Why variety selection deserves more attention than it gets
In the sequence of decisions a rice farmer makes before planting, variety selection often gets less rigorous attention than fertilizer sourcing or timing, despite being the decision that sets the biological ceiling for what the season can produce. You can apply fertilizer precisely and time your harvest well, but if the variety you chose has a maturity period that conflicts with the tail end of your rainfall window, or a lodging susceptibility that turns your plot's specific wind exposure into a crop loss, the downstream decisions are working against a structural problem that was created at the start of the season.
For most smallholder farmers in West Java, variety selection works like this: the dealer has a few options in stock, one of them has been recommended by a neighbor who had a good season, and that is the one the farmer buys. This is rational behavior given the information available, and familiar varieties do have real advantages. Farmers know how they grow, what inputs they respond to, and how to manage them. The cost of variety change is not zero.
But the information available at the counter is almost never calibrated to the specific plot. The dealer does not know whether this farmer's soil is waterlogging-prone, whether their subdistrict has had elevated tungro virus pressure in the last two seasons, or whether their plot's altitude and temperature profile better matches a mid-elevation variety than the lowland-optimized options on the shelf.
The matching variables that matter most in West Java
West Java's rice-growing geography spans several distinct agroclimatic zones. The Priangan highlands, including Cianjur, Garut, and Tasikmalaya regencies, sit between 500 and 1,200 meters above sea level. The northern coastal lowlands around Karawang and Subang are at or near sea level with different rainfall patterns and soil types. A variety that excels in one zone may be poorly suited to the other.
The three matching variables we weight most heavily in Elevarm's variety recommendation are altitude and temperature profile, water availability and drainage characteristics, and recent pest and disease pressure for the subdistrict.
Altitude matters because most IRRI-derived and BB Padi-released varieties have an optimal growing temperature range. Inpari 32, which dominates in the lowland irrigated zones around Karawang, is not well-suited to Garut's highland plots where nighttime temperatures regularly drop below 18 degrees Celsius during the main rainy season. In those plots, Inpago varieties or specific highland-adapted inbred lines perform more consistently.
Water availability splits the recommendation in a different direction. In rainfed plots with unreliable irrigation, drought-tolerant varieties like Inpara 3 or Ciherang's drought-tolerant derivatives are appropriate even if they carry a yield ceiling that is lower than an irrigated plot's best option. Recommending a high-yield variety to a farmer with an unreliable water source sets up an inevitable disappointment.
Disease pressure data from the subdistrict level
The most practically useful matching variable, and the one that takes most farmers by surprise, is current disease and pest pressure data. Blast fungus, bacterial blight, and tungro virus all have subdistrict-level incidence variation that is meaningful for variety selection.
If a subdistrict in Cianjur has had elevated blast incidence in the last two growing seasons, recommending a blast-susceptible variety like Memberamo into that environment is agronomically indefensible even if Memberamo has consistently delivered higher yields in adjacent regencies. BB Padi publishes variety resistance ratings, and we use those ratings against our subdistrict-level disease incidence signals to filter recommendations.
The challenge is that disease incidence data at the subdistrict level is not systematically collected and published anywhere in real time. We are building this layer from three sources: farmer-reported disease sightings logged in the platform, extension reports that are available through province-level agricultural agencies with a lag, and satellite-based NDVI anomaly detection that can flag potential disease stress across plot clusters. The disease pressure layer is the least complete part of the recommendation engine, but even a rough incidence flag is more information than the dealer's inventory list.
What the recommendation actually looks like for a farmer
When a Petani Pro farmer opens the planting recommendation for a new season, they see a ranked list of three varieties suited to their plot, with a short note on why each is ranked where it is. The top recommendation is the variety we estimate is best matched to their specific combination of soil class, altitude, water availability, and current disease pressure. The second and third options reflect different trade-offs: one might offer a higher potential yield at slightly more input cost, another might offer more disease resilience at a slightly lower yield ceiling.
The farmer does not have to follow the recommendation. Many farmers choose the second or third option, or stay with a variety they know well. Our job is to give them a better starting point for that conversation, not to replace their judgment with an algorithm's output.
The input availability problem
One constraint we have not fully solved is that a variety recommendation is only useful if the recommended variety is actually available from a local dealer in the week before planting. Certified seed for some BB Padi varieties has production volumes that do not match demand evenly across regencies, and the last-mile distribution of seed from provincial seed depots to village-level kios is inconsistent.
The Mitra dealer integration helps here. When an Elevarm recommendation goes out to farmers in a regency, the Mitra dealers connected to those farmers can see the aggregate demand signal before planting week arrives. A dealer who knows that twenty connected farmers are likely to want Inpari 42 in three weeks can plan their order accordingly rather than being caught short. This is one of the ways that connecting the input supply side to the advisory output side makes both sides more useful.
Seed availability is a longer-term structural problem in Indonesian smallholder agriculture, and we are not going to solve it by ourselves. What we can do is reduce the information gap between what the recommendation engine suggests and what the supply chain can actually deliver, and let farmers and dealers coordinate on that gap before planting day.