A GEE JS API implementation of the spatio-parametric Rao's Quadratic Entropy (Rao's Q).
Rao's Q is a measure of landscape diversity. It calculates inter-pixel spectral distances, which are directly related to landscape heterogeneity, a proxy for describing species habitats (Rocchini et al., 2005).
The approach is most adequately applied to (semi-)natural landscapes with low levels of fragmentation. In other words, it performs poorly over highly-fragmented agricultural landscapes, and well over closed-canopy forest, woodlands and rangelands.
Spectral indices such as NDVI may be used as input to computing the Rao's Q, in which case an accurate land cover product must be used to mask out areas of low vegetation cover. The computation of the Rao's Q has the tendency to inflate the diversity score around land cover boundaries (e.g. forest edges), hence the importance of masking out these locations.
Rao's Q example applied to NDVI data for California:
(Taken from Rocchini et al., 2021)
Since the Rao's Q based on NDVI can be derived from satellite time series such as Sentinel-2 and Planetscope (NICFI basemaps),
we can also look at the behavior of Rao's Q over time by computing the coefficient of variation (CV) and slope of change (trend),
to understand the behavior of Rao's Q over time. The code link in the Examples section below provides an example of how to compute the temporal Rao's Q metrics.
Rao's Q can take as input a canopy height model in the case of forest diversity monitoring. In such a case, Rao's Q represents the canopy height heterogeneity, which is a robust proxy of plant diversity in closed-canopy forests.
Rao's Q example applied to GEDI-derived Canopy Height data:
(Taken from Torresani et al., 2023)
The workbench provided in the Examples section combines canopy height and NDVI-derived Rao's Q into a Generalized Additive Model,
thus taking into account both canopy height and NDVI heterogeneity in the calculation.
The details about parameter setting and their associated methods is described in the main script rao.js.
To use the code, either copy-paste it to your code editor environment, or simply refer to the publicly-available script with the following line:
var diversity = require('users/soilwatchtech/biodiversityApp:rao.js');- Rao's Q applied to palm oil plantations and native forest mosaics in Kalimantan, Indonesia
The script implementing this use case is located under the examples folder, and can be accessed via this
link directly in the code editor.
The palm plantations (homogeneous, left and center) show lower Rao's Q values (run on NDVI median for 2022) when compared to other primary/secondary forests in the east of the area.
The JavaScript code for Rao's Q (rao.js) runs in the GEE code editor without installing additional packages.
In order to run the NICFI Basemaps example, one needs to request access to NICFI Basemaps data in Google Earth Engine.
The example provided (examples/rao_example.js) relies on set of utilities which can be found in the utils.js script.
Other dependencies include:
- GEE Community Datasets: Some datasets provided as community datasets by Samapriya Roy are instrumental to the example provided.
- GEE Community's EE Palettes
- SoilWatch for driving the meaningful work that lead us to discovering and using innovative algorithms for mapping biodiversity and other landscape indicators.
- Duccio Rocchini from University of Bologna, for his decades long research on monitoring biodiversity from space!
- The The Planet-NICFI partnership which is providing us amazing cloud free monthly basemaps to monitor the world's tropical landscapes.