Projects

3-meter evapotranspiration estimation based on OpenET, Landsat, and Planet information via machine learning approach.

June, 2023 - June, 2024

Given agriculture's growing challenges due to population growth, climate changes, and declining water supply, monitoring evapotranspiration (ET) at high spatiotemporal resolutions is critical to maximizing production while using limited water resources effectively. In this project, the data mining sharpener (DMS) technique was redeveloped to sharpen the 30-meter resemble ET obtained from the OpenET platform to the 3-meter level using Landsat and Planet multispectral information. The corresponding model was called the DMS-ETS model. A machine learning approach was used because the current DMS-ETS model cannot obtain daily 3-meter ET. The eXtreme Gradient Boosting (XGB) model was trained based on Landsat and the ensemble ET collected between March and October 2018 at the 30-meter resolution level. The trained XGB-ET model was then applied to the Planet’s multispectral images in 2018 at the 3-meter resolution level. To validate the 3-meter ET products from both DMS-ETS and XGB-ET models, the two-source energy balance (TSEB) model ET results at a California vineyard on July 12th, 2018, which is validated by the eddy-covariance flux tower within the corresponding footprint at a 3.6-meter resolution from previous research, were used. The results showed that the DMS-ETS model, with land surface temperature (LST) considered, could barely show the ET patterns compared to those shown on the TSEB-modeled ET. However, the ET patterns estimated by the DMS-ETS model (without LST considered) and the XGB-ET model appeared more consistent with the TSEB-modeled ET. When comparing different ET products (mm/day) at the same pixel level (10.8-meter), the DMS-ETS (without LST considered) showed the best consistency with the TSEB-modeled ET compared to other models at r=0.8, RMSE=2.6 mm/day, and Bias=2.5 mm/day. While the XGB-ET model showed the worst consistency among others at r=0.6, RMSE=4.9 mm/day, and Bias=4.9 mm/day. Although the current model performance for a California vineyard on a specific day is low, it is necessary to validate the performance of both models for different crops on different days to get more robust conclusions. 

California Vineyards water status and stress monitoring using remote sensing.

January, 2020 - June, 2023

Limited water resources can be applied in agriculture due to many factors, such as global warming, population growth, and unevenly distributed water resources. To ensure food security and security, precision irrigation is necessary, and this is especially true in California, where over a third of the country's vegetables and nearly three-quarters of the country's fruits and nuts are grown in California. We took advantage of the data resources obtained from different sensors on different platforms involved in the Grape Remote sensing Atmospheric Profile & Evapotranspiration eXperiment (GRAPEX) program. We first confirmed a hybrid machine learning model for leaf area index (LAI) estimation for four different vineyards across California. This enabled us to do accurate and efficient LAI mapping for these vineyards (and the potential for other vineyards which still need to be tested). We then documented evapotranspiration (ET) and transpiration via the two-source energy balance (TSEB) model based on the generated LAI maps. We noticed the improvement of ET estiamtion via the TSEB model compared with previous research results. However, the transpiration estimation at the field scale still needs more advanced methodologies in terms of the method for observation and modeling. We highlighted these two points in our research. At the end of this project, we used a machine-learning approach for leaf water potential and stem water potential estimation. We saw a significant potential of using collected data by unmanned aerial vehicle (UAV) coupled with the meteorological station data for leaf water potential mapping for different California vineyards at the field scale. This inspired us to collaborate with more researchers in the world to explore further the application of leaf water potential estiamtion in precision agricultural water management.

Agricultural drainage ditch system detection in low relief land using high-resolution LiDAR terrain data.

August, 2018 - January, 2020

The artificial drainage ditch plays an essential role in crop growth in the Midwestern U.S. by maintaining arable farmland and in water cycling as it accelerates flow time from farmland to the natural river. The currently available datasets and methods cannot provide the accurate location of the ditch, and manual drawing on the map is labor-intensive and time-consuming. We designed an Artificial Drainage Ditch System (ADDS) model combining platforms such as Python, MATLAB, and ArcGIS Pro Python API, which accurately delineates artificial drainage ditches based on high-resolution LiDAR data. We compared the ADDS products and the NHDPlus V2 data with our manually labeled maps in four Piatt County (Illinois) study areas. It showed that the ADDS product was more accurate.

Water resource management at the basin scale by using the Variable Infiltration Capacity (VIC) model in Xinjiang alpine regions.

September 2014 - July, 2017

Data scarcity, such as meteorological and hydrological data, is a primary challenge for water resource management at different spatial scales in Xinjiang. In this project, we built a VIC model for the Kashi River Basin (a Xinjiang alpine region) and used different reanalysis meteorological data, including the Climate Forecast System Reanalysis (CFSR) and ERA-Interim produced by ECMWF,  to force the VIC model. We saw the potential to use both CFSR and ERA-Interim data for that watershed water resource management. Additionally, we simulated the runoff under different assumed climate scenarios. The results showed a signal that the Kashi River Basin was a fragile alpine region under the climate-changing background.

(The left figure comes from https://vic.readthedocs.io/en/master/Overview/ModelOverview/)