CHIRPS Reality Checks - Revision history

Harrison: /* May 2026 */

2026-06-16T01:06:52Z

May 2026

Harrison: /* May 2026 */

2026-06-16T01:02:38Z

May 2026

Daniella: /* May 2026 */

2026-06-12T23:00:07Z

May 2026

Daniella: /* May 2026 */

2026-06-12T22:59:32Z

May 2026

Daniella: /* May 2026 */

2026-06-12T22:51:51Z

May 2026

Daniella: /* May 2026 */

2026-06-12T22:51:14Z

May 2026

Daniella: /* R-Checks Monthly Highlights */

2026-06-12T18:45:56Z

R-Checks Monthly Highlights

Daniella: /* R-Checks Monthly Highlights */

2026-06-12T17:38:42Z

R-Checks Monthly Highlights

Harrison: /* December 2025 */

2026-06-02T22:11:04Z

December 2025

Harrison: /* April 2026 */

2026-06-02T22:07:37Z

April 2026

← Older revision		Revision as of 01:06, 16 June 2026
Line 37:		Line 37:

	'''South Africa''' A series of storms impacted South Africa in early May, with persistent heavy rainfall impacting western and eastern Cape provinces, destroying infrastructure and upending lives. South African authorities declared a [https://apnews.com/article/south-africa-floods-storms-death-disaster-ee9de87f657d64586da774b3a1f0bba8 national disaster] following the severe flooding. The heavy rainfall was well captured by CHIRP3 and CHIRPS3.		'''South Africa''' A series of storms impacted South Africa in early May, with persistent heavy rainfall impacting western and eastern Cape provinces, destroying infrastructure and upending lives. South African authorities declared a [https://apnews.com/article/south-africa-floods-storms-death-disaster-ee9de87f657d64586da774b3a1f0bba8 national disaster] following the severe flooding. The heavy rainfall was well captured by CHIRP3 and CHIRPS3.
		+
		+	'''Rchecks plots''' May 2026 was characterized by exceptionally high localized rainfall totals, with maximum (wettest pixel) CHIRPS rainfall accumulations among the highest observed in the historical records for [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/rchecks/monthly_compares/chirpsV3.Southern_Africa.stats.2026.05.png Southern Africa]. Plots are available [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/rchecks/monthly_compares/?C=M&O=D here].

