Search

Soil spectroscopy is a widely used method for estimating soil properties that are important to environmental and agricultural monitoring. However, a bottleneck to its more widespread adoption is the need for establishing large reference datasets for training machine learning (ML) models, which are called soil spectral libraries (SSLs). Similarly, the prediction capacity of new samples is also subject to the number and diversity of soil types and conditions represented in the SSLs. To help bridge this gap and enable hundreds of stakeholders to collect more affordable soil data by leveraging a centralized open resource, the Soil Spectroscopy for Global Good initiative has created the Open Soil Spectral Library (OSSL). In this paper, we describe the procedures for collecting and harmonizing several SSLs that are incorporated into the OSSL, followed by exploratory analysis and predictive modeling. The results of 10-fold cross-validation with refitting show that, in general, mid-infrared (MIR)-based models are significantly more accurate than visible and near-infrared (VisNIR) or near-infrared (NIR) models. From independent model evaluation, we found that Cubist comes out as the best-performing ML algorithm for the calibration and delivery of reliable outputs (prediction uncertainty and representation flag). Although many soil properties are well predicted, total sulfur, extractable sodium, and electrical conductivity performed poorly in all spectral regions, with some other extractable nutrients and physical soil properties also performing poorly in one or two spectral regions (VisNIR or NIR). Hence, the use of predictive models based solely on spectral variations has limitations. This study also presents and discusses several other open resources that were developed from the OSSL, aspects of opening data, current limitations, and future development. With this genuinely open science project, we hope that OSSL becomes a driver of the soil spectroscopy community to accelerate the pace of scientific discovery and innovation. [ABSTRACT FROM AUTHOR]

There is a consensus about the necessity to achieve a quick soil spatial information with few human resources. Remote/proximal sensing and pedotransference are methods that can be integrated into this approach. On the other hand, there is still a lack of strategies indicating on how to put this in practice, especially in the tropics. Thus, the objective of this work was to suggest a strategy for the spatial prediction of soil classes by using soil spectroscopy from ground laboratory spectra to space images platform, as associated with terrain attributes and spectral libraries. The study area is located in São Paulo State, Brazil, which was covered by a regular grid (one per ha), with 473 boreholes collected at top and undersurface. All soil samples were analyzed in laboratory (granulometry and chemical), and scanned with a VIS-NIR-SWIR (400–2500 nm) spectroradiometer. We developed two traditional pedological maps with different detail levels for comparison: TFS-1 regarding orders and subgroups; and TFS-2 with additional information such as color, iron and fertility. Afterwards, we performed a digital soil map, generated by models, which used the following information: (i) predicted soil attributes from undersurface layer (diagnostic horizon), obtained by using a local spectral library; (ii) spectral reflectance of a bare soil surface obtained by Landsat image; and (iii) derivative of terrain attributes. Thus, the digital map was generated by a combination of three variables: remote sensing (Landsat data), proximal sensing (laboratory spectroscopy) and relief. Landsat image with bare soil was used as a first observation of surface. This strategy assisted on the location of topossequences to achieve soil variation in the area. Afterwards, spectral undersurface information from these locations was used to modelize soil attributes quantification (156 samples). The model was used to quantify samples in the entire area by spectra (other 317 samples). Since the surface was bare soil, it was sampled by image spectroscopy. Indeed, topsoil spectral laboratory information presented great similarity with satellite spectra. We observed angle variation on spectra from clayey to sandy soils as differentiated by intensity. Soil lines between bands 3/4 and 5/7 were helpful on the link between laboratory and satellite data. The spectral models of soil attributes (i.e., clay, sand, and iron) presented a high predictive performance (R2 0.71 to 0.90) with low error. The spatial prediction of these attributes also presented a high performance (validations with R2 > 0.78). The models increased spatial resolution of soil weathering information using a known spectral library. Elevation (altitude) improved mapping due to correlation with soil attributes (i.e., clay, iron and chemistry). We observed a close relationship between soil weathering index map and laboratory spectra + image spectra + relief parameters. The color composite of the 5R, 4G and 3B had great performance on the detection of soils along topossequences, since colors went from dark blue to light purple, and were related with soil texture and mineralogy of the region. The comparison between the traditional and digital soil maps showed a global accuracy of 69% for the TFS-1 map and 62% in the TFS-2, with kappa indices of 0.52 and 0.45, respectively. We randomly validated both digital and traditional maps with individual plots at field. We achieve a 75% and 80% agreement for digital and traditional maps, respectively, which allows us to conclude that traditional map is not necessarily the truth and digital is very close. The key of the strategy was to use bare soil image as a first step on the indication of soil variation in the area, indicating in-situ location for sample collection in all depths. The current strategy is innovative since we linked sensors from ground to space in addition with relief parameters and spectral libraries. The strategy indicates a more accurate map with less soil samples and lower cost. [ABSTRACT FROM AUTHOR]