Historical Delineation of Landscape Units Using Physical Geographic Characteristics and Land Use/Cover Change

2.1 Study area

The Duero River is located northwest of the state of Michoacán, Mexico at 2,000 m osl. The drainage basin has an area of 2,531.3 km². It is part of the VIII Lerma-Santiago-Pacific administrative jurisdiction, the 12 Lerma-Santiago Hydrological Region and particularly in the Lerma-Chapala River hydrological zone (Fig. 1). The shape of the basin is elongated and its main direction is southeast-northwest. The region is characterized by volcanic sedimentary environments with stratovolcano type structures, and alluvial and lacustre deposits; aspect that confers its free to semi-confined aquifers hydraulic behavior [20]. The basin is divided hydrogeologically into four sub-basins and politically into thirteen municipalities of which three have a significant proportion of indigenous population (up to 61 % in the Chilchota municipality [21]).

Figure 1

Location of the study area.

Temporal ecosystems changes over the last three decades have been described into the Duero River basin, mainly related to the aquatic system [22]. Although the basin has a positive water balance, a continuous degradation trend was reported and it has been related to unbulance human water use, with the consequent reduction in the main river flow, increased discharge of wastewater, deforestation and erosion. In addition, the presence of nutrients in the water was related, at the landscape level, to the agricultural area [23].

2.2 Steps of the analysis

For the delineation of landscape units, we analyzed the differences in landscape structure through a quantitative study of heterogeneity using the Supervised Classification method [12, 19]. We applied three procedures for the selection of a set of predictor variables or input data: a) using all bands, b) vegetation indices and c) multivariate analysis (Fig. 2).

Figure 2

Scheme of procedure for the delineation of landscape units.

For the Supervised Classification a total of 178 georeferenced points were used along the basin, corresponding to the training sites taken from the vegetation cover selected for the classification: cropland, wooded area, subtropical scrub, pastureland and human settlement, according to a dynamic criterion [24]. We assumed a lack of stability in land use exhibited in the transitional states in the main vegetation cover, characterizing the heterogeneity of the area [25].

In the first procedure that included the use of bands, we used data sets from the satellite Landsat 5 (path = 28 and row = 46 [26]), and the images of the following dates: 1983, 1990, 1995, 2000, 2011 and 2014. Here, the classification is based on the grouping of pixels with similar values and the identification of the common characteristics in the elements represented by these pixels [16]. The images were acquired as complete scenes, with all available bands, including the visible spectrum, near infrared and mid infrared with spatial resolution of 30 × 30 meters [27]. The image for the 1983 scene consists of four bands and the remaining scenes had seven bands, each recording the intensity of the energy reflected or emitted by the objects on the surface at different portions of the electromagnetic spectrum [28]. We analyzed the spectral bands using the maximum likelihood classifier that has shown acceptable results for the generation of thematic maps [29, 30, 31, 32]. The classifier uses the vector of means and the covariance matrix from pixels of the training areas of each class to assign the pixels of the image to each defined classes according to their probability of belonging to one category or another [12, 15, 33, 34]. To validate the resulting classification and to verify if a coverage map generated from remote sensing data is significantly better than a map generated by the random assignment of tags to areas, an error matrix was used according to the global accuracy and the Kappa index to know the reliability level of the predicted map; [35]. The IDRISI Selva 17.0 (Clark Labs, 2015) and Quantum Gis 2.4 software (QGIS Development Team, 2015) was used for this procedure.

In the second procedure, the Normalized Difference Vegetation Index (NDVI) was used to differentiate between plant covers. This index looks for the relationship between the red band (650 nm) and the near-infrared band (700–1300 nm), which correspond to bands 3 and 4, respectively [36]. The index has been widely used as an indicator of plant vigor that is related to the proportion of photosynthetic radiation absorbed and reflects the activity of chlorophyll in plants [27, 37].

Based on the information generated by the NDVI index in Landsat images, an analysis was made to differentiate, estimate map and zonify the vegetative cover corresponding to the classification. The hypothesis followed was that it is possible to establish a relationship between the spectral characteristics and the vegetation covers.

In the third procedure, the multivariate analysis included Principal Component Analysis (PCA) and maximum likelihood by cluster analysis, which help to identify the data set structure and to find the relationship between them, so that important information can be selected and the number of variables reduced [6]. Maximum likelihood analysis indicates groupings of samples by linking the similarities between them and shows the overall similarity of the variables within the dataset [38, 39]. The variables subjected to the PCA were soil related as K, Na, Mg, Ca, N, EC, OM, pH, CEC [40], as well as topographic characteristics gathered from the digital elevation model (DEM). We excluded the minimum and maximum temperatures in the PCA, because they are essential variables in the spatial configuration of the landscape and therefore they were incorporated directly in the subsequent cluster analysis for each of the proposed years. The components extracted from the PCA were used in the maximum likelihood analysis.

