Dataset Information

Predicting spatial and temporal variability in crop yields: an inter-comparison of machine learning, regression and process-based models.

ABSTRACT: Pervious assessments of crop yield response to climate change are mainly aided with either process-based models or statistical models, with a focus on predicting the changes in average yields, whilst there is growing interest in yield variability and extremes. In this study, we simulate US maize yield using process-based models, traditional regression model and a machine-learning algorithm, and importantly, identify the weakness and strength of each method in simulating the average, variability and extremes of maize yield across the country. We show that both regression and machine learning models can well reproduce the observed pattern of yield averages, while large bias is found for process-based crop models even fed with harmonized parameters. As for the probability distribution of yields, machine learning shows the best skill, followed by regression model and process-based models. For the country as a whole, machine learning can explain 93% of observed yield variability, followed by regression model (51%) and process-based models (42%). Based on the improved capability of the machine learning algorithm, we estimate that US maize yield is projected to decrease by 13.5% under the 2°C global warming scenario (by ~2050s). Yields less than or equal to the 10th percentile in the yield distribution for the baseline period are predicted to occur in 19% and 25% of years in 1.5°C (by ~2040s) and 2°C global warming scenarios, with potentially significant implications for food supply, prices and trade. The machine learning and regression methods are computationally much more efficient than process-based models, making it feasible to do probabilistic risk analysis of climate impacts on crop production for a wide range of future scenarios.

SUBMITTER: Leng G

PROVIDER: S-EPMC7212054 | biostudies-literature | 2020 Apr

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Predicting spatial and temporal variability in crop yields: an inter-comparison of machine learning, regression and process-based models.

Leng Guoyong G Hall Jim W JW

Environmental research letters : ERL [Web site] 20200228 4

Pervious assessments of crop yield response to climate change are mainly aided with either process-based models or statistical models, with a focus on predicting the changes in average yields, whilst there is growing interest in yield variability and extremes. In this study, we simulate US maize yield using process-based models, traditional regression model and a machine-learning algorithm, and importantly, identify the weakness and strength of each method in simulating the average, variability ...[more]

PMID: 32395176

Similar Datasets

Project description:Not all areas of a farmer's field are equal; some always produce more relative to the rest of the field, others always less, while still other areas fluctuate in their production capacity from one year to the next, depending on the interaction between climate, soil, topography and management. Understanding why the yield in certain portions of a field has a high variability over time-we call these areas unstable-is of paramount importance both from an economic and an environmental point of view, as it is through the better management of these areas that we can improve yields or reduce input costs and environmental impact. In this research, we analyzed data from 338 fields cultivated with maize, soybean, wheat and cotton in the US Midwest to understand how topographic attributes and rain affect yield stability over time. In addition to this high resolution yield monitor dataset, we used publicly available data on topography, rain and soil information to test the hypothesis that within-field areas characterized by a low topographic wetness index (proxy for areas with probability of lower water content) always perform poorly (low and stable yield) compared to the rest of the field because they are drier, and that areas of a field characterized by a mid-high wetness index (high and stable yield) always perform well relative to rest of the field because they have greater water availability to plants. The relative performance of areas of a field with a very high wetness index (e.g. depressions) strongly depends on rain patterns because they may be waterlogged in wet years, yielding less than the rest of the field, or wetter during dry years, yielding more than the rest of the field. We present three different observations from this dataset to support our hypothesis. First, we show that the average topographic wetness index in the different stability zones is lower in low and stable yield areas, high in high and stable yield areas and even higher in unstable yield areas (p?<?0.05). Second, we show that in dry years (low precipitation at plant emergence or in July), unstable zones perform relatively better compared to the rest of the field. Third, we show that temporal yield variability is positively correlated (p?<?0.05) with the probability of observing gleying processes associated with waterlogging for part of the year. These findings shed light on mechanisms underlying temporal variability of yield and can help guide management solutions to increase profit and improve environmental quality.

