when traditional time-consuming mapping processes are used. The publicly-funded GeoSat research project, of which the Lisbon City Hall is the end-user, aims at developing methods to expedite the production of geographic information for municipal planning and land monitoring using very high spatial resolution (VHR) satellite imagery. This information includes artificial and vegetated classes such as buildings, roads, sidewalks, trees, grass, and bare ground.
In the context of GeoSat, the potential of VHR satellite imagery and GEOBIA for detection, mapping, and characterization of agricultural areas in Lisbon is also being explored.
Characterizing Urban Agriculture in Lisbon
In order to test the mapping and characterize urban agriculture, a heterogeneous area located in the eastern part of the city of Lisbon was selected, covering 64 ha (158 acres) (Figure 2). Urban agriculture was characterized in comparison to the other vegetation classes present in the area: trees, grass, and natural vegetation.
GEOBIA approaches were applied using a VHR QuickBird image acquired on April 14, 2005, pan-sharpened to a spatial resolution of 0.61 m.
For extraction of features three GEOBIA-based software applications were tested: Feature Analyst (FA), ENVI Feature Extraction Module (FX), and Definiens Professional (DP). For quality assessment, these results were compared to a reference map produced by visual analysis by an independent interpreter using the same satellite image and ancillary data sources. Field work was also carried out.
Extraction results obtained with Definiens Professional (DP) present the highest agreement with the reference data set, with an overall accuracy of 52%. DP allowed detecting 90% of the area used for agriculture. All approaches over-estimate the extent of the class (which according to the reference data is 5,2 ha): by 34% (FA), by 23% (FX), and by 65% (DP).
Characterization of agriculture in the study area was performed using the features obtained by visual analysis, since our goal was to characterize the class as accurately as possible, also in order for that knowledge to assist in future extraction procedures.
Vegetation classes (including Agriculture land use) were characterized in respect to vegetation greenness (using NDVI) and topography (using terrain slope). The Normalized Difference Vegetation Index (NDVI) (Rouse et al. 1973) is a robust and widely-used indicator of presence of live green vegetation. Terrain slope was derived from a new digital terrain model (DTM) which was produced in the context of the project from large-scale base data for 1998, with a resolution of 0.5 m.
Mean NDVI and slope values were then computed for each reference agricultural polygon, and values summarized for the class. For comparison, the remaining vegetated LULC classes from the reference data set were also characterized in the same fashion.
NDVI values show that agricultural areas have a similar overall greenness (low standard deviation) with mean values lower than trees and grass but higher than those of natural vegetation.
Regarding terrain slope, agricultural use appears more characteristic: it takes place in overall more homogeneously sloping areas (lower standard deviation), and has a shorter range with a much lower maximum value than other classes (21 degrees). The fact that terrain is never flat indicates use of more marginal areas of land.
The present work is an on-going exploratory attempt, part of a wider goal of defining an expeditious method to obtain updated geographic information for municipal use. Accuracy of mapping urban agriculture using semi-automated feature extraction approaches was generally low. Semi-automated extraction of urban agriculture is complicated by the fact that the class is very heterogeneous, comprised of a mosaic of small parcels having: i) the same crops in different stages, ii) different crops, and iii) fallow or recently farmed parcels. This mapping and analysis effort has revealed availability of vacant areas with bare ground or natural vegetation, which could be used to support the expansion of urban agriculture in Lisbon in accordance with current municipal goals. Future developments include the use of another QuickBird image acquired in March 2007 to detect changes in land use. Also, more advanced contextual approaches should be explored to increase the accuracy of semi-automated detection and extraction of urban agriculture.
Hay, G.J., Castilla, G. (2008) Geographic Object-Based Image Analysis (GEOBIA): A new name for a new discipline? In: Object-Based Image Analysis – spatial concepts for knowledge-driven remote sensing applications. Eds:T. Blaschke, S. Lang, G. J. Hay. Springer-Verlag., pp. 75-89, 2008.
Kemeling, I., de Jong, S.M., van Teeffelen, P.B.M., van den Berg, L.M., and Roerink, G.J. (2003) GIS-based classification of high spatial resolution IKONOS imagery for surveying agricultural activities in the city of Ouagadougou, Burkina Faso. Proceedings of the 2nd Workshop of the EARSeL Special Interest Group on Remote Sensing for Developing Countries, Bonn, p.p.14 – 22, 2003.
Kemeling, I., de Jong, S.M., van Teeffelen, P.B.M., van den Berg, L.M., and Roerink, G.J. (2002) Remote sensing and GIS for good governance: analysis of high spatial resolution IKONOS imagery for surveying agricultural activities in the city of Ouagadougou, Burkina Faso. Proceedings of the 6th Seminar on GIS for Developing Countries, ITC Enschede, p.p 18.1 – 18.8, 2002.
Luedeling, E., Gebauer, J., Schumacher, J., El-Siddig, K., and Buerkert, A. (2007) Spatial expansion of urban agriculture in Khartoum, Sudan. International Conference on Research for Development in Agriculture and Forestry, Food and Natural Resource Management, 2007.
Netzband, M., Stefanov, W.L., and Redman, C.L., (2007) Remote sensing as a tool for urban planning and sustainability. In: Netzband, M., Stefanov, W.L., and Redman, C.L., (eds.) Applied remote sensing for Urban Planning, Governance and Sustainability. Berlin, Germany: Springer, pp. 1-23.
Thapa, R.B., Borne, F., Kusanagi, M., and Pham, Van C. (2004) Integration of RS, GIS and AHP for Hanoi peri-urban agriculture planning. Map Asia Conference, 2004.
Veenhuizen, R. van (2006) Cities farming for the future, urban agriculture for green and productive cities. RUAF Foundation, IDRC and IIRR, 2006.
Rouse, J.W., Haas, R.H., Schell, J.A., and Deering, D.W. (1973) Monitoring vegetation systems in the Great Plains with ERTS. Proceedings of the 3rd ERTS Symposium, Vol. 1, p.p. 48-62, 1973.
Sérgio Freire (firstname.lastname@example.org) is a research assistant at e-GEO, Centre of Geographical and Regional Planning Studies, New University of Lisbon, Portugal. With a master’s in geography from the University of Kansas (USA), he has worked at the National Center for Geographic Information (Portugal) and at the Portuguese Geographic Institute, namely researching land use and land cover mapping using satellite imagery and developing advanced forest fire risk methods. Current research activities include feature extraction from very high resolution satellite imagery, and mapping of disaggregated spatio-temporal population distribution.
Teresa Santos is a doctoral researcher at e-GEO, Centre of Geographical and Regional Planning Studies, New University of Lisbon, Portugal. She has a master’s in GIS from the Technical University of Lisbon (Portugal), and is presently getting a PhD in geography, with a scholarship from the Portuguese Foundation for Science and Technology. Previously she worked at the National Center for Geographic Information (Portugal) and at the Portuguese Geographic Institute. Current research is on urban feature extraction from very high resolution satellite imagery and LiDAR data.
J. A. Tenedório is associate professor at the Dept. of Geography and Regional Planning, New University of Lisbon, Portugal. His research interests are in urban remote sensing, urbanization, and geographic information systems.