Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Belgique → Assessment of fodder biomass in Senegalese rangelands using earth observation and field data

Université de Liège (2016)

Assessment of fodder biomass in Senegalese rangelands using earth observation and field data

Diouf, Abdoul Aziz

Titre : Assessment of fodder biomass in Senegalese rangelands using earth observation and field data

Evaluation de la biomasse fourragère des parcours naturels du Sénégal à partir des données de télédétection et de mesures au sol

Auteur : Diouf, Abdoul Aziz

Université de soutenance : Université de Liège

Grade : Doctorat en Sciences 2016

Senegalese livestock size has largely increased during the last three decades in relation to the population growth. The fodder biomass stock available at the end of the growing season, therefore, becomes increasingly limited to meet feeding needs of pastoral livestock which provides third of the national agricultural wealth. With the reduction of natural grazing lands mostly generated by the expansion of croplands, and the reduction of fodder biomass production due to drought effects, the increase of the livestock size leads to the rangelands overload whose persistence can lead in turn to their degradation. A technique based on a simple linear relationship between the temporal integration of the Normalized Difference Vegetation Index (NDVI) and the ground biomass data, developed in the 1980s, has been operationally applied by the Centre de Suivi Ecologique (CSE) of Dakar (Senegal) to assess the fodder biomass available in rangelands at the end of the growing season. The derived map of total biomass production enables to help pastoral livestock managers as well as national stakeholders against food insecurity and natural resources degradation. Carried out annually, this approach comprises unfortunately some uncertainties as : (1) the saturation drawback of NDVI in areas with high biomass productivity, (2) the temporal scale which is restricted to biomass data of the ongoing year not being used again in the following year, (3) the low predictive ability due to the large time gap between data collection and published results, and (4) the high costs for annual data collection. In addition, although the earth observation (EO) data have largely progressed during the last three decades, this technique has not changed over this period and consequently is not state-of-the-art. To tackle these limitations and advance the traditional method, new statistical models that include new earth observations datasets and historical in situ plant biomass data were developed for estimating and / or predicting the forage availability at the end of the growing season in Senegalese semi-arid rangelands. A backward analysis of the linear regression approach currently applied in Senegal provided evidence that nonlinear regression functions such as Exponential and Power are more suited to estimate the end-of-season total biomass in this region using annual data solely. A completely new methodology using multiple-linear models which include various phenological metrics from the time series of the Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) and 14 years of in situ total biomass samples was developed. The proposed approach provided more reliable and accurate estimates as compared to the current CSE biomass product. Multiple-linear models developed with specific metrics adapted to ecosystem properties increased the overall accuracy of the fodder biomass estimates and mitigated the saturation of FAPAR obtained with models run across the whole study area. With this new approach, timely information about possible deficits/surplus of total fodder biomass can be provided to stakeholders using phenological metrics that are available relatively early in the growing season. Another new approach based on a machine learning algorithm (i.e., Cubist) was developed, as never done before, to assess herbaceous biomass in Senegalese Sahel. Three Cubist models using FAPAR seasonal metrics and/or agrometeorological variables (i.e., soil water status indicators) were established and compared. The Cubist model including both FAPAR and agrometeorological variables provided the best estimation performance. This model enabled to mitigate the saturation affecting optical remotely sensed vegetation data in areas of high plant productivity as well as the discrepancy between herbaceous biomass and greenness, and corrected therefore for herbaceous biomass underestimations observed with the sole FAPAR based model, particularly in sparsely vegetated areas. In contrast to the date of the growing season onset retrieved from FAPAR seasonal dynamics, the rainy season onset was significantly related to the herbaceous biomass and its inclusion in models could constitute a significant improvement in forecasting risks of fodder biomass deficit. The methods developed in this research provide tools to assess Senegalese forage resources at two levels : herbaceous and total fodder biomass (Herbaceous + woody leaf biomass). They require limited data and free available software and therefore can be easily replicated in other countries of the West African Sahel.

Mots clés : fodder biomass, models, FAPAR, phenological metrics, growing season, herbaceous, forecast, food security, Senegal, Sahel

Présentation (ORBI)

Page publiée le 1er décembre 2016, mise à jour le 22 novembre 2017