Study area

The Soconusco economic region is located in the south of the state of Chiapas, in the Mazatán, Tapachula, Tuxtla Chico, Frontera Hidalgo, Suchiate, Metapa and Cacahoatán municipalities; together, the seven municipalities have a total area of 2 032 km² and cover three river basins: those of the Coatán, Cahoacán and Puerto Madero rivers (Figure 1). The area is located in an altitudinal range of 0 to 4 100 m, including mountain ranges, hills and alluvial plains (^{INEGI, 2001}). The predominant climates are warm sub-humid, and humid temperate. The combination of altitudinal and climatic elements gave rise to a wide variety of ecosystems, including mangrove swamps in the alluvial plains; medium sub-evergreen forests in the midlands and hills, and mesophilic mountain forests and pine forests in the highlands. It also favored an optimal development of agricultural and livestock activities; coffee, mango and banana crops occupy an important portion of this territory and are among the most commercialized crops both nationally and internationally (^{Flores, 2019}).

Figure 1 Location map of the study area. 

Satellite image classification

Three land cover and land use maps corresponding to 1990, 2000 and 2015 were produced; in their construction process, first an automated classification and a photointerpretation classification were carried out.

Three Landsat images downloaded from www.glovis.usgs.gov, whose acquisition dates corresponded to the dry season (January to April), were used to obtain cloud-free scenes (Path 021, Row 051): Landsat 5 TM 10, January 1990; Landsat 7 ETM, February 15, 2000, and Landsat 8 OLI 15, April 2015. Orthophotos from 1996, a digital altitude model with a pixel size of 30 m, and high-resolution images from the Google Earth platform were used as support material. Additionally, from March to May 2015, field trips were conducted at 100 control sites, selected according to fully targeted criteria, to obtain sufficient information to define the training samples with which the classification algorithms were calibrated. The information collected at each site was as follows: 1) geographic position in UTM coordinates (determined with a Garmin eTrex20x GPS navigator); 2) type of land cover and land use; 3) terrain exposure (determined with a compass, Naturehike, modelo NH15A003-E), and 4) photographs of the site (taken with a Canon Powershot Elph 18 camera). Field data were also utilized to determine the legend of the maps (Table 1).

An object-oriented approach was used for the supervised classification, which consists of two stages: in the first stage, the image was divided into a series of regions (segments or polygons), and in the second, identifiers were assigned according to the land-cover types described in the text at the foot. For the creation of the segments, the spectral and spatial characteristics of the image were considered together; the edge algorithm of ENVI 5.1 (^{ENVI, 2015}) was used; this algorithm requires the calibration of two parameters: "Scale level" and "Merge level", which range between 0 and 100. The scale level controls the size of the segments, while the merge level allows merging similar adjacent segments.

Tests were performed with different values for the two parameters. The combination of Scale Level = 50 and Merge Level = 80 was the one that visually produced the segments of the most adequate size, i.e., segments that clearly delimited the stands by maintaining a minimum number of these.

A post-processing phase was applied to the maps resulting from the supervised classification of the images for the three years (1990, 2000, and 2015), in order to identify the types of agricultural plantations, such as oil palm, banana, and mango. In this phase, the 1996 orthophotos, Google Earth images, and data from field trips were used as support material.

In order to obtain an estimator of the level of thematic error in the maps, a stratified random sampling design was used; 100 sampling points or sites were spatially distributed over the entire study region; in each of these, the type of land cover and land use were photo-interpreted with high-resolution images from Google Earth™. This data set was different and independent from the one used in the calibration of the classifiers. The error was calculated by comparing the degree of coincidence between the sampling sites and the map generated from the supervised classification, according to the analysis methodology proposed by ^{Olofsson et al. (2014)}. The thematic error estimation was applied only to the 2015 map, as the availability of high-resolution imagery for the years, 1990 and 2000, was limited.

Characterization of land-cover and land-use change

In order to obtain change matrixes, a map overlay analysis was carried out, using only two maps derived from the supervised classification, but grouped according to the general categories described in Table 1. The change matrixes obtained correspond to the complete 1990-2015 interval.

In order to have a clearer interpretation of the relevant changes and processes over the last 25 years, three landscapes were defined according to the characteristics of the relief and the productive aspects identified in the Basin (Figure 1).