Cluster analysis made it possible to group the territory into different landscape units, the resulting clusters are not necessarily continuous in space, and may be present within a larger cluster because their characteristics differ. The mean and standard deviation values of all objects that belong to each group were used to represent the variability of each landscape unit. These kind of groups with greater variability include a greater number of different objects than those with less variability, which are more compact [38].

The number of clusters chosen remains a subjective process, generally based on prior knowledge of the territory, assumptions and practical experience. Therefore, the selection of the variables is a decision based on the expert’s knowledge of the territory. The maps of the landscape units for the scenes were the result of the combination of the delineation of homogeneous areas in terms of physical-geographic variables and the maps of Land Use and Vegetation.

3.1 Land use and vegetation

The supervised classification showed that the highest overall accuracy for the scenes of 1983, 2000, 2011 and 2014 corresponds to the bands with values of 90.48, 92.06, 96.83 and 92.91 respectively. On the other hand, for the scenes of 1990 and 1995 the multinomial regression had values of 92.06 for both scenes (Table 1).

The kappa coefficient had the same precision behavior. According to the bands analysis the values 0.87, 0.89, 0.95 and 0.90 were obtained for the 1983, 2000, 2011 and 2014 scenes, whereas for 1990 and 1995 scenes, with the multinomial regression, we obtained the 0.89 value for both scenes (Table 2).

The historical analysis of the change of land use and vegetation for each of the scenes, provides a spatial description of its behavior. In general, the predominant vegetation cover was pastureland, in 1983 covered the largest area, then decreased in 1990 and from that year showed an increase to 85,919.34 ha corresponding to 32.40 % of the total basin area. Cropland remained unchanged since 1990 with an average of 29.41 % of the total area, the subtropical scrub declined to 14.93 % of the basin area and human settlement continued growing from 1983 to 2014 to reach the 1.72 % of the basin area (Table 3).

The cropland increased 12,906.90 ha (from 64,028.30 ha in 1983 to 76,935.20 ha in 2014), the forest area decreased by 6,680.09 ha, the subtropical scrub decreased by 11,969.15 ha, pastureland increased by 2,843.93 ha and human settlements increased 2,536.34 ha (Table 4).

3.2 Delineation of landscape units. Reduction of the dataset with the PCA

The use of the soil cover data extracted from satellite images and superimposed with the physical-geographical information resulting from the multivariate analysis was an efficient tool for the delineation of landscape units for the proposed years. In all scenes, the PCA constructed a two-component model with its own values > 1, these two principal components were sufficient to explain the greater variance of the original variables (> 90 %) (Annexed 1). Similarly, in all scenes the main variable associated to the first principal component was pH (0.354 on average), the following variables were Mg, then Na and finally CEC (0.340, 0.342 and 0.339 on average, respectively). The second principal component was related to organic matter (OM; 0.559 on average) (Table 5; Annexed 1).

The generation of the vegetation cover map for each scene was used as the basis for the classification of the landscape units in combination with the set of environmental variables of climate, soil and topography [8]. The analysis grouped those environmental characteristics which tend to occur together, thus the results of the cluster for the 1983 scene showed three landscape units, for the scenes of 1990, 1995, 2000 and 2011 resulted in nine scenic units and for 2014, eight scenic units (Fig. 3, annexed 2). Quantitatively there were transformations that took place in that selected period, the pixels that make up each group clearly constitute zones with similar physiography, the resulting landscape units were similar to the generated vegetation maps.

Figure 3

Land Cover VS Landscape units.

4.1 Landscape units with homogeneous characteristics

The achieved accuracy in the classification and general kappa index statistics are acceptable according to [41], accuracy assessment reports require a general classification accuracy of about 80%.

The conglomerates allowed us to define landscape units with homogeneous characteristics [44], aspect that provides a general understanding of the dynamics of the undergone changes in the basin [12]. The cluster analysis clearly defined zones with similar characteristics; the cells that created each group tend to occur together due to spatial autocorrelation in the original variables used [42].

For example, the greatest heights in the basin are located in the southeast and the lowest in the northwest, where climate, slope and vegetation conditions are different [40]. The landscape units in the greatest heights presented mainly the Andosol soil type, the greatest rainfall, OM content, acidic pH, lower maximum and minimum temperatures, forest cover and subtropical scrub. On the other hand, the central and northwest part of the basin with lower heights had Vertisol soil type, lower amounts of OM, alkaline pH, higher maximum and minimum temperatures, and agricultural and pasture cover. All the above provided elements to describe the diversity of conditions in the landscape and to visualize the patterns of territorial change [12].

In areas that share identical environmental conditions, the result of the delineation of landscape units may be due to differences in the development of political and socioeconomic factors, leading to the formation of markedly different landscapes [43]. However, in our study, the different landscape units are the result of topographic conditions, climate, soil type and vegetation cover, as well as the influence of political and socioeconomic factors.

