Examining the utility of random forest and AISA Eagle hyperspectral image data to predict Pinus patula age in KwaZulu-Natal,South Africa

High-data dimensionality is a common problem in hyperspectral data processing. Consequently, remote sensing techniques that reduce the number of bands are considered essential tools for most hyperspectral applications. The aim of this study was to examine the utility of the random forest ensemble to select the optimal subset of hyperspectral bands to predict the age of Pinus patula stands. Airborne AISA Eagle hyperspectral image data were collected over the study area. The random forest ensemble was used to test whether the forward or backward variable selection methods could identify the optimal subset of bands. Results indicate that both the selection methods produced high-predictive accuracies (root mean square error = 3.097 years). However, the backward variable selection method utilized 206 bands for the final model, while the forward variable selection utilized only a small subset of non-redundant bands (n = 9) while preserving the highest model accuracy (R ² = 0.6).     

高光谱与高空间分辨率遥感数据的亚热带森林生物量反演

精确估算森林生物量对全球碳平衡以及气候变化的研究有重要意义。以亚热带天然次生林为研究对象,借助地面实测样地数据,通过对机载LiCHy(LiDAR,CCD and Hyperspectral)传感器同时获取的高光谱和高空间分辨率数据进行信息提取和数据融合,建模反演森林生物量。首先通过面向对象分割方法进行单木冠幅提取,然后融合从高光谱数据提取的光谱特征变量和从高空间分辨率数据提取的单木冠幅统计变量,构建多元回归模型估算地上、地下生物量,最后利用地面实测生物量经交叉验证评价模型精度。结果表明,综合模型的精度(R~2为0.54—0.62)高于高光谱模型(R~2为0.48—0.57);在高光谱模型中地上生物量模型精度(R~2为0.57)高于地下生物量模型(R~2为0.48);在综合模型中地上生物量模型精度(R~2为0.62)同样高于地下生物量模型(R~2为0.54)。交叉验证结果表明,与仅使用高光谱数据(单一数据源)相比,通过集成高光谱和高空间分辨率数据的生物量反演效果有所提升,可以更加有效地估算亚热带森林生物量。     

Modelling electrical conductivity of soil from backscattering coefficient of microwave remotely sensed data using artificial neural network

Soil salinity is one of the main agricultural problems which expand to larger areas. Soil scientists categorize salinity level by electrical conductivity (EC) measurement. However, field measurements of EC require extensive time, cost and experiences. Remote sensing is one suitable option to investigate and collect spatial data in larger areas. Many researches estimated soil moisture through microwave, but there are fewer studies which mentioned about direct relationship between EC and backscattering coefficient (BC). Thus, this study aims to propose the estimation of EC directly from BC of microwave. The relationship between EC obtained from field survey and BC from microwave is non-linear, artificial neural network (ANN) is one technique proposed in this study to figure out EC and BC relationship. ANN uses multilayer of interconnected processing resulting in EC value with high accuracy which is acceptable. For this reason, ANN model can be successfully utilized as an effective tool for EC estimation from microwave.     

随机森林回归模型用于土壤重金属含量多光谱遥感反演

本文以陕西省柞水县大西沟矿区为研究区域,通过实地采集土壤样本,结合在Landsat 8多光谱遥感影像上提取的辐射亮度值和光谱衍生指数,以及从ASTER GDEM提取的3种地形因素,通过相关性分析确定了建模因子,并以K折交叉验证法建立了砷、铜、铅3种重金属元素的随机森林回归模型。试验结果表明,所建立模型的预测精度优于多元线性回归模型和CART模型,可见随机森林回归模型适用于在小样本情况下的矿区重金属含量反演。经现场调查,空间反演结果与实际情况较符合,证明了基于多光谱遥感的随机森林回归模型在矿区土壤重金属反演中的准确性。     

Predicting forest carbon stocks from high resolution satellite data in dry forests of Zimbabwe: exploring the effect of the red-edge band in forest carbon stocks estimation

