Unmanned aerial vehicles (UAVs) equipped with multispectral sensors offer high spatial and temporal resolution imagery for monitoring crop stress at early stages of development. Analysis of UAV-derived data with advanced machine learning models could improve real-time management in agricultural systems, but guidance for this integration is currently limited. Here we compare two deep learning-based strategies for early warning detection of crop stress, using multitemporal imagery throughout the growing season to predict field-scale yield in irrigated rice in eastern Arkansas. Both deep learning strategies showed improvements upon traditional statistical learning approaches including linear regression and gradient boosted decision trees. First, we explicitly accounted for variation across developmental stages using a 3D convolutional neural network (CNN) architecture that captures both spatial and temporal dimensions of UAV images from multiple time points throughout one growing season. 3D-CNNs achieved low prediction error on the test set, with a Root Mean Squared Error (RMSE) of 8.8% of the mean yield. For the second strategy, a 2D-CNN, we considered only spatial relationships among pixels for image features acquired during a single flyover. 2D-CNNs trained on images from a single day were most accurate when images were taken during booting stage or later, with RMSE ranging from 7.4 to 8.2% of the mean yield. A primary benefit of convolutional autoencoder-like models (based on analyses of prediction maps and feature importance) is the spatial denoising effect that corrects yield predictions for individual pixels based on the values of vegetation index and thermal features for nearby pixels. Our results highlight the promise of convolutional autoencoders for UAV-based yield prediction in rice.
more »
« less
Spatial-temporal Multi-Task Learning for Within- field Cotton Yield Prediction
Understanding and accurately predicting within-field spatial variability of crop yield play a key role in site-specific management of crop inputs such as irrigation water and fertilizer for optimized crop production. However, such a task is challenged by the complex interaction between crop growth and environmental and managerial factors, such as climate, soil conditions, tillage, and irrigation. In this paper, we present a novel Spatial-temporal Multi-Task Learning algorithm for within-field crop yield prediction in west Texas from 2001 to 2003. This algorithm integrates multiple heterogeneous data sources to learn different features simultaneously, and to aggregate spatial-temporal features by introducing a weighted regularizer to the loss functions. Our comprehensive experimental results consistently outperform the results of other conventional methods, and suggest a promising approach, which improves the landscape of crop prediction research fields.
more »
« less
- Award ID(s):
- 1737634
- NSF-PAR ID:
- 10128844
- Date Published:
- Journal Name:
- Pacific-Asia Conference on Knowledge Discovery and Data Mining
- Page Range / eLocation ID:
- 343-354
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)With population growth and resource depletion, maximizing the efficiency of soybean (Glycine max [L.] Merr.) and rice (Oryza sativa L.) cropping systems is urgently needed. The goal of this study was to shed light on precise irrigation amounts and optimal agronomic practices via simulating rice–rice and soybean–rice crop rotations in the Agricultural Policy/Environmental eXtender (APEX) model. The APEX model was calibrated using observations from five fields under soybean–rice rotation in Arkansas from 2017 to 2019 and remote sensing leaf area index (LAI) values to assess modeled vegetation growth. Different irrigation practices were assessed, including conventional flooding (CVF), known as cascade, multiple inlet rice irrigation with polypipe (MIRI), and furrow irrigation (FIR). The amount of water used differed between fields, following each field’s measured or estimated input. Moreover, fields were managed with either continuous flooding (CF) or alternate wetting and drying (AWD) irrigation. Two 20-year scenarios were simulated to test yield changes: (1) between rice–rice and soybean–rice rotation and (2) under reduced irrigation amounts. After calibration with crop yield and LAI, the modeled LAI correlated to the observations with R2 values greater than 0.66, and the percent bias (PBIAS) values were within 32%. The PBIAS and percent difference for modeled versus observed yield were within 2.5% for rice and 15% for soybean. Contrary to expectation, the rice–rice and soybean–rice rotation yields were not statistically significant. The results of the reduced irrigation scenario differed by field, but reducing irrigation beyond 20% from the original amount input by the farmers significantly reduced yields in all fields, except for one field that was over-irrigated.more » « less
-
more » « less
