JIA-2019-11

2637 ZHANG Xi-wang et al. Journal of Integrative Agriculture 2019, 18(11): 2628–2643 multiple regression model was chosen to assess abundance. The regression model is shown in eq. (13). 0.3511 Slope Slope + 3.7734 Slope WWAF=0.1093 EF CD AB − 0.2410 + (13) The regression model is combined with the NDVI slope images of AB, CD and EF, and the assessed winter wheat abundance is shown in Fig. 6. 4.4. Membership for winter wheat To make the training samples more representative, they should be selected from the homogeneous regional centers of each class, with the aid of field investigation. A sufficient number of training samples should be selected to meet the requirements of establishing a discriminant function. In this study, the covariance matrix and mean vector of winter wheat are counted from the training samples and substituted into eq. (5) to construct the conditional probability density function. An iterative method is used to determine the a priori probability. First, it is assumed that each class appears with an equal probability, and this probability is entered into eq. (4) to calculate the pixel membership. Then, the regional acreage proportion of winter wheat is calculated in the study area, and substituted into eq. (4), again to calculate the membership for winter wheat. The result is shown in Fig. 7. 4.5. Winter wheat identification result After the winter wheat abundance is processed according to eq. (7), the abundance value F implies that there are F pixels belonging to winter wheat in the corresponding 10×10 TM pixels. The membership was calculated and it indicates the probability of winter wheat. The processed abundance image and membership image are spatially operated using the eq. (8) to produce a threshold image. Finally, eq. (9) is used to identify winter wheat according to the threshold image and corresponding membership matrix. The identified result of winter wheat distribution is shown in Fig. 8. The total acreage of identified winter wheat in this study area is 155 538.63 ha. 4.6. Identification accuracy In the evaluation of identification results, Aa, As, OA, PA, UA and Kappa are calculated. In addition, MLC and RFC are applied based on the same Landsat image to compare the result with the identification result of the method proposed in this study. First of all, the identified winter wheat acreages are compared to the official statistics (NBSSOH 2011) for each county in the study area, as shown in Table 3. The official total winter wheat acreage is 145371.20 ha, and the identified acreages of MLC and RFC are 170 405.60 ha (Aa is 82.78%) and 164 897.29 ha (Aa is 86.57%), respectively. Whereas the identified acreage of the proposed method is 155 538.63 ha, and Aa is 93.01%. The winter wheat acreage is overestimated in most counties with only one exception, Yanshi County, where it is underestimated. Among the counties, the area accuracy is the lowest in Yichuan County, where it is 82.54%; and the highest accuracy is in Luoning County, at 95.98%. For these three methods, the RMSE values are 4.47, 3.52 and 3.07%, respectively. In addition, the identification results are validated by the randomly generated 500 validation points, and As is calculated and shown in Table 4. For the proposed method, 457 of the 500 validation points are correctly identified, andAs is 91.40%. As is the lowest in Yichuan County, at 85.22%, the highest accuracy is 95.54% in Yanshi County. Meanwhile, As of MLC and RFC is 84.65 and 88.60%, respectively. Using the 500 random validation points, confusion matrices of the three methods are produced, as shown in Table 5. Compared to MLC and RFC, OA of the proposed method is increased by 6.75 and 2.80%, respectively. The Kappa is also significantly improved. 5. Discussion 5.1. The influencing factors Identification accuracy and influencing factors Winter wheat is the primary winter crop in this study area. Its growth cycle begins in mid-October, and progresses through planting, emergence, tillering, wintering, greening, jointing, heading, booting, milky maturity, and maturity. It does not end until mid-June of the following year. During this winter wheat growing season, other crops, such as cole, spring maize, leeks and garlic, possess completely different 112°30´E 112°30´E 112°00´E 112°00´E 111°30´E 111°30´E 34°40´N N 34°40´N 34°20´N 34°20´N 0 25 50 12.5 km 1.0 0.0 Fig. 6 Abundance map for winter wheat at the 250-m scale.