A resilience sensing system for the biosphere

Abstract

We are in a climate and ecological emergency, where climate change and direct anthropogenic interference with the biosphere are risking abrupt and/or irreversible changes that threaten our life-support systems. Efforts are underway to increase the resilience of some ecosystems that are under threat, yet collective awareness and action are modest at best. Here, we highlight the potential for a biosphere resilience sensing system to make it easier to see where things are going wrong, and to see whether deliberate efforts to make things better are working. We focus on global resilience sensing of the terrestrial biosphere at high spatial and temporal resolution through satellite remote sensing, utilizing the generic mathematical behaviour of complex systems-loss of resilience corresponds to slower recovery from perturbations, gain of resilience equates to faster recovery. We consider what subset of biosphere resilience remote sensing can monitor, critically reviewing existing studies. Then we present illustrative, global results for vegetation resilience and trends in resilience over the last 20 years, from both satellite data and model simulations. We close by discussing how resilience sensing nested across global, biome-ecoregion, and local ecosystem scales could aid management and governance at these different scales, and identify priorities for further work. This article is part of the theme issue 'Ecological complexity and the biosphere: the next 30 years'.

Keywords: biosphere; climate change; ecosystems; recovery rate; remote sensing; resilience.

Figure 1.

Global maps of modelled vegetation resilience: (a) Autocorrelation function (ACF, equivalent to AR(1)) of modelled monthly NPP (2000–2019) from the LPJmL global vegetation model forced with bias-corrected climate input from the GFDL-ESM2M climate model, where high ACF (AR(1)) corresponds to low resilience (and vice versa), and negative ACF (AR(1)) suggests poor fit of an autoregressive model. (b) Trend in ACF (AR(1)) of monthly NPP (2000–2019) from the same model scenario, measured as Kendall τ rank correlation coefficient, using a 10-year sliding window. Prior to analysis, data are seasonally detrended at the pixel level (with seasonal and trend decomposition using Loess; ‘STL’) using the bfast package in R [–82]. Regions where NPP is less than 10⁻⁹ kg m⁻² s⁻¹ are filtered out and shown in dark grey.

Figure 2.

Global maps of remotely sensed vegetation resilience: (a) Mean AR(1) of monthly NDVI (2001–2020 from MODIS), where high AR(1) corresponds to low resilience (and vice versa), and negative AR(1) suggests poor fit of an autoregressive model. (b) Trend in AR(1) (Kendall τ rank correlation coefficient) of monthly NDVI 2001–2020, using a 10 year sliding window. Prior to analysis, the seasonal cycle is removed by subtracting the 20 year monthly average, then a 25-month moving average is subtracted to remove the trend (a sample of 100 random pixels were analysed to confirm that this seasonal detrending approach gives comparable results to the method used in figure 1). Regions where NDVI is less than 0.18 are filtered out and shown in grey.

Figure 3.

Zooming into trends in AR(1) of monthly NDVI (2001–2020) in South and East Asia (from figure 2b). This shows example ecoregions from the three biomes with globally the most positive AR(1) trends (montane grasslands and shrublands, tropical and subtropical dry broadleaf forests, and temperate coniferous forests). In India, the Central Deccan Plateau dry deciduous forests (τ =0.67). In Myanmar/China, a group of three coniferous forest ecoregions (west to east): Nujiang Langcang Gorge alpine conifer and mixed forests (τ = 0.60), Hengduan Mountains subalpine conifer forests (τ = 0.78), Qionglai–Minshan conifer forests (τ = 0.79). In northern China: Ordos Plateau steppe montane grassland (τ = 0.62). Regions where NDVI is less than 0.18 are filtered out and shown in grey.