In the recent past, the volume of spatial datasets has significantly increased. This is attributed to, among other factors, higher sensor temporal resolutions of the recently launched satellites. The increased data, combined with the computation and possible derivation of a large number of indices, may lead to high multi-collinearity and redundant features that compromise the performance of classifiers. Using dimension reduction algorithms, a subset of these features can be selected, hence increasing their predictive potential. In this regard, an investigation into the application of feature selection techniques on multi-temporal multispectral datasets such as Sentinel-2 is valuable in vegetation mapping. In this study, ten feature selection methods belonging to five groups (Similarity-based, statistical-based, Sparse learning based, Information theoretical based, and wrappers methods) were compared based on f-score and data size for mapping a landscape infested by the Parthenium weed (Parthenium hysterophorus). Overall, results showed that ReliefF (a Similarity-based approach) was the best performing feature selection method as demonstrated by the high f-score values of Parthenium weed and a small size of optimal features selected. Although svm-b (a wrapper method) yielded the highest accuracies, the size of optimal subset of selected features was quite large. Results also showed that data size affects the performance of feature selection algorithms, except for statistically-based methods such as Gini-index and F-score and svm-b. Findings in this study provide a guidance on the application of feature selection methods for accurate mapping of invasive plant species in general and Parthenium weed, in particular, using new multispectral imagery with high temporal resolution.

The Tree-based Pipeline Optimization Tool (TPOT) is a state-of-the-art automated machine learning (AutoML) approach that automatically generates and optimizes tree-based pipelines using a genetic algorithm. Although it has been proven to outperform commonly used machine techniques, its capability to handle high-dimensional datasets has not been investigated. In vegetation mapping and analysis, multi-date images are generally high-dimensional datasets that contain embedded information, such as phenological and canopy structural properties, known to enhance mapping accuracy. However, without the implementation of a robust classification algorithm or a feature selection tool, the large sets and the presence of redundant variables in multi-date images can impede accurate and efficient landscape classification. Hence, this study sought to test the efficacy of the TPOT on a multi-date Sentinel-2 image to optimize the classification accuracies of a landscape infested by a noxious invasive plant species, the parthenium weed (Parthenium hysterophorus). Specifically, the models created from the multi-date image, using the TPOT and an algorithm system that combines feature selection and the TPOT, dubbed “ReliefF-Svmb-EXT-TPOT”, were compared. The results showed that the TPOT could perform well on data with large feature sets, but at a computational cost. The overall accuracies were 91.9% and 92.6% using the TPOT and ReliefF-Svmb-EXT-TPOT models, respectively. The study findings are crucial for automated and accurate mapping of parthenium weed using high-dimensional geospatial datasets with limited human intervention.

Leaf area index (LAI) is a critical parameter in determining vegetation status and health. In tropical grasslands, reliable determination of LAI, useful in determining above ground biomass, provides a basis for rangeland management, conservation and restoration. In this study, interval partial least square regression (iPLSR) in forward mode was compared to partial least square regression (PLSR) to estimate LAI from in-situ canopy hyperspectral data on a heterogeneous grassland at different periods (onset, mid and end) during summer. The performance of the two techniques was determined using the least relative root mean square error to the mean (nRMSEP) and the highest coefficients of determination (R2p) between the predicted and the measured LAI. Results show that iPLSR models could explain LAI variation with R2p values ranging from 0.81 to 0.93 and low nRMSEP from 9.39% to 24.71%. The highest accuracies for estimates of LAI using iPLSR were at mid- and end of summer (R2p = 0.93 and nRMSEP = 9.39%; R2p = 0.89 and nRMSEP = 10.50%, respectively). Pooling data sets from the three assessed periods yielded the highest prediction error (nRMSEP=24.71%). Results show that iPLSR performed better than PLSR, which yielded R2p and RMSEP values ranging from 0.36 to 0.65 and from 28.44% to 33.47%, respectively. Overall, this study demonstrates the value of iPLSR in predicting LAI and therefore provides a basis for more accurate mapping and monitoring of canopy characteristics of tropical grasslands. Significance: The relationship between LAI and canopy reflectance can be used in iPLSR modelling to provide more accurate mapping and monitoring of canopy characteristics for land management and conservation.

Multidate images provide additional spectral information valuable for mapping plant species. However, correlated or redundant variables created from multiple image data and a large set of variables impede accurate and efficient landscape classification. Nevertheless, with the implementation of an appropriate feature selection method, the full potential of multidate images and subsequent increased classification accuracies may be achieved. Feature selection is a process of automatically selecting features in a dataset that contribute the most to the prediction of a target variable. It improves the classification process in terms of computation cost and predictive accuracy. Feature selection algorithms are typically classified into three groups: filters, wrappers, and embedded. Due to the inherent tradeoffs provided by different feature selection approaches, we hypothesize that a hybrid approach could optimize landscape delineation by leveraging on their complementary strengths. In this regard, we propose a new feature selection method that combines a filter ReliefF, a wrapper Support vector machines-backward (SVM-b), and the embedded random forest in mapping the noxious Parthenium weed using a Sentinel-2 multidate image. The new approach was compared to its three constituents based on the size of optimal feature subsets and classification accuracies. To make comparison and assessment of developed models from selected variables more valuable, the split between training and test datasets was implemented in three different ratios, namely, 3:1(Dataset 1), 1:1(Dataset 2); 1:3 (Dataset 3). The new approach was also evaluated against the multidate image without feature selection and a single-date image. Results showed that the new approach yielded the highest overall accuracies with the smallest optimal feature subsets. For instance, on Dataset 3, the OA was 86.6% with 22 optimal features, whereas it was 84.7% with 35 optimal features using SVM-b, which was the second most performing feature selection algorithm. The new approach yielded higher classification accuracies than the multidate image without feature selection and the single-date image. The findings of this study under-score the capability of hybrid methods to select fewer features from multidate images, with higher predictive accuracies than individual feature selection methods.