	=== April 2026 ===		=== April 2026 ===

@@ Line 30: / Line 30: @@
 '''CHIRPS v2: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/Difference_global_202605.png map] shows the difference between Prelim data and Final data, for May 2026 CHIRPS v2. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/PercentDifference_global_202605.png map] shows the ratio of values.
-'''Data notes''' We've been seeing an issue in which CHIRP3's very low values + blending procedure leads to wet stations not making their influence in CHIRPS3. We've noted this as occurring in winter-early spring and mainly in Eurasia. Here in May 2026 data we see that this has occurred to some extent in that region, but to a much lesser degree. One place we see it is in May 2026 is in central Ukraine. In central Ukraine, IMERG Late and CHIRP/S 2 show avg to abv avg rain. In CHIRP/S 3 there are a few locations with average to above average rain, though these don't align with station locations, and in between these are low values. It's not very noticable in data fields, but one can see it in anomalies. Good news: The CHIRP3 values in May 2026 here and across Eurasia are large enough (far enough from near zero) that the station values ARE having influence on CHIRPS3, ARE raising the CHIRPS3 estimates, and enabling CHIRPS3 to show a more accurate depiction of rains. Check out April 2026 vs. May 2026 data values in Rchecks, CHIRP3, IMERG. Check out the CHIRP3 anomalies vs. the CHIRPS3 Rchecks anomalies to see what a difference it makes, when CHIRP3 values are at least moderate vs. nero zero/zero, in getting stations to have an influence. There are still locations where CHIRP3 is 0 mm and so is CHIRPS3, depite stations with ~ 30mm, e.g.  in parts of northern Mongolia and northwest China, but the situation is far less extreme than in some of the previous cases. Which underscores the seasonal characteristic of the issue. The algorithm for CHIRP3, including the CHIRP dependence on cold cloud duration and the omission of an intercept term that CHIRP2 had (which enabled CHIRP2 to lean into climatology despite CCD being extremely low), means that CHIRP3 values are more often 0mm instead of drizzly in these types of cases. The current station and CHIRP blending procedure doesn't overcome such low CHIRP values.
+'''Data notes:''' We've been seeing a geographically focused issue in which CHIRP3's very low values + blending procedure leads to wet stations not making their influence in CHIRPS3, primarily in winter-early spring and mainly in Eurasia. In May 2026 CHIRPS final data we see that this occurred to some extent in that region, but to a much lesser degree. One place we see it is in May 2026 is in central Ukraine. In central Ukraine, IMERG Late and CHIRP/S 2 show average to above average rain. In CHIRP/S v3 there are a few locations with average to above average rain, though these don't align with station locations, and in between these are low values. The helpful situation is that CHIRP3 values in May 2026 here, and across Eurasia are large enough (far enough from near zero) that the station values are having a substantial influence on CHIRPS3 and are enabling CHIRPS3 to show a more accurate depiction of rains. There are still locations where CHIRP3 is 0 mm and so is CHIRPS3, despite stations with ~ 30mm, e.g.  in parts of northern Mongolia and northwest China, but the situation is far less extreme than in some of the previous cases. Based on previous cases, the issue is seasonally focused. The algorithm for CHIRP3 is in part at fault. CHIRP depends on cold cloud duration estimates, and in the CHIRPv2 to CHIRPv3 algorithm change, the intercept term was omitted in CHIRPv3. This term enabled CHIRP2 to lean into climatology, despite CCD being extremely low. In CHIRPv3, values are more often 0 mm instead of drizzly in such cases. The current station and CHIRPv3 blending procedure doesn't overcome such low CHIRP values.
 '''China''' May 2026 rainfall was well above average in Hunan, mountainous province in southwestern China, where more than 30,000 people were evacuated and several people were [https://www.nytimes.com/2026/05/20/world/asia/china-flood-rain-deaths.html reported] missing. The heavy rainfall was captured by CHIRPS and CHIRP.

