Know more

On December 9th in Montpellier, Julien Sarron defended his thesis work on spatialized estimation of mango yields in Senegal. Numerous results have been obtained, publications have been published, and several tools have been developed, including a digital tool for the mango sector that is currently being developed and transferred. There are many prospects for Julien, who has just been recruited by CIRAD to continue his research in Emile Faye's team at HortSys.

What objectives did you start your PhD with?

In the context of food insecurity in West Africa, it is essential to improve the means of monitoring agricultural production. Fruit trees play an important role in the population's diet, but their yields are low. In order to face the challenges of development, it is necessary to identify the causes. My thesis topic was applied to mango production, with the objective of producing tools for estimating mango yields at different spatial scales, the tree, the orchard and the region. In order to identify the explanatory factors of the observed yield variability, I worked on 30 orchards in the Niayes region near Dakar, divided into three types of cropping systems: intensive (monoculture), extensive, and also diversified, i.e., agro-forestry.

Agroforestry, cropping systems, remote sensing, artificial intelligence... your work covers several areas that combine agronomy and digital tools?

Yes, as an agronomist by training, I had already worked on the variability of annual crop yields, but fruit trees were new to me. I had to immerse myself in research on tropical arboriculture to understand the factors that explain yield heterogeneity in mango. This involved questions related to the ecophysiology of the tree, but also to cropping systems, which are very varied in West Africa. For example, the complex agroforestry system based on mango trees is particularly interesting.

I also invested in drone imagery and artificial intelligence (machine learning, neural networks) with a double objective: to build a land use map and to estimate the yield of mango trees at the tree level, within and between orchards.

© Louise Leroux, Cirad

Drone imagery is interesting for research in agro-forestry, to compare cropping systems...

At the orchard scale, the land use map is used to know, for each orchard, the position of the trees, their species, and even their variety in the case of mango. To establish this map, we classified the drone images with the eCognition software. We trained an algorithm to distinguish mango trees from other objects present on the plots: other trees, such as citrus and cashew trees, market gardening, bush vegetation, soil, tracks, buildings... Thus, the production of each mango tree is modeled by cross-referencing the land use map with information on the structure of the trees (i.e., their height, projected area and volume), which is also obtained by drone imagery.

• Modeling and spatialization of production shows intra- and inter-orchard yield variability: spatial distributions of tree characteristics and production appear more heterogeneous in extensive and diversified orchards than in monocultured ones.
• Drone imagery is interesting from an agronomy/agro-forestry perspective. A lot of information on the orchards has been acquired: varietal composition, number of plants, surface area in market gardening... A large number of measurements can be made and related to estimate the impact on the yield and compare the cropping systems. For example, it was found that diversified agro-forestry orchards have a higher average production per tree than intensive orchards.

A drone would not be adapted to the constraints of small producers, a smartphone would.

Drone imagery is interesting for research in agro-forestry, to compare cropping systems...

At the tree level, we have developed a model-assisted neural network image analysis tool to estimate the production of mango trees as harvesting approaches. No drone is used here, as it would not be adapted to the constraints of small producers, whereas a smartphone, for example, would be. So for the experiments, we used a digital camera of good resolution to photograph each tree and its mangoes. To detect the number of mangoes per tree, we opted, with the support of colleagues from UMR AMAP, for algorithms already used by an Australian team for monoculture orchards. Our added value is to have trained the neural network to detect 3 varieties of mango trees. These are the 3 most common varieties in the Niayes region, resulting from a wide variety of growing conditions. Inside the trees, the mangoes are hidden so I developed a statistical calibration model to link the number of visible fruits detected by the algorithm and the real number of fruits. It is calibrated for these 3 varieties of mango trees, but can be used elsewhere by re-calibrating it. I used the concept of "yield gap" to calculate the gap between the achievable production of the mango tree, defined by its structure, age and variety, and its real production. In the 30 orchards monitored, for the 3 varieties and 3 cropping systems in the region, the effects of mango variety and cropping system on yield gaps were found. We also identified the effects of certain practices. For example, irrigated mango trees have a 30% reduction in the gap between their attainable and real production, compared to non-irrigated mango trees. This neural network + calibration model is the basis of the PixFruit tool currently under development.

At the regional scale, i.e., the Niayes production basin, I addressed the regional agronomic diagnosis: what are the factors that determine yield? What are the yield variabilities between orchards? I am currently analyzing the results to compare the performance of the orchards. We can already see that the density of trees is a determining variable of the yield in the intensive system, and that the tree produces more in the diversified system. Climate, cultural practices and crop diversity are also explanatory factors of the variability of production, but their effect remains to be explored. In the future, the idea would be to have a model to extrapolate the yields at the regional level, to estimate the production of all the orchards of the basin.

Your post-thesis?

I have just been recruited at CIRAD, I will be able to continue my activity in the same team, in the HortSys unit, this time in Montpellier. At the moment, I am writing new articles on my thesis work. For PixFruit, we will start the validation of the models at the orchard scale, the development of the regional model and the development of the tool. I will also address new research questions, such as the temporal dynamics of fruit production (flowering, maturity...). In my thesis, this was not much addressed, there was already a lot to do on mature fruits. I would also like to see if the methodological concept developed on mango can be applied to other fruit species. For example, lychee in Madagascar. For this, several research collaborations will continue for the regional remote sensing aspects, with AMAP for the supervision of internships and within the PixFruit project team (see box).

Contact :  julien.sarron [AT] cirad.fr​ – Tél. +33 4 67 59 37 14 - At a glance : Julien's PhD webpage