Modeling the natural vegetation dynamic under climate change scenarios in coastal protected dryland of southeastern Tunisia

Abstract


INTRODUCTION
Vegetation is a vital component playing a crucial role in regulating the global carbon balance (Erb et al., 2018;Ge et al., 2021;Gao et al., 2022).Vegetation provides many of the essential ecosystem services.These last included soil stabilization, water regulation, carbon sequestration and grain production (Lü et al., 2012).Climate change can be considered as the primary force driving the dynamics of vegetation worldwide especially in the semiarid and arid regions (Zhang et al., 2018;Gao et al., 2022).This could lead to significant ecological and economic losses (Chen et al., 2020;Ge et al., 2021).Therefore, understanding the impacts of climate change on vegetation dynamics is crucial for developing effective strategies to mitigate these effects and ensure the long-term sustainability of our ecosystems.In addition to climate change, human activities, increasing the greenhouse gases emission into the atmosphere, such as deforestation and industry, are strongly affecting the structure and function of ecosystems (Ouled Belgacem & Louhaichi, 2013;Zhang et al., 2018;Tong et al., 2019).So, it is very interesting to protect the natural floristic resources for maintaining the ecological equilibrium/biodiversity and minimizing the impacts of climate change on wild ecosystems (Ding et al., 2016).The monitoring of vegetation dynamics becomes easier thanks to the availability of remotely sensed imagery (Jewitt et al., 2015).Remote sensing is a powerful tool that allow researcher to analyse changes in less time, with lower cost, and more precision (Maina et al., 2020).The remote sensed data provide useful information for monitoring degraded lands and environmental changes (Kumar, 2022).These later result in radiance values, which can be detected through radiometric vegetation indices (Kumar, 2022).According to Lemenkova & Debeir (2023), the vegetation index (VI) is a metric used to measure and monitor the health, density, and growth of plant cover.It can be calculated using various spectral bands of the electromagnetic spectrum, which are sensitive to the presence and the amount of chlorophyll pigments in leaves (Zeng et al., 2022).The Soil Adjusted Vegetation Index (SAVI) is recommended for dryland studies since it reduces the impact of soil brightness on spectral vegetation indices and helps to indicate the areas where vegetation is decreasing (Almutairi et al., 2013;Lemenkova & Debeir, 2023).In these contexts, the present study aims to monitor the future evolution of natural vegetation cover under two climatic change scenarios through radiometric and remotely sensed indices as well as some plant groups characteristics in the Zarat region.The main occupations are about how does the climate change influenced the vegetation cover in coastal area of North Africa?How can the Shared Socioeconomic Pathways (SSP) scenarios be used to predict the effects of climate change on radiometric vegetation indices?
Fig. 1.Geographical location of the study region.

SAVI index
The Soil Adjusted Vegetation Index (SAVI) is introduced by Huete (1988) as: where L is a correction factor which ranges from 0 (very high vegetation cover) to 1 (very low vegetation cover).These SSPs are adopted by the Fifth Assessment Report (IPCC, 2014).

