Potential of optical data (Sentinel-2) in classifying winter wheat crop

Get Complete Project Material File(s) Now! »

Study site

The Bekaa plain of Lebanon is located between 33°33’ N and 33°60’ N latitude, 35°39’ E and 36°14’ E longitude (Figure 1). The average area of the plain is around 860.25 Km² with an average elevation of 1000 above sea level. Similar to other plains in the southern Mediterranean (e.g. Plains of Sétif in Algeria (Hafsi et al., 2000), the Saïs plain in Morocco and the Medjerda plain of Tunisia (Burgers and Zoomers, 2014)), the Bekaa is characterized by a semi-arid Mediterranean climate and the average annual precipitation is around 600 mm. Agriculture is the main economic scheme at the area with the cultivation of field crops, orchards, annual and perennial plants.
Field crops areas (e.g. cereals, vegetables and legumes) of individual farmers range from 0.1 ha to up to 20 ha. 65% of the national cereal production is being produced in the Bekaa plain, while winter wheat areas in the plain correspond to 44% of the national wheat area, occupying zones ranging from 9000 to 12000 ha annually. 51% of potato, which is one of the largest tuberous crops cultivated in Lebanon, is present in the Bekaa plain as one of the most important cash crops. As for legumes, Bekaa is responsible for 20% of the national cultivation area, 16% of this area corresponds to Fava beans, occupying around 1548 ha in the plain (MoA, 2010). Wheat and fava beans are winter crops, as they are sown in November, while the potato is sown in March. As a whole, regarding irrigation management, 72% of the Bekaa cultivations are fully or supplementary irrigated. Even though Fava beans and winter wheat are grown through the winter season, however, they do receive supplementary irrigation during early spring to ensure better yield (80% of wheat in the Bekaa plain is supplementary irrigated). While potato, on the other hand, is fully irrigated (on weekly basis) from sowing to harvesting, ranging from 10 to100 mm per application, depending on the phenological phase (MoA, 2010).
Fertilization is supplied, especially nitrogen, as one of the most growth driving nutrient. Fertilization management differ among farmers, however, nitrogen is being supplied in its organic and inorganic forms. For wheat, farmers supply nitrogen of amounts up to 230 Kg/ha in the form of ammonium sulfates in February. As for potato, nitrogen is applied before planting in the form of manure (around 250 Kg N/ha), in addition to a second application of synthetic nitrogen (around 100 Kg N/ha) after flowering occurs. Fava bean receives around 50 Kg N/ha in a form of synthetic nitrogen 60 days after sowing.
Crop rotations do exist in the Bekaa plain. One of the most followed rotation type is wheat-potato rotation as it is one of the most profitable rotations. However, poor farmers do cultivate wheat in a monoculture approach to benefit from the governmental support in buying their yield with relatively good prices. 23% of wheat cultivated lands in 2016 were also wheat cultivated in 2017 (Nasrallah et al., 2018). However, among suitable lands for agriculture, around 1500 ha are left as fallow annually (around 4% of the total exploited area).Figure 2 illustrates the crop calendar of the main field crops grown in the plain.
Soils of the experimental site were both developed and deep. The available soil groups of were basically noncalcareous, clay Cambisols and Fluvisols with an average bulk density of 1.3 g/cm³. Gravimetric moisture content varied between 45.3% at field capacity and 34.8% at wilting point. Thus the water holding capacity varied between 140 mm/m and 223 mm/m (Darwish et al., 2006).

In-situ database

The in-situ measurements were done through field campaigns which took place over the period extended from 2016 to 2018. The in-situ database consists of recording of Global Positioning System (GPS) of various plots corresponding to diverse types of cultivations (2016, 2017 and 2018 seasons) in addition to in-situ measurements in 2018 cropping season, consisting of soil parameters, vegetation parameters, as well as field survey, conducted by distributing questionnaires among farmers, addressing various issues related to their practices.
The directly measured soil parameter corresponds to soil water content (Ws). The vegetation parameters correspond to above ground biomass (AGB), canopy water content (CWC), leaf area index (LAI), above ground nitrogen (AGN) and canopy height (CH). These soil and vegetation parameters were measured throughout the 2018 cropping season. More details about the in-situ database regarding the experimental design, replications, dates of measurements and values, are illustrated in Appendix A, as well as found in chapters three, four and five.

