Crop Productivity Analysis

Crop Productivity Analysis#

Introduction#

Enhanced Vegetation Index is the greenness or health of a pixel calculated by the MODIS Terra and Aqua satellites. This is typically used as a proxy for the health of a plant.

The relationship between crop yields and the Enhanced Vegetation Index (EVI), a proxy for vegetation health derived from satellite imagery, is central to agricultural forecasting [Johnson, 2016]. EVI is a spectral index that utilizes the blue, red, and near-infrared (NIR) bands of satellite data to monitor vegetation health. Relative to other vegetation indices, such as the Normalized Difference Vegetation Index (NDVI), EVI provided superior performance by offering increased sensitivity in high biomass areas and mitigating atmospheric effects [Huete et al., 2002]. Johnson [2016] provides a comprehensive overview of the use of EVI in agricultural yield prediction, highlighting its advantages over other indices like the Normalized Difference Vegetation Index (NDVI). The study concluded that EVI was among the best overall performer in predicting crop yields, showing the highest correlation in five out of nine crops studied, excluding rice.

Numerous studies have successfully utilized EVI to predict yields of various crops, including maize, soybeans, and rice. For instance, Bolton and Friedl [2013] created a linear model using EVI to forecast soybean and maize yields in the Central United States. Their findings demonstrate that EVI exhibits a stronger correlation with maize yield and yield anomalies than other vegetation indices, and incorporating phenological data significantly enhances the model’s performance. A study in Thailand by Wijesingha et al. [2015] investigated the use of MODIS EVI for rice crop monitoring and yield assessment. Their research revealed that maximum EVI achieved a high correlation of 0.95 with rice yield, indicating its effectiveness in yield prediction. Furthermore, Son et al. [2013] used MODIS EVI data for rice yield prediction in the Vietnamese Mekong Delta, reporting high correlation coefficients (R2 of 0.70 for spring-winter and 0.74 for autumn-summer in 2013, and 0.62-0.71 and 0.4-0.56 in 2014, respectively). Lastly, a study in one of the major rice-producing areas in China, Yangtze River Delta region, by Shi and Huang [2015] further confirmed that the MODIS-based area shows higher consistency with agricultural census data during the period of 2000-2010, with less than 15% error.

Both NDVI and EVI have been used to study agricultural productivity and EVI.

Satellite and Ground-Based Approaches#

Compared to ground-based methods, such as field interviews and crop cuttings, Remote sensing approach has several advantages on estimating crop yields:

  • Timeliness and Rapid Assessment: Remote sensing offers real-time or near real-time assessment of crop conditions and potential yields, which is crucial for situations like natural disasters or conflicts [Doraiswamy et al., 2003].

  • Extensive Spatial Coverage and Granularity: Satellite imagery can consistently cover large geographic areas, providing granular, field, and village-level data that can be aggregated to higher administrative levels, such as districts or provinces [Azzari et al., 2017, Becker-Reshef et al., 2010].

  • Cost-Effectiveness: Remote sensing is generally more cost-effective for collecting agricultural data over large areas compared to ground-based methods [Johnson, 2016, Rahman et al., 2009].

  • Accessibility in Difficult or Dangerous Areas: Remote sensing provides a reliable option for data collection in zones that are difficult, dangerous, or inaccessible for in-situ surveys, such as conflict areas [Jaafar and Ahmad, 2015].

  • Independent and Consistent Data Source: Remote sensing offers an independent data source that can validate or complement ground-based data, which may be subject to biases or inaccuracies.

However, remote sensing also has several limitations:

  • Spatial Resolution Limitations: While modern sensors offer higher resolutions, coarse resolution data may not accurately reflect ground situations for small agricultural clusters, small parcels, or low-intensity agriculture [Kibret et al., 2020].

  • Significant Computational Demands: Analyzing high-resolution imagery requires massive amounts of computational power and storage, which can be expensive and time-consuming [Petersen, 2018].

  • Interpretability and Bias Concerns: Satellite-derived metrics may not fully capture all determinants of crop production and may not quantitatively interpret crop growth status [Wu et al., 2023].

While remote sensing offers timely, broad-scale, and cost-effective monitoring, particularly in data-scarce or inaccessible regions, it faces challenges with resolution limitations, computational demands, and potential biases in interpreting crop conditions.