In this study, we tested whether the inclusion of the red-edge band as a covariate to vegetation indices improves the predictive accuracy in forest carbon estimation and mapping in savanna dry forests of Zimbabwe. Initially, we tested whether and to what extent vegetation indices (simple ratio SR, soil-adjusted vegetation index and normalized difference vegetation index) derived from high spatial resolution satellite imagery (WorldView-2) predict forest carbon stocks. Next, we tested whether inclusion of reflectance in the red-edge band as a covariate to vegetation indices improve the model's accuracy in forest carbon prediction. We used simple regression analysis to determine the nature and the strength of the relationship between forest carbon stocks and remotely sensed vegetation indices. We then used multiple regression analysis to determine whether integrating vegetation indices and reflection in the red-edge band improve forest carbon prediction. Next, we mapped the spatial variation in forest carbon stocks using the best regression model relating forest carbon stocks to remotely sensed vegetation indices and reflection in the red-edge band. Our results showed that vegetation indices alone as an explanatory variable significantly (p < 0.05) predicted forest carbon stocks with R² ranging between 45 and 63% and RMSE ranging from 10.3 to 12.9%. However, when the reflectance in the red-edge band was included in the regression models the explained variance increased to between 68 and 70% with the RMSE ranging between 9.56 and 10.1%. A combination of SR and reflectance in the red edge produced the best predictor of forest carbon stocks. We concluded that integrating vegetation indices and reflectance in the red-edge band derived from high spatial resolution can be successfully used to estimate forest carbon in dry forests with minimal error.     

Trade-offs between carbon,water, soil and food in Guanzhong-Tianshui economic region from remotely sensed data

It is important to ensure the efficient supply of land ecosystem services when the competition for land is increasing. In this paper we simulated the ecosystem services function under two scenarios, including carbon sequestration, agricultural production, water and soil conservation, and analyzed the tradeoffs among these ecosystem services in Guanzhong-Tianshui region from 2000 to 2050. Then the productive efficiency of ecosystem services was assessed under two scenarios and compared their production possibility frontiers (PPFs). Through the simulation analysis of their optimum allocation, we also provide the scientific evidence to the development of ecosystem. The natural rules were revealed that if these trade-offs emphasize the potential to sequester carbon in the landscape, along with very little loss of agricultural production, much more water is used. It could be identified to adhere to combine the exploitation and utilization, remediation and protection for land to promote the effective circulation of land eco-system, and meet the society’s preferences for land ecosystem service function by adjusting the use of multiple eco-services.     

Estimating temporal changes in soil carbon stocks at ecoregional scale in Madagascar using remote-sensing

Soil organic carbon (SOC) plays an important role in climate change regulation notably through release of CO2 following land use change such a deforestation, but data on stock change levels are lacking. This study aims to empirically assess SOC stocks change between 1991 and 2011 at the landscape scale using easy-to-access spatially-explicit environmental factors. The study area was located in southeast Madagascar, in a region that exhibits very high rate of deforestation and which is characterized by both humid and dry climates. We estimated SOC stock on 0.1 ha plots for 95 different locations in a 43,000 ha reference area covering both dry and humid conditions and representing different land cover including natural forest, cropland, pasture and fallows. We used the Random Forest algorithm to find out the environmental factors explaining the spatial distribution of SOC. We then predicted SOC stocks for two soil layers at 30 cm and 100 cm over a wider area of 395,000 ha. By changing the soil and vegetation indices derived from remote sensing images we were able to produce SOC maps for 1991 and 2011. Those estimates and their related uncertainties where combined in a post-processing step to map estimates of significant SOC variations and we finally compared the SOC change map with published deforestation maps. Results show that the geologic variables, precipitation, temperature, and soil-vegetation status were strong predictors of SOC distribution at regional scale. We estimated an average net loss of 10.7% and 5.2% for the 30 cm and the 100 cm layers respectively for deforested areas in the humid area. Our results also suggest that these losses occur within the first five years following deforestation. No significant variations were observed for the dry region. This study provides new solutions and knowledge for a better integration of soil threats and opportunities in land management policies.     

Estimation of aboveground biomass and carbon in a tropical rain forest in Gabon using remote sensing and GPS data

The knowledge of biomass stocks in tropical forests is critical for climate change and ecosystem services studies. This research was conducted in a tropical rain forest located near the city of Libreville (the capital of Gabon), in the Akanda Peninsula. The forest cover was stratified in terms of mature, secondary and mangrove forests using Landsat-ETM data. A field inventory was conducted to measure the required basic forest parameters and estimate the aboveground biomass (AGB) and carbon over the different forest classes. The Shuttle Radar Topography Mission (SRTM) data were used in combination with ground-based GPS measurements to derive forest heights. Finally, the relationships between the estimated heights and AGB were established and validated. Highest biomass stocks were found in the mature stands (223 ± 37 MgC/ha), followed by the secondary forests (116 ± 17 MgC/ha) and finally the mangrove forests (36 ± 19 MgC/ha). Strong relationships were found between AGB and forest heights (R² > 0.85).     

