Spatio-temporal variability of NDVI–precipitation over southernmost South America: possible linkages between climate signals and epidemics

Climate–environment variability affects the rates of incidence of vector-borne and zoonotic diseases and is possibly associated with epidemics outbreaks. Over southernmost South America the joint spatio-temporal evolution of climate–environment is analyzed for the 1982–2004 period. Detailed mapping of normalized difference vegetation index (NDVI) and rainfall variability are then compared to zones with preliminary epidemiological reports. A significant quasi-biennial signal (2.2- to 2.4-year periods, or QB) for joint NDVI–rainfall variability is revealed. From rotated EOFs, dominant NDVI patterns are partitioned according to their lead frequencies: (1) the ‘QB group’ (2.1-to 3-year periods) includes six modes over southern Brazil, Uruguay, northern-central Argentina (two modes), the southern Paraguay–northern Argentina border, and the Santa Cruz Province; (2) the QB1 (2.4- to 3-year periods) + quasi-quadrennial (QQ) mode over the Misiones Province; and (3) the QB2 (2.1- to 2.5-year periods) + QQ + inter-annual (IA) (3- to 7-year periods) two modes over south-eastern Argentina. Modes within the ‘QB group’ are positively correlated with global climate signals and SST. The Uruguayan mode is correlated with global ENSO (8-month lag) whilst the southern Entre-Rios/northern Buenos Aires provinces are correlated with central equatorial Pacific SSTs (3-month lag). The Santa Cruz (Patagonia) Province is most correlated with the Pacific South America (PSA) index and SST patterns (3-month lag) along the Antarctica circumpolar current. The spatial distribution of lead NDVI modes includes the Formosa, Misiones, Chaco and Buenos Aires provinces among others, known for being prone to vector-borne epidemics such as dengue fever, malaria, leishmaniasis (American cutaneous leishmaniasis or ACL), hantivirus, chagas and Argentine hemorrhagic fever (AHF). Some provinces also correspond to regions where lead NDVI PCs’ modes are associated with high-frequency climate signals such as the quasi-biennial oscillation in northwest Argentina. The joint preliminary results (climate–environment–public health reports) presented here for the first time are meant: (1) to contribute to a better understanding of climate–environment–epidemics process-based and modeling studies and (2) to facilitate, in the long run, the implementation of local and regional health early warning systems (HEWS) over southernmost South America. The latter is becoming crucial with ever-increasing migration, urban sprawl (re-emergence of dengue fever epidemics since the late 1990s), all embedded in a climate change context.


Introduction
The spatio-temporal variability of precipitation and temperature modulate vegetation phoenology, with feedback onto the lower-troposphere. Regional ecosystems' dynamics are directly affected by water and energy exchanges through the ground-vegetation-atmosphere interfaces (Van den Hurk et al 2003). The normalized difference vegetation index (NDVI) can thus be used as a good proxy for evaluating regional and local climate variability/changes and potential linkages with variability in regional ecosystems and local ecotones. South America is one of the continents most affected by changes in the general circulation (Minetti and Sierra 1989) and climate events such as global ENSO. During Pacific El Niño (La Niña) years, precipitation in south-eastern South America has been found above (below) normal (Ropelewski and Halpert 1987). By 'trophic cascading effects', crop yields (relying strongly upon rain-fed agriculture), public health and socio-economical issues in South American countries are found to be strongly impacted (Podesta et al 1999, Poveda et al 2001. The NDVI has been defined and analyzed first by Tucker (1979) using National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) data. Later on, Los et al (2001) studied the co-variability between NDVI, precipitation, temperature and sea surface temperature (SST) on global scales, for about a decade. In the meantime, Myneni et al (1996) showed that positive NDVI anomalies over south-eastern America were linked to tropical Pacific SST warming. Finally, del Rosario Prieto (2007) highlighted linkages between ENSO events and Paraná River floods during colonial times, whilst Barbosa et al (2006) found that vegetation patterns in north-eastern Brazil reflected impacts from enhanced aridity during the last decade of the 20th century, and Paruelo et al (2004) have proposed linkages between NDVI trends over South America and climate change since 1981.
