|TSM/TDM Science Team Meeting 2016|
|Science Team Meetings|
|How to Submit a Proposal|
|COFUR Price List (scientific use)|
|TanDEM-X Science Service System|
|Title||CROP-X BIOPAR: Multitemporal monitoring and modeling of interactionsbetween crop biophysical parameters, crop phenology, and X-band SARbackscattering|
|Investigator||Villa, Paolo - CNR - IREA, Institute for Electromagnetic Sensing of the Environment|
|Summary||Timely provision of information about agricultural crops (typology, phenology, productivity, health etc.) is crucial for a proper agronomic planning and management, especially for end users (farmers and public administrations). EO data, and in particular high resolution SAR data can provide added-value information about different crop features and their health status early during growth season. CROP-X BIOPAR project proposed here aims at: 1) fostering the physical knowledge of biophysical parameters of vegetation and soil and their influence on the backscattering response in the X-band range on agricultural targets, across the phenological cycle, and 2) implementing such knowledge into electromagnetic backscattering models, to better suit operational crop monitoring plans based on X-band satellite data during the crop growth phase, focusing on the Lomellina study area, in southern Lombardy region, Italy.
The main objectives of the proposed project are: i) to investigate the sensitivity of TerraSAR-X multitemporal data to soil condition and moisture, plant biomass and water content (a parameter strongly linked to plant biomass) and plant morphology throughout the phenological cycle of different crops (paddy rice varieties and corn varieties, mainly), ii) to exploit gathered knowledge about the interaction between microwave backscattering and crop biophysical parameters for accurately mapping crops in Lombardy region (northern Italy) during early growth stages (May-June), by using multi-temporal, inter-annual series of TerraSAR-X data, and iii) to explore TerraSAR-X capabilities in providing information not only on crop types but also on their agricultural practices, phenological stages, water stress status.
To fulfill these goals a dataset of TerraSAR-X new acquisition data are needed, both Stripmap Mode scenes in dual polarization acquired in different dates corresponding to different phenological stages of study areas crops for the current agronomic season (from April to September 2014), and multitemporal ScanSAR Mode scenes, preferably taken in the temporal interval occurring between the acquisition of the Stripmap scenes.
In situ measurement campaigns of crop biophysical parameters will be conducted simultaneously with the TerraSAR-X Stripmap scenes to be acquired over test fields cultivated with paddy rice and corn, in order to allow a more accurate correlation between the remotely sensed data and those measured on the ground. During each campaign in situ data will be collected on the characteristics of soil and vegetation characterizing each of the fields surveyed.
TerraSAR-X data and crop biophysical parameters, collected in quasi-controlled conditions will be used as dataset for testing X-band SAR crop backscattering models, from simple semi-empirical to more sophisticated ones, for assessing backscattering response of different species, crop conditions (plant morphology, water content, density), crop health status, and crop phenological stage. This information will be used for calibrating the models for simulating crop biophysical parameters starting from backscattering response derived from TerraSAR-X (on both Stripmap and ScanSAR scenes) throughout the growth season, working at filed scale through an object based approach. Finally, biophysical parameters mapping products derived from backscattering models applied to multitemporal TerraSAR-X scenes will be validated and their reliability assessed. One intermediate and one final report will be provided at 6 months and at the end of the project, in 12 months.
The work will be framed in context of two running projects: “Space4Agri” (Lombardy Region-CNR project) and “ERMES” (FP7 Collaborative Project), which will provide funding for both data and field campaign costs.
