Search

Predictor inputs and label data for crop yield forecasting are not always available at the same spatial resolution. We propose a deep learning framework that uses high resolution inputs and low resolution labels to produce crop yield forecasts for both spatial levels. The forecasting model is calibrated by weak supervision from low resolution crop area and yield statistics. We evaluated the framework by disaggregating regional yields in Europe from parent statistical regions to sub-regions for five countries (Germany, Spain, France, Hungary, Italy) and two crops (soft wheat and potatoes). Performance of weakly supervised models was compared with linear trend models and Gradient-Boosted Decision Trees (GBDT). Higher resolution crop yield forecasts are useful to policymakers and other stakeholders. Weakly supervised deep learning methods provide a way to produce such forecasts even in the absence of high resolution yield data.

This dataset includes sample data for the United States to run the weakly supervised framework as described in the paper titled A weakly supervised framework for high resolution crop yield forecasts, accessible at https://doi.org/10.48550/arXiv.2205.09016 The updated paper (including results from the US) is under review in Environmental Research Letters. The software implementation of the machine learning baseline is available at: https://github.com/BigDataWUR/MLforCropYieldForecasting/tree/weaksup. Data 1. County data (county-data.zip) for county-level strongly supervised models: * CROP_AREA_COUNTY_US.csv: County crop production area statistics (acres). Source: NASS (USDA-NASS, 2022). * CSSF_COUNTY_US.csv: Crop productivity indicators including total above-ground production (kg ha-1), total weight of storage organs (kg ha-1), development stage (0-2). Source: de Wit et al. (2022). * METEO_COUNTY_US.csv: Meteo data including maximum, minimum, average daily air temperature (℃); sum of daily precipitation (PREC) (mm); sum of daily evapotranspiration of short vegetation (ET0) (Penman-Monteith, Allen et al., (1998)) (mm); climate water balance = (PREC - ET0) (mm). Source: Boogaard et al. (2022). * REMOTE_SENSING_COUNTY_US.csv: Fraction of Absorbed Photosynthetically Active Radiation (Smoothed) (FAPAR). Source: Copernicus GLS (2020). * SOIL_COUNTY_US.csv: Soil water holding capacity. Source: WISE Soil Property Database (Batjes, 2016). * YIELD_COUNTY_US.csv: County yield statistics (bushels/acre). Source: NASS (USDA-NASS, 2022). 2. 10-km grid data (grid-data.zip) for grid-level strongly supervised models: * COUNTY_GRIDS_US.csv: Mapping between counties and grids. * CSSF_GRIDS_US.csv: Crop productivity indicators at 10km grid level (similar to county data above). * METEO_GRIDs_US.csv: Meteo data at 10km grid level (similar to county data above). * REMOTE_SENSING_GRIDS_US.csv: FAPAR at 10km grid level (similar to county data above). * SOIL_GRID, This dataset includes sample data for the United States to run the weakly supervised framework as described in the paper titled A weakly supervised framework for high resolution crop yield forecasts, accessible at https://doi.org/10.48550/arXiv.2205.09016 The updated paper (including results from the US) is under review in Environmental Research Letters. The software implementation of the machine learning baseline is available at: https://github.com/BigDataWUR/MLforCropYieldForecasting/tree/weaksup. Data 1. County data (county-data.zip) for county-level strongly supervised models: * CROP_AREA_COUNTY_US.csv: County crop production area statistics (acres). Source: NASS (USDA-NASS, 2022). * CSSF_COUNTY_US.csv: Crop productivity indicators including total above-ground production (kg ha-1), total weight of storage organs (kg ha-1), development stage (0-2). Source: de Wit et al. (2022). * METEO_COUNTY_US.csv: Meteo data including maximum, minimum, average daily air temperature (℃); sum of daily precipitation (PREC) (mm); sum of daily evapotranspiration of short vegetation (ET0) (Penman-Monteith, Allen et al., (1998)) (mm); climate water balance = (PREC - ET0) (mm). Source: Boogaard et al. (2022). * REMOTE_SENSING_COUNTY_US.csv: Fraction of Absorbed Photosynthetically Active Radiation (Smoothed) (FAPAR). Source: Copernicus GLS (2020). * SOIL_COUNTY_US.csv: Soil water holding capacity. Source: WISE Soil Property Database (Batjes, 2016). * YIELD_COUNTY_US.csv: County yield statistics (bushels/acre). Source: NASS (USDA-NASS, 2022). 2. 10-km grid data (grid-data.zip) for grid-level strongly supervised models: * COUNTY_GRIDS_US.csv: Mapping between counties and grids. * CSSF_GRIDS_US.csv: Crop productivity indicators at 10km grid level (similar to county data above). * METEO_GRIDs_US.csv: Meteo data at 10km grid level (similar to county data above). * REMOTE_SENSING_GRIDS_US.csv: FAPAR at 10km grid level (similar to county data above). * SOIL_GRID

