Improving Wheat Yield Prediction with Multi-Source Remote Sensing Data and Machine Learning in Arid Regions

Developing machine learning models with multi-source environmental data to predict wheat yield in China

Current vegetation indices and biophysical parameters derived from optical satellite data for forest monitoring are widely used in various applications but can be limited by atmospheric effects like clouds. Synthetic aperture radar (SAR) data can offer insightful and systematic forest monitoring with complete time series due to signal penetration through clouds and day and night image acquisitions. This study explores the use of SAR data, combined with ancillary data and machine learning (ML), to estimate forest parameters typically derived from optical satellites. It investigates whether SAR signals provide sufficient information for the accurate estimation of these parameters, focusing on two spectral vegetation indices (Normalized Difference Vegetation Index – NDVI and Enhanced Vegetation Index – EVI) and two biophysical parameters (Leaf Area Index – LAI and Fraction of Absorbed Photosynthetically Active Radiation – FAPAR) in healthy and disturbed temperate forests in Czechia and Central Europe in 2021. Vegetation metrics derived from Sentinel-2 multispectral data were used to evaluate the results. A paired multi-modal time-series dataset was created using Google Earth Engine (GEE), including temporally and spatially aligned Sentinel-1, Sentinel-2, DEM-based features and meteorological variables, along with a forest type class. The inclusion of DEM-based auxiliary features and additional meteorological information improved the results. In the comparison of ML models, the traditional ML algorithms, Random Forest Regressor and Extreme Gradient Boosting (XGB) slightly outperformed the Automatic Machine Learning (AutoML) approach, auto-sklearn, for all forest parameters, achieving high accuracies (R2 between 70% and 86%) and low errors (0.055–0.29 of mean absolute error). XGB was the most computationally efficient. Moreover, SAR-based estimations over Central Europe achieved comparable results to those obtained in testing within Czechia, demonstrating their transferability for large-scale modeling. A key advantage of the SAR-based vegetation metrics is the ability to detect abrupt forest changes with sub-weekly temporal accuracy, providing up to 240 measurements per year at a 20 m resolution.

Accurate regional crop classification, acreage estimation, yield prediction, and crop water requirement assessment are essential for effective agricultural planning and market forecasts. This study uses an integrated geospatial and statistical approach to assess major winter crops wheat and sugarcane cultivation in Ghotki District, Pakistan, from 2017/18 to 2022/23. It combines satellite data from Landsat 8 and Sentinel-2, ground truthing, and crop reporting records to analyze key factors such as cultivation area, crop gradients, vegetation health, normalized difference vegetation index (NDVI)-based wheat and sugarcane yield models, crop water requirements, and total irrigation water consumption. Results showed that wheat cultivation areas ranged from 15% to 19%, with the highest coverage observed in the 2021/22 winter season. Sugarcane cultivation ranged from 6% to 10%, peaking in the 2018/19 season. A strong linear association between NDVI and wheat yield (R2 = 0.86) was observed. Wheat and sugarcane yield predictions utilized linear regression, and robust linear regression models, all of which were validated by the findings. Irrigation water demand for the winter season was calculated at 1887 million cubic meters (MCM) in 2017/18, with 1357 MCM supplied by the Sindh Irrigation Drainage Authority (SIDA). By 2020/21, water demand reached 2023 MCM, while SIDA’s supply was 1357 MCM. These results highlight the significance of integrating geospatial analysis with statistical records to provide timely, reliable estimates for cropped areas, yield forecasting, vegetation dynamics, and irrigation planning. The proposed methodology contributes a scaleable solution for informed decision-making in agricultural and water resource management, applicable across other districts in Pakistan and on a global scale.

