Reference : Modélisation spatiale de la production fourragère en zone pastorale nigérienne
Dissertations and theses : Doctoral thesis
Life sciences : Environmental sciences & ecology
http://hdl.handle.net/2268/210934
Modélisation spatiale de la production fourragère en zone pastorale nigérienne
English
[en] Spatial modeling of forage production in Niger's pastoral zone
Garba, Issa mailto [Université de Liège - ULiège > > > Doct. sc. (sc. & gest. env. - Bologne)]
30-Mar-2017
Université de Liège, ​Liège, ​​Belgique
Docteur en Sciences
233
Tychon, Bernard mailto
Djaby, Bakary mailto
Ozer, Pierre mailto
Bindelle, Jérôme mailto
Toure, Ibra mailto
Hiernaux, Pierre mailto
Kaytakire, François mailto
[fr] NDVI, MEIA, BIOMASAH, SPOT VEGETATION, eMODIS, MR, Similarité, Modèle
[en] This work was carried out on the pastoral zone of Niger, the main objective was to contribute to the improvement of the methods of forage yields predicting mainly in the Sahel and especially in Niger. This is specifically to validate the model BIOMASAH of ARC; test the MEIA model; to establish a reference model by multiple linear regression; test the similarity method and finally compare the methods. The work was carried out on the one hand with the data measured on the ground by the MEIA from 2001 to 2012, reel rainfall of Niger observations network, meteorological parameters from ECMWF and also with satellite images as SPOT NDVI VEGETATION and MODIS, RFE2 of FEWS NET. Validation of BIOMASAH model was made by t and Wilcoxon tests to compare reel biomass and outputs of the model. Pearson, Kendall and Spearman correlation testing was also made. The MEIA model performance was tested by confronting the results between and within SPOT VEGETATION and MODIS sensors, by comparing R² and RMSE from the integral and maximum NDVI as a predictor of forage yield. Average comparisons by parametric and nonparametric tests were also made to compare the results. The reference model (RM) was produced by multiple linear regression with stepwise method. The selection of variables was based on adjusted R² and RMSE and the LOOCV leave one out cross validation to calculate R² for validation, we made also systematic diagnosis of residues for better characterization of the model. The similarity method was performed using the R², MAD and RMSE as a criterion, the profile of the vegetation growth period of each pixel was plotted for all years. Then we compare the profile of the target year with those of other years to identify the similar year. One hand the results of similarity were compared with actual data with the Pearson correlation test, Spearman and Kendall and secondly using t and Wilcoxon tests to compare means. Comparison of models was made on the basis of R², Adjusted R² and RMSE. Model BIOMASAH result on significant difference between average (p <0.001). Pearson correlations, Kendall and Spearman are low. Regarding the MEIA model, globally R² (0.56) is best, there’s no difference to use MODIS NDVI or SPOT vegetation, the RMSE is 367 kg.ha-1. R² and RMSE vary greatly from one year to another. On a global scale the multiple linear model gave a good R² adjusted (0.69) and RMSE (282 kg / ha) the difference between the calculated and the RMSE of validation is 2.72 kg. Comparing averages of the similarity to the real ones showed that there are no significant differences (p <0.001) for R² with the differences are significant against the same threshold for the MAD and RMSE. The Comparison of the models shows that the multiple linear regression one (reference model) is the best. Research should continue with index like LAI, FARAR and EVI.
Researchers ; Professionals ; Students ; General public
http://hdl.handle.net/2268/210934

File(s) associated to this reference

Fulltext file(s):

FileCommentaryVersionSizeAccess
Open access
PhD_IssaGARBA.pdfAuthor postprint4.54 MBView/Open

Bookmark and Share SFX Query

All documents in ORBi are protected by a user license.