The Application of Unmanned Aerial Vehicles (UAVs) and Extreme Gradient Boosting (XGBoost) to Crop Yield Estimation: A Case Study of Don Tum District, Nakhon Pathom, Thailand

Specifications	DJI Phantom 4 Pro V2.0	DJI Phantom 4 Multispectral
Satellite positioning systems	GPS/GLONASS	GPS/GLONASS
Flight range	7 km.	7 km.
Max service ceiling	500 m.	500 m.
Wind speed resistance	36 km/h	36 km/h
Operating frequency	2.40–5.85 GHz	2.40–5.85 GHz
Sensor and Camera	4K/20 MP 1-inch (12.8mm x 9.6mm) CMOS	Six cameras, including 1 RGB camera and other 5 cameras colors (red edge, near-infrared, green, red, and blue)
Aperture	f/2.8 - f/11	f/2.8 - f/11
Gimbal	3-axis (pitch, roll and pan) and stabilized gimbal	3-axis (pitch, roll and pan) and stabilized gimbal

Figure 2: Data collection towards the use of RGB and multispectral UAVs. (a) Flight path of RGB UAVs (b) Flight path of Multispectral UAV (c) Ground control points in each plot area

Table 2: Factors affecting crop yield estimation

However, the plant height factor was calculated using the raster calculator in the GIS software.Next, the relationships between relevant factors were explored in order to predict the crop yields of the sample rice fields. There were 7 factors used as independent variables, while the crop yields were the dependent variable. The crop yield prediction was analyzed by Multiple Linear Regression (MLR) [30], as explained by the following equation:

Equation 1

Where as:

β = Estimated regression coefficient

x = Factors affecting crop yields of sample rice fields

y= Outputs of sample rice fields

In addition, this study used the Extreme Gradient Boosting (XGBoost) model to predict yield using the same factors as MLR to compare yield predictions across the sample area. Extreme Gradient Boosting (XGBoost) was learning from parameters in the same way as the regression tree family approach. This model was based on the gradient descent direction of the loss function of the last established model [31], as explained by the following equation:

Equation 2

Where as:

l(ŷ,y_i)= Training loss,

Ω(f_k) = The complexity of trees

f_k = The regression trees

T = The number of leaves in the tree

w = The score of the regression tree node

Then, the results from the MLR and XGBoost analysis were compared to the crop yields of each sample plot observed by field exploration and validated by the coefficient of determination (R²). The R-Squared represents the relationship between independent and dependent variables, while the Root-Mean-Square Error (RMSE) and the Mean of Absolute value of Errors (MAE) represent the relationship between predicted crop yields and observed crop yields of 20 sample plots [32] and [33], as illustrated by the following equation:

Equation 3

Equation 4

Equation 5

Where as:

Y_obs= Crop yield observed by field exploration

Y_pre = Predicted crop yield

Y_avg= Average of crop yields observed by field exploration

N= Number of data points

Finally, the crop yield estimation results were tested by Extreme Gradient Boosting (XGBoost), a tree-based machine learning algorithm. In this algorithm, decision trees are sequentially formulated and continuously learn from the errors of previous decision trees until a stopping criterion is met, gradually increasing prediction accuracy [34].

3. Results

3.1 Analysis of Factors Affecting Sample Fields’ Crop Yields by UAVs

According to the exploration of the 20 plots (1 m² each), the multispectral UAV could analyze 3 factors, including Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation (GNDVI), and Triangular Greenness Index (TGI). The NDVI and GNDVI reflected the vegetation index values ranging from -1.0 to 1.0. Meanwhile, the TGI demonstrated the chlorophyll content of plants ranging between 15–85 mg/cm² [29]. It can be assumed that the higher index values, the greater amount of chlorophyll (as Figure 4). Moreover, there were two other factors analyzed by the RGB UAV, which included plant height and canopy cover. The RGB UAV performed in a double grid pattern, achieving UAV-captured RGB images which could be used for calculating the height data of the crop canopy and soil surface based on a geographic information system (GIS). The data gained from the previous process was analyzed to calculate the plant height in the sample fields. Regarding the canopy cover, the images were interpreted and divided into two types of pixel images representing rice areas and non-rice areas, respectively. The data was then calculated by a canopy cover estimation formula (as Figure 5). As a result, there were 5 factors affecting crop yields of the sample fields in total, as listed in Figure 6.

3.2 Estimation of Sample Fields’ Crop Yields Based on Data Collected by UAVs

To estimate the crop yields of the sample fields, crop yields were observed from the 20 plots in order to compare them with the crop yields predicted by Multiple Linear Regression based on the following factors: Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation (GNDVI), Triangular Greenness Index (TGI), plant height, and canopy cover.

