Temporal and phenological prof iles of open and dense Caatinga using remote sensing: response to precipitation and its irregularities

Janisson Batista de Jesus · Tatiana Mora Kuplich ·Íkaro Daniel de Carvalho Barreto · Cristiano Niederauer da Rosa ·Fernando Luis Hillebrand

Abstract Caatinga is a typical biome of Brazil’s semiarid regions and subject to climate changes. Research is needed on the relation of its features to climate events. This study analyzed the inf luence of rainfall and its irregularities in open and dense woody Caatinga vegetation. Phenological curves were generated by means of Normalized Difference Vegetation Index (NDVI) time prof iles in the Grota do Angico Conservation Unit study area in Sergipe State.Rainfall data from 2000 to 2018 were collected and phenological curves generated using various estimate methods that produced the following variables: [start of season, end of season, peak of season position, length of season, mean growing season and maximum seasonal]. Rainfall showed a standard intra-annual behavior, with inter-annual variations related to irregularities inf luencing Caatinga response.Dense Caatinga vegetation had higher values of NDVI in all periods, even in anomalous years compared to open Caatinga, in addition to having longer leaf coverage over the year, with an anticipated start and a more extended seasonal end. The analysis of the rainfall regime made it possible to assess its inf luence on the Caatinga and phenological prof iles proved to be fundamental to understand periods of physiological change of open and dense Caatinga. These results indicate that dense Caatinga maintains physiological activity longer, which may be associated with greater moisture maintenance in a semiarid region. In addition, because it has a greater leaf cover for longer periods, the soil may be preserved and maintain its characteristics longer, reducing the Effects of desertif ication. The results may be associated with the type of forest management and conservation in this region. The total or partial suppression of individual remnants of Caatinga should be avoided, since the most open areas have lower photosynthetic capacity, aff ected to a considerable extent from the Effects of adverse climatic conditions. Additionally, open Caatinga has a reduced capacity for regenerating naturally and its use by communities in this semiarid region should be limited.

Keywords Semiarid · Tropical dry forest · NDVI ·MODIS · Phenopix

Introduction

In South America, there are three large semiarid zones, one is the Caatinga biome in the semi-arid interior of northeastern Brazil. It is characterized by high average annual temperatures, and by high spatial and time variations in rainfall concentrated in two to four months, with strong evaporation and poor water inf iltration into the soil (Nóbrega and Santiago 2016).

Caatinga is a thorny vegetation of grasses and woody shrubs that lose their leaves in the dry season. Droughts also regulate the spatial distribution of the vegetation mainly due to the intensity and duration of water def iciencies (Cole 1960).

Forest ecosystems suff er the Effects of climate changes,and vegetation phenology is an indicator of these changes.Vegetation is directly associated to climate variations, and it is therefore critical to understand forests response to extreme climate events which vary in diff erent climate zones (Zheng et al. 2018). In the Caatinga biome, the high dependence on rainfall and its irregularities can have negative impacts on the vegetation (Salimon and Anderson 2018). This makes understanding Caatinga seasonal behavior and its relation to climate over time very relevant.

A signif icant characteristic of Caatinga vegetation is its deciduousness, the shedding or loss of leaves at the end of the growing season. This abrupt change in leaf density is a phenological marker associated with seasonal physiological events and an adaption mechanism to climate conditions.Caatinga deciduousness is concentrated at the beginning and end of canopy photosynthesis and can be detected by remote sensors (Hmimina et al. 2013).

Optical remote sensing images are one of the methods used in forest mapping and are widely used because of its ease of processing and interpretation (Qin et al. 2017).MODIS (Moderate Resolution Imaging Spectroradiometer)data is widely applied in phenological studies of diff erent types of forests (Adami et al. 2018 ; Adole et al. 2018;Chakraborty et al. 2018; Pastor-Guzman et al. 2018; Testa et al. 2018). In the Caatinga ecosystem, this has been used in vegetation studies (Formigoni et al. 2011; Redo et al. 2013;Abade et al. 2015; Schulz et al. 2017). However, there has not been research on phenological variables based on seasonal prof ile of vegetation associated with rainfall.

