diff --git a/libraries/AP_Declination/AP_Declination.cpp b/libraries/AP_Declination/AP_Declination.cpp
index 37dd1ed51c..bd50eaf0fa 100644
--- a/libraries/AP_Declination/AP_Declination.cpp
+++ b/libraries/AP_Declination/AP_Declination.cpp
@@ -37,8 +37,8 @@ bool AP_Declination::get_mag_field_ef(float latitude_deg, float longitude_deg, f
     bool valid_input_data = true;
 
     /* round down to nearest sampling resolution */
-    int32_t min_lat = static_cast<int32_t>(static_cast<int32_t>(latitude_deg / SAMPLING_RES) * SAMPLING_RES);
-    int32_t min_lon = static_cast<int32_t>(static_cast<int32_t>(longitude_deg / SAMPLING_RES) * SAMPLING_RES);
+    int32_t min_lat = static_cast<int32_t>(static_cast<int32_t>(floorf(latitude_deg / SAMPLING_RES)) * SAMPLING_RES);
+    int32_t min_lon = static_cast<int32_t>(static_cast<int32_t>(floorf(longitude_deg / SAMPLING_RES)) * SAMPLING_RES);
 
     /* for the rare case of hitting the bounds exactly
      * the rounding logic wouldn't fit, so enforce it.