Spain - Barrax / Albacete

Specific Project Objectives & Deliverables

Results

  • Ground measurements        

Figure 6 shows the distribution of vegetation types sampled during the field campaigns (data collected over 92 plots on 29th March and 146 plots on 12th July). Also shows that the around 50% of the identified crops correspond to barley fields and bare soil, in March, and bare soil and corn fields in the case of July. The other 50% correspond to a variety of crops, including alfalfa, corn, tree plantation (fruits), onion, sunflower, chickpeas, pappaver, garlic.


Figure 6. Variety and distribution of the identified crops for both campaigns, 29th March and 12th July, 2016.

 

Because of the intense agricultural activity, frequencies of the observations have a great dissimilarity, including the variance due short-term and long-term crops, harvested areas and type of crops. As an example, figure 7 shows LAI frequency distribution for both campaigns.  


Figure 7. LAI frequency distribution in March and July, 2016 field campaigns.

 

  • Ground based maps of biophysical variables:

    For the very first time, Sentinel-2A imagery was used to perform high resolution ground‐based maps of the biophysical variables over the site, and due the innovative of the process, as a first approach it was performed according to the CEOS LPV recommendations for validation.

    Sampling was evaluated based on the convex‐hull analysis performed and a quality flag image was generated. In figure 9, clear and dark blue correspond to the pixels belonging to the ‘strict’ and ‘large’ convex hulls (Martinez et al., 20095). Red corresponds to the pixels for which the transfer function is extrapolated. These maps show quite good quality, 64% on 12th March at 20x20 km2 and 69% on 23rd July, 2016.


Figure 9. Quality Flag (QF) maps based on the convex-hull test over 20 x 20 km2 areas in Barrax, Spain.

 

Transfer functions have been derived by multiple robust regressions between ESU reflectance and the biophysical variables (Martinez et al., 20094). Because the scene presents many senescent and harvested fields, we have selected the NDVI as input for the transfer function (an exponential relationship with LAIeff and LAI, and a linear relationship with FAPAR and FCOVER). NDVI assures good consistency of the maps over the whole area. The biophysical variable maps are available in geographic (UTM 30 North projection WGS‐84) coordinates at 10 m resolution. Figure 10 shows the ground-based LAI maps derived from Sentinel-2A imagery in March and July after transfer function was applied. Figure 10 also shows the difference between LAI values over both seasons, where the scene of July (Fig. 10 right), corresponding to the summer season, showed the lowest LAI values.


Figure 10. Ground-based maps (20x20 km2) retieved at the Barrax site (Sentinel-2 images)

 

The Root Mean Square Errors (RMSE) values for the several transfer function estimated are showed in Table 1.  Results showed acceptable values of the Root Mean Square Error (RMSE) for all the variables undergoing study. Results also showed lower error values in July, maybe because the scene presented many senescent and harvested fields than March scene, i.e. values of the biophysical variables close to 0.


Table 1. Root Mean Square Errors (RMSE) obtained for the LAIeff, LAI, FAPAR and FCOVER in both campaigns for Sentinel-2A images.

Methodology applied to Sentinel-2 was demonstrated as reliable. As NDVI was applied by transfer functions in an exponential and a linear mode for LAI and FAPAR, respectively, final data in biophysical maps is related as in figure 11.


Figure 11. Scatter plots LAI vs FAPAR for the two campaigns at the Barrax Site on 12th March, 2016 (left) and 23rd July, 2016 (right)

 

  • Inter-comparison with Landsat-8 imagery

    The main novelty during these campaigns was to adapt our methodology for Landsat-8 to Sentinel-2 imagery. Here, we compare the results of the ground based maps derived from both sensors (Landsat-8 and Sentinel-2).  Good correlation between NDVI values derived from both sensors Landsat-8 and Sentinel-2 are found (Fig. 12). Despite the correlations shows acceptable results (R2=0.95), it can be observed a slight higher values in the results obtained by Sentinel-2 for high values of the NDVI (NDVI> 0.6) in July. In contrast to the results obtained for low NDVI values, where Sentinel-2 showed lower values as compared to Landsat 8 results. This can be partly explained due to the differences in the spectral ranges of the bands, and partly due to the differences in the acquisition of the images (5 days). 


Figure 12. Correlation between the Sentinel-2 NDVI and Landsat-8 NDVI in the ESUs (12th March, 2016) (right), and random points over the study area (23rd - 18th July, 2016 for Sentinel-2A and Landsat 8, respectively) (left)

 

For the same date, and applying the same methodology to Landat 8 imagery, the visual analysis showed equivalent results, as shown in the figure 13, which confirms a good interoperability of both sensors for mapping biophysical parameters and monitoring its evolution from empirical approaches.  


Figure 13. Ground-based maps (20x20 km2) retrieved at the Barrax Site with Landsat-8 and Sentinel-2 images.

A statistical analysis was performed over the central area of the image (Table 2), focused of the experimental site (5x5Km). Results showed a slight overestimation of Sentinel-2 for each of the variables under study in July 2016, mainly in the estimation of LAI, where the difference between both results is close to 0.6.

Table 2: NDVI, LAI, FAPAR values over a 5x5 km site Las Tiesas(Barrax), 12th March, 2016


Table 3: NDVI, LAI, FAPAR values over 5x5 km site Las Tiesas(Barrax), 18th - 23rd July, 2016

Additionally, 28 random points representing all the NDVI range were compared around the whole image, for both analysis based on transfer functions with Sentinel-2A and Landsat-8 imagery, as shown in figures 14-15 for LAI and FAPAR for March and July images, respectively. Results showed similar FAPAR values for both sensors (Figs. 14-15-b), unlike the results obtained for LAI, where Landsat-8 showed higher values for high LAI values, especially in March, 2016 (figure 14-a). 


Figure 14. Ground-based a) LAI maps and b) FAPAR maps from Sentinel-2A and Landsat-8 versus NDVI map from Sentinel-2 at Barrax site, Spain (12th March, 2016).

 


Figure 15. Ground-based a) LAI maps and b) FAPAR maps from Sentinel-2A (23th July, 2016) and Landsat-8 (18th July, 2016) versus NDVI map from Sentinel-2 at Barrax site, Spain

 

We have accomplished most of our initial objectives, including the performing of ground acquisition and up-scaling of biophysical measurements. We are currently working on analysing methodologies for mapping biophysical variables from physically based and empirically based methods, per crop type and general.

Our guidelines for collecting LAI and FAPAR, as well as for upscaling and producing ground-based maps can be called “best practices”. It has been applied in the FP7 ImagineS to 50 field campaigns. As well to several ESA campaigns (e.g. SEN3EXP, VALSE2), and previously to many of the VALERI sites, showing good performances. It is also included as best practices in the CEOS LPV protocol for global validation of LAI products. 

©2015 Joint Experiment for Crop Assessment and Monitoring
© HER MAJESTY THE QUEEN IN RIGHT OF CANADA SA MAJESTE LA REINE DU CHEF DU CANADA (2015)