Recording of global positioning system (GPS)

As the most critical phenological phases of winter wheat lie between February and June, several field visits to record the coordinate of the reference plots, were conducted in this period throughout 2016 and 2017 seasons to serve winter wheat classification in these two seasons. In addition to winter wheat, other plots corresponding to cereals (i.e. barley and triticale) and different cultivations (i.e. potato, orchards, vineyards, alfalfa, bare soil) were also visited and their coordinates were recorded.

Soil parameters

Soil moisture

The soil moisture was measured using a tube auger following the gravimetric water content method over 18 selected reference plots located in North, Mid and West Bekaa plain of Lebanon. In each plot, the soil moisture was measured for three pedological horizons (depth of each horizons ranges from 30 to 55 cm depending on the plot). The soil moisture measurement of each horizon is replicated three times and averaged within each reference plot. For each horizon (the depth of horizon Ap and B varied among plots), a sample of soil was taken out, sealed in a bag and transported in refrigerator to the lab to be measured fresh, and then put in the oven to dry at 100 °C until constant weight. For each winter wheat reference plot, the measurement was replicated three times randomly and then averaged at each depth on five phenological phases (germination, tillering, stem elongation, heading and maturation) (Reynolds, 1970). Figure 3 shows the way soil was sampled in the field.
The soil moisture content Ws (% or g-1 / g-1) is calculated using the wet weight (Sf , soil weight after sampling) and the dry weight (Sd). The dry weight (Sd) is obtained by drying the soil sample taken at a temperature of 100° C for 24-48 hours until constant weight: % =100×[ − ] Eq. 1

Vegetation parameters

The vegetation parameters measured consist of Above Ground Biomass (AGB), Canopy Water Content (CWC), Leaf Area Index (LAI), Above Ground Nitrogen (AGN) and Canopy Height (CH). These parameters were measured on four dates corresponding to four phenological phases (tillering, stem elongation, heading and maturation) of winter wheat.
Each measurement is replicated three times and averaged within each reference winter wheat plot (18 plots). Each plot is selected based on the type of cultivar, previous crop, crop management and soil water holding capacity.

Above Ground Biomass (AGB)

Above ground biomass (AGB) was measured by a destructive method in four dates corresponding to four phenological phases of wheat (tillering, stem elongation, heading and maturation). Within each winter wheat reference plot, out of the total 18 selected reference plots, the measurement was replicated three times randomly, and then averaged over each plot. After weighing the fresh sample for each replication within each plot, the samples were quartered and a representative sample was oven dried at 70 °C until constant weight (Catchpole and Wheeler, 1992). Figure 4 shows the way the destructive above ground biomass sampling was done.

Canopy Water Content (CWC)

After collecting and weighing above ground biomass samples for each replication within each reference winter wheat plot out of the total 18 selected reference plots, the samples were quartered and a representative sample was oven dried at 70 °C until constant weight (Catchpole and Wheeler, 1992). Figure 5a represents the quartering procedure and figure 5b demonstrates the samples in the oven. Similar to AGB, the measurement was done on four dates (tillering, stem elongation, heading and maturation).

Leaf Area Index (LAI)

The leaf area index (LAI) represents the area of leaves per unit area on the ground (m² / m²). This index can be calculated by two methods, direct and indirect. The direct method is a destructive method of cutting the vegetation, and then measuring the surface of the leaves one by one using a planimeter. This technique is very difficult to achieve given the type of vegetation.
The indirect method is to acquire nadir hemispherical photos (Figure 6) using a Fish Eye lens. Thus, in each of the 18 reference plot, three images were taken for each of the replicates, resulting in 9 images averaged for each reference plot. Then, these photos are processing of hemispherical photos to calculate LAI is based on the measure of fraction of holes in vegetation. In this thesis, the indirect method was considered to measure the LAI because it is more realistic for a type of vegetation such as wheat. LAI normally evolves rapidly over time. LAI values vary between 0.1 and 6 m² / m². For most plots, the LAI reaches 4 m² / m² approximately at flowering phase. The measurement was done on five dates (germination, tillering, stem elongation, heading and maturation).
Figure 6 Hemispherical photo for the estimation of LAI. Photo acquired in the Bekaa plain of Lebanon.


Above Ground Nitrogen (AGN)

Above Ground Nitrogen (AGN) was measured following Kjeldahl‐N method (Rodriguez and Miller, 2000). In each of the 18 reference plot, on both tillering and heading phases (two dates), three biomass samples were taken. Crop N uptake was calculated from the corresponding data of dry matter production and N content.