Figure 3: Conceptual framework

Figure 4: Analysis of data received from multispectral UAV

Figure 5: Analysis of data received from RGB camera from UAV

Figure 6: Factors affecting crop yields of sample fields. (a) Plot areas (b) NDVI (c) GNDVI (d) Plant height (e) TGI (f) Canopy cover

According to the research results, the most significant factors included canopy cover, plant height, and Triangular Greenness Index (TGI), followed by Green Normalized Difference Vegetation (GNDVI) and Normalized Difference Vegetation Index (NDVI). This study found that all factors are important for rice yield estimation and could be ordered as follows: Canopy cover, NDVI, Plant height followed by GNDVI, and TGI (as the value column in Table 3). Based on the aforementioned factors, the crop yields of the sample fields were estimated by validating the crop yields observed from the 20 plots of the sample fields by the coefficient of determination (R²), Root-Mean-Square Error (RMSE), and Mean of Absolute value of Errors (MAE) as illustrated in Figure 7 and Table 4 respectively. In the Figure 7, the research results showed that the R-Squared value was 0.99, indicating that the prediction of the Multiple Linear Regression (MLR) model was valid. Then, the average difference between the predicted crop yields and the observed crop yields was measured by the Root-Mean-Square Error (RMSE = 2.506 g.), and Mean of Absolute value of Errors (MAE) = 1.778.

Table 3: Level of significance of factors affecting sample field’s crop yields

Table 4: Prediction of crop yields of 20 plots in sample fields by multiple linear regression model

Figure 7: Prediction of multiple linear regression model

Table 5: Prediction of crop yields of 20 plots in sample fields by extreme gradient boosting (XGBoost)

4. Conclusion

This research investigates the estimation of crop yields towards the application of unmanned aerial vehicles (UAVs). It was discovered that the UAVs used for estimating crop yields of the sample rice fields in Don Tum District, Nakhon Pathom, could analyze 5 factors affecting the estimation of crop yields, which included three vegetation indexes (Normalized Difference Vegetation Index, Green Normalized Difference Vegetation Index, and Triangular Greenness Index) and two physical covariates (plant height and canopy cover). According to the Multiple Linear Regression analysis, the most significant factors were canopy cover, plant height, and Triangular Greenness Index, followed by Green Normalized Difference Vegetation Index and Normalized Difference Vegetation Index, respectively (R² = 0.99). After estimating the crop yields of the sample rice fields based on the five aforementioned factors, the multiple linear regression model showed a positive relationship between the predicted crop yields and the crop yields observed by field exploration (RMSE = 2.506 g.; MAE = 1.778). Additionally, the application of Extreme Gradient Boosting (XGBoost) increased the accuracy of crop yield estimation (RMSE = 0.557 g.; MAE = 0.364). Nonetheless, since this research project was carried out in a genuine agricultural setting, the data collection process was limited by the duration of actual agricultural activities, local climate, and economic conditions of the rice fields’ owners. However, the limitation of this study used only factors derived from UAVs. Thus, future studies should include factors related to the environment, economy and farmer behavior that could lead to more accurate yield predictions.

5. Discussion

Since Thailand is one of the world’s major rice exporters, rice is a key economic plant of the country. Rice is vastly cultivated nationwide and consumed not only domestically but also internationally [8]. Nowadays, geoinformatics has been widely applied to contemporary agriculture, including rice cultivation, due to its rapid advancement [11] and [12]. Long-distance aerial exploration using optical satellites is frequently used for monitoring rice producing areas and estimating crop yield in time series based on various vegetation indexes [14]. Synthetic Aperture Radar (SAR) satellites are highly suitable for monitoring rainfed rice farming during the rainy season [15]. Apart from satellite imagery or photography, unmanned aerial vehicles (UAVs) have become a popular instrument for modern agriculture as they can provide crucial data for analytical modeling or factor analysis, for example, estimating crop yield by k-means clustering [16]. Therefore, this study employed two statistical tools such as Multiple Linear Regression and Extreme Gradient Boosting (XGBoost) to estimate crop yields. The tools provided valid research results. Furthermore, the factor analysis was performed by two different types of UAVs, including an RGB drone and a multispectral drone [35].

