Title : Developing site-specific water response functions using historical soil moisture data
Abstract:
Variable-Rate Irrigation (VRI) is an important technology in precision agriculture due to its ability to tailor water application to distinct areas of the agricultural field according to the heterogenous within-field conditions, such as soil types, landscapes, soil nutrient levels, and other crop growth conditions. To obtain the optimal results from variable-rate irrigation, a crucial step is the development of accurate site-specific crop water response functions, which form the basis for determining optimal irrigation levels across different field segments. However, comprehensive development of these functions is still lacking in existing literature. There are two principal challenges in developing site-specific water response functions. The first challenge is data collection. Unlike experiments with other farming inputs like fertilizers or lime, manipulating irrigation levels within short distances in a field is difficult. Most current experiments only manage to divide fields into a few large sections, providing insufficient spatial data for site-specific response modeling. Recently, there’s been a growing trend to utilize farmers’ center pivots for conducting on-farm irrigation experiments. However, the current commercial variable-rate center pivot systems generally only offer speed control for changing irrigation rates, where the resultant data are limited to pie-shaped sections. High-resolution, grid-based irrigation data that are essential for spatially variable water response modeling remains rare. The second challenge involves the dynamic nature of water in production field. Total available water for plants is the sum of irrigation, rainfall, and residual soil moisture, minus losses like surface runoff and evaporation. Consequently, the amount of irrigated water seldom equates to the actual water available to plants. Many crop water response studies that have used irrigated water as the independent variable have resulted in inconclusive findings, which are very likely due to this significant omission in other water sources. Recognizing the complexity of water sources, crop science research often uses plant water evapotranspiration, which reflects actual water uptake by crops, as the water variable, and often finds clear linear relationships to crop yields. However, evapotranspiration itself is an outcome of crop growth, rather than a manageable input. The evapotranspiration correlation with yield is informative but not causative, thus limiting its utility for practical farming decisions. In this study, we propose using soil water (moisture) data as the independent variable to estimate site-specific crop water response functions, considering that soil moisture level can better represent the water accessible to crops and is a manageable input factor. We gathered data from an 18-hectare production field in Brooksville, Mississippi. Soil moisture sensors were installed at 44 grid points in the field, recording hourly measurements at various depths from 2018 to 2020. These measurements, covering entire growing seasons, were coupled with grain yield maps for the corresponding crops (soybean, corn, and soybean rotations). While the number of grid points is not extensive, it surpasses most traditional irrigation experimentation data in spatial refinement. The major influencing factors of soil moisture levels, rainfall and weather conditions, can be regarded as random and exogeneous. The only source of non-randomness or endogeneity of soil moisture levels comes from the irrigation scheduling, which was based on the farmer's judgment (for instance, irrigation might be triggered when the soil is excessively dry). However, the spatial distribution of soil moisture, influenced by factors like soil texture and topography, is largely beyond the farmer's control and can still be considered spatially random. The relationship between crop growth and the timing of water supply is an extremely complicated question in agronomy research. This study employed a commonly used simple strategy to aggregate the hourly soil moisture data into average readings for several critical crop growth stages. These average readings were then used as the independent variables in crop yield response modeling. To capture the spatial variability of the response, the study employed Geographically Weighted Regression (GWR) model, which performs localized regressions using yield and soil moisture data from neighboring points. GWR’s advantage lies in its ability to estimate site-specific water response functions without requiring other site-specific data (e.g., soil testing, topography). and the results from this revealed consistent variability patterns in the crop-water relationship. The study found that both corn and soybean crops were most sensitive to water during their reproductive stage and farmers may have to optimize water use during these periods. The estimated site-specific crop yield responses to soil moisture levels by this study provide a foundation for deriving optimal soil moisture levels at different crop growth stages. Based on that, the desired irrigation amounts at each crop growth stage can be precisely calculated by matching the soil moisture levels through the total available water formula, and facilities the effective variable-rate irrigation scheduling.
