From Pixels to Preservation: Utilizing Remote Sensing to Identify Community Techniques to Save Tree Cover
By Anthony Kwabena Sarfo, Joy Heitlinger, Michael Darko Ahwireng, Basma Albanna, Robin Nowok, Tobias Kunde
The following piece was originally published by the GIZ Data Lab on 11 April 2024.
Climate change poses significant challenges for rural communities in the North of Ghana. Due to their dependence on climate-sensitive sectors like agriculture, forestry, and natural resource management, they are highly vulnerable to its effects. Implementing adaptation measures that improve resilience and address the impacts of climate change is essential for reducing their vulnerabilities. This article explores the benefits of new approaches, particularly the utilization of remote sensing data and the Data-Powered Positive Deviance (DPPD) method, in preserving Ghana’s trees and empowering local communities.
Benefits of the Data-Powered Positive Deviance method
Adaptation measures in rural agricultural areas rely on behavioral changes within communities, utilizing collective wisdom and lived experiences for sustainable outcomes and social acceptance. Community-identified adaptation measures are sustainable since they capitalise on local knowledge, resources, and local dynamics and governance for implementation. Implementation of the defined measures enhances the community’s shared responsibility for resilience and a belief in their ability to succeed. To identify community-led approaches we used the DPPD method. It addresses the subject by asking: What can we learn from outliers? Which information can we get from people who do things differently?
The method builds on the positive deviance approach which is based on the observation that in every community, there are specific individuals or groups whose uncommon behaviors and strategies enable them to find better solutions to problems than their peers, while having access to the same resources and facing similar or worse challenges. While the positive deviance approach relies mainly on primary data collection to identify and learn from outliers, the DPPD method uses preexisting non-traditional data sources instead of, or in conjunction with, traditional data sources. Non-traditional data, in this context, broadly refers to data that is digitally captured, mediated, or observed (e.g., satellite imagery).
In our case, we wanted to investigate how communities manage tree cover, trying to identify those who preserve or increase their tree cover in the community, contrary to common observations. Earth observation data was used to identify those outlier communities. After a training organized through SNRD Africa, Asia, and GIZ Data Lab, we gained first-hand experience and would like to share our learnings and approaches with you. The key question was: what do communities do differently to sustain tree cover?
Stage 1: Assess Problem-Method Fit
The first stage of the DPPD method is based on a set of questions to help decide whether or not to apply it. First, the problem has to be defined. This means to properly describe the problem, its root causes, the target group that is impacted by the problem, and the desired outcome. The second step in this stage is checking the suitability. In this method, it means to identify if there is a change in practice or behavior and a likelihood that positive deviants exist. Third, it has to be tested if it’s feasible to implement the method, regarding data availability and implementation capabilities. In the last step, the risks and benefits of applying the DPPD method have to be assessed to check its desirability. This way, unintended consequences and issues with data privacy can be avoided.
As a first step, the team defined the problem: The tree cover in northern Ghana is under threat. Forests and trees are disappearing. This is due to various reasons, including bushfires, changes in land use for agriculture, housing, animal rearing, and the growing demand for firewood. For rural and vulnerable communities, sufficient tree cover is, however, necessary for economic activities in the North (such as the production of shea butter and sustainable wood use), and it protects from floods and droughts. In northern Ghana, 97.9% of the households are engaged in crop farming, with a few involved in other forms, such as poultry and livestock. Agricultural production is the main activity, practised mainly on seasonal and subsistence levels. The sector continues to be the largest source of employment, with many working as smallholder farmers. During the dry season, charcoal burning becomes an alternative livelihood strategy and support system. There is a high demand for its sale since charcoal provides about 64% of Ghana’s domestic energy requirements. In addition to other factors such as security and bad farming practices, tree cover is at risk, leading to rising deforestation.
Amidst these trends, it has been observed that there are positive deviant communities that have successfully preserved their tree cover. Identifying and scaling up these practices could significantly impact discussions on resilience among communities. This highlights the suitability of the DPPD method. As for the feasibility of applying the method, capturing the contextual reality and outcomes of the target group is based on both traditional and non-traditional data. Traditional data from statistical services and municipal and district assemblies aided in the identification of communities with similar conditions and challenges who formed our study population. The use of remote sensing-derived vegetation indices and change detection, together with surveys, made it possible to detect forest degradation and deforestation in those communities. The methodology is desirable since positive deviant communities may have some advantage in being resilient as compared to the others.