The UAVs discovered that the most significant factors were canopy cover, plant height, and Triangular Greenness Index (TGI), followed by Green Normalized Difference Vegetation (GNDVI) and Normalized Difference Vegetation Index (NDVI), respectively. To increase the accuracy of the plant height estimation, the Light Detection and Ranging (LIDAR) system was also applied [18]. The system is still an expensive technology to date. Even though the UAVs can be used for estimating crop yields, they are high-resolution equipment which is more suitable for exploring small areas. Satellite imagery is thus suggested for monitoring large-scale rice producing areas. Furthermore, Extreme Gradient Boosting (XGBoost) is a model with higher accuracy than MLR considering the results of this study because Extreme Gradient Boosting (XGBoost) is learning from parameters in the same way as the regression tree family approach. However, MLR is a statistical method that uses two or more independent variables to estimate the dependent variable.

Acknowledgement

We would like to express our gratitude to the National Research Council of Thailand (NRCT) for supporting this research. This project is part of Technologies and knowledge transfer to develop rice cultivation potential with Geoinformatics to group of farmers in Nakhonpathom Province. We would like to extend our sincere thanks to Sam Ngam Large-Scale Rice Production Community Enterprise, Don Tum District, Nakhon Pathom, for allowing and facilitating the conduct of field experiments in their property. Lastly, we would like to express our appreciation to the Geography and Geo-Informatics Program, Faculty of Humanities and Social Sciences, Suan Suandha Rajabhat University, for providing supporting equipment and other facilities.

References

[1] Liao, P., Bell, S. M., Chen, L., Huang, S., Wang, H., Miao, J., Qi, Y., Sun, Y., Liao, B., Zeng, Y., Wei, H., Gao, H., Dai, Q. and Zhang, H., (2023). Improving Rice Grain Yield and Reducing Lodging Risk Simultaneously: A Meta-Analysis. European Journal of Agronomy, Vol.143, https://doi.org/10.1016/j.eja.2022.126709.

[2] Prathumchai, K., Nagai, M., N. K. Tripathi. and Sasaki, N., (2018). Forecasting Transplanted Rice Yield at the Farm Scale Using Moderate-Resolution Satellite Imagery and the AquaCrop Model: A Case Study of a Rice Seed Production Community in Thailand, ISPRS International Journal of Geo-Information, Vol.7. https://doi.org/10.3390/ijgi7020073.

[3] Zhang, Z., Li, Y., Xinguo, C., Wang, Y., Niu, B., Liu, D. L., He, J., Pulatov, B., Hassan I., and Meng, Q., (2023). Impact of Climate Change and Planting Date Shifts on Growth and Yields of Double Cropping Rice in Southeastern China in Future. Agricultural Systems, Vol.205. https://doi.org/10.1016/j.agsy.2022.103581.

[4] Arunrat, N., Pumijumnong, N., Sereenonchai, S., Chareonwong, U. and Wang, C., (2020). Assessment of Climate Change Impact on Rice Yield and Water Footprint of Large-Scale and Individual Farming in Thailand. Science of the Total Environment, Vol.726. https://doi.org/10.1016/j.scitotenv.2020.137864.

[5] Nyussupova, G., Tokbergenova, A., Aubakirova, G., Kelinbayeva R. and Kalimurzina, A., (2019). The Electronic Atlas of Socio-Demographic Development of the Regions of the Republic of Kazakhstan: Elaboration and Main Stages of Creation. International Journal of Geoinform atics, Vol.15, 67-75.

[6] Liu, X., Zheng, J., Yu, L., Hao, P., Chen, B., Xin, Q., Fu, H. and Gong, P., (2021). Annual Dynamic Dataset of Global Cropping Intensity from 2001 to 2019. Scientific Data, Vol.8. https://doi.org/10.1038/s41597-021-01065-9.

[7] Frazier, A. G., Yen, B. T., Stuecker, M. F., Nelson, K. M., Sander, B. O., Kantar, M. B. and Wang, D. R., (2022). Impact of Historical Climate Variability on Rice Production in Mainland Southeast Asia Across Multiple Scales, Anthropocene, Vol.40. https://doi.org/10.1016/j.ancene.2022.100353.

[8] Silva, J. V., Pede, V. O., Radanielson, A. M., Kodama, W., Duarte, A., de Guia, A. H., Malabayabas, A. J. B., Argosubekti, N., Vithoonjit, D., Hieu, P. T. M., Pame, A. R. P., Singleton, G. R. and Stuart, A. M., (2022). Revisiting Yield Gaps and the Scope for Sustainable Intensification for Irrigated Lowland Rice in Southeast Asia. Agricultural Systems, Vol. 198. https://doi.org/10.1016/j.agsy.2022.103383.

[9] Methamontri, Y., Tsusaka, T. W., Zulfiqar, F., Yukongdi, V. and Datta, A., (2022). Factors Influencing Participation in Collective Marketing through Organic Rice Farmer Groups in Northeast Thailand. Heliyon, Vol. 8. https://doi.org/10.1016/j.heliyon.2022.e11421.

