Assessment of vegetation productivity on rangelands is needed to assist in timely decision making with regard to management of the livestock enterprise as well as to protect the natural resource. Characterization of the vegetation resource over large landscapes can be time consuming, expensive and almost impossible to do on a near real-time basis. The overarching goal of this study was to examine available technologies for implementing near real-time systems to monitor forage biomass available to livestock on a given landscape. The primary objectives were to examine the ability of the Climate Prediction Center Morphing Product (CMORPH) and Next Generation Weather Radar (NEXRAD) rainfall products to detect and estimate rainfall at semi-arid sites in West Texas, to verify the ability of a simulation model (PHYGROW) to predict herbaceous biomass at selected sites (patches) in a semi-arid landscape using NEXRAD rainfall, and to examine the feasibility of using cokriging for integrating simulation model output and satellite greenness imagery (NDVI) for producing landscape maps of forage biomass in Mongolia’s Gobi region. The comparison of the NEXRAD and CMORPH rainfall products to gage collected rainfall revealed that NEXRAD outperformed the CMORPH rainfall with lower estimation bias, lower variability, and higher estimation efficiency. When NEXRAD was used as a driving variable in PHYGROW simulations that were calibrated using gage measured rainfall, model performance for estimating forage biomass was generally poor when compared to biomass measurements at the sites. However, when model simulations were calibrated using NEXRAD rainfall, performance in estimating biomass was substantially better. A suggested reason for the improved performance was that calibration with NEXRAD adjusted the model for the general over or underestimation of rainfall by the NEXRAD product. In the Gobi region of Mongolia, the PHYGROW model performed well in predicting forage biomass except for overestimations in the Forest Steppe zone. Cross-validation revealed that cokriging of PHYGROW output with NDVI as a covariate performed well during the majority of the growing season. Cokriging of simulation model output and NDVI appears to hold promise for producing landscape maps of forage biomass as part of near real-time forage monitoring systems.

Cette étude a eu pour but d'étudier la possibilité de couplage de modèles dynamiques de croissance de l'herbe avec des données de télédétection, et ce pour deux terrains contrastés : La Réunion et le Kalahari (Afrique du Sud). Deux phases se sont succédé. Une première phase exploratoire, basée sur des images SPOT5 et SPOT5take5 (satellites désorbités en cours d'étude) a permis de tirer plusieurs enseignements. A La Réunion l'ajustement d’un modèle empirique entre indices de végétation et biomasse engendre trop d'erreur. Il est en revanche possible d'estimer le Leaf Area Index (LAI) grâce au NDVI (Normalized Difference Vegetation Index). Les parcours du Kalahari, plus complexes, avec différentes strates de végétation (graminées, arbustes, arbres) n'ont pas permis d'estimer l'état du couvert de graminées. Cette phase a ouvert la voie au travail effectué sur un capteur plus pérenne dans le temps, Sentinel-2. Les données Sentinel-2 ont permis d'estimer le LAI des prairies réunionnaises avec une RMSE (Root Mean Square Error) de 0,63 (r²=0,82). Le LAI ainsi estimé a été utilisé dans le couplage du modèle dynamique permettant une baisse générale de la RMSE de l'ordre de 40% par rapport au modèle sans couplage. Ces résultats ont été obtenus durant l'hiver austral, la saison sèche. Durant la période d'été austral les pluies plus abondantes accélèrent la croissance des plantes et les cycles de pousse se raccourcissent. Les images satellites sans couvert nuageux se font plus rares. La prise en compte de cette combinaison de facteurs pouvant impacter les prédictions de biomasse fourragère fera partie des principale perspectives de ce travail. / The purpose of this study was to explore the possibility of coupling dynamic models of grass growth with remote sensing data for two contrasting countries: Reunion Island and Kalahari (South Africa). Two phases followed one another. A first exploratory phase, based on SPOT5 and SPOT5take5 images (desorbed satellites under study) allowed us to learn from this experience. In Reunion the adjustment of an empirical model between vegetation indices and biomass generates too much error. However it is possible to estimate the Leaf Area Index (LAI) thanks to the NDVI (Normalized Difference Vegetation Index). More complex Kalahari rangelands with different vegetation strata (grasses, shrubs, trees) failed to estimate grass cover conditions. This phase set the stage to work on a more durable sensor over time, Sentinel-2. Sentinel-2 data made it possible to estimate the LAI of Reunion Island grasslands with a RMSE (Root Mean Square Error) of 0.63 (r² = 0.82). The LAI thus estimated was used in the coupling of the dynamic model, allowing a general decrease of the RMSE of the order of 40% compared to the model without coupling. These results were obtained during the austral winter, the dry season. During the austral summer, the more abundant rains speed up the growth of the plants and the growth cycles become shorter. Satellite images without cloud cover are becoming scarce. Taking into account this combination of factors that may impact predictions of forage biomass will be one of the main perspectives of this work.