Analysis of stable states in global savannas: is the CART pulling the horse?

Abstract

Multiple stable states, bifurcations and thresholds are fashionable concepts in the ecological literature, a recognition that complex ecosystems may at times exhibit the interesting dynamic behaviours predicted by relatively simple biomathematical models. Recently, several papers in Global Ecology and Biogeography, Proceedings of the National Academy of Sciences USA, Science and elsewhere have attempted to quantify the prevalence of alternate stable states in the savannas of Africa, Australia and South America, and the tundra–taiga–grassland transitions of the circum-boreal region using satellite-derived woody canopy cover. While we agree with the logic that basins of attraction can be inferred from the relative frequencies of ecosystem states observed in space and time, we caution that the statistical methodologies underlying the satellite product used in these studies may confound our ability to infer the presence of multiple stable states. We demonstrate this point using a uniformly distributed ‘pseudo-tree cover’ database for Africa that we use to retrace the steps involved in creation of the satellite tree-cover product and subsequent analysis. We show how classification and regression tree (CART)-based products may impose discontinuities in satellite tree-cover estimates even when such discontinuities are not present in reality. As regional and global remote sensing and geospatial data become more easily accessible for ecological studies, we recommend careful consideration of how error distributions in remote sensing products may interact with the data needs and theoretical expectations of the ecological process under study.

Analysis of stable states in global savannas: is the CART pulling the horse? – a comment

“…In their recent paper, Hanan et al (2014) argue that the use of classification and regression trees (CARTs) to calibrate global remote sensing datasets, including the MODIS VCF tree-cover dataset, makes these data inappropriate for analysing the frequency distribution of tree cover. Specifically, multiple authors have used discontinuities (multimodality) in tree-cover distributions to suggest that forest, savanna and even grassland may be discrete biomes and may represent potential alternative stable states for equivalent environmental conditions (Hirota et al, 2011; Staver et al, 2011a, b;Ratajczak & Nippert, 2012;Scheffer et al, 2012).…”

“…Firstly, Hanan et al (2014) use pseudosatellite metrics to produce a pseudo-tree cover product. Their pseudo-satellite metrics are smoothly varying, deterministic functions of tree cover (another pseudo-dataset generated from Sankaran et al, 2005) and climate (including rainfall, seasonality, temperature and growing season length), with additional random noise.…”

“…In addition to their robust ability to delineate complex decision boundaries in our time-series image feature space, regression trees are fast in their implementation and transparent in their interpretation. These combined features led us to select tree models for our global land-cover mapping work (Hansen et al, 2000(Hansen et al, , 2003Hansen et al, 2013) and for that of the standard MODIS land-cover product (Friedl et al, 2002).Secondly, Hanan et al (2014) do not actually demonstrate that the discontinuities in the MODIS VCF tree-cover product are a product of satellite data analysis, or that bs_bs_banner Global Ecology and Biogeography, (Global Ecol. Biogeogr.)…”

“…Specifically, multiple authors have used discontinuities (multimodality) in tree-cover distributions to suggest that forest, savanna and even grassland may be discrete biomes and may represent potential alternative stable states for equivalent environmental conditions (Hirota et al, 2011; Staver et al, 2011a, b;Ratajczak & Nippert, 2012;Scheffer et al, 2012). Hanan et al (2014) use hypothetical data to demonstrate that CARTs can, in principle, produce artificial multimodality in data products. They argue that ecologists must only use remote sensing data when they fully understand the error distributions associated with these products.…”

“…They argue that ecologists must only use remote sensing data when they fully understand the error distributions associated with these products. While we agree with the most general point -that the use of remote sensing products should be informed and deliberate -their analysis overlooks the utility of the CART approach and fails to demonstrate that, in reality, despite extensive validation, MODIS VCF products have any systematic flaws in accuracy.Firstly, Hanan et al (2014) use pseudosatellite metrics to produce a pseudo-tree cover product. Their pseudo-satellite metrics are smoothly varying, deterministic functions of tree cover (another pseudo-dataset generated from Sankaran et al, 2005) and climate (including rainfall, seasonality, temperature and growing season length), with additional random noise.…”