The main objective of this study is to isolate areas (pattern-based statistical approach) and lead frequencies of NDVI variability over southernmost South America. This is accomplished by using the most recent and calibrated datasets (see section 2) with large spatio-temporal extents from remote sensing. It is then to extract the joint spatio-temporal variability of NDVI-precipitation. Possible linkages with climate signals and global SST patterns and indices for the 1982-2004 period are discussed. Climate conditions (variability and change) can impinge on human health through different pathways (see figure 2 in the ESSP GEC report, 2004). It is recognized that other non-climatic factors might also be involved with public health issues, i.e. changes in irrigation techniques and agricultural practices, deforestation and urbanization, among others. At this time, there is little understanding of disease ecology in each instance (needed to make predictions) whilst the current understanding of processes and mechanisms involved is incomplete.
In a climate and environment changes' context, modulation of natural climate signals and variability, and increased demographic processes, epidemics are re-emerging worldwide (i.e. new cases of St Louis encephalitis were reported in the . It is hoped that by working with trans-disciplinary Argentinean experts (i.e. teledetectors, entomologists, virologists, etc) through the joint French and Argentinean Spatial Agencies (CNES and CONAE) MATE (Monitoreo Argentino para Tele Epidemiologica) Project, the understanding of epidemics' outbreaks and mechanisms will be improved. This is particularly true for diseases such as dengue fever, Argentine hemorrhagic fever (AHF), malaria, leishmaniasis, hantivirus, chagas and hydatidosis, in prone regions such as the Formosa, Salta and Jujuy provinces (northwest Argentina), the Bolivian-Paraguay borders, the Entre-Rios-Corrientes-Misiones and Chaco provinces, Patagonia region and the Argentinean-Brazilian border (Gorla 2002, Yates et al 2002, Avilés et al 2003, Guerra et al 2006, EID Updates 2008.

Data and methods
The studied area is the southernmost 'South America window' (22 • S-52 • S and 75 • W-52 • W), chosen to exclude equatorial and tropical areas where NDVI can be 'saturated' from high leaf area index (LAI) and potentially 'contaminated' by cloud deep convection. The Andes Mountains in the western part of the window are masked for obvious reason. The 'window' includes data from parts of Paraguay, Uruguay, southern Brazil and southern Bolivia, and the whole of Argentina. Overall the land cover is, in general, composed of C3 grasses in the southernmost part of the 'window' and along the lee-side of the Andes Mountains. Over Argentina, C4 grasses dominate over nearly all the northern half with some sparse C4 crops (following Masson et al 2003). Deciduous and evergreen trees are present over Uruguay and southern Brazil. The 'window' is both influenced by tropical atmospheric circulation to the north, and by seasonal rainfall events and perturbation to the south during austral winter. The NDVI takes advantage of different spectral responses from chlorophyll-loaded vegetal tissues when using red and infrared channels (values range from 0.1 to 0.8 for vegetated areas). The latest NDVI dataset used here is from the Global Inventory Modeling and Mapping Studies (GIMMS) team, or NDVIg, for a 25-year period: 1982-2006. The data is derived from NOAA/AVHRR imagery using the NOAA satellite series 7, 9, 11, 14 and 16. New features from this upgraded dataset include reduced unwanted NDVI variations arising from calibration procedures, varying viewing angles, influence of volcanic aerosols and other effects not related to actual changes in vegetation cover. For example, data from NOAA-9 descending node had volcanic stratospheric aerosols ' corrections during 1982-1984 and 1991-1994. NDVI has also been improved using the empirical mode decomposition/reconstruction (EMD) technique to minimize side effects from orbital drifts. Recent studies have shown good results from minimizing the above effects (Kaufmann et al 2000, Lotsch et al 2003, Slayback et al 2003. The original NDVIg dataset is submitted to maximum value compositing procedures on a 15-day basis, with an 8 km spatial resolution (Tucker et al 2005). Here the spatio-temporal resolutions are up-scaled to a 0.5 • grid-point system with one value per month. Prior to the analyses, the monthly NDVI data is standardized, with linear trends removed using the least-squares method.
A precipitation dataset (GPCC) is available from the Deutscher Wetterdienst (DWD, National Meteorological Service of Germany). The gridded-field on a 0.5 • × 0.5 • latitude longitude resolution can be obtained at http://gpcc. dwd.de. The 'full data re-analysis products' (FDRP), used here, include rainfall data from all available stations, with nearreal-time and delayed observing modes (FDRP version 3 is for the 1951-2004 period).
