Combining High-Resolution Remote Sensing Products with a Crop Model to Estimate Carbon and Water Budget Components: Application to Sunflower

Abstract

The global increase in food demand in the context of climate change requires a clear understanding of cropland function and of its impact on biogeochemical cycles. However, although gas exchange between croplands and the atmosphere is measurable in the field, it is difficult to quantify at the plot scale over relatively large areas because of the heterogeneous character of landscapes and differences in crop management. However, assessing accurate carbon and water budgets over croplands is essential to promote sustainable agronomic practices and reduce the water demand and the climatic impacts of croplands while maintaining sufficient yields. From this perspective, we developed a crop model, SAFYE-CO2, that assimilates high spatial- and temporal-resolution (HSTR) remote sensing products to estimate daily crop biomass, water and CO2 fluxes, annual yields, and carbon budgets at the parcel level over large areas. This modeling approach was evaluated for sunflower against two in situ datasets. First, the model’s output was compared to data acquired during two cropping seasons at the Auradé integrated carbon observation system (ICOS) instrumented site in southwestern France. The model accurately simulated the daily net CO2 flux (root mean square error (RMSE) = 0.97 gC·m−2·d−1 and determination coefficient (R2) = 0.83) and water flux (RMSE = 0.68 mm·d−1 and R2 = 0.79). The model’s performance was then evaluated against biomass and yield data collected from 80 plots located in southwestern France. The model was able to satisfactorily estimate biomass dynamics and yield (RMSE = 66 and 54 g·m−2, respectively). To investigate the potential application of the proposed approach at a large scale, given that soil properties are important factors affecting the model, a sensitivity analysis of two existing soil products (GlobalSoilMap and SoilGrids) was carried out. Our results show that these products are not sufficiently accurate for inclusion as inputs to the model, which requires more accurate information on soil water retention capacity to assess water fluxes. Additionally, we argue that no water stress should be considered in the crop growth computation since this stress is already present because of remote sensing information in the proposed approach. This study should be considered a first step to fulfill the existing gap in quantifying carbon budgets at the plot scale over large areas and to accurately estimate the effects of management practices, such as the use of cover crops or specific crop rotations on cropland C and water budgets.

Forecasting Sunflower Grain Yield by Assimilating Leaf Area Index into a Crop Model

“…They differ by the nature of the actual data used for evaluating the performance of the model when assimilating remote sensing data: yield either at regional, farm, field or intra-field level, yield information from the farmer or directly measured by sensors onboard harvesters. Semi-empirical models as SAFY [17] and SAFY-CO2 [47] or statistical models (linear models, non-parametric approaches, etc.) based on the use of NDVI, LAI or leaf area duration were also evaluated for yield prediction [48,49].…”

Deep learning for photovoltaic panels segmentation

“…Others called hyperspectral sensors having the capability to capture several hundred and even thousands of bands, this allows for more levels of interpretations in the non-visible area of studies [30]. Last but not least, the temporal resolution providing the information of how much tie does it take for a satellite to fully complete an orbit and return to the same observation area such kind of information allows more expendable research most commonly agricultural applications [31,32].…”

Soil erosion assessment in the Blue Nile Basin driven by a novel RUSLE-GEE framework

“…This dataset was adopted for this study because of its extensive applications in soil erosion mapping from the local to global scales (e.g., Borrelli et al, 2017;Fenta et al, 2021;Gashaw et al, 2020;Schürz et al, 2020;Tamene and Le, 2015). It is also has been adopted in other environmental studies, for instance, predict soil bacterial biodiversity (Griffiths et al, 2016), soil type classification (Dornik et al, 2016), plant species niches (Velazco et al, 2017), yield forecast (Cunha et al, 2018), mapping global mangrove forest soil carbon (Sanderman et al, 2018), global gridded hydrologic soil groups (Ross et al, 2018), predict vegetation type (Cramer et al, 2019), soil hydraulic and thermal properties (Dai et al, 2019), dust emission probability (Effati et al, 2019), land degradation (Giuliani et al, 2020), crop models (Pique et al, 2020;Tewes et al, 2020;Wimalasiri et al, 2020), hydrological modeling (Krpec et al, 2020;Trinh et al, 2018), and land use capability analysis (Ippolito et al, 2021).…”