@@ Line 30: / Line 30: @@
 '''CHIRPS v2: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/Difference_global_202605.png map] shows the difference between Prelim data and Final data, for May 2026 CHIRPS v2. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/PercentDifference_global_202605.png map] shows the ratio of values.
-'''Data notes'''We've been seeing an issue in which CHIRP3's very low values + blending procedure leads to wet stations not making their influence in CHIRPS3. We've noted this as occurring in winter-early spring and mainly in Eurasia. Here in May 2026 data we see that this has occurred to some extent in that region, but to a much lesser degree. One place we see it is in May 2026 is in central Ukraine. In central Ukraine, IMERG Late and CHIRP/S 2 show avg to abv avg rain. In CHIRP/S 3 there are a few locations with average to above average rain, though these don't align with station locations, and in between these are low values. It's not very noticable in data fields, but one can see it in anomalies. Good news: The CHIRP3 values in May 2026 here and across Eurasia are large enough (far enough from near zero) that the station values ARE having influence on CHIRPS3, ARE raising the CHIRPS3 estimates, and enabling CHIRPS3 to show a more accurate depiction of rains. Check out April 2026 vs. May 2026 data values in Rchecks, CHIRP3, IMERG. Check out the CHIRP3 anomalies vs. the CHIRPS3 Rchecks anomalies to see what a difference it makes, when CHIRP3 values are at least moderate vs. nero zero/zero, in getting stations to have an influence. There are still locations where CHIRP3 is 0 mm and so is CHIRPS3, depite stations with ~ 30mm, e.g.  in parts of northern Mongolia and northwest China, but the situation is far less extreme than in some of the previous cases. Which underscores the seasonal characteristic of the issue. The algorithm for CHIRP3, including the CHIRP dependence on cold cloud duration and the omission of an intercept term that CHIRP2 had (which enabled CHIRP2 to lean into climatology despite CCD being extremely low), means that CHIRP3 values are more often 0mm instead of drizzly in these types of cases. The current station and CHIRP blending procedure doesn't overcome such low CHIRP values.
+'''Data notes''' We've been seeing an issue in which CHIRP3's very low values + blending procedure leads to wet stations not making their influence in CHIRPS3. We've noted this as occurring in winter-early spring and mainly in Eurasia. Here in May 2026 data we see that this has occurred to some extent in that region, but to a much lesser degree. One place we see it is in May 2026 is in central Ukraine. In central Ukraine, IMERG Late and CHIRP/S 2 show avg to abv avg rain. In CHIRP/S 3 there are a few locations with average to above average rain, though these don't align with station locations, and in between these are low values. It's not very noticable in data fields, but one can see it in anomalies. Good news: The CHIRP3 values in May 2026 here and across Eurasia are large enough (far enough from near zero) that the station values ARE having influence on CHIRPS3, ARE raising the CHIRPS3 estimates, and enabling CHIRPS3 to show a more accurate depiction of rains. Check out April 2026 vs. May 2026 data values in Rchecks, CHIRP3, IMERG. Check out the CHIRP3 anomalies vs. the CHIRPS3 Rchecks anomalies to see what a difference it makes, when CHIRP3 values are at least moderate vs. nero zero/zero, in getting stations to have an influence. There are still locations where CHIRP3 is 0 mm and so is CHIRPS3, depite stations with ~ 30mm, e.g.  in parts of northern Mongolia and northwest China, but the situation is far less extreme than in some of the previous cases. Which underscores the seasonal characteristic of the issue. The algorithm for CHIRP3, including the CHIRP dependence on cold cloud duration and the omission of an intercept term that CHIRP2 had (which enabled CHIRP2 to lean into climatology despite CCD being extremely low), means that CHIRP3 values are more often 0mm instead of drizzly in these types of cases. The current station and CHIRP blending procedure doesn't overcome such low CHIRP values.
 '''China''' May 2026 rainfall was well above average in Hunan, mountainous province in southwestern China, where more than 30,000 people were evacuated and several people were [https://www.nytimes.com/2026/05/20/world/asia/china-flood-rain-deaths.html reported] missing. The heavy rainfall was captured by CHIRPS and CHIRP.

@@ Line 29: / Line 29: @@
 '''CHIRPS v2: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/Difference_global_202605.png map] shows the difference between Prelim data and Final data, for May 2026 CHIRPS v2. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/PercentDifference_global_202605.png map] shows the ratio of values.
 '''China''' May 2026 rainfall was well above average in Hunan, mountainous province in southwestern China, where more than 30,000 people were evacuated and several people were [https://www.nytimes.com/2026/05/20/world/asia/china-flood-rain-deaths.html reported] missing. The heavy rainfall was captured by CHIRPS and CHIRP.
 '''Northern Turkey, Georgia, and southwestern Russia''' CHIRPS3 is showing above-average May 2026 rainfall in these zones, due to the wetting influence of blended stations. Heavy rains led to river overflow and flooding in northern Samsum province in Turkey in early May, according to [https://www.facebook.com/Channel4News/videos/heavy-rainfall-in-turkeys-northern-samsun-province-caused-a-river-to-overflow-fl/1639933670599518/ reports]. In Russia, the [https://oc-media.org/heavy-rains-once-again-flood-north-caucasus-and-abkhazia/ heavy rainfall] caused flooding, road damage, and power outages across the North Caucasus and Abkhazia. The CHIRP3 satellite-derived rainfall estimates showed some areas of above average rainfall around these areas, but stations reported 2x higher values and their inclusion in CHIRPS had substantial positive influence. However there are some locations where CHIRPS estimates lower rainfall values than stations due to low CHIRP3 values e.g. southern and northwestern Turkey.
 === April 2026 ===