星地多源数据的区域土壤有机质数字制图

土壤有机质(SOM)是全球碳循环、土壤养分的重要组成部分,精确估算土壤有机质含量具有重要意义。本文以中国东北—华北平原为研究区,收集了1078个土壤样本,以遥感数据(MODIS,TRMM和STRM数据)与土壤地面光谱数据为预测因子,运用基于树形结构的数据挖掘技术构建土壤有机质-环境预测因子模型进行数字土壤制图。通过不同建模样本数建模精度比较,选择300个样本数时的模型为最优模型。建模结果表明土壤光谱和气候因子是研究区SOM变异的主控因子,生物因子次之,而地形因子影响最小。预测结果经检验,RMSE为7.25,R2为0.69,RPD为1.53制图结果与基于第二次全国土壤普查数据的土壤有机质地图具有相似的分布规律,呈现SOM自东北向西南递减的趋势。通过比较分析发现,经过20年左右的土地开发与利用,研究区低SOM和高SOM含量土壤面积减少,而中等SOM含量土壤面积增加。     

A comparison of geographic datasets and field measurements to model soil carbon using random forests and stepwise regressions (British Columbia,Canada)

We used geographic datasets and field measurements to examine the mechanisms that affect soil carbon (SC) storage for 65 grazed and non-grazed pastures in southern interior grasslands of British Columbia, Canada. Stepwise linear regression (SR) modeling was compared with random forest (RF) modeling. Models produced with SR performed better than those produced using RF models (r² = 0.56–0.77 AIC = 0.16–0.30 for SR models; r² = 0.38–0.53 and AIC = 0.18–0.30 for RF models). The factors most significant when predicting SC were elevation, precipitation, and the normalized difference vegetation index (NDVI). NDVI was evaluated at two scales using: (1) the MOD 13Q1 (250 m/16-day resolution) NDVI data product from the moderate resolution imaging spectro-radiometer (MODIS) (NDVI_MODIS), and (2) a handheld multispectral radiometer (MSR, 1 m resolution) (NDVI_MSR) in order to understand the potential for increasing model accuracy by increasing the spatial resolution of the gridded geographic datasets. When NDVI_MSR data were used to predict SC, the percentage of the variance explained by the model was greater than for models that relied on NDVI_MODIS data (r² = 0.68 for SC for non-grazed systems, modeled with SR based on NDVI_MODIS data; r² = 0.77 for SC for non-grazed systems, modeled with SR based on NDVI_MSR data). The outcomes of this study provide the groundwork for effective monitoring of SC using geographic datasets to enable a carbon offset program for the ranching industry.     

A GEOBIA framework for the implementation of national and international forest definitions using very high spatial resolution optical satellite data

In recent decades, there is an increasing need for harmonised and accurate information on the status and extent of forests. However, delineating the extent of forest areas is a complex task, since the existence of more than 100 definitions of forest worldwide causes considerable discrepancies in forested area estimates. The aim of this work was to examine the potential of geographic object based image analysis (GEOBIA) and very high spatial resolution imagery to discriminate forest areas following two different definitions of forest in northern Greece. In particular, we examined the definition of forest under the Greek law as well as the United Nations Food and Agricultural Organisation definition. Our findings suggest that the developed GEOBIA approach not only performed remarkably well for the discrimination of forest areas but also allowed to estimate rapidly and reliably forest extents when the two aforementioned forest definitions were employed.     

Predicting the spatial distribution of hunted and non-hunted Sable antelope (Hippotragus niger niger) using remotely sensed woody cover in a Southern African savanna

From remotely sensed woody cover, we tested whether sables under hunting pressure preferred closed woodland habitats and whether those not under hunting preferred more open woodland habitats. We applied a two factorial logistic regression analysis to model the probability of occurrence of sable antelope in hunted and non-hunted areas of northwest Zimbabwe as a function of vegetation cover density (estimated by a normalized difference vegetation index (NDVI)). We validated the results by high-spatial resolution imagery derived tree canopy area. We subsequently compared the predictions from the two models in order to compare sable cover selection between hunted and non-hunted areas. Our results suggest that hunted sables are likely to select closed woodland, while non-hunted ones would prefer more open woodland habitats. We also established a significant positive relationship between NDVI and tree canopy cover, thus emphasizing the importance of remote sensing in studies that measure the impact of hunting on habitat selection of targeted species.     

Optimizing the spatial resolution of WorldView-2 imagery for discriminating forest vegetation at subspecies level in KwaZulu-Natal,South Africa

The objective of this study was to identify an appropriate spatial resolution for discriminating forest vegetation at subspecies level. WorldView-2 imagery was progressively resampled to coarser spatial resolutions. At a compartment level, 30 × 30-m subsets were generated across forest compartments to represent the five forest subspecies investigated in this study. From the centre of each subset, the spatial resolution of the original WorldView-2 image was resampled from 6 to 34-m, with increments of 4-m. The variance was then calculated at every resampled spatial resolution using each of the eight WorldView-2 bands. Based on the sampling theorem, the 3-m spatial resolution provided an appropriate resolution for all subspecies investigated. The WorldView-2 image was subsequently classified using the partial least squares linear discriminant analysis algorithm and the appropriate spatial resolution. An overall classification accuracy of 90% was established with an allocation disagreement of 9 and a quantity disagreement of 1.     