The main climate event to inf luence the rainfall regime in Brazil’s Northeast is the El Niño-Southern Oscillation(ENSO) which has two phases: El Niño (warm phase) and La Niña (cool phase), characterized by changes in surface temperatures of the Pacif ic Ocean. The Southern Oscillation Index (SOI) is an indication of the development and intensity of El Niño or La Niña (Freire et al. 2011). They are both important to the Northeast as El Niño abnormally warms the Equatorial Pacif ic Ocean waters, resulting in periods of drought while La Niña brings heavy rains (Freire et al. 2011).

Therefore, because of the relevance of associating Caatinga phenological characteristics to climate, this study analyzes rainfall patterns and their inf luence on open and dense Caatinga vegetation, assessing rainfall anomalies and their Effects on forest appearance. Phenological curves were obtained by means of NDVI (normalized diff erence vegetation index) time prof iles for the full time series as well as in anomalous years,testing the most adequate method of estimation for each period in the biome in Grota do Angico Conservation Unit in Sergipe.

Materials and methods

Study area

The study area is a remnant of a Caatinga biome in Sergipe state, located in the Alto Sertão region in the Grota do Angico Natural Monument Integral Protection Conservation Unit with an area of 2183 ha in Canindé de São Francisco and Poço Redondo counties (Fig. 1). Its northern border is the São Francisco River which divides Sergipe and Alagoas states.

Caatinga vegetation is categorized as Tropical Dry Forest,formed by two types of forests: dense hyperxerophilic forest and open forest in diff erent regeneration stages with presence of pastures and abandoned areas (Ribeiro and Mello 2007;FAO 2012). Two Caatinga areas were selected within the Conservation Unit, one with dense woody cover, the other open with sparse tree cover based on stand density of the remnant according to Chaves et al. ( 2008) (Fig. 1).

The region is classif ied as Bssh’ semi-arid, very hot, with seven to eight months without rain, and with an average 483.9 mm (Sousa et al. 2014). It is considered an area under risk of desertif ication (Serviço Florestal Brasileiro 2018).

Data analysis

Caatinga areas were sampled in NDVI data with pre-f iltration in coordinates: −37.6844 and −9.65729 for the dense vegetation fragment, and −37.65521 and −9.69271 for the open one.MODIS/Terra sensor MOD13Q1 product data were used, with 250 m × 250 m pixel and temporal resolution of 16 days, available in Web SATVeg tool (Embrapa 2018).

NDVI data were collected from February 2000 to September 2018. Monthly rainfall from January 2000 to September 2018 was obtained from the Tropical Rainfall Measuring Mission 3B43 V6 product (INPE 2018).

Based on the NDVI data, the physiognomies were analyzed throughout these intervals. Data referring to ENSO (El Niño-Southern Oscillation) were made available by the National Weather Service/NOAA from the Center of Weather Forecast,in order to associate rainfall and NDVI anomalies. These were calculated using monthly values according to standardized Eq. 1 (Ivits et al. 2012; Bombardi and Carvalho 2017):

where x is the monthly observed value;the monthly average for the historical series; and σ the monthly standard deviation for the historical series.

Fig. 1 Location of the study area, highlighting Sergipe State in the Northeast region, the semiarid region of Sergipe counties, and the Grota do Angico Conservation Unit. The area formed by the open and dense Caatinga fragment is highlighted with true color image (RGB:4/3/2) from Sentinel-2 (ESA 2019) of post-rain period (27-12-2018)on the left image, and drought period (28-10-2018) on the right, illustrating the vegetation state under these water conditions (on the left illustrates the vegetation after a rainy event, and on the right in the dry season)

The seasonality of each vegetation form (open or dense caatinga), and changes in physiological phases for the time series were made. Special instances of adverse rainfall were considered for selected years of anomalies (one year of drought and one rainy year) to check responses of the phenological curves of these forest forms.

The phenopix R package (Fillippa et al. 2016) was used,with the following estimate methods: Beck, Elmore, Klosterman and Gu, with 500 repetitions and by selecting the estimator with a lower root-mean-square-error (RMSE) for each period to obtain phenological curves and their variables. For a seasonal prof ile, the following were assessed: start and end of season, peak of and length of season, mean growing season and maximum seasonal.

