The Impact of Investment on Irrigated Rice, Dryland Agriculture and Afforestation in Senegal using SAR and Optical Time-Series Imagery in Data Fusion Approaches

Over the past 30 years, West Africa’s population has more than doubled; growing 2.7% annually and by 2030, 490 million people are expected to live in the region, placing additional demands on natural resources in the agricultural and forestry sectors. The West African Sahel and Sudanian region is an epicenter of land cover change hotspots which have not been sufficiently documented, due to the inability of multi-temporal moderate resolution observations to adequately resolve widespread sub-hectare changes. Documenting the complex transformation of land cover in WA has been of interest to the remote sensing and policy communities for decades, yet understanding underlying drivers has remained elusive due to a lack of adequate fine-scale observations from space. Widespread reforestation in dryland agriculture regions, expansion of irrigated rice cultivation, dryland agriculture extensification and clearing of natural savanna have been reported, but are poorly mapped and quantified.

We contend multi-source land imaging using very-high resolution imagery (VHR), a time series of multi-spectral and SAR data, as well as High-End Computing (HEC), our Team can significantly improve policy-relevant land cover change information in WA. We will inform policy and investment initiatives by quantifying land cover changes due to agricultural extensification of irrigated rice and dryland cultivation, and afforestation or tree cover loss, using VHR imagery, plus a combination of medium resolution optical and SAR time series.

Our objectives in three hotspots where rapid changes are occurring, (i) Senegal River Valley (4,600 km2), (ii) Expanded Eastern Transition Region (46,220 km2), (iii) Casamance (23,828 km2) are to:

1. Quantify change in extent and intensity of irrigated rice and dryland agriculture

2. Test Deep Learning (DL) methods for extracting agricultural fields and individual trees in comparison to recently proven, unsupervised methods, employed at regional scales; and

3.  Assess agroforestry and reforestation in degraded fields using time-series SAR (2015 to the present) and VHR imagery (2010 to the present), in predefined land parcels from existing deforestation data.

We focus on Senegal as it contains documented hotspots of change across multiple sectors that are representative of WA, but all within one country, with uniform policy and economic context. Senegal has secured large international investments to fund their ambitious national action plans in both the forestry and agriculture sectors. However, determining the impact of initiatives on land cover change has long been overdue, thus this work is highly policy relevant, both at the national and regional levels.

Our Team builds upon our prior NASA-funded successes with automatic extraction of small fields from, by fusing these methods with medium resolution SAR (Sentinel-1) and optical (Sentinel-2 and Landsat) time series data to improve the scaled-up mapping of changes in irrigated rice and dryland agriculture. In addition, we will adapt our methods to automatically identify individual trees with multi-date VHR imagery, over extensive areas. Given the rapid proliferation of Deep Learning (DL) methods, we will systematically evaluate semi-supervised, convolutional neural networks (CNN) methods for identifying individual trees and fields, by comparing them with our previously developed unsupervised methods, on NASA’s ADAPT HEC infrastructure. This work will test the feasibility of scaling up semi-supervised deep learning methods and ultimately demonstrate that our methods are scalable and transferable to similar changes in other parts of the globe. We will generate policy relevant materials and engage with USAID GeoCenter, SERVIR West Africa at Agrhymet, AfricaRice and International Rice Research Institute, and Senegal government agencies.