Sustainsahel - Eyes in the sky – Remote sensing technologies for improving crop monitoring in agroforestry parkland of Senegal

The advent of UAV and satellite systems capable of resolving smallholder agricultural plots opens new avenues for revisiting the measuring and understanding of agricultural productivity in complex smallholder systems such in the Senegalese parklands. This is the challenge that two doctoral students took up within the SustainSAHEL project.

Illustration of the high spatial heterogeneity of the millet crop induced by the presence of Faidherbia albida trees. Millet productivity is higher in the vicinity of trees. Photo: L. Leroux, Cirad.

Scattered trees in farmer fields, also known as agroforestry parkland, are integrated part of West African smallholder agricultural landscapes. However, the presence of trees inside the fields induce a huge variability that can be hardly captured by traditional approach for large scale crop monitoring relying on satellite images and ground information only. Serigne Mansour Diene and Ibrahima Diack, two Senegaleses PhD students involved in WP4 and WP7 of SustainSAHEL, have accepted to leverage this challenge by exploiting recent advances in remote sensing technologies and new spatial analytics in a Faidherbia albida parkland of Senegal. These works will be presented at the EGU23 conference in April.

UAV images to estimate millet yields at field scale

Serigne Mansour Diene works on the millet yield estimation at field scale using UAV (Drone) images and machine learning algorithms. He proposes to introduce textural data in addition to classical spectral information derived from UAV images in order to take into account the spatial relationships between pairs of pixels. He uses a Random Forest regression to calibrate a millet yield model based on vegetation, texture indices, environmental variables and ground data over the 2018-2022 period. The model provides a good estimate of yields with an accuracy of 74%. The next step is to rely on innovative geospatial approaches to evaluate the distance and directional effects of trees on crop productivity.

From field scale to landscape scale: Combining UAV and satellite images to estimate millet FCover

On his side, Ibrahima Diack takes more elevation and proposes to combine UAV and satellite images (Sentinel-2) to estimate the fraction of vegetation cover (FCover) of millet at the landscape scale. UAV images are used to derive FCover information at field scale for 6 dates during the 2021 cropping season using a thresholding approach. Then the UAV-FCover information is used to calibrate a satellite based model at landscape scale. Various scenarii of modelling were tested. The best compromise is obtained for a Random Forest model fitted on all the dates and including information on the date of acquisition. An accuracy of 73 % is reached. The next step is to move from FCover to millet yield estimation at landscape scale.

The results obtained by the two PhD students show promising opportunities for improving the crop monitoring in heterogeneous landscapes. The next step will be to better understand the influence of trees and shrubs on the millet and groundnut productivity, at the field and landscape scales.

Written by: PhD student, Louise Leroux, Remote Sensing Scientist with Cirad.