Influence of Climatic Seasonality on a Survey of Land Use and Cover in the Semi-arid Region

The dynamics of land use and land cover in watersheds of the Brazilian semi-arid region is not only influenced by human action, but also by the climatic seasonality of the region. Knowledge of the relationship between surveys of land use and land cover using geotechnology and the climatic seasonality of semi-arid regions is necessary. The aim of this study was to map and classify land use and cover in the watershed of the Orós reservoir (WSOR) with the help of geotechnology, and to identify the influence exerted by the climate on variations in the area of each class. The survey of land use and cover was carried out by means of the MAXVER method of classification of images from 2003, 2005, 2008 and 2013 from the LANDSAT 5 and LANDSAT 8 satellites. The areas of each class displayed dynamics influenced not only by human action but also by such factors as climate, topography and plant physiology. Years with high rainfall favoured classes such as thin scrub and dense scrub, with the opposite being seen in years considered as dry, when there was a considerable increase in areas of the anthropogenic class. Changes in the areas are caused by alterations in the deciduous vegetation; with leaf-fall during the dry season, these areas come to have the spectral response of areas with similar characteristics to the anthropogenic class. More-elevated regions favoured the presence of the dense-scrub class due to the microclimate and to the greater difficulty such areas present to human action.


Introduction
The conditions of forest cover are under constant change that may be subtle or abrupt, the result of both natural and anthropogenic forces (Hayes & Cohen, 2007).The seasonality of regions with a tropical semi-arid climate and two well-defined seasons, one dry and one rainy, results in major changes in the landscape of dry forests.Such changes are due to the poor distribution of rainfall during the year, where different rainfall indices are seen for the two main seasons, dry and rainy, resulting in a rapid response to environmental change (Batista & Santos, 2013).
In this way, climate is seen as the conditioning element of environmental dynamics, since it exerts a direct influence on both physical and biological processes (Silva et al., 2010).Verbesselt (2010) states that changes in ecosystems can be divided into three classes: (1) seasonal change, driven by the interaction of annual temperatures and precipitation impacting on plant phenology; (2) gradual change, such as interannual climate variability, a gradual change in land management or land degradation; and (3) abrupt change, caused by such disturbances as deforestation, urbanisation, floods and fires.
With regard to changes in ecosystems, it should be noted that the association between population growth and the development model adopted in the semi-arid region of Brazil has resulted in an increase in the pressure on natural resources due to changes in land use and cover in the region.Studies have demonstrated the negative impact that the degradation occurring over the years in the semi-arid region causes to the Caatinga Biome (Coelho et al., 2014;Pereira Filho, Silva, & CÉzar, 2013;Silva, Lima, & Mendonça, 2014;Vieira et al., 2013).
Activities such as deforestation for timber and the inclusion of new agricultural areas, intensive agriculture, overgrazing by livestock, and the expansion of urban areas, are examples of human intervention, which when carried out in a disorderly fashion result in degradation of the Caatinga.Such action has a negative effect on the  Orós reservoir (Toledo, 2013).The Orós reservoir is the largest in the river basin and the second largest in the state, with a capacity of 1,940 hm 3 , and a watershed of 202.11 km 2 (Cogerh, 2015).
The terrain of the WSOR features plateaus and tablelands as well as depressions.The highest altitudes in the basin corresponds to two geomorphological units in the area: the Sertanejo Plateau, where the Araripe Plateau is inserted (in the south of the basin), and residual plateaus that are formed by ridges remaining inside the basin (Lopes, 2013).From SRTM data, it was possible to verify the elevation of the region as varying from 183 to 951 metres.

