Irrigation

Aquifer decline and surface water depletion for agricultural irrigation have been recognized as a critical water resource management issue threatening water security. Agriculture accounts for nearly 80% of freshwater withdrawals in the U.S. The use of water in agriculture is synergistically related to whether the land is irrigated and what irrigation practice the farm operator uses. Critically important to stakeholders and managers, a comprehensive assessment of long-term irrigation status and changes in irrigation system type in the high-impact agricultural hotspots is needed to understand and manage water resources.

The overarching goal of this project is to quantify the long-term irrigation status and system type changes in the US agricultural region for an advanced understanding of irrigation-induced water resource changes by integrating multi-source earth observation data, machine learning techniques, artificial intelligence, and cloud computing. To achieve this goal, we will integrate multi-source earth observation data, artificial intelligence techniques, cloud computing, and test the method’s scalability, with three objectives:

1): Develop an annual field boundary and sub-field irrigation management boundary dataset and a quality-controlled, annotated image database CropNet for remote sensing solutions. We will generate an annual dataset that contains the delineated boundaries of relatively stable fields and the further divided sub-field irrigation management units using a two-level mapping strategy. We will also create a benchmark image database by visually annotating irrigation status and types on very-high-resolution imagery to provide high-quality training samples for classification.

2): Use multi-source land imaging strategies to quantify annual changes and long-term trends in the annual irrigation status of agricultural fields for the period 2000-2020. We will enhance recently developed techniques to perform continuous change detection on irrigation status from dense stacks of Landsat and Sentinel. Improvements will be centered on conducting classification at the field level and training the algorithm with CropNet ground truth data to achieve better performance.   

3): Apply deep-learning methods to assess irrigation types and their changes for the period 2000-2020. We will couple deep-learning analysis of very high resolution imagery with CropNet training samples to quantify the spatial-temporal pattern of irrigation practice. We will apply the developed methods to detect similar changes in different geographic regions and apply the methods to imagery acquired from other data sources. 

Publications:

Meyarian, A., Yuan, X., Liang, L., Wang, W. and Gu, L., 2022. Gradient convolutional neural network for classification of agricultural fields with contour levee. International Journal of Remote Sensing, 43(1), pp.75-94.Meyarian2022IJRS.pdf

Liang, Lu, Abolfazl Meyarian, Xiaohui Yuan, Benjamin RK Runkle, George Mihaila, Yuchu Qin, Jacob Daniels, Michele L. Reba, and James R. Rigby. "The first fine-resolution mapping of contour-levee irrigation using deep Bi-Stream convolutional neural networks." International Journal of Applied Earth Observation and Geoinformation 105 (2021): 102631.Liang2021_IJAG.pdf

Liang, L., Runkle, B.R., Sapkota, B.B. and Reba, M.L., 2019. Automated mapping of rice fields using multi-year training sample normalization. International Journal of Remote Sensing, 40(18), pp.7252-7271.   LiangIJRS2019.pdf