|Final Report||LAN2412 proposal (titled CROP-X BIOPAR) activities have been in line with project scheduling, as described in our proposal. A shifting of project starting date has occurred due to the approval being confirmed on May 9th, 2014. Therefore, the project scheduling has been shifted one month ahead from May 2014 (PM1) to April 2015 (PM12). During the first part of the project (WP1, PM1-6), already described in the intermediate report, our efforts have been focused on gathering in situ crop biophysical parameters and acquiring and processing TSX multitemporal dataset covering 2014 summer crops (rice, maize, soybean) for a study area selected in south-western Lombardy, northern Italy. In this context, specific in situ campaigns were carried out over a total of 10 crop fields (4 for rice, 4 for maize, 2 for soybean) located in Rosasco municipality (45°15’ N; 8°33’ E), show in Figure 1. A summary of crop characteristics in terms of crop type, crop variety and seeding date was given in the intermediate report and is not repeated here. Observations and crop measures were taken in 5 dates simultaneously to TSX Stripmap dual-pol HH/HV scenes acquisitions, on: May 15th, June 17th, July 9th, August 11th and September 24th., for a wide range of crop agronomic and biophysical parameters, both static and dynamic: i) Agronomic parameters (crop type, practices, seeding date, plant density); ii) Substrate parameters (rugosity, moisture, flooding conditions); iii) Crop phenology parameters (phenological stage in BBCH scale); iv) Plant morphology parameters (plant height, number of leaves, leaves size); v) Biomass/density parameters (biomass, plant water content, PAI). As for TSX data, a total of 5 Stripmap Mode dual-pol HH/HV data were acquired simultaneously to in situ campaign. In addition, 4 ScanSAR Mode HH data were acquired for both enhancing the temporal resolution of the dataset and the try to upscale the future results over a larger area of Lombardy region. TSX Stripmap dual-pol scenes were calibrated to derive Sigma0 backscattering values and stacked together in time series of HH and HV backscatter. TSX scenes were pre-processed using ESA-NEST software. Three different multilooking rates have been applied; and the best balance between spatial resolution and data filtering has been retained, consisting in: multilooking 7R X 4A, geocoding, radiometric calibration, resampling in order to produce 10meters square pixel dimension maps, followed by radiometric normalization using SRTM DEM. Figure 2 included into the attached file shows as RGB combinations the multitemporal backscatter response derived from TSX data. During the second part of the project (WP2-3, PM7-12), our efforts have been focused on the analysis of multitemporal backscattering in relation to biophysical parameters acquired in situ and the modeling tests of X-band SAR response to variable crop conditions. In situ collected data showed that crop biomass increases with time after seeding date (DAS: Day After Seeding) at different average rates: from ~15g/day (rice-soybean, from DAS 30), to ~35g/day (maize, from DAS 45). Plant Water Content (PWC), instead increases up to 3-5 times the biomass values until DAS 100, and then decreases during crop mature yellowing (and drying) phase. Plant Area Index (PAI) steadily increases until DAS 100 (120 for maize), peaking at 4-4.5 for rice, 4-6 for soybean and 5-6 for maize, and then decreases due to loss of water and plants drying up. Plant height again reaches its maximum after DAS 100-120: ~70 cm for rice (90 cm for field 338, planted with a tall cultivar), ~80 cm for soybean and in the range 300-370 cm for maize, depending on the cultivar. The analysis of sensitivity of measured backscatter (HH, HV sigma0, HV/HH ratio) is shown in Figures 3-6 with respect to different biophysical parameters (biomass, PWC, PAI, plant height). A quite typical behavior of narrow leaved crops is observed for rice, with backscatter decreasing with increasing biomass (Figure 3) and PWC (Figure 4). The sensitivity to biomass is higher for HH polarization (-4 to -12 dB), and lower for HV. Accuracy too is higher for HH than for HV. At biomass values higher than 0.3 kg/m2 a good sensitivity is shown by HV/HH ratio too (-11 to -4). For PWC, the average sensitivity is again higher for HH (-4 to -11 dB), than for HV, but accuracy is higher for HV polarization. Rice shows different behaviors in HH and HV polarizations when compared to PAI seasonal dynamics (Figure 5): at the first stage of growth, the backscatter grows (due to the interaction between the plants and the water surface over the flooded terrain, i.e. double bounce effect), then the attenuation of the canopy becomes predominant; this happens around PAI=1.5-2.0 in HH pol, and earlier in HV (PAI<1.0). The behavior observed with PAI is confirmed and better explained by plant height (Figure 6), which shows that attenuation takes prevalence over double bounce effect when plants are ~40 cm in HH pol, and for shorter plants in HV pol (<20cm). The two distinct behaviors are well separated in terms of HV/HH ratio, for which the rice height under ~50 cm show decreasing trend, while plants taller than 60 cm show a steep increment in HV/HH. For maize, a general medium-leaved behavior is observed, with backscatter increasing at the early growth stage of biomass (<1 km/m2), followed by a saturation around -8 dB in HH and -15 dB in HV for higher biomass (Figure 3), and similar trends are observed for PWC (Figure 4). The early stage rapid increment followed by saturation and loss of sensitivity is confirmed by PAI dynamics (Figure 5), where the backscatter plateau is reached for PAI>2.0, where the attenuation of the canopy balances with scattering from the leaves. In terms of height observed in out 2014 dataset (Figure 6), this situation correspond to plants taller than 200 cm, but we must note that the observed data feature a gap between 50 cm and 200 cm, due to the rapid growth of maize between May 15th and June 17th, the dates of the first two 2014 field campaigns. No particular pattern is evident for soybean, also taking into account the small number of samples collected for this crop (2 fields only, 28 and 286) and the particular conditions of field 286, which underwent a series of massive herbicide treatments and was therefore highly patchy and internally heterogeneous in terms of vegetated cover throughout the 2014 summer season. Further tests about the sensitivity of sigma0 backscatter to plant biophysical parameters, and in particular morphological features were run using a single layer radiative transfer model for microwave range developed by Paloscia et al. (2014; DOI: 10.1109/JSTARS.2014.2345475) and already used on narrow-leaved (wheat) and broad-leaved (sunflower). Here the model was tested for assessing sensitivity to plant morphology dynamics (height, stem diameter, leaf length, leaf width, average leaf angle distribution), and for comparing simulated backscatter with the experimental sigma0 measured from TSX data for both rice (narrow-leaved) and maize (medium-leaved). Figure 7 shows the results of the model runs exploring the sensitivity of HH and HV backscatter (Figure 7a), as well as the contributions of different components of the plant-soil system to total backscatter in HH polarization (Figure 7b) for crop features typical of maize. Standard conditions were assumed for soil backscatter, with roughness set to 1.5 cm, and volumetric soil moisture set to 20%, while vegetation parameters like the average leaf angle was set in the range 30-60°, and a number of 6 leaves/plant was used. Simulations revealed a general tendency towards attenuation of soil backscatter by maize plants when height is lower than 1.5 m, followed by a dominance of scattering from leaves when the plant grows. The contribution to total backscatter by maize stems is instead minor. Similarly, Figure 8 shows the results of the model runs for crop features typical of rice. Flooded conditions of paddy fields were assumed for soil backscatter , average leaf angle was set in the range 50-80°, and a fixed number of 3.3 leaves/plant was used. Figure 8a shows a good agreement with the backscatter dynamics experimentally observed with plant height (Figure 6), with an increment in HH backscatter up to 45cm followed by a decrease when the plants are taller, while the peak in backscatter is reached for lower plant height values (25cm) in HV polarization. The model seems able to simulate the complex interactions between the rice plants and flooded terrain background (including double bounce effect), even if the magnitude of the simulated backscatter (peaking at ~-8 dB) is lower than what experimentally measured in our test fields (peaking at ~-4 dB). Figure 8b shows that the predominant effect on total backscattering is due to interaction between vegetation and soil (in case of rice, in wet conditions and even covered by water), while plant parts are not playing a relevant role in directly scattering incident energy at X-band HH pol. In situ measured parameters of maize and rice plants in monitored fields throughout the 2014 season were finally used for input to the model for simulating the multitemporal evolution of HH and HV sigma0 and comparing it to the backscatter derived from TSX scenes. Figure 9 shows the results of simulation for maize fields (id: 4, 5, 26, 29), highlighting the good matching between measured and simulated sigma0 in HH (Figure 9a) in 4 out of 5 dates, with the exception of September 24th which features plants already in senescence and dry conditions; for HV pol instead (Figure 9b), the performance is not good starting with the June 17th data onwards, with serious underestimation of sigma0 HV when maize plants are higher than 0.5-2.0 m. Figure 10 shows the simulation result for rice fields (id. 118, 283, 338, 373), highlighting a certain match up to August 9th date in HH polarization (Figure 10a), with the exception of data showing very high leaf area density (>6.5 m2/m2), while in HV pol a tendency to underestimation is present, with sigma0 simulated always lower than measured values (Figure 10b). Figure 11 summarizes the simulation results by displaying the scatter plot of simulated versus measured sigma0 in HH polarization for maize and rice fields monitored in non-drying conditions (i.e. PWC<60%, excluding data acquired on September 24th, 2014). For maize (Figure 11a) a good match (R2=0.56) is observed with low RMSE (0.94 dB). For rice instead (Figure 11b), when we consider the 3 points corresponding to very high LAI (>6.5) as outliers, a fair agreement is still observed, even if with an underestimation bias around 3 dB (RMSE=3.34 dB). The results described are judged encouraging in the perspective of X-band SAR high resolution data use for estimation of biophysical crops parameters for rice and maize, up to peak of season (flowering, early stage of maturation) conditions, before the drying of plant starts. In particular, HH and HV backscatter have demonstrated sensitivity to plant biomass and the capability to follow phenological and growth dynamics especially for rice fields and could be therefore used for estimating such parameters. The test of modeling radiative transfer from rice and maize crops has been successful in capturing rice growth dynamics (including the double bounce effect, to some extent) and especially effective in simulating maize backscatter behavior during the 2014 season. Further work is needed to better assess the contribution of substrate (soil and flooded terrain) the total backscatter and therefore refine the model simulation efficiency, and to investigate the behavior of maize during the early growth stage (plant height 0.5-2.0 m), which was not covered enough in the dataset collected in 2014 and will be object of currently ongoing LAN 2934 project.|
Back to list of proposals
|© DLR 2004-2016||