Many studies have applied machine learning to crop yield prediction with a focus on specific case studies. The data and methods they used may not be transferable to other crops and locations. On the other hand, operational large-scale systems, such as the European Commission's MARS Crop Yield Forecasting System (MCYFS), do not use machine learning. Machine learning is a promising method especially when large amounts of data are being collected and published. We combined agronomic principles of crop modeling with machine learning to build a machine learning baseline for large-scale crop yield forecasting. The baseline is a workflow emphasizing correctness, modularity and reusability. For correctness, we focused on designing explainable predictors or features (in relation to crop growth and development) and applying machine learning without information leakage. We created features using crop simulation outputs and weather, remote sensing and soil data from the MCYFS database. We emphasized a modular and reusable workflow to support different crops and countries with small configuration changes. The workflow can be used to run repeatable experiments (e.g. early season or end of season predictions) using standard input data to obtain reproducible results. The results serve as a starting point for further optimizations. In our case studies, we predicted yield at regional level for five crops (soft wheat, spring barley, sunflower, sugar beet, potatoes) and three countries (the Netherlands (NL), Germany (DE), France (FR)). We compared the performance with a simple method with no prediction skill, which either predicted a linear yield trend or the average of the training set. We also aggregated the predictions to the national level and compared with past MCYFS forecasts. The normalized RMSE (NRMSE) for early season predictions (30 days after planting) were comparable for NL (all crops), DE (all except soft wheat) and FR (soft wheat, spring barley, sunflower). For example, N

Nutrient limitation is a major constraint in crop production in sub-Saharan Africa (SSA). Here, we propose a generic and simple equilibrium model to estimate minimum input requirements of nitrogen, phosphorus and potassium for target yields in cereal crops under highly efficient management. The model was combined with Global Yield Gap Atlas data to explore minimum input requirements for self-sufficiency in 2050 for maize in nine countries in SSA. We estimate that yields have to increase from the current ca. 20% of water-limited yield potential to approximately 50–75% of the potential depending on the scenario investigated. Minimum nutrient input requirements must rise disproportionately more, with N input increasing 9-fold or 15-fold, because current production largely relies on soil nutrient mining, which cannot be sustained into the future.

The Global Yield Gap Atlas project (GYGA - http://yieldgap.org) has undertaken a yield gap assessment following the protocol recommended by van Ittersum et al. (2013). One part of the activities consists of collecting and processing weather data as an input for crop simulation models in sub-Saharan African countries including Kenya. This publication covers daily weather data for 12 locations in Kenya for the years 1998-2012. The project looked for good quality weather data in areas where crops are pre-dominantly grown. As locations with good public weather data are sparse in Africa, the project developed a method to generate bias corrected weather data from a combination of observed data and other external weather data. The bias corrected weather data consist of daily TRMM rain data and NASA POWER Tmax, Tmin, and Tdew data. These data are corrected based on calibrations with short-term (<10 years) observed weather data.

Numerous studies have been published during the past two decades that use simulation models to assesscrop yield gaps (quantified as the difference between potential and actual farm yields), impact of climatechange on future crop yields, and land-use change. However, there is a wide range in quality and spatialand temporal scale and resolution of climate and soil data underpinning these studies, as well as widelydiffering assumptions about cropping-system context and crop model calibration. Here we present anexplicit rationale and methodology for selecting data sources for simulating crop yields and estimatingyield gaps at specific locations that can be applied across widely different levels of data availability andquality. The method consists of a tiered approach that identifies the most scientifically robust require-ments for data availability and quality, as well as other, less rigorous options when data are not availableor are of poor quality. Examples are given using this approach to estimate maize yield gaps in the stateof Nebraska (USA), and at a national scale for Argentina and Kenya. These examples were selected torepresent contrasting scenarios of data availability and quality for the variables used to estimate yieldgaps. The goal of the proposed methods is to provide transparent, reproducible, and scientifically robustguidelines for estimating yield gaps; guidelines which are also relevant for simulating the impact of cli-mate change and land-use change at local to global spatial scales. Likewise, the improved understandingof data requirements and alternatives for simulating crop yields and estimating yield gaps as describedhere can help identify the most critical “data gaps” and focus global efforts to fill them. A related paper(Van Bussel et al., 2015) examines issues of site selection to minimize data requirements and up-scalingfrom location-specific estimates to regional and national spatial scales.