Remote Sensing Data
Images were used from the Landsat 5-TM and Landsat 8-OLI satellites, acquired from the Image Generating Division (DGI) of the National Institute for Space Research (INPE), and from the United States Geological Survey (USGS) respectively.To cover the entire area of the WSOR, it was necessary to obtain scenes from orbits 217 and 218, points 64 and 65.
Due to the high availability of images, selection criteria were adopted, where the selected images should preferably have the least possible cloud cover, and have been produced in the second half of the year to minimise the effect of rainfall on vegetation in the region.However, it was not possible to obtain all the scenes free of clouds, and it was necessary during the post-classification process to quantify the clouds as well as any shadows they caused.
In order to identify the effect of rainfall occurring during the previous months on image classification, total rainfall was calculated for each of the three months preceding those in which the images were generated.For the period of the study, 2005 stands out for having no rainfall during any of the months surveyed.Table 1 shows the total rainfall for each month, highlighting in bold type the month in which the images were generated.Considering the above, the flowchart shown in Figure 2 illustrates the method adopted in producing the land use and cover maps for the WSOR.
Atmospheric correction of the images from the Landsat 5 satellite was carried out first.The atmospheric correction of satellite images allows characteristic reflectance values for the target to be obtained (Antunes et al., 2003).To do this, the X-6Scorr software was used (Montanher & Paulo, 2014) To evaluate the quality and precision of the classifications, the accuracy and Kappa index were used, both obtained from the confusion matrix generated with the ENVI 4.7 software.
The Kappa index is used to validate image classifications made using remote sensing techniques.Table 2 identifies the values and quality level of the classification carried out as per Landis and Koch (1977).Source: Prepared by the author.
After classification of the images, the raster files containing information on the classes of land use and cover in the basin were transformed into vectors.These vectors were then converted into shapefiles, so that the area of each class could then be calculated using the ArcMap 9.3 software.Four shapefiles were obtained, containing complete information of the areas of each class of use and cover in the WSOR for the years under analysis.All the information from the areas was exported to spreadsheets to be quantified.