The datasets described above are analyzed jointly using the multi-taper method/singular value decomposition (MTM/SVD) to highlight the first joint temporal coherence of NDVI and rainfall variability during the overlapping 1982-2004 period.
The multivariate frequency-domain decomposition technique has been successfully developed by Mann and Park (1994). The MTM/SVD technique seeks to retain shared statistical information from the NDVI and precipitation datasets, in terms of maximum variance found for narrow-frequency bands at a given location. Thus a so-called local fractional variance, or LFV spectrum, can be plotted as a function of frequencies while significance levels are obtained from 1000 Monte Carlo simulations (Mann and Park 1996). The code can be obtained at: http://www.people.virginia.edu/ ∼mem6u/mtmsvd.html.
The NDVI data alone is also submitted to rotated empirical orthogonal function (EOF) analyses using the Varimax criterion (Richman 1986) to identify key regions of variability. The rotation eliminates potential eigenvalue degenerescence (overlapping) from sampling errors (North et al 1982) and allows for identification of potential simple spatial structures. Twenty-five EOF ranked modes are retained prior to rotation, still representing ∼80% of total variance but for a simplified dataset. Thus, regions where significant NDVI signals are spatially coherent can be plotted within the studied window.
MTM spectra are obtained for the first 10 rotated principal components (PCs) obtained above. Three 'filters' are used for reasonable frequency resolution, which still provide sufficient degrees of freedom for adequate signal/noise decomposition (Mann and Park 1994). Peaks from the spectra are associated with periodic signals, and their significance levels are relative to an estimated 'red-noise' background. This null hypothesis is motivated for dynamical reasons when studying geophysical phenomena. Resulting significant climate signals associated with vegetation cover variability, combined with results from the joint NDVI-precipitation MTM/SVD analyses, are compared with epidemiological datasets.
Based upon direct regional reports from the Dirección de Epidemiología, Ministerio de Salud, Argentina, and results from lead epidemiological scientific papers, the geographic distribution for key endemo-epidemics was produced in figure 5 and compared with the location of the centers of the NDVI modes obtained from the rotated EOFs analysis (figure 2, bottom) for easier interpretation.

Results
In figure 1, individual NDVI and joint NDVI-rainfall LFV are displayed for the 1982-2004 period. It can be seen that a joint QB signal (gray-shaded area to the right) with a 2.3year peak (after polynomial smoothing) is significant at the 95% level. The other three peaks (i.e. 2.1-2.8-and 4.3-year periods) reflect power spectra for the LFV rainfall alone (not shown). A 5.8-year period is found (gray-shaded area to the left) but below the 90% level due to the relative shortness of the time series (it is highly significant though for the LFV of NDVI alone, represented by the dashed line). When the evolutive spectrum is examined with an 8-year moving window (not shown) it is found that the significant QB band is present during most of the period under investigation (with LFV values smaller than 0.72, the 95% level, prior to 1989 and between 1992 and 1995).
In figure 2, spatial distribution of explained variance (from 40% up to 80%, shaded red) from the nine out of the first ten rotated and ranked EOF modes (which represent ∼42% cumulated total variance) are partitioned based upon their dominant frequencies in their MTM PCs spectra (figure 2, top). Interestingly enough the sorted patterns correspond to specific bio-climatic domains: the first panel or 'QB group' (figure 2, bottom left) with rotated EOF modes 1, 2, 3, 10, 8 and 6 includes southern Brazil, Uruguay and central, northern and the extreme south of Argentina. Mode 4 over the Salta Province is not shown since it displays a maximum explained variance of 60%. The seasonal rainfall distribution for the first panel is slightly bi-modal with no real dry season. It is essentially covered by Sertaõ's vegetation types (Rio Grande do Sul, mode 1), fertile coastal lowlands and grasslands (mode 2), complex ecosystems of marshes, ponds, lagoons and wetlands (mode 3), and rice, soja, wheat cornfields and orchards (mode 10). Mode 8 covers the Formosa Province (east of the Salta Province) on the southern border of Paraguay, with a subtropical climate (∼1000 mm/year of rain), a jungle-type forest and the 'impenetrable' region to the west with the typical Chaco forest. Dengue and chagas epidemics are prevalent there Colors represent the percentage of explained variance for each mode, starting at 40% (blue) to 80% (red), using 10% incremental steps. QB is for quasi-biennial, QQ is for