Mitchard, Edward

Bollasina, Massimo

Disney, Mathias

Hancock, Steven

Aquino, Chiara

European Research Council

2023-03-27T17:16:28Z

2023-03-27T17:16:28Z

2023-03-27

Climate change is the greatest crisis that humanity has ever faced, threatening water supplies, food security, species and ecosystems worldwide. The tropical forest biome serves a fundamental regulatory function in the climate system, by absorbing 34% of carbon dioxide from the atmosphere every year. Moreover, tropical forests are unique biodiversity hotspots and provide an essential source of livelihood and socio-cultural identity to local communities. Their integrity is fundamental to the stability of the global climate and to the well-being of the populations that depend on them. Over the last 20 years, pressure from the extractive industries and an increasing global demand for commodities have accelerated the decline of tropical forest cover, with devastating implications for global warming, biodiversity loss and human health. In this scenario, understanding and monitoring spatio-temporal patterns of forest loss has become a complementary tool to national conservation policies to prevent further deforestation. Satellite remote sensing has emerged as the most viable system for estimating changes in forest cover at large spatial scales and in remote regions. In the last decade, thanks to the free availability of high-resolution satellite imagery, yearly maps of forest loss and forest monitoring alert systems have become operational at the national and global scale. Nevertheless, the magnitude and extent of small-scale disturbances, such as those caused by selective logging and understory fires, remain largely uncertain. In the Brazilian Amazon, it is estimated that forest degradation (as caused by selective logging, forest fragmentation, edge effects, forest fire and drought) has surpassed the rate of deforestation, contributing to 73% of the overall above ground biomass (AGB) loss in this region. It is likely that the same trend is true for other tropical areas, setting forest degradation as high priority for forest conservation policies. On the other hand, a lack of reliable ground-truth data on AGB change, together with the limitations of satellite imagery in typically cloudy and densely forested regions, poses significant challenges to reliably mapping forest degradation in the tropics. In this thesis, I use a novel biomass change dataset comprising eight 1-ha permanent sample plots that I established in logging concessions in Peru and Gabon as part of the Forest Degradation Experiment (FODEX). The plots were carefully inventoried by hand and Terrestrial Laser Scanning before and after selected number of trees were removed, giving different percentages of biomass loss values (5-30% AGB loss ha−1). My aim is to use all the available remote sensing imagery, both optical and radar, to find the most accurate methodology and sensor characteristics to detect selective logging in the field sites, with the intent of extending the results over larger spatial scales, for regional and national assessment. In Chapter 3, I focus on optical remote sensing and perform a comparative study of six multispectral satellite sensors ranging between 0.3 - 30 m in spatial resolution, namely WorldView-3, SkySat, SPOT-7, PlanetScope, Sentinel-2 and Landsat 8. For each sensor, I retrieve before/after logging differences in band reflectance, Normalized Difference Vegetation Index (NDVI) and texture parameters to find the best combination of sensor and remote sensing metrics to detect low-intensity logging over the field plots in Peru. The strength of the relationships between the change in these values and field-measured biomass loss (ΔAGB) is analysed using linear regression models. It is found that texture measures correlate more with ΔAGB than simple spectral parameters; moreover, the strongest correlations are achieved for those sensors with spatial resolutions in the intermediate range (1.5 - 10 m), while using a moving square window ranging between 9 - 14 m in length. Maps predicting ΔAGB show very promising results using a texture parameter from the infrared band of 3 m resolution PlanetScope (R2 = 0.97). Persistent cloud cover over large areas of the tropics often prevents optical remote sensing monitoring of vegetation on a continuous basis. Synthetic Aperture Radar (SAR) data from the C-band Sentinel-1 mission provides cloud-free imagery on a 6 or 12- day repeat cycle at 10 m spatial resolution, offering the opportunity to monitor forest disturbances in a timely and accurate manner. In Chapter 4, I use a novel change detection method based on the cumulative sums (CuSum) of Sentinel-1 time-series to map low intensity forest disturbances in Peru and Gabon. Using the maximum of the CuSum distribution, I develop a single change metrics to retrieve location, time and magnitude of the disturbance events. The CuSum algorithm is calibrated using 1239 ha of very high resolution (25 cm) UAV LiDAR measurements of canopy height loss collected by the FODEX project over the field plots and the surrounding forest concessions. A comparison of the CuSum results with the LiDAR reference map yields a 78% success rate for the study site in Gabon and 65% success rate for the study site in Peru, for disturbances as small as 0.01 ha and for canopy height losses as fine as 10 m. In addition to other forest monitoring systems, the methodology outlined in this thesis has the potential of retrieving the magnitude of the disturbance events. A correlation between the CuSum change metric and AGB is found with R2 = 0.95, and R2 = 0.83 for canopy height loss. Low intensity disturbances captured by the CuSum method are largely undetected by state-of-the-art forest monitoring system such as the Global Forest Watch product and the RADD Alert system. While lacking near real-time monitoring capabilities, the CuSum algorithm can be adopted for building yearly maps of forest loss, providing more accurate figures of forest degradation for long-term monitoring purposes. In Chapter 5, the method presented in Chapter 4 is employed for extending the analysis beyond the test sites, retrieving forest disturbance maps at the national scale. An image processing workflow is developed in Python to produce maps containing information on location, day of the year and biomass loss (in % ha−1) in Peru and Gabon for the year 2020. The maps are compared with current large-scale products reporting deforestation and/or degradation in humid tropical forests. Overall, the methods presented in this thesis refine the ability of both optical and radar satellite data to retrieve small-scale disturbances in dense tropical forests. The conclusions of the analysis support the usage of free-of-charge satellite data: free access to PlanetScope imagery is granted by Norway’s International Climate & Forests Initiative, while the Copernicus Open Access Hub provides free and open access to Sentinel-1 data. Therefore, all methods are available to the public, with the aim of fostering citizen science to monitor the state of tropical forests.

https://hdl.handle.net/1842/40446

http://dx.doi.org/10.7488/era/3214

en

The University of Edinburgh

SAR

optical satellite imagery

Terrestrial Laser Scanning

remote sensing imagery

optical remote sensing

small-scale disturbances in dense tropical forests

free-of-charge satellite data

Mapping small-scale tropical forest disturbances using SAR and optical satellite imagery

Thesis or Dissertation

Doctoral

PhD Doctor of Philosophy