Results

The rainfall regime of the area had a standard pattern with regards seasonal distribution for the period January 2000 to September 2018 with the lowest levels from April to July,considered a winter period in the region (Fig. 2). Occasions of higher rainfall were observed in May 2005 (221.0 mm)and 2009 (189.6 mm), followed by April 2010 (184.2 mm)and 2017 (176.9 mm).

In the months after the rainy season, there was a sharp reduction in accumulated monthly rainfalls, with low levels mainly from November to February, including an absence of rain. However, rain occurred throughout the summers:December 2000 (102.2 mm), January 2004 (370.5 mm, the highest amount of the period), February 2007 (179.9 mm)and December 2008 (74.9 mm).

Fig. 2 Distribution of accumulated monthly rainfall and NDVI for open and dense caatinga throughout the time series

Regarding the Caatinga in this study, a regular response was observed over the period following the region’s rainfall regime for both open and dense cover types (Fig. 2). Nevertheless, the response of both were similar in terms of NDVI f luctuation, both intra-annual and over the time series.

Dense Caatinga had higher normalized diff erence vegetation indices (NDVI), reaching values close to and above 0.9 on October 6th 2011, with higher values in both rainy and drought seasons, as well as in years with higher and lower rainfall. The open Caatinga had low NDVI values, usually in periods of low rainfall with values less than 0.3, reaching 0.15 January 1st, 2018.

The temporal behavior of the rains in the studied area indicates that, in some years, there was a large variation in the amount of rain, having more years of rain and other years of drought, which could be considered anomalies. NDVI of Caatinga vegetation were directly associated with rainfall,which was also observed with events of rainfall anomalies(Fig. 3).

Therefore, in years with above normal rain, Caatinga vegetation also displayed higher canopy foliage, while in drought years, the vegetation was well reduced in NDVI and in foliage cover was sparse (i.e., June 2003; April 2016 to 2017; July 2018). In spite of the Effect of rain in Caatinga development, the vegetation had an NDVI threshold (maximum limit) in years with surplus or scarce water.

In years with signif icant ENSO anomalies, vegetation ref lected atypical periods. Years of strong El Niño from October to December 2014, from June to August 2016,from May to July 2002 and from January to March 2005,a sequence of rainfall anomalies was observed, such as low NDVI. However, in years of intense La Niña (May to July 2010, and from April to July 2012) there was strong expression of Caatinga.

Over the years studied and for the complete time series,all methods for phenological curve estimates stood out as the most precise for any year analyzed, indicating that each could be appropriate, depending on the characteristics studied. However, the Gu method generated lower RMSE (rootmean square error) values, both for open and dense caatinga(Tables 1 and 2). The anomaly years 2010 and 2015 were chosen for phenological analysis because of their high and low NDVI (normalized diff erence vegetation index) values,respectively, with little variation over the year, which coincided, respectively, with El Niño and La Niña events.

Over the time series analyzed, the Gu method proved to be more effi cient to generate the phenological curves for both types of Caatinga, and for years of anomalies of the dense Caatinga. For the open vegetation, on the other hand, the Klosterman method was more precise with lower RMSEs. However, even with these methods being the best for the years assessed, they identif ied years without RMSE values. The Gu method had more absent data.

The RMSE estimates varied by year and the type of vegetation. For open Caatinga, the Beck and Elmore methods each had three periods where they were the best method to be applied, followed by the Klosterman method with seven.For the dense Caatinga vegetation, the Klosterman method was the best in terms of the number of periods, with the Elmore method with four periods, and inferior the Beck method with six.

Fig. 3 Rainfall anomalies and NDVI for open and dense Caatinga over the time series; positive values are anomalous years of heavy rain, negative values are drought years

Table 1 RMSE values for each phenological curve estimator method for the time series and each year assessed for open Caatinga

Table 2 RMSE values for each phenological curve estimator method for the time series and each year assessed for dense Caatinga

The phenological change curves indicate the seasonality of the Caatinga vegetation for the morpho-structural characteristics evaluated, according to the methods chosen based on the highest accuracy (Fig. 4). The curves show that the time prof ile along the series had diff erences in vegetation expression values and in periods with permanent canopy.

With regards to years of irregularities, there were differences in seasonal prof iles between the open and dense Caatinga. For 2010, the dense vegetation showed a phenology with a wide canopy cover, and high and nearly constant NDVIs over a long period of the year. The open caatinga,in turn, even with a season with leaf cover, did not have the same canopy density of the dense Caatinga throughout the year, having a shorter season with canopy. In comparison with the complete series, although it had an earlier onset of leaves (earlier than 100 days), there was similarity to the peak distribution of the estimated curve.

