Hotspot Detection for Monitoring New Trends in Carbon Sequestration in Systems of Trees Outside of Forests

Most forests in the tropics are losing carbon and are a net source of emissions, but there is growing evidence that landscapes of trees outside of forests (TOF) around the world are increasing tree biomass and are important potential sinks for carbon (Zomer et al. 2016). The most important hotspots for this phenomenon are TOF systems in agricultural landscapes in semi-arid zones. Africa is a particularly important region, as recent studies have identified the occurrence of farmer-mediated and promoted increases in biomass in savanna and woodland landscapes in rural areas. For example, in West Africa Brandt et al. (2018) observe elevated tree biomass around village areas compared to stocks in natural savannas. This proposed project will deploy a remote sensing method that combines pixel mixture modeling using a combination of Landsat and Sentinel-2 data with digital image analysis using very high-resolution satellite (VHR) data to map important TOF landscapes in West and East/Southern Africa. The aim is to estimate the area today, as well as examine its trend and future potential. The question is important because of the potential scale and magnitude of land area in African semi-arid lands that currently support TOF, and which could be increased through policy and economic development interventions. Furthermore, TOF systems can have significant benefits to local communities through agroforestry and other tree-based production systems that bring higher economic returns to local livelihoods, and enhanced environmental benefits for land rehabilitation.

Our proposed mapping method uses a stratified two-stage approach with medium and very high resolution remote sensing data. Landsat and Sentinel-2 data are used in Stage 1 to build a continuous fields data layer, fC, across the study area landscape based on the method of Matricardi et al. (2007, 2010, 2013) to identify areas of high TOF cover. In areas of high fC we use VHR data from WorldView, GeoEye, QuickBird, and Planet (0.6-3 m) to detect individual tree crowns as landscape objects on a sample basis using methods described by Dieng (2015) and Brandt et al. (2018). This method uses the VHR-detected trees to measure the crown projected area (CPA) of individual trees. Previous work by our team and partners at the World Agroforestry Center (ICRAF) has shown CPA to be a good predictor of tree diameter (DBH) in semi-arid regions of West Africa and East Africa (Dieng 2015, Kuyah et al. 2012). This relationship estimates an allometric model from satellite observed CPA to predict tree-by-tree biomass across the landscape on a sample or continuous basis.

This proposed project aims to examine a potentially significant but poorly documented or understood carbon sink. These landscapes have been largely ignored by most earth observation analyses because carbon stocks are relatively low and tree density is sparse. However, because the land area covered by TOF systems is expansive an increase in number of trees or productivity of trees can be important carbon fluxes in the global carbon cycle and be an important sink term in the global budget. The impact of this work on international policy is high because TOF systems directly connect to livelihoods in poor countries. Success in meeting goals for climate change mitigation and the UN Sustainable Development Goals will hinge on carbon sequestration and increasing woody perennials in Africa and throughout the developing world. We recognize that our challenge, in part, is that we cannot simply deploy the observational means directly off the bench, and that new methods of observation will need to be developed and deployed – and done in a way that minimizes detection complexities. We have reduced the intellectual risk through considerable testing at existing sites with considerable ground data prior to proposing this larger project