Insufficient Data: The Need for Standardized Monitoring Networks for Reliable Climate AI

1. Introduction

The global implementation of Artificial Intelligence (AI) and Machine Learning (ML) in environmental sciences faces a critical bottleneck: the lack of representative and standardized in situ data. Drawing from experience in managing the monitoring network in the Salado River Basin (Argentina), this article argues that, without multilateral action to expand physical infrastructure—specifically for the modeling and validation of satellite missions— automated real-time decisions will lack the physical foundation necessary to mitigate climate disasters and economic losses. These losses are projected to far exceed the $40 billion annually forecasted by the WMO for the coming years [1].

2. The AI Paradox: Advanced Algorithms vs. Low Data Curators

In the last decade, the application of ML to the analysis of processes such as drought has grown exponentially, increasing from a dozen annual publications in the early 2010s to more than one hundred in 2023 [2]. However, this quantitative growth does not reflect the maturity of the field. From my perspective as a researcher in Earth Sciences, I observe a concerning reality: countless hours are invested in applying complex models to non-representative basins using data of uncertain validity. In a future where intelligent systems will make real-time decisions without direct human intervention; the consistency and certification of measurements are the only guarantees against systemic errors.

3. Geographic Asymmetry and the Mid-Latitude Gap

The consistency of AI depends on the representativeness of its training data. Currently, essential networks such as FLUXNET show a deeply unequal distribution: North America and Europe account for 72% of the stations, while South America represents only 2% and Africa 3% [3]. This lack of data coverage in mid-latitudes is critical, as these regions are where the most intense atmospheric changes are projected and where pressure from human development is highest. Without reliable local records of air temperature and soil moisture, AI is “learning” from foreign geographical realities, which invalidates its application in local water resource management.

4. “Measuring to Decide”: The Salado River Basin Case Study

Faced with this global urgency, there are examples of concrete action. In the southern slope of the Salado River basin (Buenos Aires, Argentina), an open-access monitoring network has been established. This project is funded by national and provincial public funds (Scientific Research Commission of the Province of Buenos Aires) with support from the Inter-American Development Bank (IDB). This infrastructure is not merely a regional resource: it houses unique equipment in Southern South America designed to monitor drought and flood events across a basin of 40,000 square kilometers. Furthermore, one of the network’s stations features equipment provided by the University of Leicester for the validation of Thermal Infrared (TIR) data and products from the Sentinel-3 mission (European Space Agency) under the “Advanced Surface Temperature Radiometer and Global Network Development” project. Without these certified terrestrial reference points, satellite observations and water flow data lack a physical anchor, preventing AI from accurately predicting phenomena such as “flash droughts” caused by heatwaves, which are already altering global agricultural productivity [2].

5. The Cost of Inaction and the Multilateral Path

The urgency is both operational and economic. The yield of staple crops like maize decreases by 6% to 7.5% for every degree of warming. While organizations like the WMO estimate annual losses of $40 billion in livestock by 2100, the lack of robust AI built on standardized networks will multiply this impact [1]. Wasting time is not an option. It is imperative that governments, alongside multilateral organizations, coordinate actions to install stations in representative areas that complement current networks and enable evidence-based water management. The solution is not simply more stations, but complete ones with proper maintenance, trained professionals, and rigorous record-keeping. That is the key.

Standardizing and certifying measurements are essential so that AI results are reliable and transferable across regions. We must guarantee the continuity of records, understanding that the changing atmosphere accompanies human pressure and requires constant monitoring.

AI will only be as smart as the data that feeds it. “Measuring to decide” must be the global watchword. Projects like the Salado River network demonstrate that it is possible to generate cutting-edge science and open data to strengthen climate resilience [4]. The technology exists, but it is physical monitoring infrastructure that remains the foundation we must urgently consolidate.