4.2 Emporal change of the landscape units

The chronological delineation of landscape units in some way reflects the pressure exerted by socio-economic activities tending to exploit the resources which are ongoing today in many regions of the world [44]. In 1983, there was little modification of the vegetation cover, aspect that determined the presence of only three landscape units, each consisting of wooded area, subtropical scrub, pastureland and, to a lesser extent, cropland (Figure 3). In contrast, the scenes of 1990, 1995, 2000, 2011 and 2014, showed a significant increase in the area of crops, initially in zones close to human settlements and then extending to other areas. As consequence, the soil changed in terms of nutritional content as well as a greater break in the continuity of the vegetation cover, and the landscape units increase to nine until 2011 and eight by 2014. It is estimated that 75% of the basin experiences excessive soil erosion, which is directly related to the change of land use and vegetation cover, and the delineation of landscape units. In other words, if the land use and vegetation map is superimposed to the landscape map, a pattern of delineation is observed guided mainly by the vegetal cover.

The expansion of cropland, pastureland, and human settlements affected mainly the high altitude zones where the wooded area and subtropical scrub were dominant. The increase in croplands is, in general, the most evident transformation observed in the period analyzed and is due to an important impulse to modernize agriculture from the middle of the 1980s [45]. This aspect not only affected the agricultural surface, but also the tradition of local cultivation with the exchange of grains and some vegetables to berries and avocado. The berries and avocado plantations have intensified since the arrival of transnational corporations that require intensive agriculture from the mid-section of the basin to the highlands. This economic activity not only impacts the change of land use locally, but also regionally by the removal and dragging of soil from the upper part to the lower part of the basin, promoting erosion uphill and modifying the soil fertility and texture in the valleys [45]. In addition, this delineation of landscape units shows the loss and reduction of habitat quality in the vegetation cover, which is related to the fragmentation of the territory and has been identified as a major threat to biodiversity worldwide [46]. Some species may be restricted from moving among their habitat patches [47] and it is also considered to be an important factor leading to species extinction [48, 49]. Accordingly, the landscape of the Duero River basin has been fragmented, which makes its ecological functioning and its indigenous and cultural conservation less likely.

In addition, the units’ delineation shows a sequence that requires the application of different management and protection regimes [50]. Information on specific resources alone is not very useful, if it is not accompanied of a holistic perspective that properly manages the territory [51]. Planning and management efforts should not only be based on agricultural aptitudes, administrative issues and political or economic conveniences that often divide regions that share similar environmental characteristics. Such aspects condition the spatial configuration of the territory [43], as well as the fragmentation of social, natural and cultural units [52, 53]. The management of the territory must be done considering the basin as a whole entity composed of biotic, abiotic and cultural factors.

The delineation of landscape units shows a clear increase of cropland in areas where the physical and geographical conditions are suitable for growing berries and avocados, pastureland and human settlements also acquire greater extension so that the wooded area and the subtropical scrub decrease. The configuration dynamics of the territory showed the possibility that cropland will most likely continue to increase. In addition the boundaries of landscape units will change, some of them will merge if agricultural production and the urban extension increase, because anthropogenic and agricultural activities are the main driving forces that shape nature and landscape [54]. The change in land use that leads to the delineation of landscape units and fragmentation of the territory is related to ideological, political and market trends. Most of the distribution of crops in the study area was a result of the emergence of transnational corporations combined with those already in existence, which are in competition for the exploitation of natural resources, a situation facing many regions worldwide today [44].

This growing trend of change in vegetation cover and land use corroborates that economic forces are one of the major causes of anthropogenic land use change [43, 55]. It promotes the unequal distribution of vegetation cover, more visible to a large extent in the landscape units. Thus, wooded areas are of primary concern for the benefits they represent to keep the balance of ecosystems [56]. In the Duero River basin the wooded area decreased from 1983 to 2014 by 10.30%. Particularly, the decade from 1990 to 2000 showed the greater percentage of deforestation due to the increasing introduction of avocado cultivation with a net change area of 21.5% in the municipality of Tangamandapio and 40% in the municipality of Tangancícuaro [57]. In addition, clandestine logging, extraction of soil and firewood for the artisanal manufacture of bricks in the upper part of the basin impose a significant impact [22]. Therefore, in forest-based landscape units, the implementation of recovery and conservation programs is important, and policy management and decision-makers can execute different strategies to solve problems and predict some potential effects [58]. Although there has been a recovery of 7,148.74 ha from the year 2011 to 2014, due to reforestation activities in some areas of the upper part of the basin, deforestation is still an imminent threat.

In terms of population, there were larger fluctuations between 1980 and 1990, with an initial decrease due to migration, but with a recovery at the end of the decade [59]. In 1990 there was an accelerated population increase, and by 2011 the population doubled compared to the year 1970 [60]. This increase in human settlements is related to the change detected in the landscape, and this happened in cities where immigration related to increased employment with the arrival of multinational companies dedicated to the cultivation of berries that require intensive agriculture.