diff --git a/r_app/kpi_utils.R b/r_app/kpi_utils.R
index 4519df7..6bfebbc 100644
--- a/r_app/kpi_utils.R
+++ b/r_app/kpi_utils.R
@@ -802,9 +802,11 @@ calculate_gap_filling_kpi <- function(ci_raster, field_boundaries) {
     valid_values <- ci_values[!is.na(ci_values) & is.finite(ci_values)]
 
     if (length(valid_values) > 1) {
-      # Placeholder gap score using lowest 25% as indicator
-      q25_threshold <- quantile(valid_values, 0.25)
-      low_ci_pixels <- sum(valid_values < q25_threshold)
+      # Gap score using 2σ below median to detect outliers
+      median_ci <- median(valid_values)
+      sd_ci <- sd(valid_values)
+      outlier_threshold <- median_ci - (2 * sd_ci)
+      low_ci_pixels <- sum(valid_values < outlier_threshold)
       total_pixels <- length(valid_values)
       gap_score <- (low_ci_pixels / total_pixels) * 100
 
@@ -821,7 +823,7 @@ calculate_gap_filling_kpi <- function(ci_raster, field_boundaries) {
         gap_level = gap_level,
         gap_score = gap_score,
         mean_ci = mean(valid_values),
-        q25_ci = q25_threshold
+        outlier_threshold = outlier_threshold
       ))
     } else {
       # Not enough valid data, fill with NA row
@@ -831,7 +833,7 @@ calculate_gap_filling_kpi <- function(ci_raster, field_boundaries) {
         gap_level = NA_character_,
         gap_score = NA_real_,
         mean_ci = NA_real_,
-        q25_ci = NA_real_
+        outlier_threshold = NA_real_
       ))
     }
   }