quasi-quadrennial and IA is for inter-annual. The Andes (black contour) are masked for the analysis. (Diosque et al 2004). On its western side is a fertile valley with cardón cacti on the outskirts and tropical highland forests. To its extreme northwest on the lee-side of the Andes, the climate is very hot during summers (up to 40 • C) while cold winters dominate. The towns of Orán (Salta Province), Resistencia (Chacos Province) and Corrientes (Corrientes Province) and neighboring areas are densely populated and dengue fever epidemics re-emerge; the isolated mode 6, to the extreme south, is found over the Santa Cruz Province-Patagonia area where hydatidosis is present. The latter region is covered by shrub steppes (extreme southern part of the 'window') and is under the influence of both the Pacific and Atlantic Oceans with rainfall amounts less than 200 mm/year. The second panel includes only one rotated EOF mode (figure 2, bottom middle) which covers mostly the north of the Corrientes Province and the Misiones Province (mode 7). This is a region where dengue, malaria and leishmaniasis epidemics are also re-emerging. The third panel includes two rotated EOF modes (figure 2, bottom right) covering most of the Buenos Aires (mode 5) and the southern Buenos Aires-La Pampa-Rio Negro (mode 9) provinces with ∼500 mm/year rainfall. Grass, sorghum, corn, sunflower and a few ombú or algarrobo trees dominate there. In the Pampa, flooding can be severe and Argentina hemorrhagic fever (AHF) is endemic (Polop et al 2008) within a humid environment. Cases of hantivirus (Porcasi et al 2005) and hydatidosis (Lamberti et al 1999) have also been recorded there.
The MTM PCs spectra which allowed for associating modes into three panels, based upon lead frequencies, are displayed in figure 2 (top left for 'QB group'; top middle for single mode 7; top right for panel 3). Significant variances (99% confidence level) are found within two frequency bands: a wide QB band (2.0-to 3.1-year periods which includes QB1 or 2.4-to 3.1-year periods and QB2 or 2.2-to 2.5-year periods) only for the 'QB group'; a wide QB1 (2.4-to 3.1-year periods) plus QQ (3.1-to 4.1-year periods) band for isolated mode 7; a narrower QB2 (2.2-to 2.5-year periods) band, and a wide QQ and IA (3.1-to 7.1-year periods) bands for panel 3, with peaks at ∼3.7-year (QQ) and ∼6.8-year (IA) periods.
Lead/lag correlations between all NDVI PCs and global SST are computed and mapped.
The most significant Global SST patterns with rainfall amount over key NDVI regions and possible relationships with areas with epidemics outbreaks of a few diseases are discussed next.

Discussion and conclusion
Using the latest and best available datasets, linkages between NDVI dominant patterns of variability over regions of southernmost South America, local rainfall amount and dominant climate signals from neighboring oceans have been highlighted for the first time. For the relatively short period under investigation, the influence of ENSO with its different spatio-temporal scales on NDVI and rainfall variability are seen. Overall it is characterized by the wide 'quasi-biennial' (QB) and inter-annual/quasi-quadrennial (IA/QQ) bands, all belonging to the ENSO spectrum and distinct SST spatial patterns in the Pacific Ocean (Tourre and White 2005). The wide QB band includes two quasi-biennial sub-bands QB2 and QB1 with spectral peaks centered around periods of 2.3 and 2.7 years, as in Tourre et al (1999) and Tourre and White (2006). Early on, Angell and Korshover (1974) argued that the QB fluctuations with different frequencies are linked to the southwest-northeast movement of the so-called 'centers of action'. Later, QB signals were identified in SLP anomalies over the Northern Hemisphere (Trenberth and Shin 1984) and from global analysis of air temperature and SST (Mann and Park 1994). For the 1982-2004 period, the significant (95% level) joint NDVI-rainfall frequency variability is within a narrow QB band with a 2.3-year peak, as in the global QB climate signal (White and Allan 2001). Findings here corroborate results from Los et al (2001) but for a shorter period, when they identified a strong global NDVI signature near a 2.6-year period. Tourre et al (1999) and Tseng and Mechoso (2001) also observed the two significant QB climate signals in the Atlantic Ocean, while Jarlan et al (2005) isolated a QB NDVI signal over west Africa, among others. The presence of the QB signal in the La Pampa Province is in agreement with results by Kestin et al (1998). They also documented an increase in the amplitude of the QB signal at the end of the 20th century, probably due to climate change. It is interesting to note that, during the investigated period, the other key individual NDVI signal with a peak at 5.8 years is present and could correspond to another ENSO timescale, as in Tourre and White (2005).