The first landscape, the Coast, corresponds to the alluvial plains located at an altitude of 0 to 10 masl, characterized by the presence of wetlands. The predominant productive activity corresponds to livestock and, to a lesser extent, to agricultural plantations. The next landscape, the Plains, comprises mainly the areas located at 11 to 200 masl, with slight slopes; most of the agricultural activities ―rainfed and irrigated agricultural crops, cultivated pastures and agricultural plantations of mango, banana and oil palm― are carried out here. Finally, the Sierra is located at altitudes between 201 and 4 100 m, where the terrain slopes are often steep; the prevailing physiography in this area favors the distribution of forest remnants, and conditions are favorable for coffee cultivation practiced as an agroforestry system or under shade.

Classification of images

The overall accuracy of the 2015 map was 92 %, an acceptable level for land cover maps. Table 2 shows user and producer accuracy levels for each of the 2015 land cover classes. Most classes were identified as having a high reliability (accuracy >= 0.78), with the exception of Coffee Plantations, which often exhibited a spectral response similar to that of Disturbed Forest, resulting in low producer accuracy. Given the low level of estimated reliability, the coffee plantations may cover a surface area almost five times larger than the one estimated by the classification; in spite of this, it was decided to keep these plantations as a separate class, given that their extension is minimal.

It was not possible to assess the accuracy of the maps for the other two years (1990 and 2000) due to the scarcity of high-resolution images; moreover, the same training samples (or most of them) were used for all three images, and, therefore, a similar level of error was assumed for all three maps.

Table 3 summarizes the results of the satellite image classifications for 1990, 2000, and 2015. As may be observed, the Crops category (irrigated and annual rainfed crops and pastures) was the predominant one. This land cover had a consistent decrease in the 1990-2000 and 2000-2015 periods, but the largest decrease was recorded in the latter, reducing its occupation rate of 45 % of the area to 38 %. A similar behavior was observed in the category that includes natural and semi-natural land cover, although in this case the highest rate of loss occurred from 1990 to 2000.

Agricultural plantations and other types of plantations exhibited a steady increase in the two periods under analysis. Regarding the last category, the accelerated growth of human settlements, which have doubled in size in the last 15 years, is noteworthy.

Processes of change by type of landscape

Figure 2 shows the spatial distribution of land cover changes for the three dates of analysis. The large growth of agricultural plantations, which includes not only an increase in the occupied area, but also a greater dispersion of the area, is noteworthy. The change in the size of the city of Tapachula, the largest human settlement in the area, is also noticeable.

Figure 2 Spatial distribution of land cover and land-use 1990, 2000 and 2010. 

The change matrixes (Table 4) are the result of overlaying maps and provide information on losses and gains in land cover, but more importantly, they provide information on the transitions or types of land cover that replace the original one. Table 4 shows these matrixes for the three production landscapes; the figures represent hectares.

Figure 3 is based on the data from the matrixes and schematically shows the dominant processes of change. On the Coast, the predominant changes were from Crops to Plantations (1 789 ha) and human settlements (371 ha). Natural vegetation exhibited losses and gains, but, in net terms, there was a decrease in these areas, which were occupied by Crops. According to Table 3 and Figure 2, the vegetation types affected were hydrophilic vegetation, mangroves and some remnants of disturbed forest.

Vegetación natural y seminatural = Natural and semi-natural vegetation; Plantaciones = Plantations; Cultivos = Crops; Asentamientos humanos = Human settlements.

The thickness of the arrow indicates the magnitude of change. Figures show the percentage contribution of each transition; the total (100 %) is the sum of all areas that underwent a change in coverage between 1990 and 2015. Only the main changes are shown.

Figure 3 Dominant processes of land cover change in the three productive landscapes, in the 1990-2015 period. 

In the plains, the productive systems with the highest economic value for the Soconusco region are concentrated: mango, banana, and oil palm plantations, as well as the main population centers. The dominant process in this area was the replacement of crops by plantations (16 026 ha).

In the Sierra, fragments with forest vegetation and coffee agroforestry systems predominate; the most reported changes were the substitution of natural and semi-natural vegetation by crops.

Land cover and land use maps

The results of the maps show several aspects to be highlighted in the changes observed in the landscapes during the analysis period: deforestation, substitution of agricultural crop areas by plantations and the expansion of urban areas.

Deforestation is an important process in the study area, although not of the magnitude observed in other regions of the country (^{Ellis et al., 2021}). However, it did affect such vegetation types as mountain mesophilic forests and mangroves, which are crucial not only for their biological diversity, but also for their key role in mitigating climate change (^{Adame et al., 2018}).

