Mapping and Characterizing the Green Belt of Córdoba: Land Dynamics and the Urban-Rural Transformation Process

The work was conducted in an area of approximately 170,000 ha of the metropolitan region of Córdoba city. The limits of the study area were the Sierras Chicas (below 600 m a.s.l.) to the west, the altitude above 350 m a.s.l. to the east; the locality of General Paz to the north and the national Route C-45 to the south. The area covers the historical Green Belt of Córdoba (GBC), including three well-defined sectors: the northern sector, irrigated by Canal Maestro Norte; Villa Retiro, Villa Esquiú, El Quebrachal, part of Colonia Tirolesa; the eastern sector covers the area of Chacras de la Merced, and the southern sector includes the road to San Carlos, Ferreyra, also defined by the irrigation system –Canal Maestro Sur, which is derived from San Roque reservoir–, although they are currently supplied in part by Los Molinos reservoir. (Figure 1)

Figure 1.Study area of the GBC.

The analysis was focused both on the peri-urban area surrounding Córdoba city, where small and medium producers of fresh food that supply the local market are concentrated, and on more distant areas characterized by extensive agriculture. The study includes the productive areas to the north and south of the city, which have been characterized as herbaceous graminoids under irrigation ² in the soil cover classification map of Argentina (Year 2006-2007) performed by INTA.

Mean annual temperature in the study area is 17 ºC, with thermal amplitude of 14 ºC. Frosts occur between May and September, and the frost-free period spans 270 days. Mean annual precipitation is 750 mm, with a monsoonal seasonal distribution. Water deficit varies between 180 mm in the east and 240 mm in the west ¹⁰. Productive activities are characterized by a concentration of leaf vegetable production in the northern area, whereas the southern zone is dominated by potato and carrot production, with some cases of agroecological businesses ¹¹. The most important annual crops are soybean and maize, which are mostly distributed outside the Capital department, but with an important participation in the urban-rural interphase.

The characterization of the GBC consisted of three phases of analysis:

1. urban growth in Córdoba city (delimitation of the historical perimeter of the city and comparison with the current one)

2. present land cover types

3. dynamics of agricultural parceling (determination of the changes in parcel structure).

The historical and present boundaries of Córdoba city were delimited with a multi-temporal principal components analysis (PCA) ¹² using Landsat 1 MSS (Multi Spectral Scanner) image of 60 m of spatial resolution acquired on 12/06/1974 and a Landsat 8 OLI (Operational Land Imager) image of 30 m resolution acquired on 24/07/2014. The method consists of combining both images in a single data stack using the set of spectral bands corresponding to each sensor, except for the thermal bands. Since the images have different pixel size, the PCA was performed for each image date separately. The analysis aims at reducing spectral dimensionality of each image and generating new synthetic bands (principal components, PCs) that summarize the greatest variability of the original information. Accordingly, the urban areas were better differentiated by PC2 for 1974 and PC3 for 2014. Thus, discrimination of urban covers for each data was maximized, which allowed us to delimit their spatial distribution with greater accuracy (Figure 2). Urban perimeters were delimited using the ISODATA unsupervised classification algorithm. This type of algorithm is used to partition a spectral image into a given number of classes based on the statistical information of the image; it does not require human supervision for training ¹³. The obtained urban covers were monitored by visual inspection using the Landsat images employed in the PCA, as well as Google Earth images. The comparative analysis of the urban area was performed within the limits of the Capital department, covering a total of 58 neighborhoods distributed between the northern and southern sectors of the GBC.

Figure 2.Principal Components Analysis. The first three principal components for each year are shown.

Based on the distribution of the limits of the neighborhoods (DGCyE, 2008), we calculated the urbanized proportion within each neighborhood for the two years of study (Equation 1). Then, we used a normalized change index to evaluate the relative change in urban growth (Equation 2).

(1)

(2)

Urban development is the proportion of urban area within each analyzed neighborhood. Urb_b,cis a normalized change index between dates t1 and t2 for each neighborhood. This index provides a relative measure of the normalized change for each neighborhood.

Agricultural land use and cover types were characterized and analyzed using a 10-m spatial resolution SPOT 5 image (Satellite pour l’observation de la terre) acquired on March 11 2015 (Table 1). This type of multispectral image has a better spatial resolution than Landsat images, which allowed us to work at a lower detail scale (100 m²x pixel). Land use and land cover types were mapped via a supervised classification method using the Support Vector Machine (SVM) algorithm. SVM is a learning algorithm that allows optimal separation of a data set, each one being assigned to a class of interest, depending on the training samples added to the classifier ¹⁴.

The date selected for classification coincides with the end of the maturity cycle of summer crops and the future harvest in April/May; in contrast, winter crops cannot be characterized in March, since this in an early date. Intensive agriculture known as “Liviana” is characterized by crops of alternate short cycles of approximately 3 months; thus, production can be characterized on a continuous basis using images over the year. In March, “pesada” extensive agricultural crops (potato, carrot) are ready for harvest and, in general, local producers, particularly of potato, usually leave them in the ground for a maximum period of three months; hence in March plots will be visualized as deprived of or with low plant cover.

