Deforestation-free supply chain for cacao
How can remote sensing help for deforestation-free supply chain for cacao
Over the last five decades, world cacao cultivation has doubled through deforestation, resulting in the disappearance of about 15 million hectares (ha) of global forest cover.
Deforestation at Cacao del Peru Norte’s plantation in Tamshiyacu, Loreto, Peru. (EIA)
To minimize the deforestation footprint of cacao cultivation, deforestation and degradation-free supply chain has been implemented.
Deforestation and degradation-free supply chain contains actions stemming from commitments that private sector actors have made to eliminate deforestation along their supply chains and operations.
To measure the impact of the tool, deforestation in specific places needs to be tracked over time and the supply chain.
Remote sensing technology can offer an unbiased constant and consistent data stream on how vegetations are changing in near real-time.
The improved resolutions of map information based on Landsat and Sentinel imagery enable cacao sector stakeholders to assess deforestation risk better.
Also, it can allow farmers to achieve viable livelihoods without threatening the parks and reserves. With remote sensing science, the stakeholders can use the data to better and faster complete risk assessments and end deforestation and forest degradation in the global cocoa supply chain.
Thanks to freely available radar and optical satellite imagery from the Copernicus and Landsat programs, it is now achievable to cover global supply chains in near real-time.
About the open source and free satellites..
1. Landsat
Landsat 8, June 28, 2018 - September 12, 2019 (source: https://earthobservatory.nasa.gov)
The Landsat satellites take 30m wide resolutions and have a revisit frequency of 8-16 days. Landsat 4 was launched on July 16, 1982, and the latest Landsat 9 was launched on September 27, 2021. With its long record of continuous measurement, these characteristics help develop land monitoring algorithms that take advantage of the seasonal and long-term trends in the data. Landsat 4 and 5 acquire data in 7 bands from two separate sensors: Thematic Mapper (TM) sensor and Multispectral Scanner (MSS). Landsat 7 carries the Enhanced Thematic Mapper Plus (ETM+) sensor and has 8 bands. Landsat 8 and 9 receive data in 11 bands from two separate sensors: the Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS). (More information: https://www.usgs.gov/landsat-missions/landsat-satellite-missions)
Comparison of Landsat 7 and 8 bands with Sentinel-2 (Source: www.usgs.gov)
2. Sentinel-2
The Sentinel-2 satellites are on a multi-spectral imaging mission, taking wide swath and high-resolution images. Its data became available on the 23rd of June 2015. It has 13 spectral bands ranging from visible to near and medium infrared with a spatial resolution between 10 m and 20 m with a revisit period between 5 and 10 days, involving a constellation of two twin satellites (Sentinel-2A and Sentinel-2B). (More information: https://sentinel.esa.int/web/sentinel/missions/sentinel-2)
3. Sentinel-1
Deforestation observed by Sentinel-1B (source: digital geography)
Sentinel-1 operates in C-band (5.6 cm) wavelength and provides the single or dual-polarization capability. The Sentinel-1 mission comprises a constellation of two satellites, Sentinel-1A and Sentinel-1B, which offers a 12-day (ground track) repeat cycle for one satellite, and a 6-day (ground track) repeat for two satellites. It observes in ascending (south-to-north) and descending (north-to-south) directions. (More information: https://sentinel.esa.int/web/sentinel/missions/sentinel-1)
Problem of optical data for detecting deforestation
Sentinel-2 images over the Austrian Alps taken between July and August 2016 (© processed by EOX, contains modified Copernicus Sentinel data)
But it is very challenging to detect deforestation during rainy seasons and cloudy periods, using only optical images from satellites like Sentinel-2 or Landsat.
Sometimes, it could take months to get the cloud-free images. But don't worry! There are some solutions to tackle this problem.
(Realistic) methods for detecting deforestation
Annual composite images from my google earth engine app using Sentinel-2 (2019 vs 2021)
First, we can apply bitemporal image comparison, which is used for the classic image change detection method. However, we do not use two single images but median composite images here. Of course, there are still some major drawbacks. For example, annual composite images will have significantly fewer clouds and haze, but you can detect the changes only annually. You can use the shorter term; there is a higher possibility that the composite images contain more clouds and haze effect.
