| Summary: | Accurate and timely knowledge of crop phenology assists in planning and/or triggering appropriate farming activities. The multiple Polarimetric Synthetic Aperture Radar (PolSAR) technique shows great potential in crop phenology retrieval for its characterizations, such as short revisit time, all-weather monitoring and sensitivity to vegetation structure. This study aims to explore the potential of averaged Stokes-related parameters derived from multiple PolSAR data in oilseed rape phenology identification. In this study, the averaged Stokes-related parameters were first computed by two different wave polarimetric states. Then, the two groups of averaged Stokes-related parameters were generated and applied for analyzing averaged Stokes-related parameter sensitivity to oilseed rape phenology changes. At last, decision tree (DT) algorithms trained using 60% of the data were used for oilseed rape phenological stage classification. Four Stokes parameters (<inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>0</mn></msub></mrow></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>1</mn></msub></mrow></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>2</mn></msub></mrow></semantics></math></inline-formula> and <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>3</mn></msub></mrow></semantics></math></inline-formula>) and eight sub parameters (degree of polarization <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>m</mi></semantics></math></inline-formula>, entropy <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>H</mi></semantics></math></inline-formula>, ellipticity angle <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>χ</mi></semantics></math></inline-formula>, orientation angle <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>φ</mi></semantics></math></inline-formula>, degree of linear polarization <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>D</mi><mi>o</mi><mi>l</mi><mi>p</mi></mrow></semantics></math></inline-formula>, degree of circular polarization <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>D</mi><mi>o</mi><mi>c</mi><mi>p</mi></mrow></semantics></math></inline-formula>, linear polarization ratio <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>L</mi><mi>p</mi><mi>r</mi></mrow></semantics></math></inline-formula> and circular polarization ratio <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>C</mi><mi>p</mi><mi>r</mi></mrow></semantics></math></inline-formula>) were extracted from a multi-temporal RADARSAT-2 dataset acquired during the whole oilseed rape growth cycle in 2013. Their sensitivities to oilseed rape phenology were analyzed versus five main rape phenology stages. In two groups (two different wave polarimetric states) of this study, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>0</mn></msub></mrow></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>1</mn></msub></mrow></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>2</mn></msub></mrow></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msub><mi>g</mi><mn>3</mn></msub></mrow></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>m</mi></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>H</mi></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>D</mi><mi>o</mi><mi>l</mi><mi>p</mi></mrow></semantics></math></inline-formula> and <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>L</mi><mi>p</mi><mi>r</mi></mrow></semantics></math></inline-formula> showed high sensitivity to oilseed rape growth stages while <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>χ</mi></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mi>φ</mi></semantics></math></inline-formula>, <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>D</mi><mi>o</mi><mi>c</mi><mi>p</mi></mrow></semantics></math></inline-formula> and <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>C</mi><mi>p</mi><mi>r</mi></mrow></semantics></math></inline-formula> showed good performance for phenology classification in previous studies, which were quite noisy during the whole oilseed rape growth circle and showed unobvious sensitivity to the crop’s phenology change. The DT algorithms performed well in oilseed rape phenological stage identification. The results were verified at the parcel level with left 40% of the point dataset. Five phenology intervals of oilseed rape were identified with no more than three parameters by simple but robust decision tree algorithm groups. The identified phenology stages agree well with the ground measurements; the overall identification accuracies were 71.18% and 79.71%, respectively. For each growth stage, the best performance occurred at stage S1 with the accuracy of 95.65% for Group 1 and 94.23% for Group 2, and the worst performance occurred at stage S3 and S5 with the values around 60%. Most of the classification errors may resulted from the indistinguishability of S3 and S5 using Stokes-related parameters.
|
|---|