The training strategy of SVM was conducted based on geolocation of field samples (direct observation of crops) and by visual interpretation on a SPOT image. Field visits were made to the study area between April and June 2015 (Figure 3). For that purpose, a legend was constructed to define in the field the different land cover and use types present in the region: summer crops (soybean, maize, alfalfa), pesado horticultural crops (potato, sweet potato, carrot), light horticultural crops (lettuce, chard, spinach, etc.), mixed land use (vacant parcels, abandoned fields, roadsides, weedy plots), urban areas, forest areas, bare soil or fallow land and water. Given the small size of horticultural plots (light), very often of smaller size than the resolution of the sensor used, we decided not to include them in the classifier. For their identification, high-resolution Google Earth images were used and polygons were manually digitalized, and then added as a mask to the already classified map. This procedure helps to avoid confusion with other classes of similar phenological cycles and ensures a more accurate fit of the final horticultural area. Location and characteristics of the classes observed in the field were geolocated with GPS (Global Positioning System); a total of 96 locations were recorded, which were then used in the classification assessment phase. The result was evaluated using two indicators: overall accuracy (OA) and Kappa coefficient, both derived from the confusion matrix ¹⁵.

Figure 3.Training and validation sites used in SVM classifier

OA summarizes the overall result of the classification. The Kappa number is an alternative measure of the accuracy of classification that subtracts the effect from random accuracy. Kappa quantifies how much better a given classification is than a random classification. ¹⁶ suggests that the use of a subjective scale in which Kappa values <40% are “poor”, 40%-55% “fair”, 55%-70% “good”, 70%-85% “very good” and >85% “excellent”. The producer’s accuracy is a measure of the omission error and indicates the percentage of pixels of a given class that are correctly classified. The user’s accuracy is a measure of the commission errors and indicates the probability that a pixel classified as a particular class really represents that class in the field ¹⁵, ¹⁶.

The analysis of agricultural parceling consisted of three phases: a) texture filters, b) segmentation and c) unsupervised classification. These techniques are widely used in digital image processing with the aim of identifying homogeneous shapes, sizes and edges. Different co-occurrence texture filters were tested on different far off dates (Table 1). A Landsat 5 TM image of 30 m of resolution acquired on 11/04/1988 was used to define the historical parceling, and a Landsat 8 OLI image acquired on 24/07/2014 was used to define the current state of parceling. Texture filters are used to detect edges and shapes that follow defined patterns in an image. We tested 8 filters computed in 3 x 3 pixel moving windows applied only to the nIR bands, with the aim of identifying the maximum discrimination of the parcel borders on each of the processed dates (mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment, correlation). By visual inspection of the tests, we determined that the variance filter was the best option. In the following step, we segmented the variance images processed for both dates. Segmentation classifies the images into different segments and requires the definition of thresholds and a minimum pixel population. Here, we determined the thresholds by counting the digital values that represent at least 80% of the data for each one of the dates, which allowed us to establish the parcel borders and agricultural limits that were best represented in the image. At the final step, the segments obtained were classified using the ISODATA unsupervised method (Figure 4 a,b).

Figure 4a.Northern sector of the Green Belt of Córdoba. The image corresponds to an orthorectified mosaic generated by the Instituto Geográfico Nacional (IGN). RGB: NIR-Red-Green. (Date of aerial photographic flight: 1,2,3,4 and 5/12/14). IGN-Dirección de Sensores Remotos.

Figure 4b.Scheme of the processing for the identification of the parcel structure of a productive area of the northern sector of the Green Belt of Córdoba. The upper right quadrant (RGB: 421) represents the parceling situation in 1988. Down, the situation corresponding to the same sector in 2014 (RGB: 542).

Growth of the southern zone of the GBC was 33.2 % during the study period, which is a sensitively higher difference from that recorded for the northern zone (▲6.8%). The neighborhoods that did not exhibit urban zones in 1974 were: Ciudad Ampliacion Ferreyra, Quintas de Flores, 25 de Mayo, Country Fortin del Pozo, Quintas de Italia, Country la Santina, Rocio del Sur, Ciudad Obispo Angelelli, Villa Rivadavia, Inaudi, Ferreyra Segunda Seccion, Country Campina Del Sur, Country Los Mimbres, Posta de Vargas, Piedras Blancas and 23 de Abril. In increasing order, the neighborhoods with the greatest urban proportion at present (>50%) are: Tejas II, Nuestro Hogar III, Posta de Vargas, Piedras Blancas, Nuestro Hogar I, Villa Coronel Olmedo and 23 de Abril (Figure 6_a). The relative change with respect to the built area in 1974 was highest for the neighborhoods: Nuestro Hogar I, Villa Coronel Olmedo, Tejas II, Nuestro Hogar III, Las Canuelas, Fincas del Sur (above 90% of change) (Figure 7_ab). The only neighborhood that did not exhibit urban development over the study period is Ciudad Ampliación Ferreyra.

