TY - JOUR
T1 - Probabilistic Mapping and Spatial Pattern Analysis of Grazing Lawns in Southern African Savannahs Using WorldView-3 Imagery and Machine Learning Techniques
AU - Awuah, Kwame T.
AU - Aplin, Paul
AU - Marston, Christopher G.
AU - Powell, Ian
AU - Smit, Izak P. J.
N1 - Funding Information:
Funding: The Department of Geography and Geology, Edge Hill University, provided funding for this study. Field work was supported by the 2019 Geographical Club Award with grant reference GCA 42/19, offered by the Royal Geographical Society (RGS_IBG).
Publisher Copyright:
© 2020 by the authors. Licensee MDPI, Basel, Switzerland.
Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.
PY - 2020/10/15
Y1 - 2020/10/15
N2 - Savannah grazing lawns are a key food resource for large herbivores such as bluewildebeest (Connochaetes taurinus), hippopotamus (Hippopotamus amphibius) and white rhino(Ceratotherium simum), and impact herbivore densities, movement and recruitment rates. They alsoexert a strong influence on fire behaviour including frequency, intensity and spread. Thus, variationin grazing lawn cover can have a profound impact on broader savannah ecosystem dynamics.However, knowledge of their present cover and distribution is limited. Importantly, we lack arobust, broad-scale approach for detecting and monitoring grazing lawns, which is critical toenhancing understanding of the ecology of these vital grassland systems. We selected two sites inthe Lower Sabie and Satara regions of Kruger National Park, South Africa with mesic and semiaridconditions, respectively. Using spectral and texture features derived from WorldView-3 imagery,we (i) parameterised and assessed the quality of Random Forest (RF), Support Vector Machines (SVM),Classification and Regression Trees (CART) and Multilayer Perceptron (MLP) models for generaldiscrimination of plant functional types (PFTs) within a sub-area of the Lower Sabie landscape, and (ii)compared model performance for probabilistic mapping of grazing lawns in the broader Lower Sabieand Satara landscapes. Further, we used spatial metrics to analyse spatial patterns in grazing lawndistribution in both landscapes along a gradient of distance from waterbodies. All machine learningmodels achieved high F-scores (F1) and overall accuracy (OA) scores in general savannah PFTsclassification, with RF (F1 = 95.73 0.004%, OA = 94.16 0.004%), SVM (F1 = 95.64 0.002%,OA = 94.02 0.002% ) and MLP (F1 = 95.71 0.003%, OA = 94.27 0.003% ) forming a clusterof the better performing models and marginally outperforming CART (F1 = 92.74 0.006%,OA = 90.93 0.003%). Grazing lawn detection accuracy followed a similar trend within the LowerSabie landscape, with RF, SVM, MLP and CART achieving F-scores of 0.89, 0.93, 0.94 and 0.81,respectively. Transferring models to the Satara landscape however resulted in relatively lowerbut high grazing lawn detection accuracies across models (RF = 0.87, SVM = 0.88, MLP = 0.85and CART = 0.75). Results from spatial pattern analysis revealed a relatively higher proportion ofgrazing lawn cover under semiarid savannah conditions (Satara) compared to the mesic savannahlandscape (Lower Sabie). Additionally, the results show strong negative correlation between grazinglawn spatial structure (fractional cover, patch size and connectivity) and distance from waterbodies,with larger and contiguous grazing lawn patches occurring in close proximity to waterbodies in bothlandscapes. The proposed machine learning approach provides a novel and robust workflow foraccurate and consistent landscape-scale monitoring of grazing lawns, while our findings and researchoutputs provide timely information critical for understanding habitat heterogeneity in southernAfrican savannahs.
AB - Savannah grazing lawns are a key food resource for large herbivores such as bluewildebeest (Connochaetes taurinus), hippopotamus (Hippopotamus amphibius) and white rhino(Ceratotherium simum), and impact herbivore densities, movement and recruitment rates. They alsoexert a strong influence on fire behaviour including frequency, intensity and spread. Thus, variationin grazing lawn cover can have a profound impact on broader savannah ecosystem dynamics.However, knowledge of their present cover and distribution is limited. Importantly, we lack arobust, broad-scale approach for detecting and monitoring grazing lawns, which is critical toenhancing understanding of the ecology of these vital grassland systems. We selected two sites inthe Lower Sabie and Satara regions of Kruger National Park, South Africa with mesic and semiaridconditions, respectively. Using spectral and texture features derived from WorldView-3 imagery,we (i) parameterised and assessed the quality of Random Forest (RF), Support Vector Machines (SVM),Classification and Regression Trees (CART) and Multilayer Perceptron (MLP) models for generaldiscrimination of plant functional types (PFTs) within a sub-area of the Lower Sabie landscape, and (ii)compared model performance for probabilistic mapping of grazing lawns in the broader Lower Sabieand Satara landscapes. Further, we used spatial metrics to analyse spatial patterns in grazing lawndistribution in both landscapes along a gradient of distance from waterbodies. All machine learningmodels achieved high F-scores (F1) and overall accuracy (OA) scores in general savannah PFTsclassification, with RF (F1 = 95.73 0.004%, OA = 94.16 0.004%), SVM (F1 = 95.64 0.002%,OA = 94.02 0.002% ) and MLP (F1 = 95.71 0.003%, OA = 94.27 0.003% ) forming a clusterof the better performing models and marginally outperforming CART (F1 = 92.74 0.006%,OA = 90.93 0.003%). Grazing lawn detection accuracy followed a similar trend within the LowerSabie landscape, with RF, SVM, MLP and CART achieving F-scores of 0.89, 0.93, 0.94 and 0.81,respectively. Transferring models to the Satara landscape however resulted in relatively lowerbut high grazing lawn detection accuracies across models (RF = 0.87, SVM = 0.88, MLP = 0.85and CART = 0.75). Results from spatial pattern analysis revealed a relatively higher proportion ofgrazing lawn cover under semiarid savannah conditions (Satara) compared to the mesic savannahlandscape (Lower Sabie). Additionally, the results show strong negative correlation between grazinglawn spatial structure (fractional cover, patch size and connectivity) and distance from waterbodies,with larger and contiguous grazing lawn patches occurring in close proximity to waterbodies in bothlandscapes. The proposed machine learning approach provides a novel and robust workflow foraccurate and consistent landscape-scale monitoring of grazing lawns, while our findings and researchoutputs provide timely information critical for understanding habitat heterogeneity in southernAfrican savannahs.
KW - African savannah
KW - grazing lawns
KW - machine learning
KW - WorldView-3
KW - Support Vector Machines
KW - Random Forest
KW - Multilayer Perceptron
KW - decision trees
KW - spatial analysis
KW - Spatial analysis
KW - Grazing lawns
KW - Machine learning
KW - Decision trees
U2 - 10.3390/rs12203357
DO - 10.3390/rs12203357
M3 - Article (journal)
SN - 2072-4292
VL - 12
SP - 1
EP - 37
JO - Remote Sensing
JF - Remote Sensing
IS - 20
M1 - 3357
ER -