Presentation Type
Poster
Student
Yes
Track
Methodology
Abstract
Soil temperature is vital for agriculture, especially in the area where winter is very harsh. Previously it was assumed that the ground surface temperature may affect the surface air temperature. But now this assumption does not hold. Other factors affect the ground surface temperature. In our proposed approach we use the data from the Red River Basin of Fargo North Dakota. The Red River Basin is very fertile land for agriculture and the soil is covered with snow. The data set consists of different features like Average Air Temperature, Average Relative Humidity, Average wind speed, Average Wind Direction, Total soil Radiation, Rainfall, Snow depth and Moisture content. Our objective is to find the features most important for predicting the soil temperature at different depths. In our proposed approach we will be implementing a machine learning model. As the LSTM (Long Short Term Memory) model works best on time series datasets, the authors in paper [1] have chosen to use the LSTM model. Here in our proposed work, we argued that we could also achieve a similar or better efficiency using machine learning model. Additionally, our model will be easier to implement and understand. On the other hand, LSTM models are difficult to implement. The LSTM model only accepts datasets if they are divided into two arguments. One is the dataset and the other is lookback which is the number of previous time steps to use as the input variables to predict the next time period. When using machine learning models we do not have to restructure our dataset.
Additionally, In the paper [1] the authors only used the basic SHAP[2] framework which is the bar plot. The features are ordered from the highest to lowest effect on the prediction. It takes into account only the absolute SHAP values. Hence it is not possible to know if a feature is affecting the prediction in a
positive or negative way. In our approach, we will be using beeswarm plot to understand in detail each of the features.
That is if a feature is affecting the prediction positively or negatively. In summary, our proposed approach will be using
machine learning models to look into the features in more detail about the prediction of the snow depths.
Start Date
2-7-2025 1:00 PM
End Date
2-7-2025 2:30 PM
Predicting Soil Temperature at different depths using a Time Series Model
Volstorff A
Soil temperature is vital for agriculture, especially in the area where winter is very harsh. Previously it was assumed that the ground surface temperature may affect the surface air temperature. But now this assumption does not hold. Other factors affect the ground surface temperature. In our proposed approach we use the data from the Red River Basin of Fargo North Dakota. The Red River Basin is very fertile land for agriculture and the soil is covered with snow. The data set consists of different features like Average Air Temperature, Average Relative Humidity, Average wind speed, Average Wind Direction, Total soil Radiation, Rainfall, Snow depth and Moisture content. Our objective is to find the features most important for predicting the soil temperature at different depths. In our proposed approach we will be implementing a machine learning model. As the LSTM (Long Short Term Memory) model works best on time series datasets, the authors in paper [1] have chosen to use the LSTM model. Here in our proposed work, we argued that we could also achieve a similar or better efficiency using machine learning model. Additionally, our model will be easier to implement and understand. On the other hand, LSTM models are difficult to implement. The LSTM model only accepts datasets if they are divided into two arguments. One is the dataset and the other is lookback which is the number of previous time steps to use as the input variables to predict the next time period. When using machine learning models we do not have to restructure our dataset.
Additionally, In the paper [1] the authors only used the basic SHAP[2] framework which is the bar plot. The features are ordered from the highest to lowest effect on the prediction. It takes into account only the absolute SHAP values. Hence it is not possible to know if a feature is affecting the prediction in a
positive or negative way. In our approach, we will be using beeswarm plot to understand in detail each of the features.
That is if a feature is affecting the prediction positively or negatively. In summary, our proposed approach will be using
machine learning models to look into the features in more detail about the prediction of the snow depths.