EVI and Other Vegetation Indices#

Within the remote sensing approach itself, multiple vegetation indices have been utilized to enhance the accuracy of yield predictions. Compared to other vegetation indices, EVI offers several advantages over other vegetation indices:

  • Reduced saturation: EVI does not saturate as quickly as NDVI at higher crop leaf area or in areas with large amounts of biomass, providing improved sensitivity in dense vegetation conditions [Huete et al., 2002]. This is crucial as high saturation in indicators like NDVI can lead to unreliable yield estimates [Son et al., 2013].

  • Atmospheric and Soil Correction: EVI incorporates a soil adjustment factor and corrections for the red band due to aerosol scattering, making it more resistant to atmospheric influences and soil background noise compared to NDVI [Jurečka et al., 2018].

On the other hand, EVI also has some limitations compared to other vegetation indices:

  • Temporal Resolution and Latency: While EVI is disseminated at 250m resolution, it may only be available at 16-day time steps, compared to NDVI’s 8-day availability, which could introduce latency issues for real-time monitoring [Johnson, 2016].

  • Mixed Pixel Problems: Coarse spatial resolution of MODIS EVI (250m or 500m) can limit performance in regions with small, fragmented parcels [Kibret et al., 2020].

Climate influenced on EVI#

EVI values are influenced by several environmental factors. EVI is sensitive to rainfall patterns, as the variations of phenological stages of crops, such as the timing of planting and harvesting, are closely linked to rainfall [Kibret et al., 2020]. Additionally, while EVI is designed to be resistant to atmospheric aerosols, cloud contamination, particularly during wet seasons, can affect vegetation greenness signals and lower the accuracy of yield predictions [Son et al., 2013].

Modelling Approaches of EVI#

Several modeling approaches have been successfully employed with EVI data for crop yield prediction. Statistical models, particularly linear and quadratic regression, are widely used due to their fewer data requirements and assumptions compared to biophysical models [Ji et al., 2022]. Linear regression models are frequently applied, with EVI generally performing well and showing more linear relationships with yields than NDVI for certain crops [Tiruneh et al., 2023]. Machine learning approaches have also proven effective, with [Kibret et al., 2020] applied Random Forest algorithm to MODIS EVI time series for agricultural land use classification and cropping system identification. [Pham et al., 2022] demonstrated that integrating Principal Component Analysis (PCA) with machine learning methods (PCA-ML) on VCI data effectively addresses spatial variability and redundant data issues, enhancing prediction accuracy by up to 45% in rice yield forecasting for Vietnam.

Data#

The following datasets are utilized in this analysis for calculating and mapping crop productivity over the past years:

  1. Dynamic World Dataset:

    • Source: Dynamic World - Google and the World Resources Institute (WRI)

    • Description:The Dynamic World dataset provides a near real-time, high-resolution (10-meter) global land cover classification. It is derived from Sentinel-2 imagery and utilizes machine learning models to classify land cover into nine distinct classes, including water, trees, grass, crops, built areas, bare ground, shrubs, flooded vegetation, and snow/ice. The dataset offers data with minimal latency, enabling near-immediate analysis and decision-making.

    • Spatial Resolution: 10 meters.

    • Temporal Coverage: Data is available since mid-2015, updated continuously as Sentinel-2 imagery becomes available capturing near real-time.

  2. MODIS Dataset:

    • Source: NASA’s Moderate Resolution Imaging Spectroradiometer MODIS on Terra and Aqua satellites.

    • Description: The MODIS dataset provides a wide range of data products, including land surface temperature, vegetation indices, and land cover classifications. It is widely used for monitoring and modeling land surface processes.

    • Spatial Resolution: 250 meters.

    • Temporal Coverage: Data is available from 2000 to the present, with 16-day composite products.

  3. Administrative Boundaries (HDX):

    • Source: Humanitarian Data Exchange HDX.

    • Description: Geographic boundaries used for spatial aggregation and administrative analysis, such as calculating productivity metrics by region (e.g., governorate or district).

    • Use Case: The administrative boundaries are used to aggregate EVI statistics by region and facilitate reporting at various administrative levels.

  4. Crop production and yield statistics:

    • Source: Food and Agriculture Organization FAOSTAT.

    • Description: Data on crop production and yields used for validating remote sensing-derived productivity estimates.

    • Use Case: These statistics are used to compare and validate the EVI-based productivity estimates against actual reported yields.

