Deep learning model effectiveness in forecasting limited-size aboveground vegetation biomass time series: Kenyan grasslands case study

Authors: NOA-YARASCA, EFRAIN - Texas A&M Agrilife; OSORIO LEYTON, JAVIER - Texas A&M Agrilife; Angerer, Jay

Submitted to: Agronomy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/6/2024
Publication Date: 2/8/2024
Citation: Noa-Yarasca, E., Osorio Leyton, J.M., Angerer, J.P. 2024. Deep learning model effectiveness in forecasting limited-size aboveground vegetation biomass time series: Kenyan grasslands case study. Agronomy. 14(2). Article 349. https://doi.org/10.3390/agronomy14020349.
DOI: https://doi.org/10.3390/agronomy14020349

Interpretive Summary: The ability to predict the amount of standing vegetation biomass can provide important early warning information for livestock and fire management. However, few studies have been conducted on forecasting vegetation biomass for periods in the near-term. Additionally, artificial intelligence methods could prove to be useful in improving forecasts, yet these methods have not been fully tested for estimating near-term changes in standing vegetation. In this study, a hybrid deep learning model named the Convolutional Neural Network/Long Short-Term Memory Model (CNN/LSTM) was evaluated to assess its effectiveness in predicting standing aboveground biomass 6, 12, 18, and 24 months into the future. This model was compared to four other deep learning models and to a more traditional statistical forecasting model (Seasonal Autoregressive Integrated Moving Average [SARIMA]). Using vegetation biomass values from 5 grassland sites in Kenya, the forecasting models were run, and outputs compared. No model was found to be superior in forecasting standing vegetation biomass, and the statistical model slightly outperformed the deep learning models. As expected for all models, the accuracy decreased with increasing length of the forecast period. In this limited study, use of artificial intelligence methods did not improve forecasting ability when compared to traditional statistical methods.

Technical Abstract: Early forecasting is essential for managing aboveground vegetation biomass and ensuring food security, but studies on predicting vegetation biomass for periods in the near future are limited. Artificial intelligence (AI) methods could bridge this research gap and provide early warning to planners and stakeholders. This study evaluates the effectiveness of the CNN-LSTM hybrid deep learning (DL) algorithm for aboveground vegetation biomass prediction, examining how this model performs in an iterative forecasting procedure for four horizons. The model's performance was compared with four other DL models including the multilayer perceptron (MLP), long short-term memory (LSTM), gated recurrent unit (GRU), and the convolutional neural network (CNN), while also examining the more conventional seasonal autoregressive integrated moving average (SARIMA) statistical model. Five limited-size vegetation biomass time series from Kenyan grasslands with values at 15-day intervals over a 20-year period were chosen for this purpose. When compared, output from the model were significantly different from one another (p<0.05); however, none of the models proved superior among the five time-series and the four horizons evaluated. The SARIMA, CNN, and CNN-LSTM models performed best, with the statistical model slightly outperforming the other two. Additionally, the accuracy of all five models varied significantly according to the prediction horizon (p < .05). As expected, the accuracy of the models decreased as the prediction horizon increased, although this relationship was not strictly monotonic. Ultimately this study indicated that, in limited-size aboveground vegetation biomass time series, there is no guarantee that machine learning methods will outperform more traditional statistical methods.