Abstract Food demands are rising due to an increasing population with changing food preferences, placing pressure on agricultural production. Additionally, climate extremes have recently highlighted the vulnerability of the agricultural system to climate variability. This study seeks to fill two important gaps in current knowledge: how irrigation impacts the large-scale response of crops to varying climate conditions and how we can explicitly account for uncertainty in yield response to climate. To address these, we developed a statistical model to quantitatively estimate historical and future impacts of climate change and irrigation on US county-level crop yields with uncertainty explicitly treated. Historical climate and crop yield data for 1970–2009 were used over different growing regions to fit the model, and five CMIP5 climate projections were applied to simulate future crop yield response to climate. Maize and spring wheat yields are projected to experience decreasing trends with all models in agreement. Winter wheat yields in the Northwest will see an increasing trend. Results for soybean and winter wheat in the South are more complicated, as irrigation can change the trend in projected yields. The comparison between projected crop yield time series for rainfed and irrigated cases indicates that irrigation can buffer against climate variability that could lead to negative yield anomalies. Through trend analysis of the predictors, the trend in crop yield is mainly driven by projected trends in temperature-related indices, and county-level trend analysis shows regional differences are negligible. This framework provides estimates of the impact of climate and irrigation on US crop yields for the 21st century that account for the full uncertainty of climate variables and the range of crop response. The results of this study can contribute to decision making about crop choice and water use in an uncertain future climate.
-
null (Ed.)Abstract—The emergence of remote sensing technologies cou- pled with local monitoring workstations enables us the un- precedented ability to monitor the environment in large scale. Information mining from multi-channel geo-spatiotemporal data however poses great challenges to many computational sustainability applications. Most existing approaches adopt various dimensionality reduction techniques without fully taking advantage of the spatiotemporal nature of the data. In addition, the lack of labeled training data raises another challenge for modeling such data. In this work, we propose a novel semi-supervised attention-based deep representation model that learns context-aware spatiotemporal representations for prediction tasks. A combination of convolutional neural networks with a hybrid attention mechanism is adopted to extract spatial and temporal variations in the geo-spatiotemporal data. Recognizing the importance of capturing more complete temporal dependencies, we propose the hybrid attention mechanism which integrates a learnable global query into the classic self-attention mechanism. To overcome the data scarcity issue, sampled spatial and temporal context that naturally reside in the largely-available unlabeled geo-spatiotemporal data are exploited to aid meaningful representation learning. We conduct experiments on a large-scale real-world crop yield prediction task. The results show that our methods significantly outperforms existing state-of-the-art yield prediction methods, especially under the stress of training data scarcity.more » « less
-
Abstract Understanding the interactions among agricultural processes, soil, and plants is necessary for optimizing crop yield and productivity. This study focuses on developing effective monitoring and analysis methodologies that estimate key soil and plant properties. These methodologies include data acquisition and processing approaches that use unmanned aerial vehicles (UAVs) and surface geophysical techniques. In particular, we applied these approaches to a soybean farm in Arkansas to characterize the soil–plant coupled spatial and temporal heterogeneity, as well as to identify key environmental factors that influence plant growth and yield. UAV-based multitemporal acquisition of high-resolution RGB (red–green–blue) imagery and direct measurements were used to monitor plant height and photosynthetic activity. We present an algorithm that efficiently exploits the high-resolution UAV images to estimate plant spatial abundance and plant vigor throughout the growing season. Such plant characterization is extremely important for the identification of anomalous areas, providing easily interpretable information that can be used to guide near-real-time farming decisions. Additionally, high-resolution multitemporal surface geophysical measurements of apparent soil electrical conductivity were used to estimate the spatial heterogeneity of soil texture. By integrating the multiscale multitype soil and plant datasets, we identified the spatiotemporal co-variance between soil properties and plant development and yield. Our novel approach for early season monitoring of plant spatial abundance identified areas of low productivity controlled by soil clay content, while temporal analysis of geophysical data showed the impact of soil moisture and irrigation practice (controlled by topography) on plant dynamics. Our study demonstrates the effective coupling of UAV data products with geophysical data to extract critical information for farm management.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
