Alfalfa (Medicago sativa L.) is considered the queen of forage due to its high yield, nutritional qualities, and capacity to sequester carbon. However, there are issues with its relatively low persistency and winter survival as compared to grass. Winter survival in alfalfa is affected by diverse factors, including the environment (e.g., snow cover, hardiness period, etc.) and management (e.g., cutting timing, manure application, etc.). Alfalfa's poor winter survival reduces the number of living plants, delays plant development, and diminishes field productivity. To better understand poor winter survival and persistency in alfalfa and assess winter damage, this research aimed to develop an assessment tool for Canadian growers. In addition, a prediction model was designed to consider and account for the variability and potential risks. Both field measurements and remote sensing approaches were incorporated into the assessment tool. 

Soil samples, stem counts, and height were collected from 192 farms in four provinces – Nova Scotia, Quebec, Ontario, and Manitoba. The field sampling design used time-series vegetation indices in the k-means clustering procedure. A randomized design was implemented in each cluster. The stem count samples were measured from each site in the Spring and Fall of 2021. The soil texture was mainly loam, which varies across the provinces. A total of 1612 targeted soil samples were collected from the four regions. The sampling points were then positioned using the iPad GPS. Lab-measured soil micro-and macro-nutrients were pH, soil organic matter (SOM), phosphorus (P), potassium (K), cation exchange capacity (CEC), Magnesium (Mg), Manganese (Mn), Zinc (Zn), and Calcium (Ca). Many regions also used soil and stem characteristics for winter risk assessment grids. The initial field assessment scores were evaluated based on suitable parameters (i.e., stand age, soil pH and potassium levels, harvest frequency, and cultivar type) of all agro-ecological zones for the status of potential risks. Both historical field measurements and topographic datasets were used for the assessment model. Descriptive statistical analysis and correlation between stem characteristics and topographic variables, together with the knowledge of soil nutrients, enhanced our understanding of the spatial heterogeneity of alfalfa production areas. A random forest regression model was applied. Model parameters were developed to determine the number of essential variables and regression trees to be used in the training phase; this resulted in optimal model performance and scenario maps.

The prediction model and data-driven decisions pose challenges only with soil chemical analysis in assessing winter mortality, identifying potential agronomic and environmental factors and their potential for improvement. The emerging risk assessment tools and the application of generalized models considering all potential factors described in regional guidelines will assist Canadian forage growers in improving their productivity by using alternative management practices, including species selection and soil recommendations, using information on survival rates and persistency, to increase financial returns.