Conservation decisions, backed by evidence.

I analyzed growing-season vegetation condition across test_fields — located near Grove City in Meeker County, Minnesota — using Sentinel-2 surface reflectance, restricted strictly to the May 1 – August 31 window for each year. I composited 132 scenes with median aggregation and computed two indices: NDVI (Normalized Difference Vegetation Index, a measure of green biomass and chlorophyll activity from the contrast between red and near-infrared reflectance) and EVI (Enhanced Vegetation Index, an optimized variant that reduces soil-background and atmospheric interference and separates dense canopies better).

Baseline vs. recent comparison

Year-level detail:

The headline result is a moderate decline in period-averaged canopy density: NDVI down 0.05 and EVI down 0.13 between the two windows. But the more important finding is that this difference is not a clean restoration signal in either direction — the within-period spread is larger than the between-period gap. The baseline itself spans a strong 2018 (0.71) and a weak 2019 (0.56), and the recent period spans a weak 2024 (0.51) and a strong 2025 (0.67).

The full 2018–2025 trajectory

Because the two-period comparison is sensitive to which years land in each bin, the eight-year record is far more informative. It shows cyclical oscillation, not progressive change: high-vigor seasons in 2020 (NDVI 0.82), 2021 (0.77) and 2023 (0.80) alternate with low seasons in 2019 (0.56), 2022 (0.49) and 2024 (0.51). The fitted long-term trends are slightly negative — NDVI −0.0097 per year and EVI −0.0225 per year — but that slope is small relative to the ±0.3 swings between adjacent years.

Viewing each year as a departure from the 2018–2019 baseline makes the alternating pattern clearer, and shows that the recent period contains both one of the weaker years and a near-baseline recovery year.

Distribution and condition classes

Across-ROI statistics for the full composite are NDVI mean 0.67, median 0.69, 25th percentile 0.58, 75th percentile 0.78, with a range of −0.06 to 0.94. Year-to-year, the interquartile spread stays reasonably tight around each year's median, meaning the low years are field-wide dips rather than a few degraded patches dragging the average down.

On the condition classification for the full composite period, 90.19% of pixels fall in the healthy NDVI band and 9.14% in moderate condition, with stressed or bare-soil classes together accounting for under 1% of the area. That is a strong overall condition profile and is difficult to reconcile with a degradation interpretation.

Spatially, the NDVI and EVI composite rasters show where the higher- and lower-vigor zones sit within the parcel.

The EVI layer adds a complementary view, since EVI responds more strongly to canopy structure and is less saturated at high biomass.

An important caveat on framing

You asked about a restoration condition trajectory, and I want to be careful here rather than answer the question as posed. What I measured is growing-season greenness, and the pattern in this parcel does not look like a restoration signal at all — it looks like working cropland. The alternating high/low sequence is the classic signature of crop rotation (for example corn versus soybean, which have very different canopy closure and peak-greenness timing), combined with planting-date and seasonal-rainfall variation. A genuine restoration trajectory would typically show directional, cumulative change; this shows oscillation around a high and stable mean with under 1% of the area in stressed or bare classes.

So two conclusions, stated separately:

1. On the literal comparison: the recent period averages lower than the baseline (NDVI −0.05, EVI −0.13), and the eight-year fitted slope is mildly negative.
2. On what that means: this is very likely not evidence of declining condition or failed restoration. Both the baseline and recent windows happen to pair one strong year with one weak year, and the 2025 rebound to NDVI 0.67 after the 2024 low of 0.51 shows the field retains full productive capacity. Two-year endpoints are simply too short a window to separate a management cycle from a trend here.

If restoration assessment is genuinely the goal, I'd flag that greenness alone is a weak proxy — it can't distinguish a restored native plant community from a vigorous corn crop, since both score as “healthy.” Confirming a restoration trajectory would need the planting and management history, and ideally land-cover or species-composition evidence rather than index magnitude.

To validate the interpretation above, it's worth correlating these curves against actual planting schedules, nitrogen application records, and local rainfall for the low years (2019, 2022, 2024). Targeted soil testing and in-season scouting would be useful for any sub-zones that stayed suppressed across multiple weak years — those are the areas where a real, persistent condition problem would show up, as opposed to rotation-driven variation.

Both the NDVI and EVI composites are available as map layers and as downloadable rasters, and pixel-level datasets from the run are archived for deeper spatial review.