Model performance
The Receiver Operating Characteristic (ROC), showed by the Area Under Curve (AUC), is used to validate the predictive capacity of the model.AUC is a measure of model performance, and its higher value is close to one (1), the greater it deviates from a random pattern (Phillips et al., 2004).According to the proposal of Elith ( 2006), the AUC values are interpreted as follows: the model is "excellent" if AUC > 0.90; "good" if 0.80 < AUC ≤ 0.90; "acceptable" if 0.70 < AUC ≤ 0.80; "bad" if 0.60 < AUC ≤ 0.70; and "invalid" if AUC ≤ 0.60.The Jackknife procedure (regularized training gain) and percent variable contributions are used to determinate the predictive power of each variable and to identify those that contribute most in the distribution model produced by MaxEnt (Yost et al., 2008).(Lobo et al., 2008).It is important to mention that AUC indicates significantly higher accuracy of the model when compared to random prediction.The 0.5 value represents the completely random prediction (Massada et al., 2012).Mainly the environmental conditions are involved in such dynamic.The results of the current and projected areas of SAVI classes are showed in Tables 3 and 4. When looking at current climatic, edaphic, and topographic variables, we found that model produced 98.23ha (56.76%), 117.87ha (68.11%) and 0ha respectively for the (0.75-1) sub-class of high, medium, and low SAVI (Table 3).Under SSP245 and SSP585 climate scenarios, edaphic and topographic variables increased the subclass (0.75-1) of high SAVI with a percentage about 30% (Table 4).This indicates positive growth and potential for the conservation of plant species inside the protected area in the future.Fencing is a primary management practice for the conservation of rangeland plant communities since it promotes flora regeneration (Tarhouni et al. 2017).Such changes are followed by higher vegetation cover as indicated by SAVI in some new habitats in the protected area.However, the medium SAVI class showed a decline under SSP585 during 2041-2060 (-7.03%) and 2061-2080 (-12.61%)(Table 4).The unfavorable environmental conditions associated with the SSP585 scenario are likely to have adverse impacts on vegetation vigor and potential loss of vegetation cover.These effects could lead to a decline of SAVI.The AUC of the three plant groups (G1, G2 and G3) indicates excellent accuracy of prediction (AUC ranges from 0.98 to 1) (Table 5).Hence, the model is classified as satisfactory with the given set of training and test data.
The contribution of climatic, edaphic, and topographic variables in Maxent of G1 (Fig. 3) showed that bio1 (6.3 and 6.8% respectively for SSP245 and SSP585), bio3 (30.8 and 26.87%), bio8 (12 and 5.17%), bio14(12.1 and 8.32%) and bio18 (17.1 and 19.97%) are the highest followed by carbon stock (7.07 and 4.25%) and aspect (3.24 and 4.3%).Regarding G2, the percentage of variables contribution reveals that bio 18 is the highest (28.07 and 29.27% respectively for SSP245 and SSP585) followed by carbon stock (26.97 and 26%) and bio3 (21.27 and 22.35%).For G3, the carbon stock shows the highest percentage of contribution (47.45 and 54.3% respectively for SSP245 and SSP585).Since the distribution pattern of plants is closely linked to precipitation and temperature (Zhong et al., 2010), it seems that the soil carbon stock can be useful to predict future vegetation dynamic under climate change scenario.Currently, the group 1 (Rhanterium suaveolens and Lygeum spartum) occupies a large area of 147ha (60.2%).The predicted areas of G1 in future time-periods according to the SSP245 and SSP585 scenarios does not differ significantly from the current one (changes ranging from -2.54 to 1.68% (  6).In the future projections, percent of changes range from -16.98 to 6.75%.The oscillation in this plant group could be driven by the floristic composition which can be more adapted and tolerant to the environmental variability of the studied scenarios.Factors such as annual temperature and mean annual precipitation may influence the plant distribution.Temperature and precipitation are considered as the major determinants of plant distribution in a global scale (Zhao et al., 2018).At a local scale, the distribution of plants is consistently influenced by precipitation and temperature, with precipitation being the primary factor that controls plant productivity and composition (Ben Mariem & Chaieb, 2017).This finding would suggest that the G2 plant group is adapted to the variation of environmental conditions of the protected area.Indeed, plant groups are undergoing natural change as the ecosystem progress towards a more stable state especially in the context of climate change (Gann et al., 2019).The actual area of G3 (Imperata cylindrica) is 19.5ha (8%).This area decreased by a half in 2021-2040 (Table 6) under both SSP245 and SSP585 scenarios.From current to future projections, the area occupied by G3 remains low and never exceeding 23ha.The results highlight the effectiveness of SAVI as reliable index for predicting the dynamic of vegetation within protected areas.The differences in response of the studied plant groups to the climatic scenarios can be linked to their floristic composition basically the tolerant species.Furthermore, the present work emphasizes the importance of studying plant group communities in order to accurately predict ecosystem functioning in the future climate scenarios.By using SAVI and plant group occurrences, researchers and conservationists can make informed decisions and take proactive measures to ensure the preservation and sustainability of protected areas and their ecosystems.

Table 1 .
General description of the studied bioclimatic, soil and topographic variables.It is clear from Table2that the AUC, which measure the prediction accuracy of the model, is ranging from 0.839 to 0.857 and confirms the good performance and robustness of the model for the high class of SAVI in Zarat.The degree to which a model can successfully discriminate occurrence from background locations has been suggested to be associated with high AUC values

Table 3 .
SAVI area (ha, %) during the different time-periods and according to the SSP245 and SSP585 scenarios from the CanESM5 model.The most important sub-classes for the analysis (for each SAVI class) are indicated in Bold.

Table 4 .
Change in area (ha, %) of the most important sub-classes (0.75-1) of SAVI values (high, medium, low) from the currently to future time-periods under the SSP245 and SSP585 climate change scenarios.

Table 6 .
Change area from the currently to future time-periods according to the SSP245 and SSP585 scenarios from the CanESM5 model.G1: Rhanterium suaveolens and Lygeum spartum; G2: Lycium shawii and Nitraria retusa; G3: Imperata cylindrica.Maxent model is used to predict the potential distribution of plant groups and SAVI classes under different climate change scenarios in Zarat, coastal protected area of southern Tunisia.