Frequently, research carried out at the international (^{Bürgi et al., 2004}; ^{Belay y Mengistu 2019}; ^{Tuffour-Mills et al., 2020}) and national levels (^{Rosete et al., 2008}; ^{Mas and Flamenco, 2011}; ^{Calzada et al., 2018};) agrees that the main processes of change are due to the conversion of forests and secondary vegetation to agricultural and livestock uses. In this region, due to its climatic characteristics, another process has also occurred: deforestation from the loss of shaded coffee agroforestry systems.

Given their moderate environmental impact, it has traditionally been considered that coffee agroforestry systems can coexist in high biodiversity regions, which contributes to their conservation. However, a fragile balance in which grain prices (^{Jurjonas et al., 2016}) and pests (^{Henderson, 2019}) trigger the loss or degradation of forest cover underpins the permanence of these systems. The growth of coffee plantations without shade may be a response to the phenomenon observed by ^{Henderson (2019)} in other regions of Chiapas, where, in order to fight rust, coffee varieties that are sensitive to this fungus have been replaced by others with greater resistance, but which do not require the shade of trees.

The replacement of agricultural crop areas with plantations was another one of the changes observed; in this sense, ^{SIAP (2020)} agricultural production statistics confirm the pattern of land use change observed in the maps. Agricultural production data show a decrease in corn area (Figure 4), which went from 22 223 ha in 2003 to 18 048 ha in 2014. A similar trend is observed in soybean production. While pastures grew strongly, from 12 378 ha to 18 149 ha, perennial crops in general showed a slight increase in their extensions, except for coffee and cocoa (^{SIAP, 2020}).

Source: ^{SIAP (2020)}

Figure 4 Surface area of annual (a) and perennial (b) crops. 

This pattern of reduction of areas devoted to the production of staple crops seems to be related to the phenomenon, observed at the national level, of increasing dependence on the importation of basic grains as an effect of the Free Trade Agreement (^{Valencia et al., 2019}).

Finally, the expansion of urban areas observed in the maps also coincides with census data, since the region experienced moderate population growth from 1990 to 2010, with a population growth rate of 1.63 %, which represented 346 000 inhabitants in 1990, and 477 800 in 2010 (^{INEGI, 1990}; ^{INEGI, 2010}), concentrated in the urban areas of the country (Figure 5). At the cartographic level, which was reflected in an expansion of human settlements, the area of this category tripled from 4 000 to 12 000 ha, mainly because of the expansion of the city of Tapachula.

Source: Population and Housing Censuses (INEGO, 1990, 2000, and 2010).

Figure 5 Historical population growth in the municipalities of the study area. 

Productive transformations by type of landscape

The dominant process on the coast was the loss of natural and semi-natural vegetation due to the expansion of agricultural crops and plantations. Agricultural production data obtained from ^{SIAP (2020)} confirmed a strong growth of cultivated pastures in the municipality of Suchiate (from 600 to 5 000 ha) and a strong deforestation in the period studied. The rapid expansion of these activities into hydrophilic vegetation areas is a cause of concern because it increases the risk of deterioration or deforestation of the mangroves (^{Barrasa, 2017}).

In the Plains, the dominant process of change was the conversion of agricultural crops into agricultural plantations (banana, mango, oil palm). Oil palm is an attractive product for farmers because of the growing demand in the food industry market, the increase in prices and the public programs that support its establishment (^{Fletes-Ocón and Bonnano, 2015}; ^{Castellanos, 2018}).

At the national level, Chiapas is the state with the largest area cultivated with oil palm, with a little more than 45 000 ha (^{SIAP, 2020}), located in the economic regions: Rain Forest, Coast, and Soconusco; also, there are 11 oil extraction plants, six of them located in Soconusco (^{Castellanos, 2018}).

Apparently, the deforestation processes observed in the mountain range are due to the conversion of forest areas into annual agricultural crops (^{Cortina-Villar et al., 2012}) and to the transformation of shade coffee plantations (agroforestry system) into unshaded coffee plantations (^{Henderson, 2019}). According to the maps generated in this study, in 1990 there was an area of 200 ha of unshaded coffee, which quadrupled in 2015, ―a figure that may be even higher due to the high spectral confusion registered in this category (Table 2).

Evidence cited in other papers points in the same direction. ^{Henderson (2019)} and ^{Jurjonas et al. (2016)} find that coffee agroforestry systems are highly susceptible to price variations and diseases. In recent years, coffee rust mainly affects agroforestry systems (^{Henderson, 2019}); therefore, one of the strategies promoted to deal with it, the use of coffee plant varieties resistant to the pathogen, but that do not require the shade provided by the forest canopy (^{Jha et al., 2011}). This has resulted in a loss of tree cover, which coincides with the decrease in disturbed forests observed in this study.