@@ Line 32: / Line 32: @@
 '''China''' May 2026 rainfall was well above average in Hunan, mountainous province in southwestern China, where more than 30,000 people were evacuated and several people were [https://www.nytimes.com/2026/05/20/world/asia/china-flood-rain-deaths.html reported] missing. The heavy rainfall was captured by CHIRPS and CHIRP.
-'''Northern Turkey, Georgia, and southwestern Russia'' CHIRPS3 is showing above average May 2026 rainfall in these zones, due to the wetting influence of blended stations. Heavy rains led to river overflow and flooding in northern Samsum province in Turkey in early May, according to [https://www.facebook.com/Channel4News/videos/heavy-rainfall-in-turkeys-northern-samsun-province-caused-a-river-to-overflow-fl/1639933670599518/ reports]. In Russia, the [https://oc-media.org/heavy-rains-once-again-flood-north-caucasus-and-abkhazia/ heavy rainfall] caused flooding, road damage, and power outages across the North Caucasus and Abkhazia. The CHIRP3 satellite-derived rainfall estimates showed some areas of above average rainfall around these areas, but stations reported 2x higher values and their inclusion in CHIRPS had substantial positive influence. However there are some locations where CHIRPS estimates lower rainfall values than stations due to low CHIRP3 values e.g. southern and northwestern Turkey.
+'''Northern Turkey, Georgia, and southwestern Russia''' CHIRPS3 is showing above-average May 2026 rainfall in these zones, due to the wetting influence of blended stations. Heavy rains led to river overflow and flooding in northern Samsum province in Turkey in early May, according to [https://www.facebook.com/Channel4News/videos/heavy-rainfall-in-turkeys-northern-samsun-province-caused-a-river-to-overflow-fl/1639933670599518/ reports]. In Russia, the [https://oc-media.org/heavy-rains-once-again-flood-north-caucasus-and-abkhazia/ heavy rainfall] caused flooding, road damage, and power outages across the North Caucasus and Abkhazia. The CHIRP3 satellite-derived rainfall estimates showed some areas of above average rainfall around these areas, but stations reported 2x higher values and their inclusion in CHIRPS had substantial positive influence. However there are some locations where CHIRPS estimates lower rainfall values than stations due to low CHIRP3 values e.g. southern and northwestern Turkey.
 === April 2026 ===

← Older revision		Revision as of 18:45, 12 June 2026
Line 27:		Line 27:

	'''CHIRPS v3: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/archive/Difference_global_202605.png map] shows the difference between Prelim data and Final data, for May 2026 CHIRPS v3. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/archive/PercentDifference_global_202605.png map] shows the ratio of values.		'''CHIRPS v3: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/archive/Difference_global_202605.png map] shows the difference between Prelim data and Final data, for May 2026 CHIRPS v3. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/archive/PercentDifference_global_202605.png map] shows the ratio of values.
		+
		+	'''CHIRPS v2: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/Difference_global_202605.png map] shows the difference between Prelim data and Final data, for May 2026 CHIRPS v2. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/v2/archive/PercentDifference_global_202605.png map] shows the ratio of values.

	'''China''' May rainfall was well above average in Hunan, mountainous province in southwestern China, where more than 30,000 people were evacuated and several people were [https://www.nytimes.com/2026/05/20/world/asia/china-flood-rain-deaths.html reported] missing. The heavy rainfall was captured by CHIRPS and CHIRP.		'''China''' May rainfall was well above average in Hunan, mountainous province in southwestern China, where more than 30,000 people were evacuated and several people were [https://www.nytimes.com/2026/05/20/world/asia/china-flood-rain-deaths.html reported] missing. The heavy rainfall was captured by CHIRPS and CHIRP.

@@ Line 24: / Line 24: @@
 ''
 ==R-Checks Monthly Highlights==
 === April 2026 ===