Traditional approaches to mapping rice areas depend on detailed field surveys to gather label data. This data gathering, while thorough, is often time-consuming, and costly. Further, for long-term rice area mapping, especially at large scales, ground validation can only be made in recent years, which may lead to uncertainties in rice areas of the past. To solve this limitation, this study introduces a novel method for classifying rice cultivation areas without the need for field data by employing a self-supervised learning framework within Google Earth Engine. Using MODIS data to calculate indices such as the Normalized Difference Vegetation Index (NDVI), the Enhance Vegetation Index (EVI), and the Land and Surface Water Index (LSWI), a pseudo-label boundary was delineated. The delineation of the boundary involved marking flooded regions, where LSWI + 0.05 ≥ NDVI EVI. Within these flooded regions, instances when EVI in the first 40 days was at least half of the peak EVI during the growing season were assigned as rice. In contrast, regions that did not meet these criteria were considered for non-rice. As proof of concept, the delineated pseudo-label boundaries of rice and non-rice were randomly sampled and trained on several machine learning models like random forests, support vector machines, gradient-boosted trees, and decision trees, to classify rice areas in Punjab, India, from 2003 to 2022. The random forest model demonstrated superior performance, achieving an Area Under the Curve of receiver-operating characteristics (AUC) of 0.71, compared to other models (AUC of ~0.55). Furthermore, comparing the self-supervised models against the same machine learning models, which were traditionally trained on field survey data (228 ground points: 164 rice, 64 non-rice was collected), the self-supervised models showed ~10% higher performance than their traditionally supervised counterparts. Therefore, this study demonstrates that using this self-supervised modeling framework reduces the need for field-based annotations, while still providing reasonably accurate rice area maps.

Assessing soil productivity potential in arid region using remote sensing vegetation indices

Crop intelligence and yield prediction of potato (Solanum tuberosum L.) are important to farmers and the processing industry. Remote sensing can provide timely information on growth status and accurate yield predictions during the growing season. However, there is limited documentation on the most suitable vegetation indices (VIs) and optimal growth stages for acquiring remote sensing imagery of potato. To address this knowledge gap, a systematic review was conducted. Original scientific manuscripts published between 2000 and 2022 were identified using various databases. The findings indicate that satellite imagery is the most widely used source of remote sensing data for tuber yield prediction, whereas unmanned aerial vehicle systems (UAVs) and handheld sensors are more frequently applied for growth monitoring. The normalized difference vegetation index (NDVI), red-edge chlorophyll index (CIred-edge), green chlorophyll index (CIgreen), and optimized soil-adjusted vegetation index (OSAVI) are the most frequently used VIs for the growth and yield estimation of potato. The tuber initiation stage was found to be the most appropriate stage for remote sensing data acquisition. This review will assist potato farmers, agronomists and researchers in selecting the most suitable VIs for monitoring specific growth variables and selecting the optimal timing during the growing season to obtain remote sensing images.

Accurate prediction of food crop yield is of great significance for global food security and regional trade stability. Since remote sensing data collected from unmanned aerial vehicle (UAV) platforms have the features of flexibility and high resolution, these data can be used as samples to develop regional regression models for accurate prediction of crop yield at a field scale. The primary objective of this study was to construct regional prediction models for winter wheat yield based on multi-spectral UAV data and machine learning methods. Six machine learning methods including Gaussian process regression (GPR), support vector machine regression (SVR) and random forest regression (RFR) were used for the construction of the yield prediction models. Ten vegetation indices (VIs) extracted from canopy spectral images of winter wheat acquired from a multi-spectral UAV at five key growth stages in Xuzhou City, Jiangsu Province, China in 2021 were selected as the variables of the models. In addition, in situ measurements of wheat yield were obtained in a destructive sampling manner for prediction algorithm modeling and validation. Prediction results of single growth stages showed that the optimal model was GPR constructed from extremely strong correlated VIs (ESCVIs) at the filling stage (R2 = 0.87, RMSE = 49.22 g/m2, MAE = 42.74 g/m2). The results of multiple stages showed GPR achieved the highest accuracy (R2 = 0.88, RMSE = 49.18 g/m2, MAE = 42.57 g/m2) when the ESCVIs of the flowering and filling stages were used. Larger sampling plots were adopted to verify the accuracy of yield prediction; the results indicated that the GPR model has strong adaptability at different scales. These findings suggest that using machine learning methods and multi-spectral UAV data can accurately predict crop yield at the field scale and deliver a valuable application reference for farm-scale field crop management.