Satellite derived rainfall products are useful for drought and flood early warning and overcome the problem of sparse, unevenly distributed and erratic rain gauge observations, provided their accuracy is well known. Mozambique is highly vulnerable to extreme weather events such as major droughts and floods and thus, an understanding of the strengths and weaknesses of different rainfall products is valuable. Three dekadal (10-day) gridded satellite rainfall products (TAMSAT African Rainfall Climatology And Time-series (TARCAT) v2.0, Famine Early Warning System NETwork (FEWS NET) Rainfall Estimate (RFE) v2.0, and Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS)) are compared to independent gauge data (2001–2012). This is done using pairwise comparison statistics to evaluate the performance in estimating rainfall amounts and categorical statistics to assess rain-detection capabilities. The analysis was performed for different rainfall categories, over the seasonal cycle and for regions dominated by different weather systems. Overall, satellite products overestimate low and underestimate high dekadal rainfall values. The RFE and CHIRPS products perform as good, generally outperforming TARCAT on the majority of statistical measures of skill. TARCAT detects best the relative frequency of rainfall events, while RFE underestimates and CHIRPS overestimates the rainfall events frequency. Differences in products performance disappear with higher rainfall and all products achieve better results during the wet season. During the cyclone season, CHIRPS shows the best results, while RFE outperforms the other products for lower dekadal rainfall. Products blending thermal infrared and passive microwave imagery perform better than infrared only products and particularly when meteorological patterns are more complex, such as over the coastal, central and south regions of Mozambique, where precipitation is influenced by frontal systems.

Accurate estimation of yield gaps is only possible for locations where high quality local data are available,which are, however, lacking in many regions of the world. The challenge is how yield gap estimates basedon location-specific input data can be used to obtain yield gap estimates for larger spatial areas. Hence,insight about the minimum number of locations required to achieve robust estimates of yield gaps atlarger spatial scales is essential because data collection at a large number of locations is expensive andtime consuming. In this paper we describe an approach that consists of a climate zonation scheme supple-mented by agronomical and locally relevant weather, soil and cropping system data. Two elements of thismethodology are evaluated here: the effects on simulated national crop yield potentials attributable tomissing and/or poor quality data and the error that might be introduced in scaled up yield gap estimatesdue to the selected climate zonation scheme. Variation in simulated yield potentials among weatherstations located within the same climate zone, represented by the coefficient of variation, served as ameasure of the performance of the climate zonation scheme for upscaling of yield potentials.We found that our approach was most appropriate for countries with homogeneous topography andlarge climate zones, and that local up-to-date knowledge of crop area distribution is required for selectingrelevant locations for data collection. Estimated national water-limited yield potentials were found to berobust if data could be collected that are representative for approximately 50% of the national harvestedarea of a crop. In a sensitivity analysis for rainfed maize in four countries, assuming only 25% coverageof the national harvested crop area (to represent countries with poor data availability), national water-limited yield potentials were found to be over- or underestimated by 3 to 27% compared to estimateswith the recommended crop area coverage of =50%.

The Global Yield Gap Atlas project (GYGA - http://yieldgap.org ) has undertaken a yield gap assessment following the protocol recommended by van Ittersum et. al. (van Ittersum et. al., 2013). One part of the activities consists of collecting and processing weather data as an input for crop simulation models in sub-Saharan African countries including Kenya. This publication covers weather data for 10 locations in Kenya. The project looked for good quality weather data in areas where crops are pre-dominantly grown. As locations with good public weather data are sparse in Africa, the project developed a method to generate weather data from a combination of observed and other external weather data. One locations holds actually measured weather data, the other 9 locations show propagated weather data. The propagated weather data consist on TRMM rain data (or NASA POWER if TRMM is not available) and NASA POWER Tmax, Tmin, and Tdew data corrected based on calibrations with short-term (<10 years) observed weather data. sources (Van Wart et.al. 2015).