Soil Aggregate Stability Mapping Using Remote Sensing and GIS-Based Machine Learning Technique

Soil aggregate stability (SAS) is a critical parameter of soil quality and its mapping can help determine erosion hotspots. Despite this importance, SAS is less documented in available literature due to limited number of analyzes besides being a time consuming. For this reason, many researchers have turned to alternative methods that often use readily available variables such as soil parameters or remote sensing indices to estimate this variable. In that framework, the aim of the present study focused on the investigation of the feasibile use of adapted Leo Breiman’s random forest algorithm (RF) to mapping different mean weight diameter (MWD) tests as an index of SAS (mechanical breakdown (MWDmb), slow wetting (MWDsw), fast wetting (MWDfw) and the mean of the three tests (MWDmean)). The model was built with 77 samples distributed in the three watersheds of the study area located at Settat Ben-Ahmed, in Morocco and with the use of several environmental variables such as soil parameters (organic matter and clay), remote sensing indices (band 2, band 3, band 4, band 5, normalized difference vegetation index (NDVI) and transformed normalized difference vegetation index (TNDVI)), topography (elevation, slope, curvature plane and the topographic wetness index (TWI)) along with additional categorical variables as geological maps, land use and soil classes. The results showed a good level of accuracy for the training phase (75% of samples) for the different tests (R2 > 0.92, RMSE and MAE < 0.15) and were satisfactory for the testing phase (25% of samples, R2 > 0.65, RMSE and MAE < 0.31). Also, organic matter, topography and geology were the most important parameters in the spatial prediction of SAS. Finally, the maps build during this study could be of great use to identify areas of less stable soils in the perspective for taking the necessary measures to improve their quality.

DeepIndices: Remote Sensing Indices Based on Approximation of Functions through Deep-Learning, Application to Uncalibrated Vegetation Images

The form of a remote sensing index is generally empirically defined, whether by choosing specific reflectance bands, equation forms or its coefficients. These spectral indices are used as preprocessing stage before object detection/classification. But no study seems to search for the best form through function approximation in order to optimize the classification and/or segmentation. The objective of this study is to develop a method to find the optimal index, using a statistical approach by gradient descent on different forms of generic equations. From six wavebands images, five equations have been tested, namely: linear, linear ratio, polynomial, universal function approximator and dense morphological. Few techniques in signal processing and image analysis are also deployed within a deep-learning framework. Performances of standard indices and DeepIndices were evaluated using two metrics, the dice (similar to f1-score) and the mean intersection over union (mIoU) scores. The study focuses on a specific multispectral camera used in near-field acquisition of soil and vegetation surfaces. These DeepIndices are built and compared to 89 common vegetation indices using the same vegetation dataset and metrics. As an illustration the most used index for vegetation, NDVI (Normalized Difference Vegetation Indices) offers a mIoU score of 63.98% whereas our best models gives an analytic solution to reconstruct an index with a mIoU of 82.19%. This difference is significant enough to improve the segmentation and robustness of the index from various external factors, as well as the shape of detected elements.

Estimation of Bare Soil Moisture from Remote Sensing Indices in the 0.4–2.5 mm Spectral Range

Soil moisture content (SMC) is an important element of the environment, influencing water availability for plants and atmospheric parameters, and its monitoring is important for predicting floods or droughts and for weather and climate modeling. Optical methods for measuring soil moisture use spectral reflection analysis in the 350–2500 nm range. Remote sensing is considered to be an effective tool for monitoring soil parameters over large areas and to be more cost effective than in situ measurements. The aim of this study was to assess the SMC of bare soil on the basis of hyperspectral data from the ASD FieldSpec 4 Hi-Res field spectrometer by determining remote sensing indices and visualization based on multispectral data obtained from UAVs. Remote sensing measurements were validated on the basis of field humidity measurements with the HH2 Moisture Meter and ML3 ThetaProbe Soil Moisture Sensor. A strong correlation between terrestrial and remote sensing data was observed for 7 out of 11 selected indexes and the determination coefficient R2 values ranged from 67%– 87%. The best results were obtained for the NINSON index, with determination coefficient values of 87%, NSMI index (83.5%) and NINSOL (81.7%). We conclude that both hyperspectral and multispectral remote sensing data of bare soil moisture are valuable, providing good temporal and spatial resolution of soil moisture distribution in local areas, which is important for monitoring and forecasting local changes in climate.

Angle Effect on Typical Optical Remote Sensing Indices in Vegetation Monitoring

Optical remote sensing indices play an important role in vegetation information extraction and have been widely serving ecology, agriculture and forestry, urban monitoring, and other communities. Remote sensing indices are constructed from individual bands depending on special characteristics to enhance the typical spectral features for the identification or distinction of surface land covers. With the development of quantitative remote sensing, there is a rapidly increasing requirement for accurate data processing and modeling. It is well known that the geometry-induced variation observed on surface reflectance is not ignorable, but the situation of uncertainty thereby introduced into these indices still needs further detailed understanding. We adopted the ground multi-angle hyperspectrum, spectral response function (SRF) of Thematic Mapper (TM), Enhanced Thematic Mapper (ETM+), Operational Land Imager (OLI), Moderate-Resolution Imaging Spectroradiometer (MODIS), and Multi-Spectral Instrument (MSI) optical sensors and simulated their sensor-like spectral reflectance; then, we investigated the potential angle effect uncertainty on optical indices that have been frequently involved in vegetation monitoring and examined the forward/backward effect over both the ground-based level and the actual Landsat TM/ETM+ overlapped region. Our results on the discussed indices and sensors show the following: (1) Identifiable angle effects exist with a more elevated influence than that introduced by band difference among sensors; (2) The absolute difference of forward and backward direction can reach up to −0.03 to 0.1 within bands of the TM/ETM+ overlapped region; (3) The investigation at ground level indicates that there are different variations of angle effect transmitted to each remote sensing index. Regarding cases of crop canopy at various growth phases, most of the discussed indices have more than a 20% relative difference in nadir value except Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI), at less than 10%, and Normalized Burn Ratio (NBR) at less than 16%. For the case of wax maturity stage, the relative difference in nadir value of Enhanced Vegetation Index (EVI), Soil-Adjusted Vegetation Index (SAVI), Ratio Vegetation Index (RVI), Char Soil Index (CSI), NBR, Normalized Difference Moisture Index (NDMI), and SWIR2/NIR exceeded 50%, among which the values for NBR and NDMI reached up to 115.8% and 206.7%, respectively; (4) Various schemes of index construction imply different developments of angle effect uncertainty. The “difference” indices can partially suppress the directional influence, while the “ratio” indices show high potential to amplify the angle effect. This study reveals that the angle-induced uncertainty of these indices is greater than that induced by the spectrum mismatch among sensors, especially under the senescence period. In addition, based on this work, indices with a suppressed potential of angle effect are recommended for vegetation monitoring or information retrieval to avoid unexpected effects.