Subjects: Geosciences >> Geography submitted time 2023-09-19 Cooperative journals: 《干旱区地理》
Abstract： High- resolution soil salinity maps are urgently needed in arid and semi-arid regions to visualize thesubtle spatial variations in salinity distribution. These maps are crucial for guiding the development of land resource management policies and water resource management policies in salt-affected and potentially salt-affectedareas, aiming to prevent further soil degradation and ensure sustainable agricultural economic development andfood security. Based on PlanetScope imagery, vegetation spectral indices and soil salinity indices were extracted,resulting in a total of 21 variables. These variables were used as input for the Bagging algorithm to construct asoil salinity prediction model, referred to as ModelⅠ- . The max-relevance and min-redundancy (mRMR) methodwas employed to select relevant feature variables, which were then inputted into the Bagging algorithm to build asoil salinity prediction model, referred to as Model-Ⅱ. Field sampling data were used to assist in model buildingand validation. Model-Ⅰ and Model-Ⅱ were evaluated using model evaluation metrics. The results indicate thatthe prediction performance of Model-Ⅱ is better than that of Moedl-Ⅰ (mean R2=0.66, mean RMSE=18.00 dS·m-1,mean PRIQ=3.21 for the validation set), and that mRMR effectively reduces the multidimensional feature redundancy. PlanetScope images combined with the mRMR method successfully mapped high-resolution soil salinity,which provided more detailed information on the spatial distribution of soil salinity, and the results of the studypromoted the use of PlanetScope data to monitor soil salinity information.
Subjects: Geosciences >> Geography submitted time 2023-04-07 Cooperative journals: 《干旱区地理》
Abstract： Accurate monitoring of runoff from small and medium-sized rivers is of great significance for ecological stability in arid areas. However, it is difficult to accurately retrieve the flow of small and medium-sized rivers by remote sensing. Taking the Zhongfengchang river section of Kashi River in Nilka County, Xinjiang, China, as an example, this study constructed a power function relationship model between river width, water depth, and discharge based on the relationship fitting method and measured hydrological data, unmanned aerial vehicle data, and satellite data. Using the time series of satellite data, the runoff volume of the monitored river section was inferred 24 times in different periods. The results show that when the runoff rate is 0-50 m3 ·s −1 and 50-100 m3 ·s −1 , the inversion of the runoff rate based on the hydraulic geometry of the river width is optimal, with root mean square errors (RMSEs) of 7.15 m3 ·s −1 and 2.81 m3 ·s −1 , respectively; when the runoff rate is 100-200 m3 ·s −1 and > 200 m3 ·s −1 , the inversion of the hydraulic geometry based on water depth and river width is the best, with RMSEs of 13.37 m3 ·s−1 and 1.06 m3 ·s−1 , respectively. These findings provide a new method for the fine monitoring and management of runoff of small and medium-sized rivers in areas lacking hydrologic data and have high reference value for flood disaster prediction, hydropower resource development, and water ecosystem restoration.
Subjects: Geosciences >> Geography submitted time 2020-01-06 Cooperative journals: 《干旱区地理》
Abstract：土壤电导率 (Electrical conductivity, EC)是评价土壤盐渍化的重要指标。通过实测新疆艾比湖湿地自然保护区土壤EC及可见光—近红外光谱数据，利用波谱响应技术模拟Landsat 8 OLI、Sentinel 2、Sentinel 3卫星的宽波段数据。构建宽波段模拟数据及其5种预处理后的三维光谱指数 (Three-dimensional spectral index, TDSI)，采用梯度提升回归树算法 (Gradient boosting regression tree, GBRT) 建立3种卫星土壤EC估算模型，并比对加入TDSI后模型精度的变化。结果表明：在不同土壤EC条件下，3种卫星具有相似的光谱趋势，均在红、近红外波段附近反射率较高；TDSI与土壤EC相关性基本均在0.4以上，最大程度保留了与土壤EC敏感度高的红、绿、蓝、近红外、短波红外波段信息；GBRT对于土壤EC估算能力表现突出，3种卫星对土壤EC的最佳预测精度R2分别为0.831、0.847、0.903，在加入TDSI后，R2分别提高至0.835、0.857、0.935，综合分析发现，Sentinel 3对土壤EC估算效果最佳 (R2=0.935，均方根误差RMSE=2.986 mS·cm-1，赤池信息准则AIC=57.500)。通过利用波谱响应技术结合TDSI深度挖掘波段间的协同信息，采用GBRT验证了不同卫星对土壤R2的估算效果，二者相结合可以有效提升模型预测精度，为干旱区土壤盐渍化定量监测与防控提供有利指导。
Subjects: Geosciences >> Other Disciplines of Geosciences submitted time 2019-09-11 Cooperative journals: 《干旱区研究》
Abstract：以渭干河—库车河绿洲(渭—库绿洲)为研究区，采用在机器学习方面具有明显优势的随机森林回归算法，对绿洲内的4种典型植被(棉花、芦苇、杨树、大枣)叶片的叶绿素相对含量(soil and plant analyzer development, SPAD)进行估算和验证。首先基于“红边”处光谱信息丰富的哨兵2号(Sentinel-2)影像和由其衍生的一阶微分、二阶微分影像各提取23种对叶绿素敏感的宽波段光谱指数，加入3种影响植物生长的土壤参量(土壤含水量，土壤有机质，土壤电导率)作为影响叶片SPAD的特征变量，再根据以上特征变量对每种植被叶片各建立3种方案的SPAD估算模型，从而实现对绿洲内植被叶绿素的监测。结果表明：① 影像经一阶微分再提取的植被指数相比原位光谱植被指数，在SPAD估测模型中起到了更重要的作用，在随机森林算法的重要性排序中位居前列；② 4种植被叶片的SPAD估测模型都取得了不错的效果，芦苇叶片尤为显著，确定系数(R2)达到了0.926；③ 分析对比3种方案下模型预测能力，方案3(包含土壤参量)的预测能力卓越〔2.143方案1>方案2，土壤属性和模型预测结果有较强的非线性相关。Sentinel-2数据具有理想的估算绿洲植被叶绿素含量的潜力，提供了一种高效、低成本、潜在高精度的方案来估算叶绿素含量，可为干旱区绿洲农业、生态系统实现更有效的保护和管理提供参考。
Hits
Hits
Downloads
Comment