Central Asia’s digital agriculture ambitions face gaps in skills, data and field testing

New research reveals that digital agriculture is advancing rapidly in crop production, but much of the evidence behind its promised gains remains too narrow for direct transfer to dryland regions such as Central Asia. The review warns that sensors, drones, artificial intelligence (AI) and nano-enabled inputs are most useful when they are tied to real field decisions, not when they remain isolated technical demonstrations.

The study, titled Digital Technologies in Crop Production: A Scoping Review with Transferability Analysis for Central Asia, was published in AgriEngineering. It maps 224 empirical studies, including 205 studies from a structured Scopus search covering 2020 to 2026 and 19 additional Central Asia-focused studies, to assess how digital technologies perform in crop production and how far their results can be applied to Kazakhstan and neighbouring countries.

Digital agriculture works best when data leads to action

The review finds that four technology areas dominate current crop production research: Internet of Things (IoT) systems, drones and remote sensing, machine learning and AI, and nanostructured agrochemicals. Across those areas, the strongest results appear when data collection is connected to a complete decision chain.

IoT systems are closest to operational use when sensors are linked directly to irrigation, fertigation or environmental control. The strongest cases involve long-term field monitoring, sustained maintenance and regular calibration. The review highlights that low-cost sensors can reduce entry costs, but they also create a continuing need for data correction, recalibration and supervision. Without that support, readings can drift and decision quality can weaken.

Connectivity is another major factor. LoRaWAN is described as a practical option for remote farm settings because it supports long-range, low-power communication without relying fully on mobile operators. Cellular systems such as NB-IoT can be useful where coverage is stable, but they expose farmers to operator dependence and recurring tariff costs. The paper stresses that no single communication model is universally superior. Each system must match field geography, farm size, infrastructure and service availability.

Drones and remote-sensing systems show strong value in detecting crop stress, salinity, nutrient problems, weeds and disease. However, the authors draw a sharp line between mapping and management. A drone image has limited value unless it is converted into a timely farm action, such as a spray prescription, irrigation adjustment or variable-rate fertilizer plan. Studies that move from detection to operational prescription maps provide stronger evidence than those that stop at classification accuracy.

Machine learning and AI occupy the prediction layer of digital farming. These tools can forecast disease, estimate yields, support nitrogen recommendations and assist machinery control. But the review finds that many models perform well only within the same field, season or data structure in which they were trained. When tested across new farms, years, crop stages or climates, accuracy often declines. This makes external validation essential before AI tools can be treated as reliable decision systems.

Nanostructured agrochemicals are the least mature of the four technology groups. The review finds promising laboratory and greenhouse results for controlled nutrient or pesticide delivery, but field-level evidence remains limited. Many studies show release behaviour, residue reduction or short-term pest control, yet fewer demonstrate consistent crop benefits across seasons and real farm conditions. For Central Asia, the authors found no crop-production field study proving regional benefits from nano-enabled fertilizers or pesticides.

Central Asia can adopt monitoring tools first, but advanced AI needs local proof

The study focuses on Kazakhstan and Central Asia, where digital agriculture is viewed as a route to sustainable intensification but faces difficult field realities. The region has large cropland areas, water stress, soil salinity, variable infrastructure and limited support networks for advanced farm technologies, making simple transfer from international studies risky.

According to the review, foundational monitoring tools are the most immediately transferable technologies. Real-time water-level sensing, low-cost telemetry, salinity mapping and remote monitoring are well aligned with regional needs. In water-limited and salt-affected areas, even basic monitoring can help farmers and policymakers understand field conditions more clearly.

Remote sensing also has clear near-term relevance because salinity is a major constraint in southern Kazakhstan and other arid systems. UAV and satellite-based salinity studies show that digital sensing can support diagnosis. However, the authors caution that many salinity models still need local calibration, ground samples and field-specific adjustment. That means they should not be treated as ready-made regional solutions.

Disease sensing and AI-assisted crop-status monitoring are described as conditionally transferable. Kazakhstan-based hyperspectral wheat disease research shows that local crop disease detection is possible. But the wider regional evidence remains too narrow to support large-scale claims. The review finds that Central Asia lacks enough multi-season, multi-farm validation to prove that AI systems can consistently improve yield, profitability or resource efficiency under routine conditions.

Integrated digital platforms, digital twins and intervention-grade AI systems require even more caution. These systems depend on clean data, reliable connectivity, trained operators, local calibration, machinery compatibility and advisory support. Without those layers, advanced models may fail after pilot projects end. The authors argue that Central Asia should follow a sequential path: strengthen monitoring and calibration first, build service and advisory capacity next, and only then expand toward complex AI-driven decision systems.

The paper also identifies a major human-capital barrier. Digital agriculture requires more than the ability to use dashboards or mobile apps. Farmers, technicians and agronomists must be able to judge data quality, calibrate sensors, revise thresholds, interpret model uncertainty and decide when a system should not be trusted. This is especially important in regions where technical service networks are thin.

Data governance is another unresolved area. Farm data can support better recommendations only when it can move across devices, seasons, platforms and advisory systems. However, farmers may hesitate to share data if ownership, access rights and commercial use are unclear. The review finds that weak interoperability and incomplete trust frameworks can reduce the value of even strong technical systems.

Policy must shift from pilots to multi-season proof

Governments and investors should stop treating technical demonstrations as proof of agricultural transformation. A sensor that works in one pilot, a drone model that maps one field, or an AI system that performs well on internal data does not automatically provide scalable value.

The authors call for multi-season, multi-site field trials with external validation. IoT studies should report calibration needs, maintenance costs, sensor replacement and fallback behaviour. Drone studies should focus more on whether imagery leads to better field actions. AI studies should test uncertainty, domain shift and recalibration burden. Nano-agrochemical studies should pair crop benefits with safety, non-target effects and cost evidence.

The review also points to total cost of ownership as a recurring blind spot. Many studies report accuracy, water savings, yield effects or chemical reductions, but fewer account for software subscriptions, sensor maintenance, data cleaning, drone servicing, battery replacement, model retraining, technician labour or advisory support. These hidden costs often decide whether a technology survives beyond a funded pilot.