Predicting plant water content in Eucalyptus grandis forest stands in KwaZulu-Natal,South Africa using field spectra resampled to the Sumbandila Satellite Sensor

The measurement of plant water content is essential to assess stress and disturbance in forest plantations. Traditional techniques to assess plant water content are costly, time consuming and spatially restrictive. Remote sensing techniques offer the alternative of a non-destructive and instantaneous method of assessing plant water content over large spatial scales where ground measurements would be impossible on a regular basis. In the context of South Africa, due to the cost and availability of imagery, studies focusing on the estimation of plant water content using remote sensing data have been limited. With the scheduled launch of the South African satellite SumbandilaSat evident in 2009, it is imperative to test the utility of this satellite in estimating plant water content. This study resamples field spectral data measured from a field spectrometer to the band settings of the SumbandilaSat in order to test its potential in estimating plant water content in a Eucalyptus plantation. The resampled SumbandilaSat wavebands were input into a neural network due to its ability to model non-linearity in a dataset and its inherent ability to perform better than conventional linear models. The integrated approach involving neural networks and the resampled field spectral data successfully predicted plant water content with a correlation coefficient of 0.74 and a root mean square error (RMSE) of 1.41% on an independent test dataset outperforming the traditional multiple regression method of estimation. The best-trained neural network algorithm that was chosen for assessing the relationship between plant water content and the SumbandilaSat bands was based on a few points only and more research is required to test the robustness and effectiveness of this sensor in estimating plant water content across different species and seasons. This is critical for monitoring plantation health in South Africa using a cheaply available local sensor containing key vegetation wavelengths.     

Prediction of soil loss differences and sediment accumulation at the Nilufer creek watershed,Turkey, using multiyear satellite data in a GIS

This study integrates the RUSLE, remote sensing and GIS to assess soil loss and identify sensitive areas to soil erosion in the Nilufer creek watershed in Bursa province, Turkey. The annual average soil loss was generated separately for years 1984 and 2011, in order to expose possible soil loss differences occurred in 27 years. In addition, sediment accumulation and sediment yield of the studied watershed was also predicted and discussed. The results indicated that very severe erosion risk areas in 1984 was 13.4% of the area, but it was increased to 15.3% by the year 2011, which needs immediate attention from soil conservation point of view. Furthermore, the estimated annual sediment yield of the Nilufer creek watershed was increased from 903 to 979 Mg km^?2 y^?1 in 27 years period. The study also provides useful information for decision-makers and planners to take appropriate land management practices in the area.     

WorldView-2纹理的森林地上生物量反演

使用高空间分辨率卫星WorldView-2的多光谱遥感影像,构建植被指数和纹理因子等遥感因子与森林地上生物量的关系方程,并计算模型估测精度和均方根误差,探索高分辨率数据的光谱与纹理信息在温带森林地上生物量估测应用中的潜力。以黑龙江省凉水自然保护区温带天然林及天然次生林为研究对象,通过灰度共生矩阵(GLCM)、灰度差分向量(GLDV)及和差直方图(SADH)对高分辨率遥感影像进行纹理信息提取,并利用外业调查的74个样地地上生物量与遥感因子建立参数估计模型。提取的遥感因子包括6种植被指数(比值植被指数RVI、差值植被指数DVI、规一化植被指数NDVI、增强植被指数EVI、土壤调节植被指数SAVI和修正的土壤调节植被指数MSAVI)以及3类纹理因子(GLCM、GLDV和SADH)。为避免特征变量个数较多对估测模型造成过拟合,利用随机森林算法对提取的遥感因子进行特征选择,将最优的特征变量输入模型参与建模估测。采用支持向量回归(SVR)进行生物量建模及验证,结果显示选入模型的和差直方图均值(sadh_mean)、灰度共生矩阵方差(glcm_var)和差值植被指数(DVI)等遥感因子对森林地上生物量有较好的解释效果;植被指数+纹理因子组合的模型获得较精确的AGB估算结果(R2=0.85,RMSE=42.30 t/ha),单独使用植被指数的模型精度则较低(R~2=0.69,RMSE=61.13 t/ha)。     