Out of the 10 leading NDVI rotated EOF modes, significant QB signals (QB1 and QB2) are also present in nine modes (6 modes in the so-called 'QB group', including an isolated mode in the extreme south of Argentina or the Santa Cruz Province-Patagonia), 1 isolated mode in panel 2 with the 'lower-frequency' QB1 and QQ signals and 2 modes in panel 3, with the 'higher-frequency' QB2 and IA/QQ signals. The latter panel includes the Rio Grande do Sul region in Brazil and the Misiones Province of Argentina. The inter-annual (IA) variability of vegetation conditions there is noticeable and makes that region different from the 'QB group'. For instance, over La Pampa positive NDVI anomalies are found when positive rainfall anomalies occur during El Niño events from November to January (not shown), thus corroborating findings by Pisciottano et al (1994). The opposite holds true during La Niña events, with NDVI negative anomalies during springtime as in Grimm et al (1998). An Atlantic 4.4-year signal is being linked to local equatorial Atlantic-Ocean-atmosphere interactions. The prominent QQ signal from panel 2, located in the subtropical regions and present over the Misiones Province (∼24-30 • S), is coherent with results from Ribera and Mann (2003). Finally, the QB and IA/QQ bands found in all three panels belong to the extended ENSO spectrum as in Tourre and White (2006).
Significant positive lagged correlations (at the 99% confidence level) between NINO3 and several NDVI modes from the QB group means that an ocean warming over the NINO3 box leads to enhanced rainfall (and NDVI increase) over the considered region. This is corroborated by the correlation maps between NDVI PCs and global SST spatial patterns. The negative correlation between NDVI in mode 9 (La Pampa) on the Atlantic side with western tropical Atlantic SST with ∼3-month lag might be due to the transmission of the QQ ENSO signal from the Pacific into the Atlantic, with contemporaneously changes in the thermal state of the southern Atlantic, as identified by Tourre and White (2005). Moreover, positive lagged correlations between NDVI mode 6 (Santa Cruz-Patagonia) and PSA (the southern hemisphere equivalent of the PNA pattern in the northern hemisphere) corroborates results obtained by combining rainfall analysis and patterns, and the PSA index (Mo and Paegle 2001): the positive phase of PSA is associated with positive rainfall anomalies over the whole southern South America region during December-January-February. Over Patagonia the significant QB fluctuation could be associated with the phasing of the southern hemisphere annual mode and shifts of the circumpolar trough as modeled by Garric et al (2003). Finally, the significant correlation with global SST at the tip of South America could be as well associated with the variability of the Antarctic circumpolar wave as identified by White and Petersen (1996) and White (2004).
The complex relationships between NDVI, precipitation, land cover types, local hydrology and multi-scale climate variability can be further investigated by computing actual rainfall anomalous time series averaged at grid points from all the significant NDVI modes and compared with corresponding NDVI PCs. As an example, two areas, both from the 'QB group' with distinct temporal features, are displayed in figure 4: the Rio Grande do Sul region (mode 1, figure 4 top) and the Formosa Argentinean Province (mode 8, figure 4 bottom). In figure 4 (top), besides strong QB rainfall and NDVI variability as expected, the relationship between ENSO events (time of occurrence given on the abscissa), rainfall amount and NDVI anomalies are conspicuous, i.e. contemporaneously such as during the two large 1982-83 and 1997-98 ENSO events or during 1990when a protracted ENSO occurred (Weisberg and Wang 1997, or possibly linked to low-frequency climate variability in the South Atlantic (Venegas et al 1996), or slightly lagged like in 1985or 2002, in the Formosa Province, while a QB signal again dominates the NDVI PC (see figure 1, top left) there is evidence (not statistically significant due to the shortness of the records) of low-frequency (Tourre and White 2006) and inphase NDVI-rainfall variability. The latter is particularly true between 1986 and 1992, and after 1996 onward. For mode 7 (Corrientes and Misiones Provinces) a unique QQ NDVI signal dominates the PC and transient in-phase relationships between NDVI and rainfall anomalies can be found on these timescales (like from 1988 to 1991, not shown).
