Search

India is the second-most populous country and the second-most important producer of rice of the world. Most Indian rice production depends on monsoon timing and dynamics. In 2002, the lowest monsoon precipitation of the last 130+ years was observed. It coincided with the worst rice production anomaly recorded by FAOSTAT from 1961 to 2014. In that year, freshwater limitation was blamed as responsible for the yield losses in the southeastern coastal regions. Given the important implication for local food security and international market stability, we here investigate the specific mechanisms behind the effects of this extreme meteorological drought on rice yield at the national and regional levels. To this purpose, we integrate output from the hydrological model, surface, and satellite observations for the different rice cropping cycles into state-of-the-art and novel climate indicators. In particular, we adopt the standardized precipitation evapotranspiration index (SPEI) as an indicator of drought due to the local surface water balance anomalies (i.e., precipitation and evapotranspiration). We propose a new indicator of the renewable surface freshwater availability due to non-local sources, i.e., the standardized river discharge index (SDI) based on the anomalies of modelled river discharge data. We compare these indicators to the soil moisture observations retrieved from satellites. We link all diagnostics to the recorded yields at the national and regional level, quantifying the long-term correlations and the best match of the 2002 anomaly. Our findings highlight the need for integrating non-local surface freshwater dynamics with local rainfall variability to determine the soil moisture conditions in rice fields for yields assessment, modeling, and forecasting. [ABSTRACT FROM AUTHOR]

This study addresses the role of satellite Earth Observation (EO) indicators within an operational crop yield forecasting system for the European Union (EU) and neighbouring countries, by exploring the correlation between official yield statistics and indicators derived from fAPAR time-series at sub-national level for the period 1999–2012, and by identifying possible differences across agro-climatic conditions in Europe. A significant correlation between fAPAR and official yields ( R 2 > 0.6) was found in water-limited yield agro-climatic conditions (e.g. the Black Sea region and the Mediterranean basin) for all three crops studied. In regions where crops experience frequent water stress, most of the yield inter-annual variability is explained by substantial changes in leaf area from one year to another, and can be well captured by regional fAPAR time-series. By contrast, in regions characterized by high yields (e.g. northern Europe) – where water constraints are generally not frequent and, therefore, fAPAR inter-annual variability is low – the correlation between fAPAR and yield is weaker ( R 2 < 0.5) as yield variations tend to be explained by multiple factors other than green leaf area. These results confirm the reliability of EO time-series for operational crop yield forecasting at regional level, but also suggest that additional meteorological variables (temperature, precipitation, evapotranspiration) need to be taken into account to interpret EO products meaningfully. Moreover, specific issues related to the spatial resolution of the EO-products, and the absence of dynamic crop masks, currently impede access to crop-specific time-series in the fragmented agricultural landscapes of Europe, and restrict the use of 1-km biophysical products to major crops. [ABSTRACT FROM AUTHOR]

• We present a system to forecast crop yield with machine learning at regional level. • This fully automated system selects the best features and machine learning model. • Leaving one year out is required to rigorously assess model performance. • Yield forecasts based on machine learning are accurate even in low-yielding years. • Extensive calibration is required to deliver significant gains in predictive power. Forecasting crop yields is important for food security, in particular to predict where crop production is likely to drop. Climate records and remotely-sensed data have become instrumental sources of data for crop yield forecasting systems. Similarly, machine learning methods are increasingly used to process big Earth observation data. However, access to data necessary to train such algorithms is often limited in food-insecure countries. Here, we evaluate the performance of machine learning algorithms and small data to forecast yield on a monthly basis between the start and the end of the growing season. To do so, we developed a robust and automated machine-learning pipeline which selects the best features and model for prediction. Taking Algeria as case study, we predicted national yields for barley, soft wheat and durum wheat with an accuracy of 0.16–0.2 t/ha (13-14% of mean yield) within the season. The best machine-learning models always outperformed simple benchmark models. This was confirmed in low-yielding years, which is particularly relevant for early warning. Nonetheless, the differences in accuracy between machine learning and benchmark models were not always of practical significance. Besides, the benchmark models outperformed up to 60% of the machine learning models that were tested, which stresses the importance of proper model calibration and selection. For crop yield forecasting, like for many application domains, machine learning has delivered significant improvement in predictive power. Nonetheless, superiority over simple benchmarks is fully achieved after extensive calibration, especially when dealing with small data. [ABSTRACT FROM AUTHOR]

• A comparison of crop masks to spatially aggregate NDVI for yield forecasting in EU. • Forecasts improve when using annual crop group-specific than generic crop masks. • Improvements concern both accuracy and timeliness of yield forecasts. • Annual crop group-specific masks allow better interpretation of NDVI time series. • Agricultural Decision Support Systems (DSS) can benefit from the obtained results. Operational crop yield forecasting services typically provides crop yield forecasts based on regression models between official yields and agro-environmental variables, among which meteorological data, crop simulation model or satellite-derived indicators. The reliability with which these variables infer on yields depends, among many other factors, also on their aggregation in the space domain, for example on the type of the crop masks utilized in the aggregation process from point to regional scale. This work investigates how the yield explanatory power of satellite-derived indicators is changing moving from time-stable arable land masks to annual crop group-specific masks. We compare in particular the linkage between time series of regional crop yield in Europe and the Normalized Difference Vegetation Index (NDVI) from Moderate Resolution Imaging Spectroradiometer (MODIS), when generic arable land masks or crop group-specific and year-specific information are applied to aggregate pixel values at regional level. Regional (Eurostat level NUTS-2) crop yield statistics were collected from official databases for the period 2003–2019, while NDVI data were derived from MODIS daily products at 250 m spatial resolution for the same reference period. Regional NDVI profiles were retrieved by averaging single pixels time series according to the information of five crop masks, including generic arable land masks, crop group-specific static masks (separating winter and spring crops from summer crops), and annual crop group-specific masks (distinguishing between the two crop groups and varying in time). A compared correlation analysis between yield data and regional temporal NDVI profiles was performed assuming a linear regression model. Coefficient of determination R2 and Root Mean Squared Error (RMSE) were computed to assess the models' errors and to analyze the effect of the aggregation on the different crop masks. Results indicated an improvement in yield estimation when using annual crop group-specific indicators with respect to generic and static products. Although not homogeneously distributed throughout Europe, advantages have been highlighted both in terms of accuracy and timeliness of the prediction. In most regions, the introduction of annual masks allowed to reduce RMSE values by 0.3 t/ha and advanced forecast times by up to 30 days. The added value in the use of annual crop group-specific masks concerned the two European most cultivated crops, namely soft wheat and grain maize. [ABSTRACT FROM AUTHOR]