In 2015, with low rainfall irregularities, there was a signif icant diff erence in the initial portion of the phenological curves up to approximately 160 days. Moreover, the open Caatinga had higher variations of repetitions for the estimate of seasonal rainfall distribution, with two cluster curves and greater variations in the intervals for the beginning and end of the seasons.

Fig. 4 Open and dense Caatinga phenological curves for the time series and for years of anomalies 2010 and 2015, respectively, rainy and drought; sos: start of season; eos: end of season; pop: peak of season; replications of estimator (gray)

Phenological curves showed distinct characteristics for each period analyzed, despite the similarities observed between the open and dense Caatinga. This may be observed while analyzing the curves’ variables (Table 3). Despite showing equal start and peak seasons across physiognomies assessed, there are diff erences in the end of season, length of season, season maximum value and growth average, all being higher for dense vegetation.

For the years of rainfall anomalies, in 2010, the dense and open Caatinga vegetation expression values were superior to those of 2015, with higher maximum values of seasonal and average growth. They also gave the anticipated start as well as the extended end of the season, with longer periods with a canopy cover. With regard to the seasonal peak, there is a period of almost 30 days between times of the estimated curve, however, with a higher interval in the dense Caatinga.

Similar results were observed when comparing variables of open and dense Caatinga in the selected years of irregularities, showing that dense vegetation had higher NDVI values than those of open vegetation in both periods, with a more prolonged leaf cover in the rainy year. In 2015, there was similarity as to the end of season, however the start of the season was shorter in open Caatinga. Nevertheless, the duration of foliage did not vary across physiognomies.

Discussion

There was similar rainfall distribution with regards to monthly rainfall over the year as to that found by Lundgren et al. ( 2017a). They also noted a strong spatial dependence of these climate events on the distance from the ocean, with the most distant region having the lowest rainfall. These accumulated monthly amounts were also similar to the pattern and yearly rainfall accumulation observed by Lundgren et al. ( 2017b).

For the semiarid region, the variations in rainfall and number of rainy days are considerable and much greater than in the Northeast coast and the Agreste region. The pluviometric characteristics in this portion of the sergipano Sertão indicate higher average rainfall from April to July. It is also a sharp contrast between rainy and drought periods(Silva et al. 2011), as observed in the series for Canindé de São Francisco county where the highest average monthly accumulated was due to the rainy season.

There was a signif icant variation in rainfall events in months with a low water regime (October to February),with more days without rain. There were also diff erences in rainfall amounts for some years, particularly rainfall peaks as in January 2004 (370.5 mm). According to Silva et al. ( 2011), this characteristic was seen in Sergipe statealong with Alagoas, which have extreme variation coeff icients compared to other Northeast states; the lowest variations occur in June and the highest in November or December. Furthermore, the highest average of the number of rainy days is in the Brazilian Northeast Zona da Mata and Agreste, decreasing towards the interior. This situation, along with a lower number of rainy days is characteristic of this semiarid region (Silva et al. 2012).

Table 3 Estimated values of conf idence interval (95% level)for phenological variables assessed for open and dense Caatinga for 2000 to 2018, and 2010 (rain anomaly) and 2015(drought anomaly)

In Sergipe state, the Alto Sertão semiarid region of this study, has the fewest rainy days, with high variations in rainfall. This is observed in rainfall temporal distribution with the rainiest period usually in the f irst half of the year due to the Intertropical Convergence Zone and Easterly Wave Disturbances, while the driest period occurs in the second half of the year (Pereira et al. 2011).

The climate patterns that inf luence the Northeast region and consequently the study area are characterized by irregular ocean and atmospheric standards caused by the ENSO, which results in rainfall anomalies (Kousky and Cavalcanti 1984). The irregularities in rainfall observed in this study are also inf luenced by the presence of ENSO events, as observed from May 2010 to June 2012, where a strong La Niña occurred (NOAA 2019), resulting in continuous positive values for irregularities. During El Niño abnormalities from July 2002 to May 2005, and from October 2014 to August 2016 (NOAA 2019), negative values characteristic of this event was also observed.

