Smarter and Faster Soil Health Checks with a Cloud-Based Spectroscopy System

Reliable and accurate information about soil health is fundamental for promoting
sustainable land use, improving agricultural productivity, and supporting effective
ecosystem restoration. However, conventional soil analysis methods are
inherently slow, costly, and highly dependent on chemical reagents, which
generate hazardous waste and require specialized laboratory infrastructure.
These limitations hinder large-scale soil monitoring and timely decision-making,
especially in regions with limited technical capacity or infrastructure. In the face
of growing global challenges, including land degradation, climate change, and
food insecurity, there is an urgent demand for innovative, rapid, and
environmentally sustainable solutions that democratize access to soil
information. This study introduces a fully automated, cloud-based platform
designed to predict the Soil Health Index (SHI) and five key soil functions using
visible and near-infrared (VNIR) spectroscopy integrated with deep learning. By
eliminating the need for wet chemistry, the platform offers a fast, low-cost, and
reagent-free alternative for soil diagnostics, requiring minimal technical expertise
from users. The system was developed from a robust dataset of 46,000 soil
samples, each analyzed through standardized laboratory procedures and
scanned with VNIR spectroscopy. These paired datasets were used to train a
Long Short-Term Memory (LSTM) neural network, capable of capturing complex,
nonlinear relationships between spectral reflectance and soil properties. The SHI
was formulated as an integrative indicator, combining physical, chemical, and

biological attributes into a single, comprehensive measure of soil condition.
Model performance was robust across all outputs. For the five individual soil
functions, RMSE values were close to 0.01, with R2

ranging from 0.42 to 0.78.
Specifically for the SHI, the model achieved an RMSE of 0.07, an R2 of 0.67, and
an RPIQ (Residual Predictive Interquartile Range) of 2.82, indicating strong
predictive accuracy and reliability. The platform operates entirely online. Users
simply upload VNIR spectral data from unknown soil samples and instantly
receive predictions for the SHI and the five functional indicators. This eliminates
dependence on laboratory infrastructure, reduces analysis costs, and provides
real-time, evidence-based support for decision-making. Its architecture supports
continuous learning, allowing the model to evolve and improve as new data are
incorporated. By significantly improving the speed, scalability, and sustainability
of soil diagnostics, this technological innovation directly supports global
sustainability targets. It contributes to the achievement of Sustainable
Development Goals (SDGs) 2 (Zero Hunger), 12 (Responsible Consumption and
Production), and 15 (Life on Land), offering a transformative tool for advancing
regenerative agriculture, climate-resilient land management, and soil
conservation worldwide.