融合物理机理与随机森林算法的FY-4A AGRI数据晴空大气可降水量遥感反演

作为中国发射的风云四号系列首颗卫星,FY-4A搭载了先进的静止轨道辐射成像仪AGRI。本文利用AGRI传感器的中红外到热红外波段观测数据、ERA5水汽再分析产品以及全球无线电探空数据集IGRA等数据进行晴空大气可降水量反演研究。在海洋表面,分别利用回归分析法与随机森林算法反演大气可降水量,反演结果与ERA5水汽产品相比,均方根误差为0.493 cm和0.247 cm。在陆地表面,利用随机森林模型反演大气可降水量,反演结果与ERA5水汽产品相比,均方根误差为0.155 cm;与IGRA水汽产品相比,均方根误差为0.215 cm。结果表明本文使用的随机森林算法可以有效地提升热红外遥感数据反演大气可降水量的精度。     

Modeling soil parameters using hyperspectral image reflectance in subtropical coastal wetlands

Developing spectral models of soil properties is an important frontier in remote sensing and soil science. Several studies have focused on modeling soil properties such as total pools of soil organic matter and carbon in bare soils. We extended this effort to model soil parameters in areas densely covered with coastal vegetation. Moreover, we investigated soil properties indicative of soil functions such as nutrient and organic matter turnover and storage. These properties include the partitioning of mineral and organic soil between particulate (>53 μm) and fine size classes, and the partitioning of soil carbon and nitrogen pools between stable and labile fractions. Soil samples were obtained from Avicennia germinans mangrove forest and Juncus roemerianus salt marsh plots on the west coast of central Florida. Spectra corresponding to field plot locations from Hyperion hyperspectral image were extracted and analyzed. The spectral information was regressed against the soil variables to determine the best single bands and optimal band combinations for the simple ratio (SR) and normalized difference index (NDI) indices. The regression analysis yielded levels of correlation for soil variables with R² values ranging from 0.21 to 0.47 for best individual bands, 0.28 to 0.81 for two-band indices, and 0.53 to 0.96 for partial least-squares (PLS) regressions for the Hyperion image data. Spectral models using Hyperion data adequately (RPD > 1.4) predicted particulate organic matter (POM), silt + clay, labile carbon (C), and labile nitrogen (N) (where RPD = ratio of standard deviation to root mean square error of cross-validation [RMSECV]). The SR (0.53 μm, 2.11 μm) model of labile N with R² = 0.81, RMSECV= 0.28, and RPD = 1.94 produced the best results in this study. Our results provide optimism that remote-sensing spectral models can successfully predict soil properties indicative of ecosystem nutrient and organic matter turnover and storage, and do so in areas with dense canopy cover.     

Estimating fractional cover of crop,crop residue,and soil in cropland using broadband remote sensing data and machine learning

The fractional vegetation cover (FVC), crop residue cover (CRC), and bare soil (BS) are three important parameters in vegetation–soil ecosystems, and their correct and timely estimation can improve crop monitoring and environmental monitoring. The triangular space method uses one CRC index and one vegetation index to create a triangular space in which the three vertices represent pure vegetation, crop residue, and bare soil. Subsequently, the CRC, FVC, and BS of mixed remote sensing pixels can be distinguished by their spatial locations in the triangular space. However, soil moisture and crop-residue moisture (SM-CRM) significantly reduce the performance of broadband remote sensing CRC indices and can thus decrease the accuracy of the remote estimation and mapping of CRC, FVC, and BS. This study evaluated the use of broadband remote sensing, the triangular space method, and the random forest (RF) technique to estimate and map the FVC, CRC, and BS of cropland in which SM-CRM changes dramatically. A spectral dataset was obtained using: (1) from a field-based experiment with a field spectrometer; and (2) from a laboratory-based simulation that included four distinct soil types, three types of crop residue (winter-wheat, maize, and rice), one crop (winter wheat), and varying SM-CRM. We trained an RF model [designated the broadband crop-residue index from random forest (CRRF)] that can magnify spectral features of crop residue and soil by using the broadband remote sensing angle indices as input, and uses a moisture-resistant hyperspectral index as the target. The effects of moisture on crop residue and soil were minimized by using the broadband CRRF. Then, the CRRF-NDVI triangular space method was used to estimate and map CRC, FVC, and BS. Our method was validated by using both laboratory- and field-based experiments and Sentinel-2 broadband remote-sensing images. Our results indicate that the CRRF-NDVI triangular space method can reduce the effect of moisture on the broadband remote-sensing of CRC, and may also help to obtain laboratory and field CRC, FVC, and BS. Thus, the proposed method has great potential for application to croplands in which the SM-CRM content changes dramatically.