Irregularities in rainfall markedly inf luence the response by semiarid vegetation, with f luctuation of both NDVI values and variations of this index over the study period for both vegetation types. Dense Caatinga vegetation has higher NDVI values and is also ref lected on phenological curves. With a similar response over the years, dense vegetation had a more extended season of foliage, considering years of irregularities as well, which suggests that this dense ecosystem may have the capacity to resist or adapt to climate variations.

The response by vegetation of the Northeast region ref lects a strong inf luence of rainfall anomalies caused by ENSO based on NDVI variation, as verif ied by Barbosa and Kumar ( 2016) based on the time series from 1982 to 2001. Their study verif ied that El Niño years are associated with NDVI (normalized diff erence vegetation indices)below average conditions, while La Niña events occur with NDVI above normal. They also observed that rainfall is the dominant causative factor in vegetation drought dynamics,and that there is correlation between the amount of rain and NDVI. This was verif ied in the time series analyzed in the Canindé de São Francisco Caatinga for both cover types, ref lecting not only NDVI variations but also anomalies in vegetation prof iles.

However, irregularities in rainfall levels do not always result in the same vegetation response, and this was also observed by Erasmi et al. ( 2014) while analyzing ENSO events and NDVI anomalies of a time series from 1982 to 2010. They reported a close relationship between ENSO hot periods and reduction in vegetation greenness but observed that the presence of an ENSO hot period did not always result in NDVI anomalies, remarking on the inf luence of this climate occurrence on the Caatinga.

During a drought period, Barbosa et al. ( 2019) assessed the response by the Caatinga vegetation using NDVI data from 2008 to 2016, and verif ied the drought impact on Caatinga dynamics, indicating that, even being xerophytic, this type of vegetation strongly responds to rain, with one to three months delay. They also suggest that an increase in rainfall will not result in more vegetative activity because the area is covered with climax vegetation or that a decrease in rainfall will not result in more degradation because the area is already covered with minimum vegetation. These results were also observed in this study because, in some periods, NDVI responses reached a limit for both positive and negative anomalies and, despite a continuous increase or reduction of rainfall, vegetation expression has a threshold.Moreover, dense Caatinga vegetation, which is close to a climax vegetation, resulted in lower NDVI variations even with the presence of anomalies.

The relation of rainfall and Caatinga vegetation was also analyzed by Santos and Brito ( 2007) who demonstrated a strong correlation of extreme rain indices, except for consecutive dry days, with Caatinga biome vegetation dynamics, identifying that Sertão NDVI depends more on rainfall extremes. They also conf irmed ENSO inf luence on the semiarid climate regime, resulting in extremes of rainfall and consequently in the dynamics of vegetation in Rio Grande do Norte and Paraíba states, while analyzing the relations of climate variables with NDVI obtained from Advanced Very High Resolution Radiometer (AVHRR).

Adami et al. ( 2018) examined an NDVI time series from 2000 to 2008 of Caatinga vegetation associated with monthly tropical rainfall measuring mission data. They observed high seasonality of hyper-xerophyte and hypoxerophyte vegetation types; hyper-xerophyte vegetation had the lowest normalized diff erence vegetation indices while hypo-xerophyte vegetation had high vegetation indices in peak periods of rainy seasons. The transition area,on the other hand, had high NDVI values and less seasonal variation resulting from the semi-deciduousness nature of the vegetation in periods of low rainfall. Since Canindé de São Francisco caatinga is characterized as hyper-xerophyte, strong seasonal variation was observed, particularly for open forest cover which had lower NDVI values, while the dense cover type had the highest values, in addition longer foliage cover, as shown in the phenological curves.

Barbosa et al. ( 2006), while studying NDVI variation based on AVHRR data from a 20-year series (1982-2001)in the northeast, observed oscillations in Caatinga vegetation development, with peaks associated with highest rainfall and lower NDVI values associated with drought. This has been observed in other studies and is typical for this type of vegetation. This was also detected in this study for both Caatinga ecosystems. The open caatinga, besides having the lowest NDVI values, also had the lowest NDVI to that of the dense Caatinga in high rainfall events. Barbosa et al. ( 2006) also suggested that the upward and downward vegetation greening would have an inter-annual pattern of seven to eight years and that, during the years assessed,may be due to spatial-temporal impacts of extreme climate variation like ENSO. In this study, this tendency over time was not observed for rain or vegetation anomalies.