Figure 6a.Map representing the urban area in 1974 (in black), and in 2014 (overlapped in red). B. Percent change in the urban area per neighborhood.

Figure 7a.Average urban development level for the neighborhoods in the southern zone. b) Changes (%) in the urban area per neighborhood between 1974 and 2014.

From a comparative perspective, for both zones, in most of the studied neighborhoods an overall greater urban development is observed in those neighborhoods that were rural areas in the past, whereas in the historically urban areas, the urban process was lower. In other words, urban growth occurred in new areas rather than in already urbanized sectors.

Mapping results are presented in Figure 10. Soybean was the crop covering the greatest cultivated area (59500 ha), followed by maize (17477 ha), alfalfa (9724 ha), potato (6850 ha), and light vegetables (1780 ha) (Figure 8). Soybean was the prevailing crop, covering an area 24.5% greater than that of maize. The “mixed use lands” class (urban, rural and indeterminate zone areas), with an area of 34600 ha, and “urban areas”, with 24740 ha, ranked second and third, respectively, in the general land cover ranking in the study area.

Figure 8.Areas of the mapped land cover and use classes.

Mixed use lands covered 20 % of the total area; These zones are around the city, in which the rural and urban activities are distributed.

Production of intensive crops (light horticulture) covered 1% of the study area, whereas heavy intensive production covered 4% (with dominance of potato) (Figure 9, Figure 10)

Figure 9.Cover types grouped according to type of crops and land uses in the Green Belt of Córdoba.

Figure 10.Land cover types and uses: Above (left): Distribution of soybean crops. Above (right): Distribution of intensive production. Below (left): Urban and mixed use zones. Below (left): Complete map of cover types.

(Table 2) presents the summary of the calculated assessment measures, assigning 80 % of accuracy, which indicates an acceptable general result for the aim of this work. The Kappa obtained coefficient value defined the classification result as “very good” (76%). The variation of the results of the Kappa coefficient for each class indicates very good and excellent results for all the classes, except for the potato crops, forest and fallow, which were scarcely or not represented in the field sampling (Table 2).

In comparative terms, the “urban” class had the best classification assessment, without confusion due to loss or overestimation of data. The highest omission error corresponded to maize class, and was due to the confusion recorded with the potato class and “mixed used lands” (data not shown). These results mean that the classifier recorded a smaller maize area than the true area size. Likewise, potato crop had the highest commission value of the classification, since this class was mostly confused with light horticultural crops, i.e. the classifier falsely depicted a potato crop.

The approach used to detect changes in size, distribution and number of agricultural parcels present in 1988 and 2014 was based on two procedures: 1) visual interpretation by combining both dates in false color composite (RGB; nIR-Red-Blue), and 2) analysis of size and shapes of segments associated with agricultural parcels using the previously described procedure (see 2.2). Notable differences between parcel distribution and size were visually differentiated for the northern GBC. Although the dates of analysis have a three-month difference for each year, it is evident that parcel distribution and size were reduced at the temporal scale analyzed. To estimate the relationship between size and number of parcels in the years analyzed, we applied the segmentation process, which allowed us to evaluate those differences without the need to classify the land uses of the past. Here, they were used to define the characteristics of the agricultural parceling. Figure 11_a,b shows a higher number of small farm parcels (in red) in 1988; by contrast, in 2014, a smaller number of larger parcels were recorded, which were devoted to extensive production, showing a land tenure concentration process. The reasons behind are related to technological adoption (mainly soybean monoculture) and its related socioeconomic impact. The concentration of land tenure has been the result of free market policies, which have defined the search for bigger production scales to reduce average production costs, implying that the larger the area, the greater the productive competitiveness.

Figure 11.Results of segmentation and classification of agricultural parcels. a) Distribution and size of parcels for 1988 and b) for 2014.

Although the segment size may be over- or underestimated, the results of the analysis conducted provide a basis for making an estimation and support the hypothesis postulating the loss of production spaces devoted to intensive agriculture at the expense of the expansion of extensive agriculture due to soyization. Figure 12 illustrates that relationship based on the segments classified for each year.

Figure 12.Size and number of agricultural parcels represented by homogeneous segments.

In general, society has put an emphasis on the importance of urban green spaces, especially because recreational and landscape aspects are associated with a better quality of life. However, at present there is a new challenge: protecting peri-urban areas to ensure food production near the cities and to enable the continuity of traditional production practices.

This work evidences how the urban growth process in Córdoba city, through the expansion of the urban area, has had direct effects on the reduction of the productive agricultural areas. The urban sprawls and the change processes recorded in the productive systems have displaced small and medium scale productive businesses, which were replaced with residential plots, social housing and gated neighborhoods, as well as agro-industrial businesses. Accordingly, local producers have been forced to sell or rent their lands and migrate to produce in other localities, or to definitively abandon food production, which threatens the continuity of agricultural activity in the Green Belt of Córdoba. Future work is needed in order to assess the quality and continuity of quantification and classification of peri-urban horticulture. More complex procedures are needed to be developed such as machine learning techniques. The information must be immediately available to end users and the institutions who are involved in urban and rural planning.