Canopy Height (CH)

At each sampling date (on the four phenological phases dates), the canopy height of winter wheat was measured in each reference winter wheat plot (out of the 18 reference plots)
(Figure 7). The measurement was replicated three times in each plot and averaged. The canopy height (CH) was determined using a measuring tape.
Figure 7 Canopy height (CH) measurements at the Bekaa plain using a measuring tape.

Survey (questionnaires)

In 2017 winter season, a survey was conducted within the area of interest (i.e. Bekaa plain of Lebanon). Through the distribution of the questionnaires, we targeted farmers, especially those involved in winter wheat cultivation. The profile, including the name, profession (if other than agriculture), telephone number, location, age and gender, was created. Further, the farmer was asked about the crop types they grow, management strategy (e.g. irrigation, fertilization…etc.), effect of climate and inputs on their production’s evolution, economic situation and expenses, and the roles of governmental and non-governmental entities. Figure 8 shows meeting with farmers to ask questions related to the questionnaire filling within the study area.
More than 40 farmers, distributed within the area of interest, were interviewed. The output of the survey was used afterwards to: (1) understand the winter wheat-based cropping systems within the area of interest, including the rotation types and management, (2) calculate the expenses as well as the net profit of each of the systems and (3) evaluate the socio-economic situation of the farmers to better perceive and weigh the outputs.

Remote sensing (satellite) database

Within the framework of this thesis, satellite imageries (optical and radar) were utilized. The images downloaded cover two winter wheat cropping seasons (2016, 2017) for crop classification and one cropping season (2018) for phenological phases mapping. Each winter wheat cropping season extends from November to July, next year. More details about the remote sensing database regarding the number of images, dates of acquired images, calibrations and corrections, are found in chapters three, four and five.

Optical images

The optical data was acquired by Sentinel-2 satellites (2A and 2B). Sentinel-2 are the second generation Earth Observation (EO) satellites operated by the European Space Agency (ESA) (Drusch et al., 2012). The launching of Sentinel-2A and Sentinel-2B was in June 2015 and March 2017 respectively, as an integral part of Europe’s Copernicus program aiming at independent and continued global observation capacities (Immitzer et al., 2016). Sentinel-2 offers a fine spectral, spatial and temporal resolutions (i.e., 13 bands ranging from 10 m to 60 m with a revisit time of five days). Datasets produced by this satellite could be downloaded free of charge from Europe’s Copernicus website (Https://scihub.copernicus.eu).
For winter wheat classification in 2016 and 2017 cropping seasons, eight low cloud cover Sentinel-2 images (covering the main phenological phases) were downloaded for each of the cropping seasons, during the period between January and May, in each season (2016 and 2017). This period is enough as the classification is executed before the end of the cropping cycle (July).
For winter wheat monitoring, in the cropping season of 2018, fifty-eight images were downloaded from November, 2017 through August, 2018. These images were used to calculate the NDVI over the whole cropping season (before the start of the wheat cycle until a period after harvesting), in order to analyze the whole crop dynamic and to compare it with the SAR data.
The images were initially downloaded at L1C (or L2A when available) level (Top of Atmosphere or TOA reflectance). The pre-processing of these Sentinel-2 images (L1C) including ortho-rectification, cloud removal (using cloud mask produced by Sen2Cor/SNAP), radiometric calibration and atmospheric correction, was produced using SNAP/Sentinel-2 toolbox. The output of the pre-processing corresponds to L2A (Bottom of Atmosphere or BOA reflectance).
In this study, as we relied on the Normalized Difference Vegetation Index (NDVI) derived from Sentinel-2 satellite, only bands in the visible and infrared wavelengths were used, benefiting from their fine spatial resolution (i.e. 10 m).

