- 期刊
基於多來源資料和神經網路的灌區土壤鹽分變化分析
Analysis of soil salinity change in irrigation area based on multi-source data and neural network
摘要
土壤鹽化是由灌區長期大水漫灌，排水不暢所導致的一個重要生態問題，同時也是節水改造措施實施後，評價灌區生態環境是否有所改善的重要指標。本文主要針對內蒙古河套灌區沈烏灌域，於2020年4月1至10日期間，開展了不同土壤鹽化程度原狀土壤樣本採樣工作，並同步採集了原狀土的光譜資訊；結合環境衛星遙測影像資料，採用神經網路演算法反演獲取了研究區節水改造前（2012年）和節水改造後（2020年）4月份同期的土壤鹽化分佈，探討了節水改造前後沈烏灌域土壤鹽漬程度的變化情況。結果表明：受土壤質地、土壤水分和植被覆蓋等因素的影響，土壤鹽分與光譜反射率之間呈複雜的非線性關係；以環境一號衛星光譜資訊＋常態化差異植生指標＋垂直乾旱指數＋土壤質地等多來源資料為輸入變數，建立神經網路反演模型能夠很好地識別輸入變數和土壤鹽分的關係，具有很強的非線性擬合能力；相較于僅使用衛星光譜資訊，多來源資料結合將模型R2率定期由0.931提高至0.998，驗證期由0.715提高至0.813；節水改造後，沈烏灌域土壤鹽化程度明顯改善，非鹽土面積增加了29.76%，輕度鹽化土減少了69.98%，中度鹽化土減少了76.52%，重度鹽化土減少了30.90%，鹽鹼地減少了49.79%。研究成果可為灌區節水改造後的生態環境效應分析提供參考。
並列摘要
Soil salinization is an important ecological problem caused by long-term flood irrigation and poor drainage in irrigation area. It is also an important index to evaluate whether the ecological environment of irrigation area has been improved after the implementation of water-saving reform project. In this paper, the sampling of undisturbed soil samples with different soil salinization degrees was carried out in Shenwu irrigation area of Hetao irrigation district of Inner Mongolia from April 1 to 10, 2020, and the spectral information of undisturbed soil was synchronously collected.Combined with the environmental satellite remote sensing image data, the neural network algorithm was used to obtain the soil salinization distribution in the same period in April before (2012) and after (2020) water-saving reform in the study area, and the change of soil salinization in Shen-Wu irrigation area before and after water-saving renovation was discussed.The results showed that due to the influence of soil texture, soil moisture, vegetation cover and other factors. There is a nonlinear relationship between soil salinity and spectral reflectance. The neural network inversion model is established by using multi-source data such as Spectral Information of HJ-1 satellite + Normalized Difference Vegetation Index + Perpendicular Drought Index + Soil texture as the input has strong nonlinear fitting ability. Compared with only using satellite spectral data, the combination of multi-source data regularly increases the R2 from 0.931 to 0.998 in the calibration periods, from 0.715 to 0.813 in the validation periods. After water-saving transformation, the degree of soil salinization in Shenwu irrigation area was significantly improved. The area of non-saline soil increased by 29.76 %, mild saline soil decreased by 69.98 %, moderate saline soil decreased by 76.52 %, severe saline soil decreased by 30.90 %, and saline-alkali soil decreased by 49.79 %. The research results can provide reference for the ecological environment effect of water-saving transformation in irrigation areas.