This research offers a fresh understanding of desertification in Turkmenistan by utilizing satellite remote sensing data and machine learning techniques. With 80% of its area covered by desert, Turkmenistan has particular difficulties as a result of the harsh effects of desertification, which are made worse by climate change and irresponsible land use. Despite the fact that desertification has been the subject of numerous studies conducted worldwide, this study is among the first to use a multi-index approach to specifically focus on Turkmenistan. It does this by integrating six important desertification indicators within machine learning models like random forest (RF), eXtreme Gradient Boosting (XGBoost), naïve Bayes (NB), and K-nearest neighbors (KNN). These indicators include the Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Moisture Index (NDMI), Bare Soil Index (BSI), Enhanced Vegetation Index (EVI), and land surface temperature (LST). Based on Sentinel-2 satellite data processed by the Google Earth Engine (GEE) platform, the findings show that the country’s northern, central, and eastern regions are undergoing severe desertification. Moreover, RF and XGBoost performed better than the straightforward models like NB and KNN in terms of accuracy (96% and 96.33%), sensitivity (both 100%), and kappa (0.901 and 0.9095). By concentrating on Turkmenistan, this study fills a significant gap and provides a framework for tracking desertification in similar regions around the world.

Modelling species distributions with remote sensing data: bridging disciplinary perspectives

Climate change and human activities have increased droughts, especially overgrazing and deforestation, which seriously threaten the balance of terrestrial ecosystems. The ecological carrying capacity and vegetation cover in the arid zone of Xinjiang, China, are generally low, necessitating research on vegetation response to drought in such arid regions. In this study, we analyzed the spatial and temporal characteristics of drought in Xinjiang from 2001 to 2020 and revealed the response mechanism of SIF to multi-timescale drought in different vegetation types using standardized precipitation evapotranspiration index (SPEI), solar-induced chlorophyll fluorescence (SIF), normalized difference vegetation index (NDVI), and enhanced vegetation index (EVI) data. We employed trend analysis, standardized anomaly index (SAI), Pearson correlation, and trend prediction techniques. Our investigation focused on the correlations between GOSIF (a new SIF product based on the Global Orbital Carbon Observatory-2), NDVI, and EVI with SPEI12 for different vegetation types over the past two decades. Additionally, we examined the sensitivities of vegetation GOSIF to various scales of SPEI in a typical drought year and predicted future drought trends in Xinjiang. The results revealed that the spatial distribution characteristics of GOSIF, normalized difference vegetation index (NDVI), and enhanced vegetation index (EVI) were consistent, with mean correlations with SPEI at 0.197, 0.156, and 0.128, respectively. GOSIF exhibited the strongest correlation with SPEI, reflecting the impact of drought stress on vegetation photosynthesis. Therefore, GOSIF proves advantageous for drought monitoring purposes. Most vegetation types showed a robust response of GOSIF to SPEI at a 9-month scale during a typical drought year, with grassland GOSIF being particularly sensitive to drought. Our trend predictions indicate a decreasing trend in GOSIF vegetation in Xinjiang, coupled with an increasing trend in drought. This study found that compared with that of the traditional greenness vegetation index, GOSIF has obvious advantages in monitoring drought in the arid zone of Xinjiang. Furthermore, it makes up for the lack of research on the mechanism of vegetation GOSIF response to drought on multiple timescales in the arid zone. These results provide strong theoretical support for investigating the monitoring, assessment, and prediction of vegetation response to drought in Xinjiang, which is vital for comprehending the mechanisms of carbon and water cycles in terrestrial ecosystems.

The Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) are vegetation indices widely used in remote sensing of above-ground biomass. Because both indexes are based on spectral features of plant canopy, NDVI and EVI may suffer reduced accuracy in estimating above-ground biomass when flower signals are mixed in the plant canopy. This paper addresses how flowers influence the estimation of above-ground biomass using NDVI and EVI for an alpine meadow with mixed yellow flowers of Halerpestes tricuspis (Ranunculaceae). Field spectral measurements were used in combination with simulated reflectance spectra with precisely controlled flower coverage by applying a linear spectral mixture model. Using the reflectance spectrum for the in-situ canopy with H. tricuspis flowers, we found no significant correlation between above-ground biomass and EVI (p = 0.17) or between above-ground biomass and NDVI (p = 0.78). However, both NDVI and EVI showed very good prediction of above-ground biomass with low root mean square errors (RMSE = 43 g m−2 for NDVI and RMSE = 43 g m−2 for EVI, both p < 0.01) when all the flowers were removed from the canopies. Simulation analysis based on the in-situ measurements further indicated that high variation in flower coverage among different quadrats could produce more noise in the relationship between above-ground biomass and NDVI, or EVI, which results in an evident decline in the accuracy of above-ground biomass estimation. Therefore, the study suggests that attention should be paid both to the flower fraction and the heterogeneity of flower distribution in the above-ground biomass estimation via NDVI and EVI.

Vegetation indices (VIs) such as the Normalized Difference Vegetation Index (NDVI) are widely used for assessing vegetation cover and condition. One of the NDVI's significant disadvantages is its sensitivity to aerosols in the atmosphere, hence several atmospherically resistant VIs were formulated using the difference in the radiance between the blue and the red spectral bands. The state‐of‐the‐art atmospherically resistant VI, which is a standard Moderate Resolution Imaging Spectroradiometer (MODIS) product, together with the NDVI, is the Enhanced Vegetation Index (EVI). A different approach introduced the Aerosol‐free Vegetation Index (AFRI) that is based on the correlation between the shortwave infrared (SWIR) and the visible red bands. The AFRI main advantage is in penetrating an opaque atmosphere influenced by biomass burning smoke, without the need for explicit correction for the aerosol effect. The objective of this research was to compare the performance of these three VIs under smoke conditions. The AFRI was applied to the 2.1 µm SWIR channel of the MODIS sensor onboard the Earth Observing System (EOS) Terra and Aqua satellites in order to assess its functionality on these imaging platforms. The AFRI performance was compared with those of NDVI and EVI. All VIs were calculated on images with and without present smoke, using the surface‐reflectance MODIS product, for three case studies of fires in Arizona, California, and Zambia. The MODIS Fire Product was embedded on the images in order to identify the exact location of the active fires. Although good correlations were observed between all VIs in the absence of smoke (in the Arizona case R 2 = 0.86, 0.77, 0.88 for the NDVI–EVI, AFRI–EVI, and AFRI–NDVI, respectively) under smoke conditions a high correlation was maintained between the NDVI and the EVI, while low correlations were found for the AFRI–EVI and AFRI–NDVI (0.21 and 0.16, for the Arizona case, respectively). A time series of MODIS images recorded over Zambia during the summer of 2000 was tested and showed high NDVI fluctuations during the study period due to oscillations in aerosol optical thickness values despite application of aerosol corrections on the images. In contrast, the AFRI showed smoother variations and managed to better assess the vegetation condition. It is concluded that, beneath the biomass burning smoke, the AFRI is more effective than the EVI in observing the vegetation conditions.