Crop area statistics#

  Name Crop Area (ha) in 2023 Crop Area (ha) in 2025 % Change in Crop Area (2023-2025)
1 Aleppo 829,480 508,074 -38.75%
0 Al-Hasakeh 754,855 572,731 -24.13%
2 Ar-Raqqa 432,211 272,511 -36.95%
8 Homs 414,471 198,439 -52.12%
7 Hama 399,303 246,211 -38.34%
9 Idleb 265,526 183,460 -30.91%
5 Dar'a 186,564 130,346 -30.13%
3 As-Sweida 161,659 102,869 -36.37%
6 Deir-ez-Zor 159,704 116,377 -27.13%
12 Rural Damascus 130,318 72,270 -44.54%
11 Quneitra 94,322 90,957 -3.57%
13 Tartous 26,948 25,117 -6.79%
10 Lattakia 20,064 21,233 +5.83%
4 Damascus 397 252 -36.48%

Crop Seasonality#

Using this time series dataset of EVI images, we apply several pre-processing steps to extract critical phenological parameters: start of season (SOS), middle of season (MOS), end of season (EOS), length of season (LOS), etc. This workflow is heavily inspired by the TIMESAT software.

Pre-processing steps

  1. Remove outliers from dataset on per-pixel basis using median method: outlier if median from a moving window < or > standard deviation of time-series times 2.

  2. Interpolate missing values linearly

  3. Smooth data on per-pixel basis (using Savitsky Golay filter, window length of 3, and polyorder of 1)

Phenology Process
We then extract crop seasonality metrics using the seasonal amplitude method from the phenolopy package.

The chart below shows the result of this process for a single crop pixel. The blue dots represent the raw EVI values, the black line represents the processed EVI values, and the dotted lines represent season parameters extracted for that pixel: start of season, peak of season, and end of season.

Based on the phenology process, we identified that the season starts in February and ends in June with the peak being in March. However, the seasonality can vary depending on the crop type and geographic region.

In the figure below, we run the phenology process for each governorate. While the majority of the governorates follow the same seasonality pattern, some governorates such as Ar-Raqqa and Deir ez-Zor have two distinct peaks, which could indicate multiple cropping seasons or different crop types being cultivated. Also, governorates such as As-Sweida and Tartous show a slightly delayed seasonality pattern compared to other governorates.

The figure below shows the EVI time series from 2010 to 2025, aggregated at the national level. EVI in 2025 appears to be lowest since 2010, indicating potential reduced vegetation health this year. However, it is important to note that EVI for 2025 only includes data up to October.

Pre and Post Regime Change#

The figure below shows the EVI time series from January 2023 to October 2025, aggregated at the national level. The vertical dashed line indicates the regime change in November 2024. There is a noticeable decline in EVI values following the regime change. The decline in EVI could be attributed to various factors, including changes in agricultural practices, policy shifts, or environmental conditions following the regime change.

The figure below zooms into the EVI time series during growing season in 2024 and 2025 and it shows that the EVI values during the growing season in 2025 are significantly lower compared to the same period in 2024.

In the figure below, we add rainfall data to the EVI time series from January 2023 to October 2025, aggregated at the national level. Rainfall in Syria is mainly between October and May. We can see that the rainfall in 2025 is significantly lower compared to 2024, which could explain the lower EVI values observed during the growing season in 2025. There’s also a lagged effect where reduced rainfall in earlier months impacts EVI in subsequent months.

Hide code cell outputs

We run a regression analysis to quantify the relationship between EVI and regime change, while controlling for rainfall. The results indicate a negative impact of the regime change on EVI values, even after accounting for rainfall variations. However, the impact itself is not statistically significant, suggesting that other factors may also be influencing EVI changes post-regime change.

Dependent variable: evi_log
Based on data from November 2023 to September 2025
(1)
Post-Regime Change-0.194
(-0.494 , 0.106)
Intercept-1.965***
(-2.217 , -1.712)
Rainfall (log mm)0.064*
(-0.007 , 0.135)
Observations23
R20.235
Adjusted R20.159
Residual Std. Error0.342 (df=20)
F Statistic3.080* (df=2; 20)
Note:*p<0.1; **p<0.05; ***p<0.01

Hide code cell outputs

The figure below shows the EVI and rainfall time series from January 2023 to October 2025, aggregated at the governorate level. There is a noticeable decline in rainfall across most governorates compared to previous years, while in Lattakia and Tartous, rainfall appears to be relatively stable.

Crop yield#

In the figure below, we compare EVI with FAO-reported crop-related metrics in Syria from 2010 to 2023. Given that Wheat is the dominant crop in Syria, we focus on wheat yield for this comparison. The EVI trend generally aligns with the wheat yield and production trends reported by FAO, indicating that EVI can be a useful proxy for monitoring crop productivity in Syria.

