A recent article published in the journal AgriEngineering explored using machine learning algorithms to monitor and predict agricultural drought and its impact on the crop yield and leaf area index (LAI) of Argania spinosa, a tree native to Morocco. The researchers introduced a combined drought index (CDI) based on four environmental parameters from satellite data to evaluate agricultural drought in Argane forest areas using remote sensing data.
Study: Machine Learning Aids Argane Tree Conservation. Image Credit: tolobalaguer.com/Shutterstock.com
Background
Argania spinosa, commonly known as Argane, is a unique tree found only in southwestern Morocco. It is important for maintaining the ecosystem, fighting desertification, and supporting the local economy. However, its distribution and productivity have dropped significantly due to global climate change, especially chronic droughts. Therefore, assessing drought stress and predicting the crop yield and LAI of Argane trees is crucial for sustainable farming and conserving these valuable tree species.
About the Research
In this paper, the authors aimed to develop a flexible and efficient model for predicting crop yield and LAI of Argane trees using machine learning and various condition indices derived from multisource remote sensing data.
The datasets used include climate hazards group infrared precipitation with station data (CHIRPS) precipitation data, moderate resolution imaging spectroradiometer normalized difference vegetation index (MODIS NDVI), land surface temperature (LST), and evapotranspiration data, soil moisture active passive (SMAP) soil moisture data, shuttle radar topography mission (SRTM) digital elevation model (DEM) data, and rain gauge data. Field measurements of crop yield and LAI were collected from 300 sample plots across 10 study areas in southern Morocco.
The study employed seven machine learning models: extreme gradient boosting (XGBoost), gradient boosted decision trees (GBDT), random forest (RF), decision tree (DT), support vector regressor (SVR), lasso regression (LR), and artificial neural network (ANN), to predict crop yield and LAI using condition indices such as precipitation condition index (PCI), vegetation condition index (VCI), temperature condition index (TCI), evapotranspiration condition index (ETCI), and soil moisture condition index (SMCI).
The RF model was used to downscale CHIRPS precipitation data from 5 km to 1 km resolution using topographic and vegetation variables as predictors. Additionally, the RF model assessed the relative importance of the condition indices and constructed the CDI by combining four key parameters: VCI, PCI, ETCI, and TCI. The CDI was validated by comparing it with other drought indices such as vegetation health index (VHI), and standardized precipitation index (SPI).
Research Findings
The outcomes showed that the XGBoost model outperformed the other models in predicting crop yield and LAI of Argane trees, with high accuracy and low error. The XGBoost-based crop yield model achieved a coefficient of determination (R²) value of 0.94 and a root mean squared error (RMSE) of 6.25 kg/ha, while the LAI model achieved an RMSE of 0.67 and R² of 0.62. The condition indices positively correlated with LAI and crop yield, indicating their usefulness as predictors.
The RF model effectively downscaled CHIRPS precipitation data, enhancing spatial resolution and accuracy. The downscaled CHIRPS data showed good agreement with rain gauge observations, with an R² of 0.89, an RMSE of 5.94 mm, and a mean absolute error (MAE) of 3.69 mm.
The CDI proved a reliable and comprehensive index for assessing and monitoring agricultural drought in Argane forest areas. It showed a positive correlation with SPI, VHI, and crop yield, with a strong correlation with VHI (r = 0.83). The CDI also identified five distinct classes of drought severity, from no drought to exceptional drought.
Applications
The study has several applications for managing and conserving Argane forest ecosystems and supporting the local population's livelihoods. It provides a robust approach for enhancing precipitation data and improving agricultural drought monitoring, which is essential for better land and water management. The study offers valuable insights for farmers, aiding productivity and informed agricultural decision-making. It also provides a reliable method for predicting crop yield and LAI of Argane trees, crucial indicators of vegetation health and productivity.
Additionally, the study contributes to understanding Argane trees' responses and adaptations to changing environmental conditions. It can help develop early warning systems and mitigation strategies for drought events and optimize production and decision-making for the Argane oil industry.
Conclusion
In summary, the XGBoost model proved effective for predicting crop yield and LAI of Argania spinosa using condition indices from multisource remote sensing data. Similarly, the CDI was effective for assessing and monitoring drought in Argane forest areas. Moving forward, the researchers suggested extending the methods to other regions and crops. They recommended incorporating additional remote sensing data to improve model accuracy and robustness.
They also proposed integrating other indicators, such as solar-induced chlorophyll fluorescence, into the CDI model to enhance its sensitivity and accuracy for early drought detection. Additionally, they advised investigating drought impacts on other aspects of Argane forest ecosystems, including biodiversity, carbon sequestration, and soil quality.
Journal reference:
- Mouafik, M.; Fouad, M.; El Aboudi, A. Machine Learning Methods for Predicting Argania spinosa Crop Yield and Leaf Area Index: A Combined Drought Index Approach from Multisource Remote Sensing Data. AgriEngineering 2024, 6, 2283-2305. DOI: 10.3390/agriengineering6030134, https://www.mdpi.com/2624-7402/6/3/134