Stage 2: Determine Positive Deviants
In this stage of the DPPD method, the positive deviants are identified. Using non-traditional data, communities are evaluated regarding their performance and subsequently clustered. Before moving to the next stage, in which underlying factors are uncovered, the identified positive deviants are validated in a detailed process.
For the identification process, our team used satellite imagery to detect forest degradation in the communities in the project area. To detect those changes, we chose the Normalized Difference Vegetation Index (NDVI), a remote sensing-based vegetation index. To compute NDVI for each pixel, Sentinel 2 data were sourced through the Google Earth Engine platform and used to create NDVI images per pixel via band combination techniques. Subsequently, we clustered the pixels within a community into homologues based on land cover (Dynamic World) and precipitation data to control for climatic and biophysical variability.
To identify communities where trees are intact, we used a technique called local scaling (Prince et al. 2009). This method assesses the health of vegetation by comparing it to a baseline value that reflects the expected natural state unaffected by extreme climate and human activity. In simple terms, without human impact or severe weather conditions, there should be no difference from this reference value. By using this approach, we adjusted for differences in vegetative growth caused by varying local factors, as productivity is measured against its maximum potential within each homologue.
To assess tree cover changes between 2018 and 2023, the research team focused on the dry seasons and acquired images in December for the study period to maximize the contribution of trees and omit the contribution of crops to the NDVI signal. Using those images, Monthly median NDVI values were calculated for each pixel. To identify significant changes in NDVI, we calculated z-scores from the slope, rate of change of the difference in the NDVI value of the respective pixel, and 90th percentile value for the cluster the pixel belongs to. The pixels with a z-score greater than 2 were identified as Positive Deviant (PDs) pixels. In addition, we used the mean value of all z-scores within a 1km buffer zone surrounding each community to rank the communities. To offer a balanced insight into the trends observed, we chose the top 20 as the PDs and the tail 20, as the Negative Deviants (NDs). In summary, based on their mean z-score within a 1 km buffer zone, positive deviants represent communities with substantial tree cover gains, while negative deviants highlight areas experiencing significant deforestation.
Prior to starting stage 3, in which the communities are visited and underlying factors are identified, a preliminary validation process was conducted. Utilizing high-resolution satellite imagery, we manually analyzed the top 20 communities exhibiting the highest z-scores. Focusing on the 1km buffer zone, distinctive structural or behavioral patterns present in each were identified manually by experts. For instance, the integration of tree cover into agricultural areas over time or the presence of particular tree formations. We applied a similar procedure to the 20 communities showing the highest negative z-scores, aiming to pinpoint negative behavioral activities contributing to reduced tree cover, such as burnings or desertification.
Stage 3: Discover Underlying Factors
The team is presently at this stage of the methodology, intending to unearth the underlying factors contributing to the maintenance and increase in tree cover amidst diverse factors contributing to deforestation in the project area. Based on this rigorous process, decisions were made regarding which positive and negative deviant communities would be visited. Nevertheless, before starting the fieldwork, the list of communities will be validated by a local expert and additional information regarding their socio-demographic and structural attributes will be collected.
The work on the ground, elementary to identify local patterns and behaviors, is in the preparatory steps and should start soon. Initial hypotheses are strong bylaws, agroforestry practices, a heightened perception of climate risks, and how tree cover is connected, among others. By understanding these factors, communities can adopt effective adaptation strategies.
Background of the Project
The Resilience Against Climate Change project aims to build resilience in 200 rural communities across 14 districts in Northern Ghana. Through participatory community action planning and the adoption of conservation agriculture, the project empowers 200 rural communities to identify and implement locally appropriate adaptation strategies.
Utilizing land use maps generated through joint efforts, communities develop climate-resilient action plans. Spatial data and mapping play a critical role in facilitating community engagement, knowledge sharing, participatory decision-making processes, and adaptation strategies generation that make sense in the particular setting.