Around 95% of the global cocoa production originates from smallholders who work on land plots of 1 to 3 hectares. One hectare means 10,000 m 2 . So if the cocoa forestry has a square form, Sentinel-2 images will show deforestation with ten by ten pixels in true color composite, and Landsat-9 images will deliver it with only 3.3 by 3.3 pixels in the true-color composite. Of course, spatial resolution is not the only factor affecting deforestation monitoring, but it still significantly impacts change detection accuracy.
Second, time series analysis techniques can be applied to detect deforestation. Time series analysis is a popular and effective method for abrupt change detection. It uses repeat data in the locations (pixels) where the forest might have suffered disturbance at a specific time.
Time series analysis in remote sensing can be based on variables derived from the original data prior to analysis. Vegetation indices such as the Normalized Difference Vegetation Index (NDVI), the enhanced vegetation index (EVI), the Leaf Area Index (LAI), the soil water index (SWI), and many other indices or feature space components, such as Landsat Tasseled Cap components are commonly used variables in time series analysis for forest monitoring. Thematic variables are variables, for example, land-cover and land-use information, that are derived from classification or regression approaches prior to time series analysis. Data sets from thematic variables are usually binary data sets with two classes (water/non-water, forest/non-forest).
Remote sensing time series analysis contains three components; trend, seasonal, and residual components. The trend component focuses on a long directional term and sometimes requires several decades. The remote sensing time series analysis of the forest's change dynamics contains a seasonal component (i.e., monthly, seasonal, annual, or multi-annual data) and short-term fluctuations.
Here I would like to mention one of the famous algorithms for time series analysis using landsat: LandTrendr (Landsat-based Detection of Trends in Disturbance and Recovery). This algorithm aims to temporally segment a time series into distinct periods representing stable conditions, disturbance periods, and recovery. LandTrendr was originally implemented in IDL(Interactive Data Language), but now it has been implemented to the GEE platform.
LandTrendr pixel time series segmentation. Image data is reduced to a single band or spectral index and then divided into a series of straight line segments by breakpoint (vertex) identification. Source: LT-GEE Guide ( https://emapr.github.io/LT-GEE/landtrendr.html)
As an example of time series analysis in remote sensing, it is impossible not to mention the "Global Forest Change" from the University of Maryland ( https://www.globalforestwatch.org/, Hansen, et al., 2013) , which is also implemented as a dataset in Google Earth Engine.
What is the secret behind Global Forest Watch?
First, they process Landsat Level 1 images.
Important procedures here: Coverting integer digital numbers into top-of-atmosphere relfectance, Cloud and cloud-shadow masking, implementing reflectance normalization procedures.
Next, they aggregate individual Landsat images into 16-day composite and store the composites in geographic coordinates and organize in the form of 1 x 1 degree tiles. The images are resampled by nearest neighbor resampling method and the images with best QF are selected.
Then, they select the observations with the best available quality from four years of data are selected for change detection from the corresponding year.
C: the corresponding year time-series
P: the proceeding year time-series
After selecting clear-sky observations and constructing data time-series, reflectance and its change distribution statistics are extracted from the time-series data.
The inter-annual spectral reflectance difference can be visualized by combining the same statistics extracted from different years.
This feature shows the composites of selected 2018 annual change detection metrics.
Global Forest Watch is absolutely powerful.
Now they implement Sentinel-2, GEDI, Sentinel-1 and even Planet data.
Are there still any problems?
But time series analysis also has challenges. Because it detects the changes within the complete long-term data sets involving seasonal variation, it still includes errors derived from geometric errors, sensor errors, atmospheric scattering, and cloud effects.
So, what can we do to avoid cloud effects?
Third, we can apply radar data, which has the advantage of all-weather capability for monitoring deforestation and degradation. Microwave remote sensors (radar) are essentially cloud-penetrating and can guarantee continuous monitoring through clouds. For tropical nations, this is particularly meaningful as constant cloud cover severely limits the availability of optical data (Flores et.al., 2019).
But the problem is that the existing open source and free radar satellite is Sentinel-1, which operates in C-band. At the C-band wavelength, the radar signal scatters directly on the leaves at the top of the canopy without penetrating significantly through the foliage. Also, after deforestation, rough soil conditions and remnant debris can produce a strong backscatter.