References#

AKV21

Serkan Arslanalp, Robin Koepke, and Jasper Verschuur. Tracking trade from space: an application to pacific island countries. IMF Working Papers, 2021(225):A001, 2021. URL: https://www.elibrary.imf.org/view/journals/001/2021/225/article-A001-en.xml, doi:10.5089/9781513593531.001.A001.

AJL17

George Azzari, Meha Jain, and David B. Lobell. Towards fine resolution global maps of crop yields: Testing multiple methods and satellites in three countries. Remote Sensing of Environment, 202:129–141, December 2017. doi:10.1016/j.rse.2017.04.014.

BRVLJ10

I. Becker-Reshef, E. Vermote, M. Lindeman, and C. Justice. A generalized regression-based model for forecasting winter wheat yields in Kansas and Ukraine using MODIS data. Remote Sensing of Environment, 114(6):1312–1323, June 2010. doi:10.1016/j.rse.2010.01.010.

BF13

Douglas K. Bolton and Mark A. Friedl. Forecasting crop yield using remotely sensed vegetation indices and crop phenology metrics. Agricultural and Forest Meteorology, 173:74–84, May 2013. doi:10.1016/j.agrformet.2013.01.007.

CKomaromiLS20

Diego A. Cerdeiro, András Komáromi, Y. Liu, and Mamoon M. Saeed. World seaborne trade in real time: a proof of concept for building ais-based nowcasts from scratch. IMF Working Papers, 2020.

DMCS03

Paul C. Doraiswamy, Sophie Moulin, Paul W. Cook, and Alan Stern. Crop Yield Assessment from Remote Sensing. Photogrammetric Engineering & Remote Sensing, 69(6):665–674, June 2003. doi:10.14358/PERS.69.6.665.

GTS+14

Dirk Geudtner, Ramón Torres, Paul Snoeij, Malcolm Davidson, and Björn Rommen. Sentinel-1 system capabilities and applications. In 2014 IEEE Geoscience and Remote Sensing Symposium, volume, 1457–1460. 2014. doi:10.1109/IGARSS.2014.6946711.

HT04

Ramaswamy Hariharan and Kentaro Toyama. Project Lachesis: Parsing and Modeling Location Histories. In Geographic Information Science. Volume 3234, volume 3234, 106–124. 2004. doi:10.1007/978-3-540-30231-5_8.

HDM+02(1,2)

A Huete, K Didan, T Miura, E.P Rodriguez, X Gao, and L.G Ferreira. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83(1-2):195–213, November 2002. doi:10.1016/S0034-4257(02)00096-2.

JA15

Hadi H. Jaafar and Farah A. Ahmad. Crop yield prediction from remotely sensed vegetation indices and primary productivity in arid and semi-arid lands. International Journal of Remote Sensing, 36(18):4570–4589, September 2015. doi:10.1080/01431161.2015.1084434.

JPZ+22

Zhonglin Ji, Yaozhong Pan, Xiufang Zhu, Dujuan Zhang, and Jinyun Wang. A generalized model to predict large-scale crop yields integrating satellite-based vegetation index time series and phenology metrics. Ecological Indicators, 137:108759, April 2022. doi:10.1016/j.ecolind.2022.108759.

JPS19

Haiying Jia, Vishnu Prakash, and Tristan Smith. Estimating vessel payloads in bulk shipping using ais data. International Journal of Shipping and Transport Logistics, 11:25, 01 2019. doi:10.1504/IJSTL.2019.096864.

Joh16(1,2,3,4)

David M. Johnson. A comprehensive assessment of the correlations between field crop yields and commonly used MODIS products. International Journal of Applied Earth Observation and Geoinformation, 52:65–81, October 2016. doi:10.1016/j.jag.2016.05.010.

JLH+18

František Jurečka, Vojtěch Lukas, Petr Hlavinka, Daniela Semerádová, Zdeněk Žalud, and Miroslav Trnka. Estimating Crop Yields at the Field Level Using Landsat and MODIS Products. Acta Universitatis Agriculturae et Silviculturae Mendelianae Brunensis, 66(5):1141–1150, October 2018. doi:10.11118/actaun201866051141.

KMC20(1,2,3,4)

Kefyalew Sahle Kibret, Carsten Marohn, and Georg Cadisch. Use of MODIS EVI to map crop phenology, identify cropping systems, detect land use change and drought risk in Ethiopia – an application of Google Earth Engine. European Journal of Remote Sensing, 53(1):176–191, January 2020. doi:10.1080/22797254.2020.1786466.