Project description:Attaining high crop yields and increasing carbon storage in agricultural soils, while avoiding negative environmental impacts on water quality, soil erosion, and biodiversity, requires accurate and precise management of crop inputs and management practices. The long-term analysis of spatial and temporal patterns of crop yields provides insights on how yields vary in a field, with parts of field constantly producing either high yields or low yields and other parts that fluctuate from one year to the next. The concept of yield stability has shown to be informative on how plants translate the effects of environmental conditions (e.g., soil, climate, topography) across the field and over the years in the final yield, and as a valuable layer in developing prescription maps of variable fertilizer rate inputs. Using known relationships between soil health and crop yields, we hypothesize that areas with measured constantly low yield will return low carbon to the soil affecting its heath. On this premises, yield stability zones (YSZ) provide an effective and practical integrative measure of the small-scale variability of soil health on a field relative basis. We tested this hypothesis by measuring various metrics of soil health from commercial farmers' fields in the north central Midwest of the USA in samples replicated across YSZ, using a soil test suite commonly used by producers and stakeholders active in agricultural carbon credits markets. We found that the use of YSZ allowed us to successfully partition field-relative soil organic carbon (SOC) and soil health metrics into statistically distinct regions. Low and stable (LS) yield zones were statistically lower in normalized SOC when compared to high and stable (HS) and unstable (US) yield zones. The drivers of the yield differences within a field are a series of factors ranging from climate, topography and soil. LS zones occur in areas of compacted soil layers or shallow soils (edge of the field) on steeper slopes. The US zones occurring with high water flow accumulation, were more dependent on topography and rainfall. The differences in the components of the overall soil health score (SHS) between these YSZ increased with sample depth suggesting a deeper topsoil in the US and HS zones, driven by the accumulation of water, nutrients, and carbon downslope. Comparison of the field management provided initial evidence that zero tillage reduces the magnitude of the variance in SOC and soil health metrics between the YSZ.

Project description:Systematic monitoring of pasture quantity and quality is important to match the herd forage demand (pasture removal by grazing or harvest) to the supply of forage with adequate nutritive value. The aim of this research was to monitor, assess and manage changes in pasture growth, morphology and digestibility by integrating information from an Unmanned Aerial Vehicle (UAV) and two process-based models. The first model, Systems Approach to Land Use Sustainability (SALUS), is a process-based crop growth model used to predict pasture regrowth based on soil, climate, and management data. The second model, Morphogenetic and Digestibility of Pasture (MDP), uses paddock-scale values of herbage mass as input to predict leaf morphogenesis and forage nutritive value. Two field experiments were carried out on tall fescue- and ryegrass-based pastures under rotational grazing with lactating dairy cattle. The first experiment was conducted at plot scale and was used to calibrate the UAV and to test models. The second experiment was conducted at field scale and was used to test the UAV's ability to predict pasture biomass under grazing rotation. The Normalized Difference Vegetation Index (NDVI) calculated from the UAV's multispectral reflectance (n = 72) was strongly correlated (p < 0.001) to plot measurements of pasture biomass (R2 = 0.80) within the range of ~226 and 4208 kg DM ha-1. Moreover, there was no difference (root mean square error, RMSE < 500 kg DM ha-1) between biomass estimations by the UAV (1971±350 kg ha-1) and two conventional methods used as control, the C-Dax proximal sensor (2073±636 kg ha-1) and ruler (2017±530 kg ha-1). The UAV approach was capable of mapping at high resolution (6 cm) the spatial variability of pasture (16 ha). The integrated UAV-modeling approach properly predicted spatial and temporal changes in pasture biomass (RMSE = 509 kg DM ha-1, CCC = 0.94), leaf length (RMSE = 6.2 cm, CCC = 0.62), leaf stage (RMSE = 0.7 leaves, CCC = 0.65), neutral detergent fiber (RMSE = 3%, CCC = 0.71), digestibility of neutral detergent fiber (RMSE = 8%, CCC = 0.92) and digestibility of dry matter (RMSE = 5%, CCC = 0.93) with reasonable precision and accuracy. These findings therefore suggest potential for the present UAV-modeling approach for use as decision support tool to allocate animals based on spatially and temporally explicit predictions of pasture biomass and nutritive value.

Dataset Information

Predicting spatial and temporal variability in crop yields: an inter-comparison of machine learning, regression and process-based models.

Publications

Predicting spatial and temporal variability in crop yields: an inter-comparison of machine learning, regression and process-based models.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets