Satellite Remote Sensing Retrieval Of Nitrogen And Phosphorous Nutritional Status In Apple Tree Leaves/Canopies At Blossom Stage

Nitrogen and phosphorous are essential nutrients for apple trees, so it is very important todiagnose their content accurately in time for adequate fertilization and efficient management.The physiological features of leaves and flowers would change with the different contents ofnitrogen and phosphorous, and the reflectance changes caused by physiological featuresprovide possibility to obtain nutrient information by remote sensing. In recent years, as one ofthe research hot spot in the field of agriculture, quantitative remote sensing has been a usefulmeans to monitor vegetation biochemistry compositions. However, most researchers paidmore attention to the crops but less to fruit trees, and furthermore most researches about fruittrees were based on hyperspectral data, few satellite imageries were used.Taking Qixia City as the study area and Red Fuji apple trees at prosperous blossom stage asthe study objects, this study started with the reflectance retrieval of apple tree canopy basedon Landsat-5TM and ALOS AVNIR-2images using apple tree hyperspectral reflectancedata, DEM and6S Models with the soft support of Erdas Image,ENVI,ArcGIS,SPSS,Libsvm,Access and so on. Then the spectral indices for nitrogen and phosphorus in appletrees were composed based on canopy retrieval reflectance, and the sensitive indices werechosen as independent variables for three kinds of retrieval models. Furthermore, the bestmodels were used to retrieval nitrogen and phosphorous nutrient status in apple trees locatedin the study area.(1) Apple tree canopy reflectance were retrievedGround surface reflectance was retrieved from TM and ALOS images through radiometriccorrection based on6S models. It showed that radiometric correction effectively weakenedthe effects caused by atmosphere and topography, recovered the ground objects in the shadowof the hills, and obviously enhanced the analytical ability of ground surface retrievalreflectance images.The ratio of canopy flower to leaf was put into pixel unmixing model, and the reflectanceof apple tree canopy was further retrieved by the model. Then the retrieval effects andaccuracy were assessed by the comparison of retrieved reflectance with measured canopyreflectance and apparent reflectance of30sample apple orchards. Errors between canopy retrieval reflectance and measured reflectance of apple tree canopy were very small either onTM or ALOS imagery, the relative errors of all bands was consistent, and its variation trendamong30samples was also consistent with the measured reflectance, which showed thenecessary of pixel unmixing.Moreover, the change of reflectance showed similar characteristics when the retrievalmethods were used for different resolution imageries. Image with higher resolution was a bitsuperior, while it would be better to integrate with other moderate resolution images becauseof band limitation.(2) Sensitive spectral indices for nitrogen and phosphorus were chosenMore than100spectral indices were constructed based on canopy retrieval reflectance andmeasured canopy reflectance. Then correlations were analysized between nitrogen,phosphorous content and retrieval reflectance of30sample orchards, measured reflectance of90sample trees. Those indices with higher correlation coefficients based on all kinds ofreflectance were chosen as the sensitive spectral indices.The results showed that the whole sensibility of sensitive indices either for nitrogen or forphosphorous was: canopy＞leaf＞flower, all sensitive indices for canopies, parts for leaveswere significant at α=0.05level, but none for flowers were significant. And the wholesensibility of sensitive indices for nutrients either in canopies or in leaves was: nitrogen＞phosphorous. The bands composing sensitive indices were maily4,3and2, which wasconsistent with the sensitive bands of vegetation. The sensibility based on ALOS data was alittle better than TM at canopy level, but the situation was the opposite at leaf level.(3) Retrieval models of nitrogen and phosphorous nutrients were constructedTaking the most sensitive spectral indices as independent variables, three methods, one-variable regression, multi-variable stepwise regression and support vector regression, wereused to construct the models of nitrogen and phosphorous nutrients in apple tree leaves andcanopies based on TM ans ALOS canopy retrieval reflectance. The whole accuracy of modelswas: support vector regression＞multi-variable stepwise regression＞one-variable regression,coefficients of determination between measured values and predicted values from supportvector regression models were all above0.799.So support vector regression models were chosen to retrieve nutrient status in the studyarea. Similar accuracy characteristics were showed when the models were used for differentresolution imageries: canopy＞leaf, and nitrogen＞phosphorous.(4) Nitrogen nutrient of apple tree in the study area was retrieved The leaf retrieval nitrogen content and canopy retrieval nitrogen content index wereobtained by spatial retrieval methods using support vector regression models.Retrieval nitrogen content in leaves based on TM and ALOS were mainly Grade3-4. Themean content based on TM was1.231g/kg higer than that based on ALOS. The leaf retrievalnitrogen content comparison of based on TM with based on ALOS showed that84.96%pixelshad similar retrieval content at15%permissible error level,71.32%at10%permissible errorlevel, and48.37%at5%permissible error level.Retrieval nitrogen content index in canopies based on TM was different from that based onALOS because of the difference of the ratio of canopy flower to leaf. But the nutrient statuswas mainly Grade2-4with consideration of mean difference. The mean canopy nutrientcontent index based on TM was still a little higer than that based on ALOS. The canopyretrieval nitrogen content index comparison of based on different imageries showed that92.76%pixels had similar retrieval value at15%permissible error level,78.51%at10%level,47.93%at5%level. The retrieval accuracy of canopies was better than that of leaves.The correlation between predicted nutrient contents and measured ones in flowers of30samples was significant at α=0.05level. The predict accuracy based on TM and ALOS were73.33%and86.67%respectively at5%permissible error level,76.67%and90.00%at10%permissible error level, which showed that the accuracy based on ALOS was a bit better thanthat based on TM.(5) Phosphorous nutrient of apple tree in the study area was retrievedThe leaf retrieval phosphorous content and canopy retrieval phosphorous content indexwere obtained by spatial retrieval methods using support vector regression models.Retrieval phosphorous content in leaves based on TM and ALOS were mainly Grade3-4.The mean content based on TM was0.394g/kg higer than that based on ALOS. The leafretrieval phosphorous content comparison of based on TM with based on ALOS showed that50.80%pixels had similar retrieval content at15%permissible error level,36.11%at10%level, and18.80%at5%level.Retrieval phosphorous content index in canopies based on TM was different from thatbased on ALOS because of the differences of the ratio of canopy flower to leaf. But thephosphorous level was also mainly Grade2-4with consideration of mean difference. Themean canopy nutrient content index based on TM was still a little higer than that based onALOS. The canopy retrieval phosphorous content index comparison of based on twoimageries showed that76.77%pixels had similar retrieval value at15%permissible error level,60.92%at10%level,35.69%at5%level. The retrieval accuracy of canopies was alosbetter than that of leaves, as same as the retrieval nitrogen accuracy.The correlations between predicted phosphorous content and measured values in flowers of30samples were significant at α=0.05level either based on TM or ALOS. The predictaccuracy based on TM and ALOS were both76.67%at5%permissible error level,80.00%and76.67%at10%permissible error level. It still showed that the accuracy based on ALOSwas a bit better than that based on TM, but one or two samples had higer error based onALOS.In total, nitrogen retrieval accuracy was better than phosphorous either for leaves andcanopies. With consideration of mean difference between TM and ALOS, the canopy retrievalaccuracy of either nitrogen or phosphorous was better than leaves. The mean nutrient statusbased on TM was a little higer than that based on ALOS to any kind of nutrient.(6) Spatial distribution of nitrogen and phosphorous nutrients in apple trees wasknownFrom the spatial distribution it could be known that nutrient status was mainly from the2ndto4th grade, especially from the3rd to4th grade, the sum of pixel counts of these threegrades was more than80%. Either leaves or canopies, the high grade pixels of both nitrogenand phosphorous were located at Zangjiazhuang Town and Guanli Town in the south of QixiaCity, Sujiadian Town and Songsha Town in the northwest, Zangjiazhuang Town and TingkouTown in the northeast, which was consistent with the distribution of well-developed towns forapples. The nutrient grades of the analysis area based on ALOS image showed that thenutrient content index was higer in the northeast than that in the southwest, and grades besidethe rivers was also higer, but the lower grades distributed scattered and mixed with higerones.In conclusion, this study provided a feasible method and process for the acquisition ofreliable apple tree canopy reference and nitrogen and phosphorous nutrient status for appletree management, and also provided reference for other similar retrieval researches.