[10] Office of Agricultural Economics, (2022). Agricultural Economic Data, [Online]. Available: Office of Agricultural Economics, https://www.oae.go.th. [Accessed December 31, 2022].

[11] Zhou, X., Zheng, H., Xu, X., He, J., Ge, X., Yao, X., Cheng, T., Zhu, Y., Cao, W. and Tian, Y., (2017). Predicting Grain Yield in Rice Using Multi-Temporal Vegetation Indices from UAV-Based Multispectral and Digital Imagery. ISPRS Journal of Photogrammetry and Remote Sensing, Vol.130, 246-255. https://doi.org/10.1016/j.isprsjprs.2017.05.003.

[12] Shen, Y., Mercatoris, B., Cao, Z., Kwan, P., Guo, L., Yao, H. and Cheng, Q., (2022). Improving Wheat Yield Prediction Accuracy Using LSTM-RF Framework Based on UAV Thermal Infrared and Multispectral Imagery. Agriculture, Vol.12. https://doi.org/10.3390/agriculture12060892.

[13] Worachairungreung, M., Thanakunwutthirot, K. and Kulpanich, N., (2023). A Study on Oil Palm Classification for Ranong Province Using Data Fusion and Machine Learning Algorithms. Geographia Technica., Vol.18, 161-176. http://doi.org/10.21163/GT_2023.181.12.

[14] González, J. S., Angelats, E., Eixarch, M. M. and Alcaraz, C., (2022) Monitoring Rice Crop and Yield Estimation with Sentinel-2 Data, Field Crops Research. Vol.281, https://doi.org/10.1016/j.fcr.2022.108507.

[15] Hashemi, M. G., Abhishek, A., Jalilvand, E., Jayasinghe, S., Andreadis, K.M., Siqueira, P. and Das, N. N., (2022). Assessing the Impact of Sentinel-1 Derived Planting Dates on Rice Crop Yield Modeling, International Journal of Applied Earth Observation and Geoinformation, Vol114. https://doi.org/10.1016/j.jag.2022.103047.

[16] Reza, M. N., Na, I. S., Baek, S. W. and Lee, K. H., (2019). Rice Yield Estimation Based on K-means Clustering with Graph-Cut Segmentation Using Low-Altitude UAV Images, Biosystems Engineering, Vol.177, 109-121. https://doi.org/10.1016/j.biosystemseng.2018.09.014.

[17] Ge, H., Ma, F., Li, Z. and Du, C., (2022). Estimating Rice Yield by Assimilating UAV-Derived Plant Nitrogen Concentration into the DSSAT Model: Evaluation at Different Assimilation Time Windows, Field Crops Research, Vol.288. https://doi.org/10.1016/j.fcr.2022.108705.

[18] Phan, A. T. and Takahashi, K., (2021). Estimation of Rice Plant Height from a Low-Cost UAV- Based LIDAR Point Clouds. International Journal of Geoinformatics, Vol.17, 89–98. https://doi.org/10.52939/ijg.v17i2.1765.

[19] Kulpanich, N., Worachairungreung, M. and Chaysmithikul, P., (2020). Surveying for Built-Up Areas in Nakhon Pathom Province with Remote Sensing from Landsat 8 and Sentinel 2 Imagery. Suan Sunandha Academic & Research Review, Vol.14, 44-55.

[20] Land Development Department, (2022). Soil Service - Land Use.

[21] DJI, (2018). Phantom 4_Pro and Pro Plus Series User Manual.

[22]Kulpanich, N., Worachairungreung, M., Waiyasusri, K., Saengow, P. and Pinasu, D., (2022). Height Measurement and Oil Palm Yield Prediction Using Unmanned Aerial Vehicle (UAV) Data to Create Canopy Height Model (CHM). Geographia Technica, Vol.17, 164-178. https://doi.org/10.21163/GT_2022.172.14.

[23] DJI, (2020). P4 Multispectral User Manual v1.4 EN.

[24] Kong, J., Ryu, Y., Huang, Y., Dechant, B., Houborg, R., Guan, K. and Zhu, X. (2021). Evaluation of Four Image Fusion NDVI Products Against In-Situ Spectral-Measurements Over a Heterogeneous Rice Paddy Landscape. Agricultural and Forest Meteorology, Vol.297. https://doi.org/10.1016/j.agrformet.2020.108255.