Upcoming satellites for monitoring deforestation
NISAR (2023)
NISAR radar instrument from NASA and ISRO provides all-weather, day and night imaging of nearly the entire land, considering ascending and descending orbits. NISAR utilizes two SAR instruments operating at different frequencies; L-band and S-band. Depending on the operating mode, NISAR’s orbiting radar can image at resolutions of 3-50 meters.
Sensitivity of SAR measurements to forest structure and penetration into the canopy at different wavelengths used for airborne or spaceborne remote sensing observations of the land surface. Credit: NASA SAR Handbook.
L-band is especially useful for mapping activities underneath canopies in dense forests due to its ability to penetrate vegetation covers. Longer wavelengths like L-band consequently penetrate through the forest canopy (since the small leaves are transparent) and interact with the larger structures such as the trunks and larger branches of trees. On the other hand, the shorter wavelength, such as C-band, is more sensitive to sparse and low biomass vegetation.
Comparison between ALOS-2 PALSAR-2 images (L-band) and Sentinel-1 images (C-band).
Backscatter at L-band shows the secondary growth after deforestation very vividly. Since the structure of leaves and small twigs in the top canopy of mature forests resemble that of secondary growth of forests, both provide similar backscatter responses at C-band.
Reference
Flores, Africa & Herndon, K. & Thapa, Rajesh & Cherrington, Emil. (2019). The SAR Handbook: Comprehensive Methodologies for Forest Monitoring and Biomass Estimation. 10.25966/nr2c-s697
Gao, Y., Skutsch, M., Paneque-Gálvez, J., & Ghilardi, A. (2020). Remote sensing of forest degradation: A review. Environmental Research Letters, 15(10), 103001. https://doi.org/10.1088/1748-9326/abaad7
Hansen, M.C. & Potapov, Peter & Moore, Rebecca & Hancher, M & Turubanova, Svetlana & Tyukavina, Alexandra & Thau, D & Stehman, Stephen & Goetz, Scott & Loveland, Thomas & Kommareddy, Anil & Egorov, Alexey & Chini, L & Justice, C.O. & Townshend, J.. (2013). High-Resolution Global Maps of 21st-Century Forest Cover Change. Science (New York, N.Y.). 342. 850-853. 10.1126/science.1244693.
Kennedy, R. E., Yang, Z., & Cohen, W. B. (2010). Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr — Temporal segmentation algorithms. Remote Sensing of Environment, 114(12), 2897–2910. https://doi.org/10.1016/j.rse.2010.07.008
Kuenzer, C., Dech, S., & Wagner, W. (2015). Remote Sensing Time Series Revealing Land Surface Dynamics: Status Quo and the Pathway Ahead. In C. Kuenzer, S. Dech, & W. Wagner (Eds.), Remote Sensing Time Series: Revealing Land Surface Dynamics (pp. 1–24). Springer International Publishing. https://doi.org/10.1007/978-3-319-15967-6_1
Potapov, & Hansen, & Kommareddy, Anil & Turubanova, Svetlana & Pickens, & Adusei, Bernard & Tyukavina, & Ying, Qing. (2020). Landsat Analysis Ready Data for Global Land Cover and Land Cover Change Mapping. Remote Sensing. 12. 426. 10.3390/rs12030426.
Potapov, Peter & Tyukavina, Alexandra & Turubanova, Svetlana & Tolero, Yamile & Hernandez-Serna, Andres & Hansen, Matthew & Saah, David & Tenneson, Karis & Poortinga, Ate & Aekakkararungroj, Aekkapol & Chishtie, Farrukh & Towashiraporn, Peeranan & Bhandari, Biplov & San Aung, Khun & Ngyuen, Quyen. (2019). Annual continuous fields of woody vegetation structure in the Lower Mekong region from 2000‐2017 Landsat time-series. Remote Sensing of Environment. 232. 10.1016/j.rse.2019.111278.
Verbesselt, J., Hyndman, R., Newnham, G., & Culvenor, D. (2010). Detecting trend and seasonal changes in satellite image time series. Remote Sensing of Environment, 114(1), 106–115. https://doi.org/10.1016/j.rse.2009.08.01
Wessel, M., & Quist-Wessel, P. F. (2015). Cocoa production in West Africa, a review and analysis of recent developments. NJAS-Wageningen Journal of Life Sciences, 74, 1-7
