基于植被指数的藏北牧区土壤湿度反演

土壤湿度的遥感动态监测在农牧业生产中具有重要意义。近年来,多种基于遥感指数的土壤湿度监测方法被提出并得到广泛关注,但当前对不同深度土壤湿度的反演及植被指数反映土壤湿度滞后性的研究较少。该文针对遥感指数反演土壤湿度的精度问题,对MODIS(moderate resolutionimaging spectroradiometer)的2种植被指数产品归一化差异植被指数(normalized difference vegetation index,NDVI)和增强型植被指数(enhanced vegetation index,EVI)与土壤湿度实测值进行相关分析,并利用在其中一个样点得到相关系数最高的回归模型对距离较远的其它点进行土壤湿度值估算,最后用土壤湿度实测值对模型的精度进行验证。结果表明,2种植被指数均与土壤湿度值呈现出较强的相关性,且利用植被指数估算土壤湿度的延迟天数为5~10 d。在相同气候模式、土壤类型和植被类型的条件下,高程为影响回归模型精度的主要因素。该研究可为牧区多层深度土壤湿度反演方法的选择和监测提供参考依据。

Soil moisture inversion in pasture of northern Tibet based on vegetation index

Dynamic monitoring of soil moisture by remote sensing can play a significant role in agricultural production.In recent years, continual attention has been focused on the thought that soil moisture information can be extracted by remote sensing indices.However, the majority of the studies on soil moisture estimation are applied without considering different depths and the time lag that vegetation index(VI) responds to soil moisture.This study investigated the potential of using the moderate resolution imaging spectroradiometer(MODIS) products, including normalized difference vegetation index (NDVI) and enhanced vegetation index(EVI), to estimate soil moisture at distant in situ measured sites.In this study, 30 sites were sampled under the same climatic setting, with the same soil type and the same vegetation type.The MOD13Q1 series data were selected to receive both NDVI and EVI products, which were 16 day composites with 250 meter spatial resolution.We alsoobtainedthe in situ soil moisturedata that were measured once every 30 min from Soil Moisture/Temperature Monitoring Network (SMTMN) in the pasture of northern Tibet.Daily soil moisture was the average of soil moisturesthat were measured once every 30 min.To study the correlation between VIs and soil moisture, the daily time series of soil moisture data had to be processed to match the 16 day VIs.In order to move autocorrelation of most time series data, a simple moving averagemethod was used to identify the seasonal components: 47 point moving average for the daily soil moisture and 3 point moving average for the 16 day VIs.Deseasonalized time series was then produced by subtracting seasonal time series from raw time series.Collocatethe deseasonalized time series of soil moisture at 4 depths(0~5, 10, 20and 40 cm) and the NDVI, EVI in 2012 were used for correlation analysis.Similar analysis was also conducted for the comparison.Pearson Product Moment correlation coefficients were calculated during the growing season (from May to October)for 4 depths.Our hypothesis was that the soil moisture VI regression model developed at one site could be used to estimate soil moisture using VIs at a distant site, providing that other sites had similar soil type, vegetation, and climate regime.Wetested the hypothesis by developing a regression model(at No.2 site) using the deseasonalized NDVI with a 5 day time lag as the independent variable and the deseasonalized soil moisture as the dependent variable at 4 native sites(No.1, No.3, No.4 and No.5) within the growing season.Results showed that the deseasonalized time series and the raw time series had the consistent results between NDVI, EVI and soil moisture at the 30 sites.Both NDVI and EVI needed longer time to respond to soil moisture change.Correlation based on raw time series of VIs and soil moisture was consistent with that based on deseasonalized time series at every depth.The maximum correlation value between deseasonalized NDVI,EVI and soil moisture was from 0.50 to 0.95, and the correlation was significant at the 99% level.Most correlation reached the maximum value whenVIslaged soil moisture by 5-10 days.Regression analysis was conducted using the deseasonalizedsoil moisture time series and the deseasonalizedNDVI time series with a 5 day time lag at No.2 site.Regression models developed at one site and applied to a similar distant site could estimate soil moistures.The correlation coefficient values between estimated and in situ measured soil moisture at different depths varied from 0.64 to 0.68 at No.1 site, from 0.42 to 0.55 at No.3 site, from 0.88 to 0.94 at No.4 site, and from 0.67 to 0.92 at No.5 site, and higher elevationhad smaller correlation coefficient.Thus, elevation is the main factor that affects the accuracy of the regression model.This research can provide valuable information for method selection in pasture soil moisture estimation at different depths by remote sensingindices.