As a major disturbance to forest ecosystems, wildfires pose a serious threat to the ecological environment. Monitoring post-fire vegetation recovery is critical to quantifying the effects of wildfire on ecosystems and conducting forest resource management. Most previous studies have analyzed short-term (less than five years) post-fire recovery and limited the driving factors to temperature and precipitation. The lack of long-term and multi-faceted observational analyses has limited our understanding of the long-term effects of fire on vegetation recovery. This study utilized multi-source remote sensing data for a long time series analysis of post-fire vegetation recovery in China based on Google Earth Engine (GEE) cloud computing platform. Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Normalized Burn Ratio (NBR), and Normalized Difference Moisture Index (NDMI) were selected to quantify the low, moderate, and high severity of burned areas. Ridge Regression Model (RRM) was used to analyze the relationship between 15 driving factors and the vegetation regeneration process. The results show that it took at least 7–10 years for the vegetation index to recover to the pre-fire level after a forest fire. The recovery rate of high severity combustion areas was the fastest within the first two years. From the results of Ridge Regression, it came out that the overall fitting degree of the model with NDVI as the dependent variable was superior than that with EVI. The four variables of temperature, precipitation, soil temperature, and soil moisture were able to explain the change in more detail in vegetation indices. Our study enriches the research cases of global forest fires and vegetation recovery, provides a scientific basis for the sustainable development of forest ecosystems in China, and provides insight into environmental issues and resource management.

Spatio-temporal change of winter wheat yield and its quantitative responses to compound frost-dry events – An example of the Huang-Huai-Hai Plain of China from 2001 to 2020

<strong class="journal-contentHeaderColor">Abstract.</strong> The AnisoVeg product consists of monthly 1 km composites of anisotropy (ANI) and nadir-normalized (NAD) surface reflectance layers obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor over the entire South American continent. The satellite data were preprocessed using the multi-angle implementation atmospheric correction (MAIAC). The AnisoVeg product spans 22 years of observations (2000 to 2021) and includes the reflectance of MODIS bands 1 to 8 and two vegetation indices (VIs), namely the normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI). While the NAD layers reduce the data variability added by bidirectional effects on the reflectance and VI time series, the unique ANI layers allow the use of this multi-angular data variability as a source of information for vegetation studies. The AnisoVeg product has been generated using daily MODIS MAIAC data from both Terra and Aqua satellites, normalized for a fixed solar zenith angle (SZA <span class="inline-formula">=</span> 45<span class="inline-formula">^∘</span>), modeled for three sensor view directions (nadir, forward, and backward scattering), and aggregated to monthly composites. The anisotropy was calculated by the subtraction of modeled backward and forward scattering surface reflectance. The release of the ANI data for open usage is novel, and the NAD data are at an advanced processing level. We demonstrate the use of such data for vegetation studies using three types of forests in the eastern Amazon with distinct gradients of vegetation structure and aboveground biomass (AGB). The gradient of AGB was positively associated with ANI, while NAD values were related to different canopy structural characteristics. This was further illustrated by the strong and significant relationship between EVI<span class="inline-formula">_ANI</span> and forest height observations from the Global Ecosystem Dynamics Investigation (GEDI) lidar sensor considering a simple linear model (<span class="inline-formula"><i>R</i>²=0.55</span>). Overall, the time series of the AnisoVeg product (NAD and ANI) provide distinct information for various applications aiming at understanding vegetation structure, dynamics, and disturbance patterns. All data, processing codes, and results are made publicly available to enable research and the extension of AnisoVeg products for other regions outside of South America. The code can be found at <a href="https://doi.org/10.5281/zenodo.6561351">https://doi.org/10.5281/zenodo.6561351</a> (Dalagnol and Wagner, 2022), EVI<span class="inline-formula">_ANI</span> and EVI<span class="inline-formula">_NAD</span> can be found as assets in the Google Earth Engine (GEE; described in the data availability section), and the full dataset is available from the open repository <a href="https://doi.org/10.5281/zenodo.3878879">https://doi.org/10.5281/zenodo.3878879</a> (Dalagnol et al., 2022).