[25] Basso, M., Stocchero, D., Henriques, R. V. B., Vian, A. L., Bredemeier, C., Konzen, A. A. and de Freitas, E. P., (2019). Proposal for an Embedded System Architecture Using a GNDVI Algorithm to Support UAV-Based Agrochemical Spraying. Sensors, Vol.19. https://doi.org/10.3390/s19245397.

[26] Hunt, Jr. E. R., Doraiswamy, P. C., McMurtrey, J. E., Daughtry, C. S., Perry, E. M. and Akhmedov, B., (2013). A Visible Band Index for Remote Sensing Leaf Chlorophyll Content at the Canopy Scale. International Journal of Applied Earth Observation and Geoinformation, Vol.21, 103-112. https://doi.org/10.1016/j.jag.2012.07.020.

[27] Wan, L., Cen, H., Zhu, J., Zhang, J., Zhu, Y., Sun, D., Du, X., Zhai, L., Weng, H., Li, Y., Li, X., Bao, Y., Shou, J. and He, Y., (2020). Grain Yield Prediction of Rice Using Multi-Temporal UAV-Based RGB and Multispectral Images and Model Transfer – A Case Study of Small Farmlands in the South of China. Agricultural and Forest Meteorology, Vol.291. https://doi.org/10.1016/j.agrformet.2020.108096.

[28] Malambo, L., Popescu, S., Murray, S., Putman, E., Pugh, N., Horne, D., Richardson, G., Sheridan, R., Rooney, W., Avant, R., Vidrine, M., McCutchen, B., Baltensperger, D. and Bishop, M., (2018). Multitemporal Field-Based Plant Height Estimation Using 3D Point Clouds Generated from Small Unmanned Aerial Systems High-Resolution Imagery. International Journal of Applied Earth Observation and Geoinformation, Vol.64, 31-42. https://doi.org/10.1016/j.jag.2017.08.014.

[29] Hunt, E. R., Hively, W. D., McCarty, G. W., Daughtry, C. S. T., Forrestal, P. J., Kratochvil, R. J., Carr, J. L., Allen, N. F., Fox-Rabinovitz, J. R. and Miller, C. D., (2011). NIR-Green-Blue High-Resolution Digital Images for Assessment of Winter Cover Crop Biomass. GIScience & Remote Sensing, Vol.48, 86-98. https://doi.org/10.2747/1548-1603.48.1.86.

[30] Abrougui, K., Gabsi, K., Mercatoris, B., Khemis, C., Amami, R. and Chehaibi, S., (2019). Prediction of Organic Potato Yield Using Tillage Systems and Soil Properties by Artificial Neural Network (ANN) and Multiple Linear Regressions (MLR). Soil and Tillage Research, Vol.190, 202-208. https://doi.org/10.1016/j.still.2019.01.011.

[31] Wang, J., Wu, B., Kohnen, M., Lin, D., Yang, C., Wang, X., Qiang, A., Liu, W., Kang, J., Li, H., Shen, J., Yao, T., Su, J., Li, B. and Gu, L., (2021). Classification of Rice Yield Using UAV-Based Hyperspectral Imagery and Lodging Feature. Plant Phenomics, Vol.2021, 1-14. https://doi.org/10.34133/2021/9765952.

[32] Li, P., Hao, H., Bai, Y., Li, Y., Mao, X., Xu, J., Liu, M., Lv, Y., Chen, W. and Ge, D., (2022). Convolutional Neural Networks-Based Health Risk Modelling of Some Heavy Metals in a Soil-Rice System. Science of the Total Environment, Vol.838. https://doi.org/10.1016/j.scitotenv.2022.156466.

[33] Chu, Z. and Yu, J., (2020). An End-to-End Model for Rice Yield Prediction Using Deep Learning Fusion. Computers and Electronics in Agriculture, Vol.174. https://doi.org/10.1016/j.compag.2020.105471.

[34] Abdikan, S., Sekertekin, A., Narin, O. G., Delen, A. and Sanli, F. B., (2022). A Comparative Analysis of SLR, MLR, ANN, XGBoost and CNN for Crop Height Estimation of Sunflower Using Sentinel-1 and Sentinel-2. Advances in Space Research, Vol.71(7), 3045-3059. https://doi.org/10.1016/j.asr.2022.11.046.

[35] Bascon, M. V., Nakata, T., Shibata, S., Takata, I., Kobayashi, N., Kato, Y., Inoue, S., Doi, K., Murase, J. and Nishiuchi, S., (2022). Estimating Yield-Related Traits Using UAV-Derived Multispectral Images to Improve Rice Grain Yield Prediction. Agriculture, Vol. 12(8). https://doi.org/10.3390/agriculture12081141.