How USDA Predicts the Navel Crop Before a Single Orange Is Picked: Inside the Objective Measurement Survey

Long before a single navel orange reaches a packing house, a USDA field crew has already counted it — or, more precisely, counted a mathematical stand-in for it. Every summer, the National Agricultural Statistics Service sends surveyors into hundreds of California groves to walk individual tree branches, count fruit by hand, and convert those counts into the crop forecast that growers, packers, and the futures side of the industry treat as the season’s opening number. The 2025-26 California Navel Orange Objective Measurement Survey ran from June 12 to August 21, 2025, and its published result — an 80.0 million-carton forecast released that September — rested on a method that has almost nothing to do with satellites, sensors, or aggregate acreage math, and almost everything to do with counting oranges one branch at a time.

Choosing the Groves

The survey begins with random selection, not convenience. NASS drew a sample of 804 navel orange groves across California’s citrus-growing counties for the 2025-26 season, of which 753 were ultimately used in the survey — a distinction that matters, since groves can drop out of a sample for reasons ranging from access refusal to a block having been pulled out since the sample frame was built. The usable sample split heavily toward the San Joaquin Valley, with 420 groves in Tulare County, 174 in Kern County, and 133 in Fresno County, plus a smaller number scattered across the state’s other navel-growing regions. That distribution mirrors where California’s navel acreage actually sits today, weighted toward the valley rather than the historic Southern California citrus belt.

Within each selected grove, surveyors do not attempt to examine every tree. Two trees are randomly chosen per grove — a sample within a sample, designed so that the fruit-count data reflects the range of vigor and yield found across an orchard block rather than whichever tree happens to be nearest the row’s edge.

Counting a Tree Branch by Branch

The counting method itself is the survey’s most distinctive feature, and it is built around a simple problem: an adult navel tree can carry thousands of oranges, far too many to count individually across an entire grove without the work taking months. NASS’s answer is a randomized branching walk that samples a single, defensible path through the tree rather than surveying the whole canopy.

Starting at the trunk, a surveyor uses a random number table to select one of the tree’s main scaffold branches, then follows that choice out to the next fork, using the same random selection method to choose again among the branches that split off from it. That process repeats — branch, fork, random choice, branch again — moving progressively further from the trunk and toward the tree’s outer canopy, until the surveyor reaches what the protocol calls a terminal branch: a branch small enough that it has no further significant limbs splitting off from it.

At that terminal branch, the actual counting happens. The surveyor counts every piece of fruit attached to that final branch, and every piece of fruit along the entire path traced back from the trunk to reach it — the full chain of branches selected at each random fork along the way.

Turning One Branch Into a Whole Tree

A single traced path obviously does not capture a tree’s total fruit count on its own. What it captures is a statistically known slice of the tree, because each branch selected at each fork carried a specific, calculable probability of being the one chosen. NASS’s forecasting model uses that probability — how likely this exact path was to have been selected out of every possible path through the tree’s branching structure — to convert the fruit actually counted along it into an estimated total fruit count for the entire tree.

That estimate is only the first of several inputs. Surveyors also measure fruit diameter on a subsample of the counted oranges, since size directly affects how many pieces of fruit fill a standard 40-pound carton. The 2025-26 survey recorded an average fruit set of 375 oranges per tree, down 9 percent from the prior year, against an average diameter of 2.179 inches, up 6 percent — a combination that illustrates why a forecast needs both numbers rather than either alone. Fewer oranges on the tree does not necessarily mean a smaller crop if the ones that are there are measurably larger.

From Tree Counts to a Statewide Number

The per-tree estimate is the foundation, but it is not the forecast. NASS combines the fruit-set-per-tree figure with the measured fruit diameter, an estimate of trees per acre drawn from grove-spacing data, and a statewide bearing-acreage figure — 109,000 acres for the 2025-26 navel crop — to build a projected yield in cartons per acre, and from there a full statewide production forecast. For 2025-26, that chain of calculations produced a yield estimate of 734 cartons per acre and a total forecast of 80.0 million 40-pound cartons, up roughly 2 million cartons from the previous season’s initial estimate.

The mandarin side of California’s citrus industry has only recently been folded into this same rigor. NASS’s objective-measurement forecast for the Tango and W. Murcott Afourer varieties is a comparatively new addition to the survey program, reflecting how quickly easy-peel mandarins have grown from a specialty planting into a crop large enough to warrant the same field-counting discipline long applied to navels.

A Method Built for a Crop That Cannot Be Guessed At

The branch-by-branch counting protocol looks laborious because it is, and that is precisely the point. Citrus is not a crop where a drone flyover or a satellite index can substitute for physically counting fruit, because navel oranges hang inside a dense, uneven canopy where a large share of the crop is not visible from above or from the row’s edge. The randomized branch selection exists to make sure that invisibility does not bias the count — a surveyor cannot simply tally the easy-to-see fruit on the tree’s exterior, because the random path is just as likely to lead to buried interior wood as to an exposed outer limb.

For anyone trying to make sense of the season’s first crop number when it lands every September, the useful habit is remembering what actually produced it: not a model extrapolated from acreage and history, but several hundred trees, chosen at random, counted branch by branch by a field crew working through the California heat months before the fruit itself is ready to pick.