Improving fodder biomass modeling in the sahelian zone of Niger using the multiple linear regression method

This study was carried out in Niger and aims to propose an improved fodder biomass estimate model using the Multiple Linear Regression (MRM) method. The work was carried out with measurements of herbaceous mass (in situ) made
from 2001 to 2012 by the Ministry of Livestock and Animal Industry of Niger MEIA); rainfalls observed by the Niger Meteorological Office and the meteorological variables from the European Center for Medium - Range Weather Forecasts (ECMWF), processed in AgrometShell (AMS) to derive the agro-
meteorological variables; the SPOT VEGETATION NDVI satellite images processed in the "Vegetation Analysis in Space and Time" (VAST) program to derive biophysical variables from the annual NDVI decadal series and finally the estimated rainfall known as RFE from the American institution "Famine Early
Warning Systems Network "(FEWSNET) for the calculation of annual rainfall totals. The model was performed by multiple linear regressions with the ascending step – by - step procedure for the selection of variables based on the adjusted R² and the RMSE. Leave One Out Cross Validation (LOOICV) was used to calculate the validation R² and a systematic diagnosis of residues to better characterize the model. Throughout the (national) study area, MRM performed an
adjusted R² of 0.68 and a RMSE of 282 kg. Ha-1, the difference between the RMSE of the calibration and that of the validation is 3.72 kg.ha-1. However, it is necessary to continue this research with other indices such as LAI and FAPAR and EVI. Also, it would be interesting to explore ways such as: taking into account the foliage of the trees, adjusting the metrics to the phenology of the herbaceous plants, and those of the woody ones. This work will improve the quality of information used to plan development actions in favor of Niger society in order to protect it against pastoral crises.