Climate variability is only one component associated with epidemics outbreaks (i.e. new, existing with some periodicity and/or re-emerging). Ecosystem responses to climate variability can be seen as 'mediators' on diffusion of infectious vector-borne and rodent-borne diseases (following Root et al 2003). Woodland savannas, tropical forests and ecotones are indeed areas with plant and animal biodiversity with complex distribution (Patz et al 2004) and variability. Rain-fed agriculture highly dependent on climate variability is also directly linked to public health (Trewavas 2002). Results presented here on climate-environment variability are meant to contribute first to a better understanding of mechanisms possibly linked with outbreaks and diffusion of diseases and distribution of pathogens' reservoirs. The distribution of reported disease cases in the studied 'window' can be found in figure 5. The centers of lead NDVI modes are given to compare easily with disease-prone areas. It is interesting to note that Argentine hemorrhagic fever (AHF), as identified by Polop et al (2008), prevails where the NDVI modes 10 and 5 are found. Mode 10 is also in a zone where hantivirus dominates and can include preferred areas for rodents' habitats and reservoirs of zoonoses (Sergio Sosa-Estani et al 2001, Porcasi et al 2005. Mode 1 which belongs to the QB group (essentially the State of Rio Grande do Sul, Brazil) encompasses a region where leishmania (Viannia) is encountered (Fagundes et al 2007) and deserves further investigation for possible physical mechanisms involved. American Cutaneous Leishmaniasis (ACL, Córdoba-Lanúsa et al 2006), dengue fever occurrences and malaria epidemics events have been recorded where modes 8 (Formosa Province) and 7 (Misiones Province) display strong QB and QQ variability. Hydatidosis cases are reported over La Pampa-Rio Negro provinces (Lamberti et al 1999), a region covered by mode 10. All of the above remarks warrant further statistical and quantitative investigations on the role climateenvironment variability can play in terms of spatio-temporal appearance and diffusion of diseases.
It is acknowledged that epidemics such as dengue fever and malaria in northern Argentina (Salta, Formosa and Misiones provinces, among others) are also emerging Figure 5. Mapping of key endemic and endemo-epidemic zones in the southernmost South America (except Chile). The geographic distribution is to be compared to those of NDVI modes, highlighted in figure 2 (bottom) and whose centers are given here. The center for NDVI mode 4 is also given with a maximum explained variance of 60%.
where urban sprawl, intense mobility of population through neighboring countries and climate change become congruous processes. For example, new and early detection of circulation of dengue fever virus have occurred only since 1997 (Avilés et al 2003). Malaria is re-emerging in rural areas of Jujuy and Salta provinces (along the Bolivian border) and the Corrientes and Misiones provinces (along the Paraguayan border). Nevertheless it has been recently observed (since 1998 onward, data from the Dirección de Epidemiología, Ministerio de Salud, Argentina) that dengue outbreaks seem to follow QB and QQ cycles over northern Argentina (region 8, in figures 2 and 5), and as such could be associated with natural variability of climate signals. This is particularly true for the Salta and Misiones provinces. Further and more comprehensive analyses are required since understanding sequences of the above events should help in predicting the re-emergence of such diseases.
The advanced knowledge of the phasings of climate signals (for high and low frequencies such as the QB and IA/QQ signals as identified in this paper) which are going to modulate local seasonal variability like the South American Monsoon System (SAMS, see figure 2 in Ma and Mechoso 2007) is a prerequisite to study and mitigate consequential impacts on public health. Working closely with entomologists and virologists to outline some mechanisms contributing to epidemics outbreaks also becomes a necessity.
All relevant information (in near real-time) is to be distributed to stakeholders (including health officials) involved with health early warning systems (HEWS). Applied results should help in improving decision-making practices for best prevention and adaptation strategies of climate impact on human health and welfare (McMichael andKovats 2000, Noireau et al 2005). The latter is very important in the climate change context, modifying the so-called natural climate signals. In a forthcoming paper downscaled processes and mechanisms for epidemics outbreaks, possibly linked with climateenvironment variability/change, will be assessed.