@@ Line 94: / Line 94: @@
 '''California''' Six stations in northern California that were initially excluded by the automated screening step were included in CHIRPS Final. These were in Mendocino area and had relatively high values of 200-500 mm. These appeared realistic given the atmospheric river conditions that impacted this region during Christmas week, and these were similar to reported amounts at other stations. CHIRP likely underestimated rainfall at these locations.
-'''Data notes''' In a swath across central Eurasia there were numerous stations indicating above-average precipitation, but the CHIRPS predominantly estimated below-average amounts due to the influence of CHIRP, which estimated very low (< 10 mm) amounts for many locations. In this case, the blending of stations did lead to higher CHIRPS Final values than preliminary CHIRPS estimates across many locations- a good thing, but still too much adverse influence from CHIRP could be seen in the Rchecks. This type of situation has been observed before, periodically, in the same general region and around winter-early spring season. The issue and potential solutions for a future version of CHIRPS was again discussed. At a given location, the CHIRPS estimate is in part due to satellite (CHIRP) and nearby stations, and in cases of extreme difference and when CHIRP is very low, the final estimate will closely follow the low value of CHIRP. However, there are modifications that could be made. Such as to the degree of influence from each respective source, by changing a capping factor or by incorporating a separate layer of gridded station data. Or the CHIRP estimation procedure could be modified so that the geostationary satellite brightness temperature-based 'cold cloud duration' technique is replaced by one that better captures these type of events.
+'''Data notes: Eurasia''' In a swath across central Eurasia there were numerous stations indicating above-average precipitation, but the CHIRPS predominantly estimated below-average amounts due to the influence of CHIRP, which estimated very low (< 10 mm) amounts for many locations. In this case, the blending of stations did lead to higher CHIRPS Final values than preliminary CHIRPS estimates across many locations- a good thing, but still too much adverse influence from CHIRP could be seen in the Rchecks. This type of situation has been observed before, periodically, in the same general region and around winter-early spring season. The issue and potential solutions for a future version of CHIRPS was again discussed. At a given location, the CHIRPS estimate is in part due to satellite (CHIRP) and nearby stations, and in cases of extreme difference and when CHIRP is very low, the final estimate will closely follow the low value of CHIRP. However, there are modifications that could be made. Such as to the degree of influence from each respective source, by changing a capping factor or by incorporating a separate layer of gridded station data. Or the CHIRP estimation procedure could be modified so that the geostationary satellite brightness temperature-based 'cold cloud duration' technique is replaced by one that better captures these type of events.
 '''CHIRPS v3: Final versus Prelim''' This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/archive/Difference_global_202512.png map] shows the difference between Prelim data and Final data, for December 2025 CHIRPS v3. This [https://data.chc.ucsb.edu/products/CHIRPS/v3.0/diagnostics/chirps_minus_prelim/archive/PercentDifference_global_202512.png map] shows the ratio of values.

@@ Line 32: / Line 32: @@
 ''' Data notes: New SSOD station archive''' This month CHIRPS began blending station reports from the Synoptic Summary of the Day (SSOD), a new source that is intended to replace the Global Summary of the Day (GSOD). The GSOD archive ended in August 2025. Like other global sources, and some regional sources, we put SSOD through an automated quality control step that screens for false zeros, then we blend the QCed data (fSSOD) into CHIRPS v3 Final. Currently, for April 2026 CHIRPS v3 data, the SSOD reports are in eleven countries, and this number is expected to increase.
-'''Data notes: Eurasia''' CHIRP v3 and CHIRPS v3 again underestimated precipitation in the region around eastern Europe to central Asia, including in western Russia, Ukraine, northern Turkiye, Kazakhastan, based on stations. The very low CHIRP values and blending algorithm resulted in the stations, though included, having little ability to raise CHIRPS v3 values. This is an issue that has been seen before during this time of year in CHIRP/S v3. e.g. December 2025.
+'''Data notes: Eurasia''' CHIRP v3 and CHIRPS v3 again underestimated precipitation in the region around eastern Europe to central Asia, including in western Russia, Ukraine, northern Turkiye, Kazakhstan, based on stations. The very low CHIRP values and blending algorithm resulted in the stations, though included, having little ability to raise CHIRPS v3 values. This is an issue that has been seen before during this time of year in CHIRP/S v3. e.g. December 2025.
 '''Cambodia''' This month, with the inclusion of the new SSOD station dataset in the CHIRPS v3 blending process, Cambodia gained four new stations in CHIRPS v3 Final. This is helpful as this is the oblu source with stations located in Cambodia. In the SSOD archive there is metadata for 23 stations, though only these 4 stations had available reports to blend into CHIRPS.