Table of contents :

CHAPTER 1: Résumé analytique
1. Contexte général
2. Etat de l’art
2.1. Utilisation des satellites de télédétection pour la classification des cultures
2.2. Potentiel des données RSO dans la surveillance du cycle de croissance des cultures
2.3. Capacité à modéliser des cultures pour une gestion efficace
3. Problématique
4. Démarche
4.1. Approche générale
4.2. Potentiel des données optiques (Sentinel-2) à répertorier les cultures de blé d’hiver
4.3. Potentiel des données RSO à identifier les principales phases phénologiques du blé et à prévoir leurs dates
4.4. Modélisation des cultures pour évaluer la performance des systèmes de culture du blé et réduire de façon durable le risque économique dans la plaine de la Bekaa au Liban
5. Conclusion générale
CHAPTER 2: Study site and database description
1. Study site
2. In-situ database
2.1. Recording of global positioning system (GPS)
2.2. Soil parameters
2.3. Vegetation parameters
2.4. Survey (questionnaires)
3. Remote sensing (satellite) database
3.1. Optical images
3.2. Radar images
CHAPTER 3: Potential of optical data (Sentinel-2) in classifying winter wheat crop
1. Objectives
2. Study site
3. Datasets
4. Methods
4.1. Satellite and ground data
4.2. SEWMA Generation
5. Results
6. Discussion
7. Conclusions, strengths, limitations and future directions
Article one: A Novel Approach for Mapping Wheat Areas Using High Resolution Sentinel-2 Images
1. Introduction
2. Study area
3. Material and methods
3.1. Datasets and preprocessing
3.2. SEWMA Generation
4. Results
4.1. Crops’ temporal profiles
4.2. SEWMA First phase preliminary results
4.3. SEWMA Accuracy assesment
4.4. Wheat spatial distribution
5. Discussion
5.1. Crops’ temporal profiles
5.2. SEWMA First phase preliminary results
5.3. SEWMA Accuracy assessment
5.4. Wheat spatial distribution
5.5. Strengths, limitations and future directions
6. Conclusions
CHAPTER 4: Potential of SAR data in monitoring the winter wheat phenology
1. Objectives
2. Study site
3. State of art
4. Datasets (satellite and ground data)
5. Methods
6. Results
6.1. Optimal configuration
6.2. Accuracy assessment and quantitative analysis
6.3. Towards near real time monitoring
7. Discussion
7.1. S1 polarizations versus NDVI temporal behavior
7.2. Influence of incidence angle
7.3. Mapping outputs and quality indicators
8. Conclusions and future directions
Article two: Sentinel-1 Data for Winter Wheat Phenology Monitoring and Mapping
1. Introduction
2. Material and methods
2.1. Study site
2.2. Remote sensing data
2.3. In Situ observations (reference plots)
2.4. Meteorological data
2.5. Software employed and statistical analysis
2.6. Methodological approach
3. Results
3.2. NDVI Temporal profiles
3.3. Sentinel-1 temporal profiles
3.4. Smoothing and Gaussian fitting
3.5. Germination, heading, soft dough, and harvesting mapping
3.6. Toward near-real time phenology monitoring
4. Discussion
4.2. S1 versus NDVI temporal behavior
4.3. Influence of S1 incidence angle
4.4. Wheat phenology mapping
5. Conclusions
CHAPTER 5: Crop modelling for assessing wheat-based cropping systems’ performance and economic risk
1. Motivations and objectives
2. Study site
3. Methodological approach
4. Results
4.1. Model (CropSyst) performance
4.2. Wheat grain yield as altered by the effects of rotation, management, and soil type
4.3. Nitrogen and water apparent recovery efficiency by difference (ARED)
4.4. Rotations’ performance (productivity and efficiency) and the economic sustainability risk
5. Discussion
6. Conclusions and recommendations
Article three: Performance of wheat-based cropping systems and economic risk of low relative productivity assessment in a sub-dry Mediterranean environment
1. Introduction
2. Methods
2.1. Study site and crop management
2.2. Simulation model
2.3. Developing the scenarios to be simulated by the CropSyst model
2.4. Calculation of the productivity and efficiency indicators for assessing the performance of wheat-based cropping systems
2.5. Calculation of the “economic risk of low relative productivity”
3. Results
3.1. Calibration and validation of the CropSyst model
3.2. Wheat grain yield as altered by the effects of rotation, management system, and soil type
3.3. Nitrogen and water Apparent Recovery Efficiency by Difference (ARED)
3.4. Trends of the crops’ yields (10 rotations) over the simulation period
3.5. Rotation performance (productivity and efficiency) and economic risk of low relative productivity
4. Discussion
5. Conclusions
CHAPTER 6: Conclusions and perspectives
1. General context and main methodological challenges
2. Main results
2.1. Wheat classification. Accuracies and areas
2.2. First experience with SAR in mapping wheat phenology
2.3. Which wheat-based cropping systems to be promoted?
3. Research perspectives
3.1. Methodology related perspectives
3.2. Application related perspectives


Related Posts