Page 107 - AJWEP-22-4

Conservation of the vulnerable Vitellaria Paradoxa

the rainfall of the driest month ranging from 0.20 to cover the entire subregion. 55,56 For example, a previous
0.30 mm. study has mapped and identified savanna woodland
1
ecosystems as the primary habitat for V. paradoxa
4. Discussion in Kwara State in north-central Nigeria. In contrast,
another study described its distribution around the
57
Both BRTs and RFs are the two best-performing savanna national parklands/reserves in Africa, where
machine learning algorithms in this study. They have naturally occurring tree stands are protected by locals
been described as top-performing machine learning or government authorities. Besides the protection of the
algorithms and applied in several predictive modeling tree species in a few national game reserves, no extensive
38
in ecology and biogeography 35,36 for conservation programs have been made toward the conservation of
purposes. In contrast, the ANN model failed to the tree species in the study area. These earlier studies
produce meaningful predictions in this study. It has aligned with the model predictions from this research,
also been reported to be the least-performing model in which indicate high habitat suitability and the presence
comparable SDM studies. One potential explanation of viable seed sources for V. paradoxa in forest clusters
38
for this poor performance is the relatively small sample and protected areas within the Guinea savanna zone. The
size used (n = 100), as ANN performance has been high suitability areas, mostly located in forest clusters,
shown to improve with larger sample size and is more parks, and game reserves, could be indicative of tree
47
suitable for classification tasks than for regression- exploitations and habitat degradation in unprotected
based predictions. It is therefore recommended that areas over the years, leading to a significant decline in
38
ANN be used primarily for classification purposes with V. paradoxa populations across much of its natural range.
a relatively larger sample size. Climate variables, particularly precipitation and its
A weighted average of the two best-performing associated variables, contributed more than 25% to
algorithms, BRT and RF, was applied as a consensus the SDM predictions for V. paradoxa. This finding is
approach for the final SDM prediction. This ensemble consistent with previous research that identified rainfall
40
58
method helps to reduce uncertainty or bias associated and soil fertility as key determinants of phenotypic
with relying on a single algorithm. traits, such as leaf size, fruit weight, and nut size in
Several plant phenotyping studies in ecology and V. paradoxa. In regions where rainfall is abundant and
agriculture have utilized vegetation indices, such as the soil is fertile, the trees tend to produce larger nuts
NDVI, GNDVI, NDWI, and SAVI to characterize with higher oil content. Although soil variables were
phenological changes in plants and crops. 48,49 not included in the present model, vegetation indices
Understanding these essential vegetation indices representing plant greenness may serve as effective
provides insights into plant health, nutrient status, and proxies for assessing soil nutrient availability.
the stress level, which could be a major step toward tree The inclusion of vegetation phenology parameters,
conservation strategies and rational choices guiding such as NDWI, GNDVI, and NDVI, not only enhanced
the selection of seed sources for propagation. Earlier model performance but also facilitated the identification
studies have applied remote sensing technologies to of V. paradoxa populations with viable/healthy seeds.
assess health status, such as nitrogen and chlorophyll This has significant implications for assessing tree
contents of various plants and crops, 50-52 which have growth, nutrient status, water stress levels, productivity,
proven valuable in assessing plant productivity, yield, and restoration efforts.
and vulnerability to environmental stress. The positive responses of V. paradoxa to most
While most habitat suitability/SDM studies have vegetation indices (e.g., NDWI/MNDWI, EVI, and
focused on the use of climate variables as predictors, 53,54 NDVI) signify the importance of these variables to the
and occasionally land use or habitat types, 36,42 this study growth and health status of the species, both of which
integrates both climate variables and remote sensing directly influence seed quality. An optimal prediction
technology. By applying machine learning algorithms, was reached at approximately 1,050 mm for total annual
this approach not only predicts suitable habitats for precipitation and around 250 mm during the wettest
V. paradoxa but also identifies priority locations for month. Previous studies have also shown that the leaf
viable seed sources (SZPI) within seed zones that could and fruit quality of V. paradoxa, as well as the trees’
be useful for tree propagation and restoration. circumference, are positively correlated with rainfall. 9
Although V. paradoxa is mostly found in West Africa, Water stress is another important environmental
its distribution has often been erroneously mapped to variable that not only affects the distribution of

Volume 22 Issue 4 (2025) 99 doi: 10.36922/AJWEP025210160