Gurgel and Ferreira ( 2003) identif ied NDVI variations by statistical technique of principal component analysis(PCA), and found strong negative anomalies of vegetation that included the Caatinga biome, indicating a dry period of three to six months, sometimes up to 10 months.The relation between the increase and decrease of this forest response associated with the rainfall regime with high NDVI values, reached positive anomalies in specif ic months. This variation in periods of anomalies, being negative or positive, was also found in this study or both dense and open Caatinga. Months with higher inf luence in each one, particularly the typical drought months with strong negative anomalies, resulted in irregular seasonal NDVI.

The Caatinga biome displays a typical seasonality as seen both in inter- and intra-annual variations, as well as in the characterization of phenological prof iles of dense and open vegetation with well-def ined periods of canopy presence and changes in canopy cover, and is therefore, a reliable characteristic of the semiarid vegetation of Brazil. This explicit seasonal variation was also reported by Formigoni et al. ( 2011) who identif ied the Caatinga biome one that had higher seasonal variation compared to the Amazonia, Cerrado and Atlantic forests.

Schucknecht et al. ( 2013), while assessing variation and tendencies of Brazil’s Northeast vegetation using AVHRR and MODIS NDVI data, also found high temporal variations of NDVI for the Caatinga biome compared with other regional biomes, verifying the lowest average NDVI values for this forest type which results from low annual rainfall and prolonged dry periods. They also observed high interannual seasonality as a function of rainfall variability. In this study in Sergipe State, this inter-annual variation is directly associated to El Niño and La Niña events which may result in years of variable rainfall, conditioning the Caatinga vegetation to express according to this climate phenomena with high variations for open caatinga.

Rainfall plays an important role in the development of Caatinga vegetation since its intra-annual variation inf luences the spatial and temporal heterogeneity of this biome’s seed bank richness and density (Santos et al. 2013). However, through available ecophysiological data, some Caatinga species endure periods of severe drought and there is the possibility that they could not survive the current unsustainability level of the biome (Santos et al. 2014). Rainfall and vegetation development may also be associated with the type of Caatinga, as shown by dense and open phenological data, where dense cover had a higher presence of leaves during rainfall variations and in years of irregularities, with higher NDVI values and longer seasons with leaf cover.

It is important to conserve Caatinga vegetation as it is located in a vulnerable region according to projections of climate change, with temperature increases, reduction in evapotranspiration and rainfall, and increasing aridity (Guimarães et al. 2016).

Conclusion

This study analyzed the rainfall regime and its inf luence on the semiarid Caatinga vegetation of Sergipe State through a time series, verifying the highest normalized diff erence vegetation indices for the fragment of dense Caatinga compared to the open one. Moreover, it was observed that rainfall irregularities, including periods with strong El Niño and La Niña events, aff ected the response of Caatinga vegetation.

The analysis of phenological prof iles were fundamental to understand the periods of physiological change of open and dense Caatinga. Dense caatinga had higher normalized diff erence vegetation indices as well as longer periods with leaves. This ref lects the capacity of Caatinga vegetation to withstand and indicates the importance of conservation areas as a measure to protect and maintain the biome.The results of this study are relevant as they conf irm that dense Caatinga, even as a preserved fragment, maintains physiological activity for a longer period, which may be associated with greater moisture conservation. In addition,because the dense Caatinga t has more leaf cover longer, the soil can maintain its qualities longer, reducing the Effects of desertif ication. The results may be associated with the type of forest management and conservation in this region.The total or partial suppression of individual trees in the Caatinga remnants should be avoided since most open areas have lower photosynthetic capacity as a result of climatic conditions. Additionally, open Caatinga vegetation has a reduced capacity for natural regeneration, and its exploitation by local communities should be limited.

This study shows that further research is needed on diff erent meteorological variables and types of semiarid ecosystems. Understanding the phenology of the Caatinga biome is essential to its management and conservation.

Acknowledgements The authors thank the National Council for Scientif ic and Technological Development (CNPq) and the Coordination of Superior Level Staff Improvement-Brazil (CAPES) for supporting this research.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affi liations.