Climatic Seasonality in the WSOR
The rainfall of the region for the years under study (2003, 2005, 2008 and 2013) ranged from 453 mm to 932 mm, with an average of 635 mm.Table 3 shows the total annual rainfall, total rainfall during the rainy season (January to May), and the percentage of total rainfall in the watershed during the rainy season, for each year considered in the study.The rainy season mentioned above refers to the period of predominantly more rains in the study area, which includes January, February, March, April and May.Source: Prepared by the author.
Analysing this information it is clear that a large part (at least 71%) of the annual rainfall depth in the watershed for all the years surveyed is concentrated from January to May (rainy season).It is worth mentioning that the year with the lowest total annual rainfall, 2013, also had the lowest percentage of incident rainfall during the rainy season, 71%.Although the total rainfall depth occurring that year was lower, the rainfall occurred more evenly throughout the year (Figure 3D).The opposite was seen in 2008, when the greatest total annual rainfall for the time series occurred, 932 mm.That year presented the highest percentage of incident rainfall during the rainy season, 96%, representing 891 mm.
As for rainfall during the rainy season, Alves, Souza, and Repelli (1998) found that on average during extremely dry seasons, rainfall during the rainy season is less than 40% of that expected, and changes of more than -40% may occur.Further, according to those authors, each year of drought sees a different configuration to the rainfall distribution in the various regions of the State of Ceará, i.e. for each year of drought, different areas of the state are affected by the phenomenon.
There is a strong interannual variability to the rainfall in the region (Figure 3) associated with drought, where the maximum monthly rainfall remained below 100 mm (Figure 3D).It can also be seen that for March 2008 (Figure Source: Prepared by the author.
It was found that the classes with the highest CV were those that suffered the greatest influence of the rainfall regime in the region, except for the cloud, shadow and temporary-wetland classes.The interannual variation in total value for the areas of cloud and shadow resulted in the CV of these areas presenting the highest value among all the remaining classes.This happened because these two classes were present during only three of the four years of the study in addition to the large presence of and shadow in 2008, which favoured an increase in CV for these two classes.
The temporary-wetland class showed a gradual decrease in area for each year under analysis, resulting in the high value seen for CV.The decrease in area in 2008 was expected due to the high rainfall index that year causing the watersheds of the reservoirs to rise, and reducing the area of this class.In 2013, there should have been an increase in area, but the opposite, a decrease, was seen.This class can be considered as the only class that was not directly influenced by the rainfall regime of the watershed, as discussed below.
The dense-scrub class had the lowest value for CV (12.73%) among all the classes under analysis, suffering the least influence of the rainfall in the watershed during the period analysed.It maintained a trend for constant growth throughout the period analysed, even between one year considered rainy and another considered dry (2008 and 2013 respectively).The opposite was seen for the dense-scrub class, which displayed a greater value for coefficient of variation, 36%.The area of this class was greatly reduced in 2013, a year considered dry, noting that it suffered with the low rainfall in the watershed that year.In addition, Coelho et al. (2014) emphasise that areas of caatinga vegetation can show a progressive substitution by such activities as agriculture or animal farming.
Another class of use and cover in the basin also influenced by the rainfall regime was riparian forest.Again, the influence of two years of extremes in rainfall, 2008 and 2013, can be seen.The rainfall depth in 2008 favoured the recovery of this class of area in the watershed, with the opposite happening in 2013, the year in which there was a drastic reduction in the area of this class.The degradation suffered by these areas in the watershed of the Upper Jaguaribe should also be taken into account, a fact already verified by Sousa, Melo, and Da Silva (2013), who observed the removal of riparian vegetation in several stretches along the River Jaguaribe, which, according to the authors, can cause silting and erosion of the river margins.
There was a marked drop in the area of the anthropogenic class in 2008, which favoured an increase in its CV value.This reduction in area is due to the response of the vegetation in the watershed to the rains that occurred during that year, which filled the previously exposed soil with vegetation or with types of usage characteristic of the anthropogenic class.In this way, an area previously classified as anthropogenic was classified that year as thin scrub or dense scrub.However, there is strong human pressure on the natural resources of the semi-arid region of the Northeast; thus, Ribeiro et al. (2008) affirm that anthropogenic action has resulted in the continuous degradation of natural resources, causing irreversible damage to the environment of the semi-arid region.
The water class had the third highest coefficient of variation among all the classes of soil use and cover in the watershed, 41.98%.Comparing the rainfall regime of 2003 and 2013 (Figure 3), it can be seen that even though there was less rainfall in the watershed in 2013, with 453 mm, this was well distributed throughout the year.In 2003, the total rainfall was 618 mm, tending to increase from January to March, with an abrupt drop from April to June.The dynamics displayed by this class is explained not only by the behaviour of the rainfall in the years of study (2003, 2005, 2008, and 2013), but also in the preceding years.This was because the incident rainfall in the watershed during the preceding years influenced a greater recharge of the reservoirs, causing the volume of water stored during the years of the study to reach a higher level.
In 2003, a high value was seen for the anthropogenic class, this being an indication that the WSOR displays a high rate of substitution of native plant composition.Silva, Lima, and Mendonça (2014) carried out a study in the semi-arid region of the Northeast, with the aim of mapping changes in plant cover in the sub-basin of the Espinharas River from 2000 to 2010.Those authors found that most of the sub-basin under study, 80% of the total, presents marked anthropic activity with highly degraded stretches, showing such areas to be the result of a framework of long-established degradation in the region.
Further in relation to human pressure exerted in areas of caatinga, Silva, Lima, and Mendonça (2014) state that practices such as cutting down vegetation to produce cuttings, charcoal and firewood for domestic or commercial use, and to prepare the area for agriculture, favour a reduction in caatinga vegetation in the semi-arid region.Pereira Filho, Silva, and Cézar (2013), point out that overgrazing by livestock, and itinerant agriculture have been identified as the main factors in the degradation of Caatinga ecosystems.
In relation to the classifications made, and the data obtained, several factors may have acted on the dynamics displayed by the antropogenic, thin-scrub and dense-scrub classes during the study period, these being basic to the behaviour of these classes in 2008.The presence of a large number of clouds in 2008 negatively influenced the total of the anthropogenic class that year, underestimating its total area.In this way, variations in the cloud class from one year to the next can influence certain classes positively or negatively.
Accordingly, the absence of clouds positively influenced the data obtained by Sousa et al. (2007) in a study carried out with the aim of comparing land use and occupation in the watershed of the Upper Piauí during 2004 and 2005 for the rainy and dry season respectively, through satellite image processing.In that study, it was found that the anthropogenic class occupied about 10.8% of the area of the watershed in 2004, increasing to 18.1% in 2005, showing an increase of 68%.According to the authors, this significant increase occurred due to the images of 2005 being practically free from clouds, contributing to the total of that class.In the above study, the cloud class occupied 18.1% of the watershed in 2004 against 1.1% in 2005.
Another factor worth noting in 2008 is that the total rainfall in the watershed, 932 mm, had a positive influence on the increase of both the thin scrub and dense scrub classes.For the response of vegetation to the rainfall regime in semi-arid regions, Cunha et al. (2012) state that plant cover in dry areas is highly correlated with rainfall.This is the reason for expecting that the years showing a higher rainfall index before passage of the sensor should be exactly those that show a greater increase in plant cover.Further, according to that author, regions with a semi-arid climate have vegetation that is more susceptible to change, responding quickly to the presence or absence of rainfall.
Similarly, Batista and Santos (2011) state that in semi-arid regions, studies of changes in land use and cover using satellite images over fairly short periods, show that seasonality may have a greater influence on changes in classes of land use and cover in a given area, rather than the temporal distance between the years under analysis.
Deciduous vegetation, being one of the main characteristics of the Caatinga, directly influences identification of the classes of land use and cover in semi-arid regions.When this vegetation loses its leaves during the dry season, the soil remains exposed for a certain length of time, altering reflectance values in the band of the electromagnetic spectrum, and displaying the spectral behaviour of bare soil.In this way, an area with a predominance of caatinga vegetation may be classified as exposed soil during the dry season, and be included in the anthropogenic class.
While observing changes in class area between the years of study, Sousa et al. (2007) state that native vegetation consisting of mostly hyperxerophilic caatinga, which has the characteristic of leaf-fall during the dry period, causes these areas to reflect the characteristics of exposed soil in the image, i.e. they may display in a characteristically anthropogenic colour tone.
The inverse can be found in studies such as that of Batista and Santos (2011), who analysed land use and occupation in the town of Teofilândia in the State of Bahia.The authors verified changes to the classes between years, where areas of the exposed-soil class were classified the following year as pasture and caatinga.
According to the authors, these changes occurred because the vegetation that was dry and senescent, and previously classified as exposed soil, sprouted and flourished, turning areas of vegetation green, and changing reflectance values due to the action of chlorophyll in the plants.
However, it must be emphasised that areas previously classified as anthropogenic can undergo natural regeneration and later be classified as thin scrub, the opposite also possibly occurring.There is therefore a   Although the above factors may be related to decreases in the area of the wetland class during the study period, it should also be pointed out that this trend may be related to gradual abandonment and the lack of interest on the part of producers to continue this practice.Other more profitable and less laborious activities or even government aid may be helping to reduce this practice in the WSOR over time.