Kun14

Robin Kunaka, Charles; Carruthers. Trade and Transport Corridor Management Toolkit. World Bank Training, World Bank, 2014. License: CC BY 3.0 IGO. URL: http://hdl.handle.net/10986/18241.

Pet18

Lillian Kay Petersen. Real-Time Prediction of Crop Yields From MODIS Relative Vegetation Health: A Continent-Wide Analysis of Africa. Remote Sensing, 10(11):1726, November 2018. Number: 11 Publisher: Multidisciplinary Digital Publishing Institute. doi:10.3390/rs10111726.

PAK+22

Hoa Thi Pham, Joseph Awange, Michael Kuhn, Binh Van Nguyen, and Luyen K. Bui. Enhancing Crop Yield Prediction Utilizing Machine Learning on Satellite-Based Vegetation Health Indices. Sensors, 22(3):719, January 2022. doi:10.3390/s22030719.

RRK+09

Atiqur Rahman, Leonid Roytman, Nir Y. Krakauer, Mohammad Nizamuddin, and Mitch Goldberg. Use of Vegetation Health Data for Estimation of Aus Rice Yield in Bangladesh. Sensors, 9(4):2968–2975, April 2009. doi:10.3390/s90402968.

SH15

Jingjing Shi and Jingfeng Huang. Monitoring Spatio-Temporal Distribution of Rice Planting Area in the Yangtze River Delta Region Using MODIS Images. Remote Sensing, 7(7):8883–8905, July 2015. doi:10.3390/rs70708883.

SCC+13(1,2,3)

N. T. Son, C. F. Chen, C. R. Chen, L. Y. Chang, H. N. Duc, and L. D. Nguyen. Prediction of rice crop yield using MODIS EVI−LAI data in the Mekong Delta, Vietnam. International Journal of Remote Sensing, 34(20):7275–7292, October 2013. doi:10.1080/01431161.2013.818258.

TMA+23

Gizachew Ayalew Tiruneh, Derege Tsegaye Meshesha, Enyew Adgo, Atsushi Tsunekawa, Nigussie Haregeweyn, Ayele Almaw Fenta, Tiringo Yilak Alemayehu, Temesgen Mulualem, Genetu Fekadu, Simeneh Demissie, and José Miguel Reichert. Mapping crop yield spatial variability using Sentinel-2 vegetation indices in Ethiopia. Arabian Journal of Geosciences, 16(11):631, November 2023. doi:10.1007/s12517-023-11754-x.

VKH21a

Jasper Verschuur, Elco E Koks, and Jim W Hall. Global economic impacts of COVID-19 lockdown measures stand out in high-frequency shipping data. PLoS One, 16(4):e0248818, April 2021.

VKH21b

Jasper Verschuur, Elco E. Koks, and Jim W. Hall. Observed impacts of the covid-19 pandemic on global trade. Nature Human Behaviour, 5(3):305–307, Mar 2021. URL: https://doi.org/10.1038/s41562-021-01060-5, doi:10.1038/s41562-021-01060-5.

WDS15

J. S. J. Wijesingha, N. L. Deshapriya, and L. Samarakoon. Rice Crop Monitoring and Yield Assessment with MODIS 250m Gridded Vegetation Products: A Case Study of Sa Kaeo Province, Thailand. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XL-7/W3:121–127, April 2015. doi:10.5194/isprsarchives-XL-7-W3-121-2015.

WZZ+23

Bingfang Wu, Miao Zhang, Hongwei Zeng, Fuyou Tian, Andries B Potgieter, Xingli Qin, Nana Yan, Sheng Chang, Yan Zhao, Qinghan Dong, Vijendra Boken, Dmitry Plotnikov, Huadong Guo, Fangming Wu, Hang Zhao, Bart Deronde, Laurent Tits, and Evgeny Loupian. Challenges and opportunities in remote sensing-based crop monitoring: a review. National Science Review, 10(4):nwac290, March 2023. doi:10.1093/nsr/nwac290.

WorldBank21

World Bank. World Development Report 2021 : Data for Better Lives. World Bank, 2021. License: CC BY 3.0 IGO. URL: http://hdl.handle.net/10986/35218.

WorldBank22

World Bank. Syria Economic Monitor - Spring 2022 : Lost Generation of Syrians. Economic Updates and Modeling, World Bank, 2022. License: CC BY 3.0 IGO. URL: http://hdl.handle.net/10986/37617.

Contents