Published in


Soil Erosion Watch — A Bootstrapped Approach to Identify the World’s Degrading Soils

A deep dive into the underbelly of Earth Observation (EO) environmental monitoring applications: Soils. And why they matter in the fight against climate change

Identifying Bare Soil in Space and Time

I lied. Bare soil can be observed occasionally in the middle of the rainforest, in the form of mineral-rich clearings, also referred to as “Bais”, and which, contrary to bare soil in farmland and rangeland, has tremendous ecosystem value (read here why). This one was spotted in the Sangha-Mbaéré district of the Central African Republic. Not even the Google satellite mosaics generated from years of data manage to be fully cloud-free in that part of the World…
Bare soil Synthetic Composite using GEOS3 for the year 2020 in the Nakuru County of Kenya. These are the pixels we are interested in, and no cloudy pixels in sight ☀️😎

Modelling Soil Erosion Hazard

R — Rainfall erosivity factor (

K — Soil erodibility factor (−1)

LS — Slope length (L) and steepness factor (S) (dimensionless)

Rills represent the first step of water erosion on slopes, and can eventually lead to more severe erosion in the form of gullies
  • The covariates they are derived from are of relatively low resolution, ranging from 30m (Digital Elevation Model) to 1km (Rainfall Erosivity).
  • Their temporal variability is limited to long-term trends (multi-year). So unless, through some divine powers, we would be able to control global rainfall patterns in clacks of fingers (R), or flip the full soil column upside down (K), or even move mountains the literal sense (LS), the estimates of these factors should hold for a few years. In fact, the R-factor used is essentially based on rainfall records from the 2000–2010 period, and the K-factor on covariates representative of the year 2017 at best. It isn’t ideal, but is the best available to derive this global baseline estimation.
The mountains that Mr Montague moves are metaphorical, and does not have an impact on the LS-factor estimation

V — Vegetation factor (dimensionless)

L — Landscape factor (dimensionless)

Result of the 3x3 pixell (30m) Sobel filter, highlighting within-field punctual and linear features, which has a positive impact on soil erosion

A — Average annual soil erosion rate in soil mass per unit area per year (t.ha−1.year−1)


  1. Bare Soil RGB composite: This is the layer resulting from the GEOS3 bare soil synthetic data generator from Demattê et al., 2020. The default data displayed is that of Kenya for the year 2020.
  1. Bare Soil Frequency BSf: The number of times the cloud-free sentinel-2 observations were determined to be bare soil by GEOS3 in the time series, divided by the total number of cloud-free observations. Since bare soil most likely occurs in the drier season when cloud cover is less likely, this frequency may be over-estimated. This is more likely to be the case in sub-tropical regions with a pronounced rainy season.
  1. The sustainability factor S: This factor combines the effect of both the cover vegetation factor V and the landscape factor L, and plots its inverse (S = 1 / (V*L)), so as to be in the range [0–1], with 0 being favorable to sustainably managed land, and 1 being unfavorable. Many of the areas with a high sustainability factor value may be open mining pits or rocky outcrops, which may impact the aggregated statistics on soil loss for a given jurisdiction. To mitigate this effect, values above 0.9 are masked as these correspond to areas of total vegetation absence and imperviousness, and most likely do not correspond to cropland or grassland soils.
  1. Soil Erosion Hazard A: This layer is the result of the revised RUSLE equation presented in this post. It is the best possible approximation we can make of annual soil loss using an empirical model like RUSLE in combination with global datasets made available by the open-source community. The novelty here is the integration of high-resolution multi-temporal Sentinel-2 covariates, namely Sentinel-2 derived Sobel convolution Sf and the temporal, bare soil frequency-weighted (BSf) fraction of green vegetation cover FCoverm.


Global Charts

  1. The proportion of the total area observed bare/non-bare in the selected time period. In the case of the arid area of Al-Fasher, almost every pixel in the image has been observed bare, and the 1.7% exception are likely sparsely located woody perennials.
  2. The bare soil frequency histogram, summarizing the distribution of bare soil pixel’s bare frequency. Unsurprisingly, the distribution is ramping up towards the high frequency values, as the majority of the pixels observed correspond to land degraded beyond arable.
  3. The annual soil loss rate histogram. The distribution was normalized using the natural logarithm function because the majority of the soil erosion rates typically fall between 0 and 10 t.ha-1.year-1, with a runaway effects of a few values peaking at 30+ t.ha-1.year-1, which are invisible in a plot if not normalized. The further to the right the peak of the distribution is, the more significant the soil loss hazard is in the area. The skewness of the distribution towards the right is also an important indicator of soil erosion hazard. All values below 0 represent minor erosion (< 1 t.ha-1.year-1), so if the histogram barely exceeds 0 on the x-axis, the region is generally speaking not prone to erosion.

On-Draw FCover plots



Remote monitoring | Soil Regeneration

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
William Ouellette

Regenerating degraded 🌍 and eroded soils 🌱, one pixel at a time🛰️