Accuracy of the Classifications
The confusion matrix to evaluate the accuracy of each classification was generated by the ENVI 4.7 software from the Regions of Interest (ROI) defined during the image classification stage, the result being the accuracy and Kappa index.Table 5 shows the values for accuracy and Kappa index obtained for the classification of the 4 years (4 images) used in this study.The values for accuracy show that between 95.69% and 99.71% of pixels were classified correctly.All the classifications made were considered excellent according to the Landis and Koch (1977) classification, as the values for the Kappa index were within the range of 0.8 to 1.

Conclusion
The climatic seasonality in the watershed had a direct influence on the classes of use and cover.Higher rainfall in the watershed favoured an increase in the thin-scrub and dense-scrub classes, the opposite being seen in the areas of the anthropogenic class.
The changes that took place during the period under analysis are not only the result of human intervention in the environment, but also of climatic factors.The time the images were generated should be taken into account, so that the influence of the climate on image classification is avoided or reduced.
Land use in the region was seen to be inadequate, considering the high degree of degradation of the areas of riparian forest in the watershed.
Figure 7. repared by the

Table 1 .
Rainfall behaviour in the three months prior to the month the images were obtained

Table 2 .
Quality of classification according to the Kappa index

Table 3 .
Values for total annual rainfall, total rainfall during the rainy season, and percentage of total incident rainfall during the rainy season in the WSOR

Table 4 .
Total area (km²) and coefficient of variation (%) for each class of use and land cover in the WSOR

Table 5 .
Values for Accuracy and Kappa Index obtained with the classifications Source: Prepared by the author.