Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

Abstract

Vegetation maps are models of the real vegetation patterns and are considered important tools in conservation and management planning. Maps created through traditional methods can be expensive and time‐consuming, thus, new more efficient approaches are needed. The prediction of vegetation patterns using machine learning shows promise, but many factors may impact on its performance. One important factor is the nature of the vegetation–environment relationship assessed and ecological redundancy. We used two datasets with known ecological redundancy levels (strength of the vegetation–environment relationship) to evaluate the performance of four machine learning (ML) classifiers (classification trees, random forests, support vector machines, and nearest neighbor). These models used climatic and soil variables as environmental predictors with pretreatment of the datasets (principal component analysis and feature selection) and involved three spatial scales. We show that the ML classifiers produced more reliable results in regions where the vegetation–environment relationship is stronger as opposed to regions characterized by redundant vegetation patterns. The pretreatment of datasets and reduction in prediction scale had a substantial influence on the predictive performance of the classifiers. The use of ML classifiers to create potential vegetation maps shows promise as a more efficient way of vegetation modeling. The difference in performance between areas with poorly versus well‐structured vegetation–environment relationships shows that some level of understanding of the ecology of the target region is required prior to their application. Even in areas with poorly structured vegetation–environment relationships, it is possible to improve classifier performance by either pretreating the dataset or reducing the spatial scale of the predictions.

Community patterns and environmental drivers in hyper‐diverse kwongan scrub vegetation of Western Australia

“…Our study did identify some of the environmental factors as important drivers, but, on the whole, the unexplained variability remained very high. These assertions find additional support through Macintyre, Van Niekerk, Dobrowolski, Tsakalos, and Mucina () who implemented spatially explicit modelling to demonstrate the high predictability of well‐structured vegetation‐environment relationships of the Northern Kimberley woodlands compared to our Eneabba kwongan data. Therein, Macintyre et al.…”

“…Therein, Macintyre et al. () found that the vegetation of Eneabba was difficult to spatially separate owing to the high redundancy of the environment‐vegetation relationship, particularly true at the Community spatial scale. We, therefore, hypothesised that processes other than ecological sorting, albeit working along long‐term evolutionary scales lead to the formation of vegetation structures showing poor correspondence with the current ecological fabric of the landscape.…”

Composition and ecological drivers of the kwongan scrub and woodlands in the northern Swan Coastal Plain, Western Australia

“…Spatially explicit modelling of that system also found a weak relationship (low predictability) between the vegetation and the environment when compared to the Northern Kimberly Eucalyptus Woodland which was well structured along environmental gradients (Macintyre et al . ).…”

Ecology, biodiversity and mining: science and solving the challenges

“…This mapping was done in times where the only accessible remote-sensed data were aerial photos. The world of vegetation mapping has long moved on and embraced tools such as the use of vegetation plot data and formalised classifications, satellite imagery interpretation, and machine-learning based predictive modelling (Mucina & Daniel 2013;Macintyre et al 2018).…”

“…The maps of PNV can have limitations. For example, in the production of maps B and C (Macintyre et al 2018), it was shown that at the coarse scale, the predictions were robust, while at the finer scales, additional factors of ecological redundancy made these predictions unreliable. In contrast, map D shows a supervised per-pixel classification of the vegetation of the area using multi-temporal, remotely sensed data from the Sentinel 2 platform (equivalent to ~ 1:12,500 scale) which returned a highly reliable classification.…”